System for predicting at least one characteristic parameter of a fuel

Abstract

A system comprising a distribution grid (2) for a fuel, combustion engines (3), which are coupled with the distribution grid (2) and are configured to combust the fuel, and a computer system (4) comprising data connections (5) to the combustion engines (3) and a data storage device (6), wherein the computer system (4) is configured to receive engine operation parameters stemming from an operation of the combustion engines (3) at a first time and/or during a first time period via the data connections (5) and geographical data of the combustion engines (3) are stored in the data storage device (6), wherein the computer system (4) has a processor (7) which is configured to compute a prediction for at least one characteristic parameter of the fuel at a second time and/or during a second time period later than the first time and/or the first time period and with respect to a geographical location, and the computation of the prediction being based on the geographical data and the engine operation parameters of the combustion engines (3).

Claims

1. A system, comprising: a distribution grid for a fuel, combustion engines, which are coupled with the distribution grid and are configured to combust the fuel, and a computer system comprising data connections to the combustion engines and a data storage device, wherein the computer system is configured to receive engine operation parameters stemming from an operation of the combustion engines at a first time and/or during a first time period via the data connections and geographical data of the combustion engines are stored in the data storage device, wherein the computer system has a processor configured to compute a prediction for at least one characteristic parameter of the fuel at a second time and/or during a second time period later than the first time and/or the first time period and with respect to a geographical location, and the computation of the prediction being based on the geographical data and the engine operation parameters of the combustion engines.

2. The system according to claim 1, wherein the prediction being based on the engine operation parameters is based on changes of the engine operation parameters at the first time or during the first time period.

3. The system according to claim 1, wherein the at least one characteristic parameter of the fuel comprises: a first parameter indicative of the combustion energy of the fuel, wherein the first parameter comprises a gross calorific value of the fuel, or a second parameter indicative of a tendency for uncontrolled combustion of the fuel, wherein the second parameter comprises a methane number.

4. The system according to claim 1, comprising at least one further combustion engine coupled to the distribution grid and configured to combust the fuel.

5. The system according to claim 4, wherein set points for a closed and/or open loop control of the at least one further combustion engine and/or a time of operation of the further combustion engine are based on the prediction of the at least one characteristic parameter of the fuel for the second time and/or the second time period with respect to the geographical location of the at least one further combustion engine.

6. The system according to claim 1, wherein the engine operation parameters comprise at least one of: mechanical power output of the combustion engine and/or a load of the combustion engine, thermal output of the combustion engine and/or a load of the combustion engine, electric power output of a genset comprising the combustion engine and a generator driven by the combustion engine, charge pressure, charge temperature, efficiency of a genset comprising the combustion engine and a generator driven by the combustion engine, volume of the fuel, pressure of the fuel, temperature of the fuel, ignition timing point, wastegate position, throttle valve position, compressor bypass valve position, variable valve timing parameters, variable turbo charger geometry position, emission parameters of the combustion engine and/or changes over time of these quantities.

7. The system according to claim 1, wherein the processor is configured to convert the engine operation parameters to estimations of the at least one characteristic parameter of the fuel and/or to estimations of changes of the at least one characteristic parameter of the fuel.

8. The system according to claim 7, wherein the engine operation parameters and/or the estimations of the at least one characteristic parameter of the fuel and/or the estimations of the changes of the at least one characteristic parameter of the fuel on the one hand and the geographical data of the combustion engines on the other hand are provided in the form of a data grid, wherein positions of entries (Q.sub.ij) within the data grid represent the geographical data and values of the entries (Q.sub.ij) represent the engine operation parameters and/or the estimations of the at least one characteristic parameter and/or the estimations of changes of the at least one characteristic parameter, and the data grid is provided for and/or as an input layer (IL) of a neural network.

9. The system according to claim 8, wherein the engine operation parameters and/or the estimations of the at least one characteristic parameter and/or the estimations of changes of the at least one characteristic parameter and/or data from chemical analysis of the fuel are provided in two or more channels of the data grid.

