Statistical Process Control Method of Demand Side Management

Abstract

The present invention is a method of “Demand Side Management” (DSM) that is intended to manage the increasingly chaotic nature of the power grid, and is designed to adapt to the future impact to the grid patterns caused by the ongoing introduction of renewable power generation sources, battery charging associated with the increasing number of electric vehicles, and future unforeseen developments, by utilizing “Statistical Process Control” (SPC) techniques. SPC is typically a method for controlling manufacturing process variation in a factory, analogously the present invention adapts SPC methods to help monitor the quality of the grid by treating said grid as if said grid were a process with varying levels of quality, to which the present invention can detect, anticipate and respond by making immediate and adaptive future scheduling decisions for the control of device loads, or generation, for the benefit of consumers as well as power companies.

Claims

1. The “demand side management” method of continuously optimizing the stability of the AC grid which comprises the steps of: a. characterizing the status of the grid utilizing “statistical process control” techniques comprising: i. the first monitoring of the grid voltage and frequency; ii. logging the data of the first monitoring; iii. analyzing said data of the first monitoring using statistical process control techniques; iv. utilizing said statistical process control techniques, determine the status of said first monitoring; v. utilizing said status determinations of said first monitoring to assist in current load operation decisions such that said load operation is advantageous to the stability of the grid; vi. saving said status determinations of said first monitoring; b. characterize the status determinations of said first monitoring comprising: i. utilizing said status determinations, forecast the future status of said grid; ii. utilizing said future status forecast to assist in scheduling future load operation decisions; iii. comparing said future status forecasts of the grid status of the first monitoring, with the second monitoring of the actual grid status, at the forecasted time, and adjusting said forecast algorithms to optimize the accuracy of future projected characterizations.

2. The methods of claim 1 further comprising: a. utilizing said status determinations of said first monitoring to assist in current generation operation decisions, such that said generation operation is advantageous to the stability of the grid; b. utilizing said future status forecast to assist in scheduling future generation decisions.

3. The methods of claim 1 further comprising: analyzing said status determinations looking for deviation of size of individual bins; using said deviations look for a 7-day sequence; using the 7-day sequence, determine week days and weekend days; utilizing said determinations to assist in current load operation decisions; utilizing said determinations to assist in current generation operation decisions; saving said status determinations in 7-day sequences in a plurality of bins.

4. The methods in claim 1 further comprising; looking for a pattern indicative of any periodic number of days including 7 days.

5. The methods of claim 1 further comprising; a. utilizing error checking and correction for line dropouts and grid interruptions comprising; i. utilizing the running average to fill in erroneous data; ii. utilizing a running clock to fill in erroneous data.

6. The methods of claim 1 further comprising; a. the first monitoring of the grid, the device operation and the power consumption over time; b. logging the data of the first monitoring; c. analyzing said data of the first monitoring using statistical process control techniques; d. utilizing said statistical process control techniques, determine the status of said first monitoring; e. utilizing said status determinations of said first monitoring to assist in current load operation decisions; f. saving said status determinations of said first monitoring.

7. The methods of claim 1 further comprising; a. characterizing the status of the grid utilizing “statistical process control” techniques comprising multiple levels of response, based on multiple levels of deviation.

8. The methods of claim 1 further comprising; a. characterizing the status of the grid utilizing “statistical process control” techniques using a rule set of responses.

9. The methods of claim 1 further comprising; characterizing the status of the grid utilizing “statistical process control” techniques to follow the grid condition as it changes over time.

10. The methods of claim 1 further comprising; The option of keeping the exact location of the device secure, if preferred by the user.

11. The methods of claim 1 further comprising; The option of keeping the data of the user secure, regarding the operation of the device, if preferred by the user.

12. The methods of claim 1 further comprising; a method of real time immediate emergency shut down capability in the event of serious stress on the grid, such as a brown out, or sag, to protect the grid from a blackout.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0016] FIG. 1: Is a flow chart of the daily tracking into bins to be used for predictive scheduling of future loading.

[0017] FIG. 2: Is a flow chart of medium-term tracking to use for real time load decisions as they happen.

[0018] FIG. 3: Is a flow chart of short-term tracking looking of high grid stress events requiring quick response.

[0019] FIG. 4: Shows a 15-day AC frequency average of 45-minute bins over 24 hours. This in a Summertime period starting at midnight. AC line frequency on the vertical scale, and the 45-minute bins shown on the horizontal scale.

[0020] FIG. 5: Shows similar data as in FIG. 4 but for wintertime.

[0021] FIG. 6: Shows plus and minus one Standard deviation as calculated for each bin for the same 15-day period as in FIG. 4. Dashed line is average and two irregular solid lines are plus and minus one standard deviation.

