Embodiments implement non-intrusive load monitoring using machine learning. A trained convolutional neural network (CNN) can be stored, where the CNN includes a plurality of layers, and the CNN is trained to predict disaggregated target device energy usage data from within source location energy usage data based on training data including labeled energy usage data from a plurality of source locations. Input data can be received including energy usage data at a source location over a period of time. Disaggregated target device energy usage can be predicted, using the trained CNN, based on the input data.

Embodiments implement non-intrusive load monitoring using ensemble machine learning techniques. A first trained machine learning model configured to disaggregate target device energy usage from source location energy usage and a second trained machine learning model configured to detect device energy usage from source location energy usage can be stored, where the first trained machine learning model is trained to predict an amount of energy usage for the target device and the second trained machine learning model is trained to predict when a target device has used energy. Source location energy usage over a period of time can be received, where the source location energy usage includes energy consumed by the target device. An amount of disaggregated target device energy usage over the period of time can be predicted, using the first and second trained machine learning models, based on the received source location energy usage.

Embodiments implement non-intrusive load monitoring using a novel learning scheme. A trained machine learning model configured to disaggregate device energy usage from household energy usage can be stored, where the machine learning model is trained to predict energy usage for a target device from household energy usage. Household energy usage over a period of time can be received, where the household energy usage includes energy consumed by the target device and energy consumed by a plurality of other devices. Using the trained machine learning model, energy usage for the target device over the period of time can be predicted based on the received household energy usage.

A method for controlling an operating condition of an electric power grid having an intermittent power supply coupled thereto, comprising: using an energy variability controller, controlling variability of a delivered power output of the intermittent power supply to the grid by: monitoring an actual environmental value for a location proximate the intermittent power supply, an available power output of the intermittent power supply being dependent on the actual environmental value; when the actual environmental value is increasing and hence the available power output is increasing, increasing the delivered power output according to a predetermined rate of increase; monitoring a forecast environmental value for the location; when the forecast environmental value is decreasing, decreasing the delivered power output according to a predetermined rate of decrease; and, limiting the delivered power output to below a predetermined threshold. The electric power grid may be or may include an electric power microgrid.

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for shaping compute load using virtual capacity. In one aspect, a method includes obtaining a load forecast that indicates forecasted future compute load for a cell, obtaining a power model that models a relationship between power usage and computational usage for the cell, obtaining a carbon intensity forecast that indicates a forecast of carbon intensity for a geographic area where the cell is located, determining a virtual capacity for the cell based on the load forecast, the power model, and the carbon intensity forecast, and providing the virtual capacity for the cell to the cell.

Systems include one or more critical datacenter connected to behind-the-meter flexible datacenters. The critical datacenter is powered by grid power and not necessarily collocated with the flexboxes, which are powered “behind the meter.” When a computational operation to be performed at the critical datacenter is identified and determined that it can be performed art a lower cost at a flexible datacenter, the computational operation is instead routed to the flexible datacenters for performance. The critical datacenter and flexible datacenters preferably shared a dedicated communication pathway to enable high-bandwidth, low-latency, secure data transmissions.

A power usage prediction system implements a long short term memory (LSTM) neural network to receive power usage inputs and generate predicted values of power consumption for a plurality of devices. A user provides configuration input regarding the time steps at which the predicted values are to be generated and the various devices for which the predicted values of power consumption are desired. Whenever a power usage input is received, the LSTM neural network outputs the corresponding hidden state values for a plurality of time steps as the predicted values. The hidden state values are each compared to a final cell state value corresponding to a power consumption threshold of the time interval which includes the time steps. Based on the comparison, a power usage condition is recorded. Various actions to mitigate the high power consumption can be implemented in response to recording the power usage condition.

A battery charge and discharge cycling with predictive load and availability control system has a server and a load control device. The server predicts usage and future solar capacity at the specific sites where the battery packs are in use. The batteries are cycled near the lower end of capacity (state of charge or SoC), which extends life. Further, the load control device can reduce a battery's SoC if the weather forecast is sunny. The load control device may turn on appliances in the home to use excess energy that is generated by the solar system.

A site management system has a site management device located on a fielded site, which has a controller unit integral with a power provision unit, and the power provision unit receives an input voltage via a conductor cable and delivers power to one or more receptacles. Additionally, the system has a plurality of remote devices communicatively coupled to the site management device over a wireless network and at least one off-site computing device communicatively coupled to the site management device. Further, the system has a processor on the controller unit that communicatively couples with at least one remote device, receives data indicative of a unique identifier from the wireless remote device, and determines whether the unique identifier correlates with a remote device of an individual who is permissively on the fielded site. In addition, the processor transits data indicative of the individual and data indicative of whether the individual is permissively on the fielded site to the off-site computing device or a site manager's remote device.

A controller for building equipment that operate to provide heating or cooling for a building or campus. The controller includes a processing circuit configured to perform an optimization of an objective function subject to an override constraint to determine amounts of one or more resources to be produced by the building equipment and control the building equipment to produce the amounts of the one or more resources determined by performing the optimization subject to the override constraint. The override constraint overrides an output of the optimization by specifying an override amount of a first resource of the one or more resources to be produced by a first subset of the building equipment.