10. The system according to claim 9, wherein the processor is configured to calculate the prediction using a machine learning model having at least one neural network.

11. The system according to claim 10, wherein the processor is configured to receive and/or represent the data grid as the input layer (IL) of the at least one neural network, and the at least one neural network comprises a convolutional neural network.

12. The system according to claim 11, wherein the convolutional neural network contains at least one first layer (CL1, CL2) comprising convolution operations with filters of smaller size than the data grid, where a number of channels of the filters equals a number of the channels of the data grid.

13. The system according to claim 12, wherein the at least one neural network contains at least one second layer (DL1, DL2) comprising deconvolution operations and weights for the at least one second layer, which weights are learned during training of the at least one neural network such that an output of the at least one neural network is the prediction of the at least one characteristic parameter of the fuel at the second time or during the second time period and with respect to the geographical location.

14. The system according to claim 13, wherein the at least one neural network contains at least one third layer realising an LSTM network with weights, which are learned during training of the at least one neural network such that an output of the at least one neural network is the prediction of the at least one characteristic parameter of the fuel at the second time or during the second time period and with respect to the geographical location.

15. A computer system for predicting at least one characteristic parameter of a fuel supplied by a distribution grid, comprising: a signal input device configured to receive engine operation parameters stemming from an operation of combustion engines at a first time and/or during a first time period, a data storage device configured to store geographical data of the combustion engines, and a processor configured to compute a prediction for the at least one characteristic parameter of the fuel at a second time and/or during a second time period later than the first time and/or the first time period and with respect to a geographical location, the prediction being based on the operation parameters and the geographical data of the combustion engines.

16. A computer program product for predicting at least one characteristic parameter of a fuel supplied by a distribution grid comprising instructions which cause a computer executing the computer program product to: receive operation parameters of combustion engines stemming from an operation of the combustion engines with a fuel from a distribution grid at a first time and/or during a first time period, access a data storage device to obtain geographical data of the combustion engines, and compute a prediction for the at least one characteristic parameter of the fuel at a second time and/or during a second time period later than the first time and/or the first time period and with respect to a geographical location, the prediction being based on the operation parameters and the geographical data of the combustion engines.

17. The computer program product according to claim 16, wherein the computer program product is configured to use at least one neural network, wherein the computer program product is configured to use training data to obtain a learned parameter set and compute the prediction for at least one characteristic parameter of the fuel.

18. The computer program product according to claim 17, wherein the at least one neural network comprises a convolutional neural network and/or a recurrent neural network.

19. The computer system according to claim 15, wherein the processor is configured to computer the prediction using a machine learning model having at least one neural network, wherein the at least one neural network comprises a convolutional neural network and/or a recurrent neural network.

20. The system according to claim 1, wherein the processor is configured to calculate the prediction using a machine learning model having at least one neural network, wherein the at least one neural network comprises a convolutional neural network and/or a recurrent neural network.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Further advantages and embodiments are apparent from the accompanying figures and accompanying description thereof. The figures show:

(2) FIG. 1a illustrates a system according to the invention,

(3) FIG. 1b illustrates a schematic diagram of a combustion engine,

(4) FIGS. 2a and 2b illustrate an example geographical disposition of combustion engines without and with a grid overlay, respectively,

(5) FIG. 2c illustrates an example of a data grid representing the geographical data and the engine operation parameters of the combustion engines,

(6) FIG. 3 illustrates a schematic diagram of a neural network for calculating the prediction according to the invention,

(7) FIG. 4 illustrates a schematic diagram of a further embodiment of the invention making use of LSTM cells,

(8) FIGS. 5a and 5b illustrate two schematic diagrams regarding a further embodiment of the invention, and

(9) FIG. 6 illustrates a deconvolution operation.

DETAILED DESCRIPTION

(10) FIG. 1a schematically shows a system 1 according to the invention. It comprises a distribution grid 2 for gaseous fuel, such as natural gas (which comprises methane as a main part), biogas, propane and/or molecular hydrogen.

