A method and device for adjusting a molten salt heat storage system

Abstract

The present invention proposes a molten salt heat storage system adjustment method and device, wherein the method comprises: acquiring thermal power unit data at multiple historical moments of the thermal power unit, and training a long short-term memory (LSTM) network according to the thermal power unit data at each historical moment, so as to predict the thermal storage power generation power, heat release power generation power, heat storage duration and heat release duration at subsequent moments after the corresponding historical moment for at least one combination of values ​​of heat storage power generation power, heat release power generation power, heat storage duration and heat release duration using the trained LSTM network for the input thermal power unit data, so as to obtain corresponding predicted values, and for any combination, determining the value of the objective function according to each predicted value, and after determining the target combination according to the value of the objective function, adjusting the molten salt heat storage system according to the values ​​of the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration in the target combination.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal power generation, and in particular to a method and device for adjusting a molten salt heat storage system. Background Art

[0002] In recent years, with the gradual increase in the installed capacity of my country's new energy power generation grid-connected equipment and the proposal of the "carbon neutrality" goal, traditional coal-fired thermal power has ushered in a transformation. Thermal power units need to be frequently used for frequency regulation and peak regulation, and the high-frequency deep peak regulation of thermal power units will greatly reduce their service life, and thus cannot balance the overall benefits of new energy consumption losses. In order to resolve the above contradictions, the energy storage system, with its flexible and controllable advantages, participates in the peak and frequency regulation of thermal power units. As a sensible heat storage technology, molten salt heat storage technology can enable the solar thermal power generation system to have the ability to generate electricity continuously and stably, fully meeting the grid's requirements for power quality.

[0003] In order to meet the flexibility requirements of the new power system, a molten salt heat storage system needs to be configured to cooperate with thermal power units for deep peak regulation. Therefore, how to optimize the operation mode of the molten salt heat storage system to maximize the system benefits has become an urgent problem to be solved in the industry. Summary of the invention

[0004] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0005] To this end, the first purpose of the present invention is to propose a molten salt heat storage system adjustment method to adjust the molten salt heat storage system according to the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration in the target combination, optimize the operation mode of the molten salt heat storage system, and improve the overall benefits of the molten salt heat storage system and the thermal power unit.

[0006] The second object of the present invention is to provide a molten salt heat storage system adjustment device.

[0007] A third objective of the present invention is to provide an electronic device.

[0008] A fourth object of the present invention is to provide a non-transitory computer-readable storage medium.

[0009] A fifth object of the present invention is to provide a computer program product.

[0010] To achieve the above-mentioned purpose, a first embodiment of the present invention provides a molten salt heat storage system adjustment method, comprising:

[0011] Obtaining thermal power unit data at multiple historical moments of the thermal power unit; wherein the thermal power unit data includes heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, automatic generation control AGC frequency modulation instruction and peak load regulation instruction;

[0012] According to the thermal power unit data at each of the historical moments, a long short-term memory (LSTM) network is trained;

[0013] For at least one combination of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration, a trained LSTM network is used to predict the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration at subsequent moments after the corresponding historical moment based on the input thermal power unit data, so as to obtain corresponding predicted values;

[0014] For any of the combinations, the value of the objective function is determined according to each of the predicted values, so as to determine the target combination according to the value of the objective function; wherein the objective function is used to indicate the cycle electrical efficiency of the thermal power unit;

[0015] The molten salt heat storage system is adjusted according to the values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration in the target combination.

[0016] Optionally, as a first possible implementation of the first aspect, the training of a long short-term memory (LSTM) network according to the thermal power unit data at each of the historical moments includes:

[0017] The thermal power unit data at each of the historical moments is used as training data and input into the LSTM network to perform multiple rounds of iterative training;

[0018] Whenever a round of iterative training is performed, the LSTM network is used to predict the thermal storage power generation, the heat release power generation, the heat storage duration and the heat release duration at subsequent moments after the corresponding historical moment for the input thermal power unit data at at least one historical moment, so as to obtain corresponding predicted values;

[0019] The LSTM network is trained based on the difference between each predicted value and the actual value at a subsequent moment after the corresponding historical moment.

[0020] Optionally, as a second possible implementation of the first aspect, the thermal power unit data at each of the historical moments is used as training data and input into the LSTM network to perform multiple rounds of iterative training, including:

[0021] Normalizing the thermal power unit data at each of the historical moments using a normalization method to obtain the training data;

[0022] Generate an input matrix of the LSTM network according to the training data; wherein each column in the input matrix corresponds to a historical moment, and each element in the same column in the input matrix is ​​used to indicate the thermal power unit data collected at the historical moment corresponding to the column;

[0023] In the order of each of the historical moments, the thermal power unit data of at least one column in the input matrix is ​​input into the LSTM network to determine the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration according to the output of the LSTM network.