[0022] FIG. 7: Shows average from FIG. 4 and standard deviation from FIG. 6.

[0023] FIG. 8: Converts data from FIG. 7 to control chart format. Sigma's are shown on the vertical scale (standard deviations from each bin) over a 15-day period on the horizontal scale.

[0024] FIG. 9: Each datum point shows a bin over a 2-day period in summer in a control chart format. Dashed line is the long-term average. Dotted lines are the sigma levels and long-term standard deviations.

[0025] FIG. 10: Is an actual event with a large sudden drop in frequency at approximately 11:30.

[0026] FIG. 11: Is same data as FIG. 10 in a control chart format. Solid line is overall average while dashed and dotted lines are standard deviations.

[0027] FIG. 12: Is bin data for UK over 12 days in summertime. Average is dashed line.

[0028] FIG. 13: is a device usage pattern for a typical home appliance showing a 30-day average of 45-minute bins over 24 hours.

[0029] FIG. 14: Shows standard deviation data for the same bins as in FIG. 13.

[0030] FIG. 15: Converts data from FIG. 14 to control chart format. Vertical scale standard deviations from each bin.

[0031] FIG. 16: data form FIG. 13 one bin over 30 days.

[0032] FIG. 17: Bin from FIG. 16 averaged in 7-day cycle, Monday to Sunday.

[0033] FIG. 18: combines AC grid control chart as in FIG. 8 and individual device usage patterns as in FIG. 15 (inverted) into a new control chart.

[0034] FIG. 19: The “Duck Curve” graph.

DETAILED DESCRIPTION

[0035] While this invention is susceptible to embodiment in many different forms, there are shown in the drawings and will be described herein in detail, specific embodiments Thereof, with the understanding that the present disclosure is to be considered as an exemplification of the principles of the invention and is not intended to limit the invention to the specific embodiments illustrated.

[0036] The present invention is the DSM method of sensing the status of the grid by monitoring and tracking the voltage and/or frequency of said grid, storing data over time, to utilize SPC techniques to make adaptive future scheduling decisions for loads or generation. The present invention will also use real time monitoring to adjust operational decisions and will also monitor the grid for critical stress events and enable drastic load reduction when deemed necessary. Said invention can be applied to any device with an existing, or added micro controller, or equivalent. Said invention is not dependent on external communication to make operational decisions, instead regards the appliance and the grid as a machine process, using frequency and, or voltage, over time, as the variables to make operational decisions and predictions utilizing SPC methods.

[0037] A plurality of data bins and a data matrix are used to track and decipher patterns as shown in FIG. 1. Said bins can be of any length, but in this exemplification, 32 bins per 24 hours, or 45 minutes for each bin, are used. The line frequency (and/or voltage) average is calculated for each bin and logged. Many methods can be used to determine the average, but in this exemplification the method is to count zero crosses over the entire bin and log the count. Said count will be an inverse of the frequency. Said frequency value and running averages are logged for a given number of days. Said averaging is necessary to account for occasional aberrant values that are not representative of the actual grid operation. Said Averaging can be done by keeping a matrix of the daily counts over many days, for instance, 16 days, in general, longer is better, however too long would not allow algorithm to move between the changing seasonal patterns. The transition of summer to fall and winter and spring grid profiles are different. Presently, the peak load for summer is in the afternoon, while winter has morning and evening peak load periods, but grid conditions are changing rapidly. Presently the grid in the western United States is seeing the effects of an increasing number of solar power farms coming on line, which only generate electricity during daylight hours and then quickly drop off at sunset, requiring traditional power generation to come on line quickly and take over, creating the so-called “duck curve” when looking at the power production demand chart. Solar power is pushing the summer afternoon peak into an evening stress peak. Stress is monitored for the combination of load and generation effects. In choosing bin size the limitation of the particular device and processing power as well as storage must be considered. Since bins are circular in a 24-hour day no real time clock is needed.

[0038] The accuracy of the micro controller time base is a consideration. Even the simplest RC timer bases are now over 98% accurate. Enough accuracy for simplest devices, such as a refrigerator. Accuracy needs to be over many days, for instance 15 days. Longer term time drift is taken up by the overall binning averages which move with time and do not need to be synchronous with the actual time of day. Critical applications can use more accurate time base if required.