(11) The grid provides this gaseous fuel for combustion engines 3 which are geographically located in a service area 21 of the distribution grid 2. An example of the geographical disposition of the combustion engines is given in FIGS. 2a and 2b. FIG. 1a does not reflect this geographical disposition, as its purpose is to depict the schematic functional relations within the system 1 according to the invention.

(12) The combustion engines 3 are connected to a computer system 4 via data connections 5. Through the data connections 5 the combustion engines 3 provide engine operation parameters from an operation of the combustion engines to the computer system 4.

(13) The combustion engines 3 may comprise control units 18 for open and/or closed loop control of the engines. These may be used to provide the operation parameters which are sent to the computer system 4.

(14) In order to ensure a clear view of the structure of the system 1, the reference numerals for the control units 18 and the data connections 5 have only been provided once each in the depiction of FIG. 1a.

(15) The computer system 4 comprises a signal input device 17 for receiving the operation parameters from the combustion engines 3 via the data connections 5.

(16) The computer system 4 comprises furthermore a data storage device 6 in which the geographical data about the combustion engines 3 is stored. The geographical data may be stored permanently in the data storage device 6 or it may be stored transiently, in which case the geographical data may be provided together with the engine operation parameters via the data connections 5.

(17) The computer system 4 comprises furthermore a processor 7 which is configured (e.g., through appropriate software) to calculate a prediction of at least one parameter characteristic for the gaseous fuel based on the engine operation parameters and the geographical data. An example for calculating the prediction is given in connection with FIGS. 2a to 2c and 3.

(18) The prediction can for example be used to determine a well-suited location for deploying an additional combustion engine 3 or to determine a quality of the gaseous fuel at the geographic location of a further combustion engine 8, which is also supplied by the distribution grid 2 for gaseous fuel.

(19) In the latter case, the prediction can be used to determine or correct set points for open or closed loop control (using the control unit 18) for an operation of the further combustion engine 8.

(20) The prediction can also be used to determine when an operation of the further combustion engine 8 can advantageously be started or stopped, if no full-time operation of the further combustion engine is intended.

(21) As has already been mentioned, the role of the further combustion engine 8 can be changed depending on what prediction according to the invention is desired and depending on its intended use. For example, if the further combustion engine 8 is started (maybe even because of a prediction of good fuel quality at a certain time), it may then be used as part of the combustion engines 3 the operation parameters of which are used as basis for the prediction according to the invention.

(22) Additionally, expenses for the fuel and/or CO.sub.2 emissions can be reduced because the properties of the fuel can be predicted at any outlet port of the distribution grid 2. The fuel for a specific combustion engine or other consumers can then for example be optimised at the given location for the desired application for example through admixture of other gaseous fuels (e.g., from renewable sources), like molecular hydrogen.

(23) FIG. 1b schematically depicts a combustion engine 3. It comprises combustion chambers 19, which may be in the form of piston cylinder units. The combustion engine 3 may, of course also be turbine or a heating plant or may comprise combinations thereof.

(24) The combustion engine may comprise any number of combustion chambers 19.

(25) The combustion engine 3 of FIG. 1b comprises furthermore a turbo charger 20 which includes an exhaust turbine and a compressor which is mechanically and/or electrically coupled to the turbine. Using the energy provided by the turbine the compressor provides charge air which is charged into the combustion chambers 19—together with the gaseous fuel or separate.

(26) For controlling the combustion engine 3 at least one of the following can be provided: a throttle valve 13, a compressor bypass valve 14, wastegate 12, fuel valve 25. Additionally or alternatively, further control measures can be taken with regards to ignition timing for example.

(27) Other actuators or different actuators may be present depending on the particular combustion engine 3, e.g., for determining emissions (like for example an NOx-Sensor).

(28) As mentioned, the control of the combustion engine 3 can be realised with a control unit 18. The control unit is of course connected in an appropriate way to the different elements of the combustion engine 3 which are controlled. These connections have been omitted in the schematic drawing of FIG. 1b in order to keep a clearly viewable depiction.