[0024] Optionally, as a third possible implementation manner of the first aspect, the LSTM network includes a plurality of neurons, the neurons include a forget gate, an input gate, and an output gate, and each time a round of iterative training is performed, the LSTM network is used to predict the thermal storage power generation power, the heat release power generation power, the heat storage duration, and the heat release duration at a subsequent moment after the corresponding historical moment for the input thermal power unit data at at least one historical moment, so as to obtain corresponding predicted values, including:

[0025] Whenever a round of iterative training is performed, for each neuron, data of a thermal power unit at least at the historical moment is used as input data corresponding to each neuron;

[0026] For any one of the neurons, using the forget gate to process the output data of the previous neuron and the input data to determine the discarded data corresponding to the input data;

[0027] Using the first activation function of the input gate to update the output data and the input data of the previous neuron to obtain first updated data, and using the second activation function of the input gate to perform alternative updating on the output data and the input data of the previous neuron to obtain second updated data;

[0028] Determine third update data according to the first update data, the second update data, the output data of the previous neuron and the third update data of the previous neuron;

[0029] The output gate is used to process the output data of the previous neuron, the input data and the third update data to obtain corresponding output data.

[0030] Optionally, as a fourth possible implementation manner of the first aspect, the using the output gate to process the output data of the previous neuron, the input data and the third update data to obtain corresponding output data includes:

[0031] Using the third activation function of the output gate to perform screening processing on the output data and the input data of the previous neuron to obtain screened data;

[0032] The third updated data and the screened data are processed using a fourth activation function of the output gate to obtain corresponding output data.

[0033] Optionally, as a fifth possible implementation manner of the first aspect, for any one of the combinations, determining a value of an objective function according to each of the predicted values, so as to determine a target combination according to the value of the objective function; wherein the objective function is used to indicate the cycle electrical efficiency of the thermal power unit, including:

[0034] For any one of the combinations, determining the value of the objective function according to the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration;

[0035] In response to a value of the objective function being greater than or equal to a set threshold, determining the corresponding combination as a reference combination;

[0036] In response to the number of reference combinations being multiple, determining the target combination according to the magnitude of the value of the objective function corresponding to each of the reference combinations;

[0037] In response to the reference combination being one, the reference combination is determined as the target combination.

[0038] The molten salt heat storage system adjustment method provided in the embodiment of the present invention obtains the thermal power unit data of multiple historical moments of the thermal power unit, and realizes training the long short-term memory LSTM network according to the thermal power unit data at each historical moment, so as to predict the thermal power generation power, heat release power generation power, heat storage duration and heat release duration of the subsequent moment after the corresponding historical moment for the input thermal power unit data by using the trained LSTM network, so as to obtain the corresponding predicted value, and for any combination, according to each predicted value, determine the value of the objective function, and after determining the target combination according to the value of the objective function, adjust the molten salt heat storage system according to the values ​​of the thermal power generation power, heat release power generation power, heat storage duration and heat release duration in the target combination. Thus, by adjusting the molten salt heat storage system, the operation mode of the molten salt heat storage system is optimized, and the overall benefits of the molten salt heat storage system and the thermal power unit are improved.

[0039] To achieve the above-mentioned purpose, the second embodiment of the present invention proposes a molten salt heat storage system adjustment device, comprising:

[0040] An acquisition module is used to acquire the thermal power unit data at multiple historical moments of the thermal power unit; wherein the thermal power unit data includes heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, automatic generation control AGC frequency modulation instruction and peak load regulation instruction;

[0041] A training module, used for training a long short-term memory (LSTM) network according to the thermal power unit data at each of the historical moments;

[0042] A prediction module, for predicting, for at least one combination of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration, the heat storage power generation, the heat release duration and the heat release duration at a subsequent moment after a corresponding historical moment for the input thermal power unit data, using a trained LSTM network to obtain a corresponding predicted value;

[0043] A determination module, used for determining the value of the objective function according to each of the predicted values ​​for any one of the combinations, so as to determine the target combination according to the value of the objective function; wherein the objective function is used to indicate the cycle electrical efficiency of the thermal power unit;

[0044] An adjustment module is used to adjust the molten salt heat storage system according to the values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration in the target combination.

[0045] Optionally, as a first possible implementation manner of the second aspect, the training module includes:

[0046] An input unit, used to input the thermal power unit data at each of the historical moments as training data into the LSTM network to perform multiple rounds of iterative training;

[0047] A prediction unit, configured to predict the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration at subsequent moments after the corresponding historical moment by using the LSTM network for inputting the thermal power unit data at at least one historical moment each time a round of iterative training is performed, so as to obtain corresponding prediction values;

[0048] A training unit is used to train the LSTM network according to the difference between each predicted value and the actual value at a subsequent moment after the corresponding historical moment.

[0049] Optionally, as a second possible implementation manner of the second aspect, the input unit is further used to:

[0050] Normalizing the thermal power unit data at each of the historical moments using a normalization method to obtain the training data;

[0051] Generate an input matrix of the LSTM network according to the training data; wherein each column in the input matrix corresponds to a historical moment, and each element in the same column in the input matrix is ​​used to indicate the thermal power unit data collected at the historical moment corresponding to the column;

[0052] In the order of each of the historical moments, the thermal power unit data of at least one column in the input matrix is ​​input into the LSTM network to determine the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration according to the output of the LSTM network.

[0053] Optionally, as a third possible implementation manner of the second aspect, the LSTM network includes a plurality of neurons, the neurons include a forget gate, an input gate, and an output gate, and the prediction unit is further used to:

[0054] Whenever a round of iterative training is performed, for each neuron, data of a thermal power unit at least at the historical moment is used as input data corresponding to each neuron;

[0055] For any one of the neurons, using the forget gate to process the output data of the previous neuron and the input data to determine the discarded data corresponding to the input data;

[0056] Using the first activation function of the input gate to update the output data and the input data of the previous neuron to obtain first updated data, and using the second activation function of the input gate to perform alternative updating on the output data and the input data of the previous neuron to obtain second updated data;

[0057] Determine third update data according to the first update data, the second update data, the output data of the previous neuron and the third update data of the previous neuron;

[0058] The output gate is used to process the output data of the previous neuron, the input data and the third update data to obtain corresponding output data.