[0039] Since the grid does experience occasional short line dropouts, which could cause errors in bin averages, error detection and correction is necessary. Short line dropouts, or voltage sags, would trigger an algorithm to fill in the blank time. Most dropouts are seconds, so the devices power supply will need enough hold up time to keep the controller running until the data is saved, or the power restored. Longer dropout times would trigger an algorithm to disregard that bin, in time frame effected. Another option is an algorithm to fill in the lost bin with the current average, thus simplifying the overall analysis with little loss of information. More critical devices can include a power supply with a longer hold up time. With some level of error detection, nearly all power disruptions can be handled.

[0040] Each Bin of 45 minutes is averaged and the standard deviation calculated. The total time covered by all bins in the matrix are averaged. Alternate methods can be used in smaller devices with less processing power, such as calculating the running average minimum and maximum values and estimating the standard deviation. Said methods are scalable to meet the needs of each device. Any given bin has an average near, above, or below the total average result which would look something like FIG. 4, which is a 15-day period captured during the summer, compared to FIG. 5, which is a wintertime sample. Patterns are generally not simple morning, or afternoon peaks, but instead an irregular pattern reflecting the effects of load and generation switching. Load is somewhat sinusoidal with a summer afternoon peak and morning and evening wintertime peaks. Generation generally ramps up ahead of the anticipated load increase and ramps down after as larger generators shut down. Since Generation is scheduled ahead of time, it is not in full sync with load, to avoid shortages. This creates peaks, or valleys, in frequency and voltage during ramping periods. The introduction of renewable energy sources increases the amount of variation during the day light hours due to the inherent variability due to weather and inexact forecasts. Even with averaging over many days, it is still difficult to predict when best to schedule loads, as shown in FIGS. 4 and 5.

[0041] The present invention regards the grid as a machine and uses SPC and control chart techniques to monitor the variation over a daily pattern. The standard deviation is calculated for the captured frequency data in each bin. Using the same data as presented in FIG. 4, standard deviation is calculated for each bin as shown in FIG. 6. FIG. 6 shows +1 and −1 sigma (one standard deviation) around the long-term average. A modified control chart is used to predict grid stress resulting from both load and generation. FIG. 7, shows both bin frequency and bin standard deviation (Chart is corrected to start at midnight with bin 1, but in practice bin 1 could be any of the 32 bins). The period from early morning to near noon is the low stress period with lowest deviation, or daily variation of line frequency. Said period also has the highest average frequency period, except for bin 28 (peak late in day FIG. 4), which is also high. Bin 28, while higher in frequency, also has a high deviation. This makes bin 28 less desirable to schedule loads than if we look at frequency alone. The variability of bin 28 is possibly due to the ramp down of loads that are moving quickly with the variation and ramp down of generation, which is lagging to avoid shortages.

[0042] FIG. 8 is a control chart created from the sigma calculations of each bin. This is done using the long-term average, each bins average, and each bins standard deviation. Since the vertical axis is on the same scale as in FIG. 7, the deviation can be plotted in a control chart fashion. The FIG. 8 chart now considers both frequency average and variability of the bins and takes on a more sinusoidal look. The FIG. 8 sigma values are now easily understood numbers and can now be compared, analyzed, and then used to make decisions about when to schedule loads and when to avoid scheduling loads. In FIG. 8, early morning to mid-morning are all positive values and as high as 1.6 sigma. Said positive values tend to occur during low grid stress periods, while a string of low and negative sigma values occur from mid-morning to early evening, occur during higher grid stress periods where operating loads are to be avoided. Said values can be collected, compiled and used make predictions about future grid stress and use that information to schedule loads and offsets in devices.

[0043] Control chart rules such as “Shewhart control chart rules” (Shewhart, W. A. (1939). Statistical Method from the Viewpoint of Quality Control) can be used to make detailed decisions. For instance, look at the oscillating bins as seen in mid-day in FIG. 8 and avoid that period. Smaller bins would be helpful to make decisions and allow finer control.

[0044] The control chart rules can be setup so that periods of higher than average frequency are considered better times to operate loads than periods of lower than average frequency, and use said rules to make decisions of when to schedule high load usages within known low load demands of the grid. For example, home refrigeration defrost typically happens about once a day during the summer. The present invention has the ability to look forward and predict the best low usage time (bin) to run the next defrost cycle. In this exemplification, the control makes the determination to pick early morning to run the next defrost cycle. Other devices might run large loads over a few days, or more than once a day. The control can anticipate the best time (or bin) to run the loads by using grid data and load data. The designers of each particular device can scale these techniques to meet the requirements of the application using known operating behaviors.

[0045] Although said higher frequency periods are better for scheduling loads, this is not true for generation, which is best centered on the line frequency. This also applies to local renewable generation such as wind and solar installations which are becoming more prevalent, and can be a disruptive force if allowed to feed the grid during a period of high traditional generation, or during ramp up and ramp down periods. Conversely the grid can benefit from local renewable generation if controlled by the present invention, which is programmed to help balance some of the ramp up and down problems and can be scheduled to feed the grid during low frequency, or low voltage periods.