(29) The combustion engine 3 usually drives a load 9, which can for example be a generator 11 for generating electrical energy which can then be supplied to an electrical supply grid. The generated electrical energy can also be used to drive an electrical load directly.

(30) Of course, the mechanical power provided by the combustion engine 3 can also be used directly to drive a mechanical load.

(31) The combustion engine may for example be a gas engine, a dual fuel engine, or a gas turbine.

(32) The arrangement of a gas engine driving a generator for generating electric energy is called a genset.

(33) As already mentioned, FIGS. 2a and 2b depict a (fictional) example of the geographical disposition of the combustion engines within a service area 21 in which the distribution grid 2 supplies the gaseous fuel. Each of the crosses represents a combustion engine 3.

(34) In FIG. 2b, there is a rectangular grid overlay covering the service are 21 of the distribution grid 2. The geographical data of the combustion engines 3 may be stored in the data storage device 6 in accurate manner such as in FIG. 1a or in a collected manner where only the presence of the combustion engine 3 in the respective tiles of the grid is stored.

(35) The geographical data and the engine operation parameters can be provided in the form of a data grid 15, which is shown in FIG. 2c. The individual entries Q.sub.i j are indexed by the position in the i-th row and j-th column within the data grid 15.

(36) The values of the entries Q.sub.i j in the data grid 15 are at least one operation parameter for each of the combustion engines 3, or estimations of the at least one characteristic parameter for the gaseous fuel, or estimations of changes of the at least one characteristic parameter of the gaseous fuel.

(37) The placement of the entries Q.sub.i j within the data grid 15 encodes the geographical data of the combustion engines 3. That is, an entry Q.sub.i j in a specific position within the data grid 15 corresponds to one or more of the combustion engines situated in the corresponding tile in the grid of FIG. 2b.

(38) The entries Q.sub.i j for grid tiles where no combustion engine 3 or no engine operation parameter is available can be set to 0 or to some other string or value indicating this fact.

(39) There can be several operation parameters for each combustion engine 3. In the data grid 15 this can be realised as two or more channels of the data grid 15. These channels can for example be realised as copies of the grid depicted in FIG. 2c for each of the channels, such that the whole data grid 15 then has an additional index k for the different channels. The entries in the data grid can then be denoted by Q.sub.ij.sup.(k).

(40) Additional channels in the data grid can also be used if there are more than one combustion engines 3 in one of the tiles of the grid overlay of FIG. 2b. Alternatively or additionally, the entries Q.sub.i j can be averages or median values of the operation parameters of the single combustion engines 3 in each tile.

(41) The values for the entries Q.sub.i j can be estimations for the at least one characteristic parameter which can for example be calculated according to the following example (for a gas engine):

(42) Input Variables: Gas Pressure (P, bar) Gas Temperature (T, ° C.) Gas Volume (V, m.sup.3) Load (L, kWh) (can be measured as electrical power output in case of a genset or as mechanical power output)

(43) Output Variable: Gas Quality (Gross Calorific Value, kWh/Nm.sup.3)

(44) Calculation:

(45) $\mathrm{Efficiency}\eta =f\left(L\right)\mathrm{Normal}\mathrm{Volume}\left(\mathrm{NV}\right)=\frac{P\times 273.15}{\left(273.15+T\right)\times 1.01325}V\left({\mathrm{Nm}}^3\right)\mathrm{Gross}\mathrm{Calorific}\mathrm{Value}Q=\frac{L}{\eta \times \mathrm{NV}}\left(\mathrm{kWh}/{\mathrm{Nm}}^3\right)$

(46) Other ways for estimating the at least one characteristic parameter of the gaseous fuel can of course also be used. This concerns both different ways of estimating the Gross Calorific Value (depending e.g., on the specific combustion engine) and other characteristic parameters of the gaseous fuel, like a methane number or the like. Estimations for the methane number in and of themselves are known in the prior art.

(47) The data grid 15 can be provided as or for an input layer IL of a neural network 16 for calculating the prediction according to the invention. General information and information on examples of how such a neural network 16 can be set up can be found under the respective sections above in the general part of the description.