[0059] Optionally, as a fourth possible implementation manner of the second aspect, the prediction unit is further configured to:

[0060] Using the third activation function of the output gate to perform screening processing on the output data and the input data of the previous neuron to obtain screened data;

[0061] The third updated data and the screened data are processed using a fourth activation function of the output gate to obtain corresponding output data.

[0062] Optionally, as a fifth possible implementation manner of the second aspect, the determining module includes:

[0063] a first determining unit, configured to determine, for any one of the combinations, a value of an objective function according to the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration, and the heat release duration;

[0064] A second determining unit, configured to determine the corresponding combination as a reference combination in response to a value of the objective function being greater than or equal to a set threshold;

[0065] A third determining unit, configured to determine the target combination according to the magnitude of the value of the objective function corresponding to each of the reference combinations in response to the number of the reference combinations being multiple;

[0066] The fourth determining unit is configured to determine the target combination by using the reference combination in response to the reference combination being one.

[0067] The molten salt heat storage system adjustment device provided by the embodiment of the present invention obtains the thermal power unit data of multiple historical moments of the thermal power unit, and realizes training the long short-term memory LSTM network according to the thermal power unit data at each historical moment, so as to predict the thermal power generation power, heat release power generation power, heat storage duration and heat release duration of the subsequent moment after the corresponding historical moment for the input thermal power unit data by using the trained LSTM network, so as to obtain the corresponding predicted value, and for any combination, according to each predicted value, determine the value of the objective function, and after determining the target combination according to the value of the objective function, adjust the molten salt heat storage system according to the values ​​of the thermal power generation power, heat release power generation power, heat storage duration and heat release duration in the target combination. Therefore, by adjusting the molten salt heat storage system, the operation mode of the molten salt heat storage system is optimized, and the overall benefits of the molten salt heat storage system and the thermal power unit are improved.

[0068] To achieve the above object, a third aspect of the present invention provides an electronic device, including:

[0069] at least one processor; and

[0070] a memory communicatively connected to the at least one processor; wherein,

[0071] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

[0072] In order to achieve the above-mentioned purpose, the fourth aspect of the present invention proposes a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the method described in the first aspect.

[0073] In order to achieve the above-mentioned purpose, a fifth aspect of the present invention provides a computer program product, including a computer program, which implements the method described in the first aspect when executed by a processor.

[0074] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0076] Figure 1A schematic flow chart of a molten salt heat storage system adjustment method provided in an embodiment of the present invention;

[0077] Figure 2 A schematic diagram of electrical wiring of a molten salt heat storage system provided by an embodiment of the present invention;

[0078] Figure 3 A schematic flow chart of another molten salt heat storage system adjustment method provided in an embodiment of the present invention;

[0079] Figure 4 A schematic flow chart of another molten salt heat storage system adjustment method provided in an embodiment of the present invention;

[0080] Figure 5 A schematic flow chart of another molten salt heat storage system adjustment method provided in an embodiment of the present invention;

[0081] Figure 6 A schematic diagram of the structure of a single neuron in an LSTM network provided by an embodiment of the present invention;

[0082] Figure 7 A schematic diagram of the structure of an LSTM network provided by an embodiment of the present invention;

[0083] Figure 8 A schematic flow chart of another molten salt heat storage system adjustment method provided in an embodiment of the present invention;

[0084] Fig. 9 A schematic diagram of the principle of a molten salt heat storage system adjustment method provided in an embodiment of the present invention;

[0085] Fig.10 A schematic diagram of the structure of a molten salt heat storage system adjustment device provided by an embodiment of the present invention; and

[0086] Fig.11 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0087] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0088] The following describes a molten salt heat storage system adjustment method and device according to an embodiment of the present invention with reference to the accompanying drawings.

[0089] Figure 1A schematic flow chart of a molten salt heat storage system adjustment method provided in an embodiment of the present invention.

[0090] As traditional coal-fired power generation undergoes transformation, in order to meet the flexibility requirements of the new power system, a molten salt heat storage system needs to be configured to cooperate with thermal power units for deep peak regulation. Therefore, it is urgent to optimize the operation mode of the molten salt heat storage system to maximize the system benefits.

[0091] To address this problem, an embodiment of the present invention provides a molten salt heat storage system adjustment method, so as to adjust the molten salt heat storage system according to the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration in the target combination, optimize the operation mode of the molten salt heat storage system, and improve the overall benefits of the molten salt heat storage system and the thermal power unit, such as Figure 1 As shown, the molten salt heat storage system adjustment method comprises the following steps:

[0092] Step 101, obtaining thermal power unit data of the thermal power unit at multiple historical moments.

[0093] It should be noted that the molten salt heat storage system adjustment method provided in the embodiment of the present invention can be performed by a molten salt heat storage system adjustment device. The molten salt heat storage system adjustment device can be an electronic device, or can be configured in an electronic device. Among them, the electronic device can be any stationary or mobile computing device capable of data processing, such as a mobile computing device such as a laptop computer, a smart phone, a wearable device, or a stationary computing device such as a desktop computer, or a server, or other types of computing devices, etc., which are not limited in the embodiment of the present invention.