[0046] Described below is one possible exemplification of a two step approach to device load mitigation. Device loads can be scheduled to avoid lower frequency periods with the understanding that 100% forecast accuracy cannot be expected due to daily variability. In addition to scheduling loads to avoid said high demand periods, devices with larger loads will also need to monitor grid stress in real time to make operational and load use decisions. It may be necessary to immediately and aggressively reduce loads if, for example a brown out were to occur (see FIG. 2). Devices with larger loads and less tolerance for operational changes, need an accurate method of real time measurement. Devices with smaller loads and tolerant of operational changes, can use a simple, less accurate real time measurement method. Analysis of the preceding minutes, or seconds, can be used to make final operational decisions and judge real-time grid stress. Many grid connected devices automatically turn on and off loads and even the simplest devices can monitor and assess the grid condition before turning on. Complex devices with large loads can assess the real time grid condition to determine if it is advantageous to turn on, or off loads immediately, or reschedule.

[0047] The grid frequency and/or voltage data is monitored for a given period of time and used to calculate the average and standard deviation, See FIG. 2. Said calculations can be stored as a matrix of data, or a running average, depending on the accuracy required by the device, and used to develop an out of control specification and reaction plan. For this exemplification the specification has a rule that if the last 10 minutes are below −2 sigma, the reaction plan requires the control to wait to turn on a load, until the grid condition improves above −2 sigma, or until the delay limit in the reaction plan is exceeded. The standard, or modified “Shewhart control chart rules” can be applied if more critical decisions are required. The rules have a ranking of severity from low, to high. Said decisions could be a more complex set of decisions such as aggressive load shedding, or to avoid energizing loads. Moderate reduction for medium shift indicators. Operate normally if the process measurement is in control. It is also possible to modify said rules, so that for above average positive sigma, loads could be run while not necessary, to store energy, following severity rankings. Each device will make its own decisions based on its original design, programming and operational tolerances. For example, a refrigerator can change its temperature set point within a narrow range, without effecting its operation. The ability to make ice could be deferred with no noticeable effect on the user. Pools and Spas also have set points and clean-up operations that can be changed within a narrow band but tolerate shift for a longer period of time.

[0048] The present invention can generate commands that conform to standards such as CTA-2045. See CTA-2045 standard which specifies said ranking commands: ANSI/CTA-2045 specifies a modular communications interface (MCI) to facilitate communications with residential devices for applications such as energy management. The MCI provides a standard interface for energy management signals and messages to reach devices. The present invention will conform to the MCI standard, but does not require said MCI, or any other form of external communication.

[0049] FIG. 9 shows a 2-minute sampling of the grid frequency, with an obvious sag at the beginning. A device attempting operation in this period could delay operation waiting for better conditions with a maximum time limit set for each device. For example, a pool can delay most functions for several hours, and a refrigerator might be able to delay 20 or 30 minutes without any negative effects. Well known control chart rules can be used to make more detailed decisions depending on the complexity and flexibility of the device.

[0050] Typical control chart rule examples;

Any data point excursion beyond 3 sigma is considered out of control and immediate action is required.
8 data points below -1 sigma indicates grid stress.
4 out of 5 data points below −1 sigma are warnings of grid stress, defer loads if possible.
4 out of 5 data points above 1 sigma is an opportunity to increase loads.
The data point sampling can be done over a few minutes and would be simple to implement in smaller devices. For larger devices, or even complex smaller devices additional time frames can be added to monitor unexpected grid events such as detecting a generator dropping out anywhere on the grid. By sampling over a very short timeframe of a few seconds, larger events can be detected from any point on the grid, using the SPC control chart approach (see FIGS. 3, 10 and 11). FIG. 10 at 11:30 shows a generator dropping out hundreds of miles from the monitoring point. A simple device monitoring a short running average, for example 1 second, can detect unusually high grid stress events (below 3 sigma). More complex devices can be configured to detect the oscillations that typically precede a sudden drop in grid output (see FIG. 11). “Shewhart control chart rules” can be applied flag oscillations above and below 1 or 2 sigma. Standard control chart rules can be configured to detect precursors to other grid events and respond with action favorable to grid stability.

[0051] While these conditions are rare immediate load reduction could help protect the grid. The grid needs about 10 minutes to adjust so most devices have the capability to turn off for a short period of time without a noticeable reduction in performance.