(48) An advantageous embodiment for an architecture of a neural net 16 is depicted in FIG. 3.

(49) It is noteworthy that the data grid 15 does not have to be provided to the neural net 16 in the exact form depicted in FIG. 2c. It is for example common to provide the data grid 15 in vectorised form to convolutional layers CL1, CL2 of the network on the level of the software employed to set up the neural network.

(50) The actual data structures provided for the processor 7 (i.e., after compiling) can, of course, differ significantly and is in many cases specially adapted to the hardware (including the processor 7) used.

(51) The different layers of the neural network 16 are successively computed from the input layer IL in the upper left to the output layer OL in the lower right. These (hidden) layers are in order: a first convolution layer CL1 (one of the at least one first layers in this embodiment), a first pooling layer PL1 (max-pooling), a second convolution layer CL2 (one the at least one first layers in this embodiment), a second pooling layer PL2 (max-pooling), a first deconvolution layer DL1 (one the at least one second layers in this embodiment), and a second deconvolution layer DL2 (one the at least one second layers in this embodiment).

(52) FIG. 4 shows another embodiment of a neural network according to the invention. LSTM cells are used as at least one third layer at each time step for “evolving” the fuel properties to the next time step. Convolutional layers CL1 and deconvolutional layers DL1 are used at each time step for “encoding” and “decoding” these properties—or their spatial distributions—much in the same way as in the embodiment according to FIG. 3. Note that the input and output layers IL and OL are in this example given as data grids 15 (symbolised as little versions of FIG. 2b). General ways of implementing LSTM cells have been described above.

(53) FIG. 5a shows a simple embodiment of the invention where only a convolutional neural network is used at different times t.sub.1 and t.sub.2. The convolutional neural network according to FIG. 5 includes at least one fully connected layer (not shown) in order to determine given features of the distribution of the gaseous fuel in the distribution grid at or near the geographical location of interest (i.e., in the vicinity 22) in a manner which is in principle known in the prior art.

(54) At each time t.sub.1 and t.sub.2 this results in time specific outputs O1 and O2 of the convolutional neural network 16 which represent these features. Of course it is also possible to use more than two times t.sub.1 and t.sub.2 (i.e., t.sub.i with i indexing the number of time steps used). The times t.sub.i represent the first time (of which there is then more than one in this embodiment) or the first time period according to the invention.

(55) The output of the convolutional neural network at times t.sub.1 and t.sub.2 according to FIG. 5a could be visualised as depicted in FIG. 5b. The actual output of the convolutional network will of course not be in the form of a picture as in FIG. 5b. FIG. 5b is only drawn in this way as a simple example in order to make it clear for the reader how this embodiment of the invention works.

(56) As can be seen there is an area (hatched) of different gas quality (at least one characteristic parameter of the gaseous fuel) compared to the rest of the service are 21 of the distribution grid 2. If the hatched area increases in size so that its new boundary 24 overlaps the vicinity 22 of geographical location of interest (visualised as a cross), this can be used as an indicator that the gas quality at the geographical location is about to change.

(57) Computationally, in the embodiment of FIGS. 5a and 5b the outputs for times t.sub.1 and t.sub.2 are compared (subtracted from each other in a suitable way) and if the difference is above a given threshold value an alarm is triggered. The prediction according to the invention is then simply that at the location of interest there will be a change in the value of the at least one characteristic parameter of the gaseous fuel in the next time step (i.e., at the second time or during the second time period according to the invention). This is visualised as computation block 23.

(58) FIG. 6 depicts a deconvolution operation (⊕) as superposition of the filter F.sub.k.sup.(j) with factors from the input matrix h.sup.(j−1). As has been mentioned in the earlier section “Deconvolution Operations” this deconvolution operation can be implemented as a matrix operation with the transposed sparse matrix. Note, that the vector for .sup.(j−1) is the vectorised form of the input matrix h.sup.(j−1) of FIG. 6.