[0094] In this embodiment, the thermal power unit, i.e., the thermal power generating unit, refers to a unit that uses coal, oil or combustible gas as fuel, heats the water in the boiler to increase the temperature, and then uses steam with a certain pressure to drive the turbine to generate electricity. For example, it generally refers to a small thermal power unit with a generating unit capacity of less than 50,000 kilowatts, etc. Among them, a small thermal power unit generally consists of three major equipments: boiler, steam turbine, and generator, as well as related supporting boiler water treatment, fuel supply, ash treatment, steam turbine heat recovery system, etc.

[0095] In this embodiment, the thermal power unit data may include heat storage power generation, heat release power generation, base load power generation, heat storage duration, heat release duration, AGC (Automatic Generation Control) frequency modulation instructions and peak load modulation instructions. Optionally, the thermal power unit data may be represented by the following symbols and units: The symbol for heat storage power generation is P C , the unit is MW (megawatt); the symbol for heat generation power is P D, in MW; the symbol for base load power generation is P Base , in MW; the symbol for heat storage duration is Δt D , the unit is h (hour); the symbol for the duration of heat release is Δt C , the unit is h; the symbol of AGC frequency modulation instruction is P TP , in MW; the symbol for the peak load instruction is P TF , unit is MW.

[0096] In this embodiment, the molten salt heat storage system adjustment device can obtain the thermal power unit data of multiple historical moments of the thermal power unit and execute subsequent steps. As a possible implementation method, the thermal power unit data within the set period can be determined as the thermal power unit data of a corresponding historical moment, so that the thermal power unit data of multiple historical moments in the historical period of the thermal power unit can be obtained. Among them, the specific values ​​of the set period and the historical period are not limited in this embodiment. Optionally, they can be set according to manual experience. For example, the set period can be set to one day, and the historical period can be set to the past three years, that is, the thermal power unit data of any day in the past three years can be determined as the thermal power unit data of a corresponding historical moment. Obtaining the thermal power unit data of multiple historical moments in the historical period of the thermal power unit is to obtain the thermal power unit data of each day in the past three years. Alternatively, it can be dynamically adjusted according to actual application needs. This is not limited in this embodiment. Optionally, after obtaining the thermal power unit data of the thermal power unit in the past three years, the trend of the thermal power unit data in the past three years over time can be represented by a curve, so as to obtain a curve chart of the thermal storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, AGC frequency regulation instruction and peak load regulation instruction over time. C -t(MW), heat generation power P D -t(MW), basic load power generation P Base -t(MW), heat storage duration Δt D -t(h), heat release duration Δt C -t(h), AGC frequency modulation instruction P TP -t(MW) and peak load instruction P TF -t(MW) refers to the thermal storage power generation power P C , heat release power P D , basic load power generation P Base , heat storage duration Δt D , heat release duration Δt C , AGC frequency modulation instruction P TP and peak load instruction P TFThe change curve with day as the time unit, that is, one day corresponds to a thermal storage power generation power P C The value of a heat release power P D The value of a basic load power generation power P Base The value of a heat storage duration Δt D The value of a heat release duration Δt C The value of an AGC frequency modulation instruction P TP The value of a peak load instruction P TF It is understandable that the value range of the heat storage duration and the heat release duration corresponding to one day can be 0 to 24 hours.

[0097] It should be noted that the molten salt heat storage system adjustment device in this embodiment can obtain the thermal power unit data of multiple historical moments of the thermal power unit through various public, legal and compliant methods. In a possible implementation method, the molten salt heat storage system adjustment device can obtain the thermal power unit data of multiple historical moments of the thermal power unit by online collection or offline collection. For example, the molten salt heat storage system adjustment device can collect the data of each thermal power unit of the thermal power unit in real time at a historical moment after authorization, so as to obtain the thermal power unit data of multiple historical moments of the thermal power unit, or, after authorization, it can also obtain the thermal power unit data of multiple historical moments of the thermal power unit through offline manual methods, etc., which is not limited in this embodiment. In another possible implementation method, the molten salt heat storage system adjustment device can obtain the thermal power unit data of multiple historical moments of the thermal power unit through network transmission or physical copying. For example, the molten salt heat storage system adjustment device can obtain the thermal power unit data of multiple historical moments of the thermal power unit from other devices through network transmission or physical copying after authorization.

[0098] Step 102: training a long short-term memory (LSTM) network based on the thermal power unit data at each historical moment.

[0099] In this embodiment, after obtaining the thermal power unit data of multiple historical moments of the thermal power unit, the LSTM (Long Short-Term Memory) network can be trained according to the thermal power unit data of each historical moment, so that the LSTM can accurately predict the heat storage power generation, heat release power generation, heat storage duration and heat release duration of the subsequent moments after the corresponding historical moment based on the input thermal power unit data. As a possible implementation method, the thermal power unit data of each historical moment can be used as a training sample, and the LSTM network can be trained for multiple rounds of iterations to obtain a trained LSTM network.

[0100] Step 103, for at least one combination of the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration, a trained LSTM network is used to predict the heat storage power generation, heat release power generation, heat storage duration and heat release duration at subsequent moments after the corresponding historical moment for the input thermal power unit data, so as to obtain corresponding predicted values.