[0052] Due to the autonomous, stand-alone nature of the present invention, there is no risk of a hacker commandeering many devices to destabilize the grid. Instead each device makes independent decisions with some variation in sensitivity. The present invention derives it instructions from the grids frequency and voltage, both of which are nearly impossible to hack. Even the frequency of a small local area would be difficult to change for even a second.

[0053] When a device, using the present invention, recovers from a grid event, there would be a random component to the resumption of normal operation. Over a period of several minutes some loads would gradually resume normal operation and other loads would resume operation only when needed, as dictated by the specific load rescheduling instructions of each device, which makes it unlikely that all loads would turn back on all at once after the resumption of power.

[0054] A combination of frequency and voltage are useful in larger complex devices where load, generation and transmission lines are a consideration. Frequency is favored for predictive scheduling and voltage is favored for real time decisions, but some decisions consider input from a combination of both. Control and warning levels are considered as generated from “Shewhart control chart rules” or other similar set of SPC decisions.

[0055] One Advantage of the present invention is that is adapts to the different grid conditions present in the three different US grids (East, West and Texas). The 3 sigma levels of the present invention are similar to the three alarm levels presently used in these three US grids even though the alarm levels vary slightly, from grid to grid, but will be generally proportional to the normal value of each grid. The present invention will operate effectively in any region without any setup or calibration.

[0056] A devices typical usage pattern is an important consideration in determining optimum load scheduling. FIG. 13 is a device usage pattern, or load side pattern, for a typical home appliance, that shows a 3-day average over 24 hours for this example, the data starts at midnight. A similar pattern, as taught by Lacey in U.S. Pat. No. 9,032,751 of the running duty cycle pattern of a refrigerator. As in FIG. 7, the deviation is taken for each bin in FIG. 14. The chart shows a larger variation in data during the morning period. In this case, could be attributed to changing patterns due higher usage and differences between weekdays and weekends. FIG. 15 converts data into a control chart plotting usage deviation from full average and bin standard deviation, in the same way as FIG. 8. A high average deviation in a stable period is considered a higher usage event. In this example it would be best to run loads during the low use period and avoid the high deviation period of which the usage is uncertain. This differs from FIG. 8 since the negative usage pattern is inverted. The least desirable times are the inverse of the frequency, therefore would avoid scheduling defrost in the higher number periods.

[0057] Consider both the AC grid control chart as in FIG. 8 and the load Control chart in FIG. 15 and combine the period standard deviation numbers to create a new control chart result as shown in FIG. 18 . Said new control chart data is used to make predictions based on grid history and stability, as well as the load usage pattern. In this case late morning and afternoon are less desirable than with AC grid alone. In this case we find current morning and afternoon peaks. As those condition change, we can follow them and use control chart to allow adaptive predictions of when to change operating parameters and schedule large loads.

[0058] This invention takes advantage of the average and deviation to be able to plot data against standard deviation and finding best and worst points for adjusting load and running high load like defrost and making Ice. This technique can be applied to any device with an expected daily, or periodic usage pattern. Converting frequency and load into standard deviation values allows us to accumulate differing values into a standard form and simplifying analysis. In the case of a refrigerator we might find the best period for grid could be worse case for load.

[0059] Other periodic usage patterns can be considered. Referring to FIGS. 16 and 17 a weekly pattern is evident, which could be used in some devices where a weekly pattern is expected and can be used to schedule or defer loads. Pools for example might have a higher weekend loading and we could schedule cleanup for just before and after weekend usage. FIG. 16 is a plot of the daily average of the bin with the highest standard deviation. In this case early morning is bin 10. A pattern can be easily seen in FIG. 17 where every 7.sup.th day from bin 10 is averaged. To make data easier to see, day 1 is set to Monday. Bins 6 and 7 (weekend) have the lowest average in the case of a refrigerator. Other 7-day patterns are possible and might not coincide with the weekend. The algorithm looks for patterns, even if not synchronous with the weekdays, just looking for lower and higher usage patterns in a weekly cycle. One could go further and look for patterns other than 7 days that might emerge from, for example, first responder schedules, possibly 3-day cycles. Depending on how advantageous this information is to each device, determines how much processing power should be dedicated.

[0060] While the present invention can be considered independent of a devices control functions, it is better to consider each devices response using the information provided in the present invention, considering each devices' operating principles and boundaries to optimize responses.

[0061] Said techniques are scalable to each application allowing low additional complexity to individual devices. Some of today's light bulbs contain enough processing power to look at the real time grid status and made decisions to lower loads individually. Other devices that run continuously, such as refrigeration, can add tracking to make predictive decisions to change load behavior. Devices with even larger loads or more tolerance for change can run all said techniques to accomplish autonomous demand response such as pools and water heaters.