[0101] In this embodiment, the heat storage power generation, heat release power generation, heat storage duration and heat release duration in the combination can take any value, so that at least one combination of the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration can be obtained, and then for at least one combination of the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration, a trained LSTM network can be used to predict the heat storage power generation, heat release power generation, heat storage duration and heat release duration at the subsequent moment after the corresponding historical moment for the input thermal power unit data to obtain the corresponding predicted value. In other words, for multiple different combinations of the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration, a trained LSTM network can be used to predict the heat storage power generation, heat release power generation, heat storage duration and heat release duration at the subsequent moment after the corresponding historical moment for the input thermal power unit data, so that the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration at the subsequent moment after the historical moment corresponding to each combination can be obtained.

[0102] Step 104, for any combination, determine the value of the objective function according to each predicted value, and determine the target combination according to the value of the objective function.

[0103] In this embodiment, after obtaining the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration at the subsequent moments after the historical moments corresponding to each combination, the value of the objective function can be determined for any combination according to the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration at the subsequent moments after the historical moments corresponding to the combination, thereby obtaining the value of the objective function corresponding to each combination, and thus determining the target combination according to the value of the objective function corresponding to each combination. The objective function is used to indicate the cycle electrical efficiency of the thermal power unit.

[0104] Step 105, adjusting the molten salt heat storage system according to the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration in the target combination.

[0105] In this embodiment, after the target combination is determined, the operation mode of the molten salt heat storage system can be adjusted according to the values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration in the target combination. It should be noted that the values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration here are not the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration, but the values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration in the combination of the values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration.

[0106] In order to clearly illustrate the relationship between the molten salt heat storage tank and the thermal power unit in the molten salt heat storage system, this embodiment provides the following Figure 2 The schematic diagram of the electrical wiring of the molten salt heat storage system is shown in Figure 2 As shown, there are two molten salt heat storage tanks, and each molten salt heat storage tank is connected to a double-winding transformer, so that the two double-winding transformers connected to the molten salt heat storage tanks can be connected to the same three-winding transformer, and connected to the thermal power unit through the three-winding transformer. In this way, the molten salt heat storage system can be configured to cooperate with the thermal power unit for deep peak regulation.

[0107] The molten salt heat storage system adjustment method provided in the embodiment of the present invention obtains the thermal power unit data of multiple historical moments of the thermal power unit, and realizes training the long short-term memory LSTM network according to the thermal power unit data at each historical moment, so as to predict the thermal power generation power, heat release power generation power, heat storage duration and heat release duration of the subsequent moment after the corresponding historical moment for the input thermal power unit data by using the trained LSTM network, so as to obtain the corresponding predicted value, and for any combination, according to each predicted value, determine the value of the objective function, and after determining the target combination according to the value of the objective function, adjust the molten salt heat storage system according to the values ​​of the thermal power generation power, heat release power generation power, heat storage duration and heat release duration in the target combination. Thus, by adjusting the molten salt heat storage system, the operation mode of the molten salt heat storage system is optimized, and the overall benefits of the molten salt heat storage system and the thermal power unit are improved.

[0108] Through the above analysis, it can be known that in the embodiment of the present invention, the long short-term memory LSTM network can be trained according to the thermal power unit data at each historical moment. In order to clearly illustrate how the present invention trains the long short-term memory LSTM network according to the thermal power unit data at each historical moment, the present invention also provides a molten salt heat storage system adjustment method.

[0109] Figure 3 A schematic diagram of another molten salt heat storage system adjustment method provided in an embodiment of the present invention. Figure 3 As shown, the molten salt heat storage system adjustment method may include the following steps:

[0110] Step 301, obtaining the thermal power unit data of the thermal power unit at multiple historical moments.

[0111] It should be noted that the execution process of this step can refer to the execution process of step 101 in the previous embodiment, and the principles are the same, which will not be repeated here.

[0112] Step 302: The thermal power unit data at each historical moment is used as training data and input into the LSTM network to perform multiple rounds of iterative training.

[0113] In this embodiment, the thermal power group data of multiple historical moments of the thermal power group obtained can be used as training data and input into the LSTM network to perform multiple rounds of iterative training. Optionally, for each iterative training, the thermal power group data of a set number of consecutive historical moments in multiple historical moments can be taken and input into the LSTM network for training. Among them, the specific value of the set number is not limited in this embodiment. Optionally, it can be set according to manual experience. For example, the set number can be set to 10, or it can be set to 100, or it can be dynamically adjusted according to actual application requirements. This is not limited in this embodiment. It can be understood that although the specific value of the set number is not limited in this embodiment, the value of the set number should not be greater than the number of multiple historical moments obtained.

[0114] Step 303, each time a round of iterative training is performed, the LSTM network is used to predict the thermal storage power generation, heat release power generation, heat storage duration and heat release duration at subsequent moments after the corresponding historical moment for the thermal power unit data input at least one historical moment, so as to obtain the corresponding predicted values.

[0115] In this embodiment, the thermal power group data of at least one historical moment input may be the thermal power group data of a set number of consecutive historical moments among multiple historical moments. Among them, the value of the set number should not be greater than the number of multiple historical moments obtained. For example, assuming that the thermal power group data of multiple historical moments of the thermal power group are the thermal power group data of the thermal power group for nearly three hundred days, then the thermal power group data input for each iterative training may be the thermal power group data of the first one hundred days among the thermal power group data of the three hundred days, or may be the thermal power group data of the first fifty days among the thermal power group data of the three hundred days, but may not be the thermal power group data of more than three hundred days among the thermal power group data of the three hundred days.

[0116] In this embodiment, each time a round of iterative training is performed, an LSTM network can be used to predict the thermal storage power generation, thermal release power generation, thermal storage duration and thermal release duration at subsequent moments after the corresponding historical moment for the thermal power unit data at at least one historical moment input, so as to obtain the corresponding predicted value. As a possible implementation method, when the thermal power unit data within a set period is determined as the thermal power unit data at a corresponding historical moment, and obtaining the thermal power unit data at multiple historical moments of the thermal power unit is to obtain the thermal power unit data at multiple historical moments within the historical period of the thermal power unit, the subsequent moment after the historical moment can be the most recent set period after the historical moment. For example, assuming that the set cycle is one day and the historical period is nearly three hundred days, obtaining the thermal power unit data of multiple historical moments of the thermal power unit is to obtain the thermal power unit data of each day in the nearly three hundred days of the thermal power unit, and the thermal power unit data of at least one historical moment input is the thermal power unit data of the first one hundred days of the thermal power unit data of the thermal power unit for these three hundred days. Then, the subsequent moment after the historical moment is the day after the historical moment, so that the thermal power unit data of the first one hundred days of the thermal power unit data of the thermal power unit for these three hundred days can be input into the LSTM network, so that the LSTM network can predict the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration of the one hundred and first day to obtain the corresponding prediction values.

[0117] Step 304 , training the LSTM network based on the difference between each predicted value and the actual value at a subsequent moment after the corresponding historical moment.

[0118] In this embodiment, after obtaining the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration at the subsequent moment after the corresponding historical moment, the LSTM network can be trained according to the difference between each predicted value and the actual value at the subsequent moment after the corresponding historical moment. For example, in the above example of predicting the heat storage power generation, heat release power generation, heat storage duration and heat release duration of the 101st day, after obtaining the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration of the 101st day, the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration of the 101st day can be compared with the actual values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration of the 101st day, which is the day after the first 100 days in the thermal power unit data obtained in the past 300 days, that is, the 101st day, so as to train the LSTM network according to the difference between each predicted value and the corresponding actual value.

[0119] Step 305: for at least one combination of the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration, a trained LSTM network is used to predict the heat storage power generation, heat release power generation, heat storage duration and heat release duration at subsequent moments after the corresponding historical moment for the input thermal power unit data, so as to obtain corresponding predicted values.

[0120] Step 306, for any combination, determine the value of the objective function according to each predicted value, and determine the target combination according to the value of the objective function.

[0121] Step 307, adjusting the molten salt heat storage system according to the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration in the target combination.

[0122] It should be noted that the execution process of steps 305-307 can refer to the execution process of steps 103-105 in the previous embodiment, and the principles are the same, which will not be repeated here.

[0123] The molten salt heat storage system adjustment method provided by the embodiment of the present invention uses the thermal power unit data at each historical moment as training data, inputs it into the LSTM network to perform multiple rounds of iterative training, and realizes that each time a round of iterative training is performed, the LSTM network is used to predict the thermal storage power generation, heat release power generation, heat storage duration and heat release duration at the subsequent moment after the corresponding historical moment for the thermal power unit data input at least one historical moment, so as to obtain the corresponding predicted value, and then train the LSTM network according to the difference between each predicted value and the actual value at the subsequent moment after the corresponding historical moment. As a result, the trained LSTM network can more accurately predict the thermal storage power generation, heat release power generation, heat storage duration and heat release duration at the subsequent moment after the corresponding historical moment, thereby improving the accuracy of the prediction result.

[0124] From the above analysis, it can be seen that in an embodiment of the present invention, the data of thermal power units at historical moments can be used as training data and input into the long short-term memory LSTM network to perform multiple rounds of iterative training. In order to clearly explain how the present invention uses the data of thermal power units at historical moments as training data and inputs into the long short-term memory LSTM network to perform multiple rounds of iterative training, the present invention also provides a molten salt heat storage system adjustment method.

[0125] Figure 4 A schematic diagram of another molten salt heat storage system adjustment method provided in an embodiment of the present invention. Figure 4 As shown, the molten salt heat storage system adjustment method may include the following steps:

[0126] Step 401, obtaining the thermal power unit data of the thermal power unit at multiple historical moments.

[0127] It should be noted that the execution process of this step can refer to the execution process of step 301 in the previous embodiment, and the principles are the same, which will not be repeated here.

[0128] Step 402, normalizing the thermal power unit data at each historical moment using a normalization method to obtain training data.

[0129] In this embodiment, a normalization method can be used to normalize the thermal power unit data at each historical moment to obtain training data. As a possible implementation method, the normalization method can adopt a standardized normalization method, and its formula is as follows:

[0130]

[0131] Among them, x is used to represent any kind of thermal power unit data at any historical moment, Min is used to represent the minimum value of any kind of thermal power unit data at multiple historical moments of the thermal power unit, and Max is used to represent the maximum value of any kind of thermal power unit data at multiple historical moments of the thermal power unit.

[0132] It should be noted that, since the thermal power unit data can include seven types of data, namely, heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, AGC frequency regulation instruction and peak regulation instruction, and the thermal power unit data within a set period can be determined as the thermal power unit data at a corresponding historical moment, obtaining the thermal power unit data at multiple historical moments of the thermal power unit is to obtain the thermal power unit data at multiple historical moments within the historical period of the thermal power unit. Therefore, when the formula of the standardized normalization method is used to normalize the thermal power unit data, different types of data need to be processed separately. For example, assuming that the set cycle is one day and the historical period is nearly three hundred days, obtaining the thermal power unit data of multiple historical moments of the thermal power unit is to obtain the thermal power unit data of each day in the nearly three hundred days. Then, when using the formula of the standardized normalization method to normalize the thermal power unit data of each day in the nearly three hundred days, for each type of data, the maximum and minimum values ​​of the thermal power unit data of each day in these three hundred days can be first determined, and then the thermal power unit data of each day can be normalized using the formula of the above-mentioned standard normalization method to obtain the corresponding training data.

[0133] Step 403: Generate an input matrix of the LSTM network according to the training data.

[0134] In this embodiment, after obtaining the training data, an input matrix of the LSTM network can be generated according to the training data. Among them, each column in the input matrix corresponds to a historical moment, and each element in the same column of the input matrix is ​​used to indicate the thermal power unit data collected at the historical moment corresponding to the column. In other words, the number of rows of the input matrix is ​​consistent with the number of historical moments obtained, and the number of columns of the input matrix is ​​consistent with the type of thermal power unit data obtained. And because the thermal power unit data can include seven types of data, namely, heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, AGC frequency modulation instructions and peak load instructions, the number of rows of the input matrix of the generated LSTM network can be seven rows. Optionally, assuming that the number of historical moments obtained is n, the input matrix H of the generated LSTM network 7×n The expansion of can be:

[0135]

[0136] Among them, P D (1) , P D (2) ,…,P D (n) It is used to represent the heat release power at the first, second, ..., nth historical moments; P C (1) , P C (2) ,…,P C (n) It is used to represent the thermal storage power generation power at the first, second, ..., and nth historical moments; P Base (1) , P Base (2) ,…,P Base (n) It is used to represent the basic load power generation at the first, second, ..., and nth historical moments; P TP (1) , P TP (2) ,…,P TP (n) AGC frequency modulation instructions used to indicate the first, second, ..., nth historical moments; P TF (1) , P TF (2) ,…,P TF (n) Used to indicate the peak load instruction of the first, second, ..., nth historical moment; Δt D (1) , Δt D (2) ,…,ΔtD (n) Used to indicate the duration of heat release at the first, second, ..., nth historical moment; Δt C (1) , Δt C (2) ,…,Δt C (n) Used to indicate the heat storage duration at the first, second, ..., nth historical moments.

[0137] Step 404, inputting the thermal power unit data of at least one column in the input matrix into the LSTM network in the order of each historical moment, so as to determine the predicted values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration according to the output of the LSTM network.

[0138] In this embodiment, at least one column of thermal power unit data in the input matrix can be input into the LSTM network in the order of each historical moment, so as to determine the predicted values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration according to the output of the LSTM network. Among them, the thermal power unit data in one column of the input matrix is ​​the thermal power unit data at a historical moment. It should be noted that when multiple columns of thermal power unit data in the input matrix are input into the LSTM network, the multiple columns of thermal power unit data in the input matrix can be continuous multiple columns of data, so that at least one column of thermal power unit data in the input matrix is ​​input into the LSTM network, that is, the thermal power unit data of a set number of continuous historical moments in multiple historical moments are input into the LSTM network.

[0139] It can be understood that, since the thermal power unit data of at least one column in the input matrix can be input into the LSTM network in the order of each historical moment, the thermal power unit data of a set number of columns in the input matrix can be input into the LSTM network based on the order of each historical moment, and multiple rounds of iterative training can be performed. Among them, the specific value of the set number of columns is not limited in this embodiment. Optionally, it can be set according to manual experience. For example, the set number of columns can be set to 5, or it can be set to 10, or it can be dynamically adjusted according to actual application requirements, which is not limited in this embodiment. For example, assuming that the number of columns is set to 5, the first round of iterative training can be to input the thermal power unit data of the first 5 columns in the input matrix into the LSTM network, so as to determine the predicted values ​​of the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration in the 6th column according to the output of the LSTM network, and perform the second round of iterative training after training the LSTM network based on the difference between the predicted value of the 6th column and the corresponding actual value, and the second round of iterative training can be to input the thermal power unit data of the 2nd to 6th columns in the input matrix into the LSTM network, so as to determine the predicted values ​​of the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration in the 7th column according to the output of the LSTM network, and perform the third round of iterative training after training the LSTM network based on the difference between the predicted value of the 7th column and the corresponding actual value, and so on, and perform multiple rounds of iterative training.

[0140] Step 405, whenever a round of iterative training is performed, the LSTM network is used to predict the thermal storage power generation, heat release power generation, heat storage duration and heat release duration at subsequent moments after the corresponding historical moment for the thermal power unit data at at least one historical moment input, so as to obtain the corresponding predicted values.

[0141] Step 406, training the LSTM network based on the difference between each predicted value and the actual value at a subsequent moment after the corresponding historical moment.

[0142] Step 407: for at least one combination of the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration, a trained LSTM network is used to predict the heat storage power generation, heat release power generation, heat storage duration and heat release duration at subsequent moments after the corresponding historical moment for the input thermal power unit data, so as to obtain corresponding predicted values.

[0143] Step 408, for any combination, determine the value of the objective function according to each predicted value, and determine the target combination according to the value of the objective function.

[0144] Step 409, adjusting the molten salt heat storage system according to the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration in the target combination.

[0145] It should be noted that the execution process of steps 405-409 can refer to the execution process of steps 303-307 in the previous embodiment, and the principles are the same, which will not be repeated here.

[0146] The molten salt heat storage system adjustment method provided in the embodiment of the present invention uses a normalization method to normalize the thermal power unit data at each historical moment to obtain training data, thereby generating an input matrix of the LSTM network according to the training data, and then inputting the thermal power unit data of at least one column in the input matrix into the LSTM network according to the order of each historical moment, so as to determine the predicted values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration according to the output of the LSTM network. Thus, after data processing is performed on the thermal power unit data of multiple historical moments of the acquired thermal power unit, multiple rounds of iterative training can be performed on the LSTM network based on the order of each historical moment.

[0147] Through the above analysis, it can be known that in the embodiment of the present invention, each time a round of iterative training is performed, an LSTM network can be used to predict the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration at the subsequent moments after the corresponding historical moment for the thermal power unit data input at least one historical moment, so as to obtain the corresponding predicted values. In order to clearly illustrate how the present invention uses an LSTM network to predict the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration at the subsequent moments after the corresponding historical moment for the thermal power unit data input at least one historical moment, the present invention also provides a molten salt heat storage system adjustment method.

[0148] Figure 5 A schematic flow chart of another molten salt heat storage system adjustment method provided in an embodiment of the present invention. In this embodiment, the LSTM network may include a plurality of neurons, each of which may include a forget gate, an input gate, and an output gate.

[0149] like Figure 5 As shown, the molten salt heat storage system adjustment method may include the following steps:

[0150] Step 501, obtaining the thermal power unit data of the thermal power unit at multiple historical moments.

[0151] Step 502: The thermal power unit data at each historical moment is used as training data and input into the LSTM network to perform multiple rounds of iterative training.

[0152] It should be noted that the execution process of steps 501-502 can refer to the execution process of steps 301-302 in the above embodiment, and the principles are the same, which will not be repeated here.

[0153] Step 503: Whenever a round of iterative training is performed, for each neuron, the thermal power unit data at at least one historical moment is used as the input data corresponding to each neuron.

[0154] It can be understood that, since the thermal power unit data may include seven types of data, namely, heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, AGC frequency modulation instruction, and peak shaving instruction, the LSTM network may be provided with seven input ports for inputting heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, AGC frequency modulation instruction, and peak shaving instruction. And since the LSTM network predicts the heat storage power generation, heat release power generation, heat storage duration, and heat release duration at subsequent moments after the corresponding historical moment based on the input thermal power unit data, the LSTM network may be provided with four output ports for outputting the predicted values ​​of heat storage power generation, heat release power generation, heat storage duration, and heat release duration at subsequent moments after the corresponding historical moment.

[0155] In this embodiment, the LSTM network may include multiple neurons, so that each time a round of iterative training is performed, for each neuron, the thermal power unit data at at least one historical moment may be used as the input data corresponding to each neuron. Optionally, the input and output corresponding to the thermal power unit data at each historical moment may correspond to a group of LSTM neurons.

[0156] In order to clearly illustrate the structure of a single neuron in the LSTM network, this embodiment provides the following Figure 6 A schematic diagram of the structure of a single neuron in an LSTM network is shown in Figure 6 As shown, each neuron can include a forget gate, an input gate, and an output gate. The forget gate is used to determine the discarded data for the LSTM network neuron; the input gate is used to update the input data of the current neuron; the output gate is used to determine the corresponding output data based on the input data of the current neuron. Therefore, the LSTM network can achieve data protection and control through the input gate, forget gate, and output gate.

[0157] Step 504: for any neuron, use a forget gate to determine the corresponding discarded data based on the output data and input data of the previous neuron.

[0158] In this embodiment, after the input data corresponding to each neuron is determined, the corresponding discarded data can be determined for any neuron by using the forget gate in the neuron to determine the output data of the previous neuron and the input data of the current neuron. As a possible implementation, the following calculation formula can be used to determine the corresponding discarded data for the output data of the previous neuron and the input data of the current neuron:

[0159] f n =σ(W f ·[h n-1 ,x n ]+b f )

[0160] Among them, f n Used to indicate the corresponding discarded data; W f Used to represent weights, which can be determined by the LSTM network; h n-1 Used to represent the output data of the previous neuron; x n Used to represent the input data of the current neuron; b f Used to represent the bias, which can also be determined by the LSTM network; σ(x) is used to represent the sigmoid function.

[0161] It should be noted that the data of thermal power units at a historical moment include seven types of data, namely, heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, AGC frequency modulation instruction and peak shaving instruction. A neuron can only process one type of data of thermal power units at a historical moment. That is to say, for any neuron, its input data can be the heat storage power generation at a historical moment, or the heat release power generation at a historical moment, or the basic load power generation at a historical moment, or the heat storage duration at a historical moment, or the heat release duration at a historical moment, or the AGC frequency modulation instruction at a historical moment, or the peak shaving instruction at a historical moment.

[0162] Optionally, the determined discarded data can be any value between 0 and 1, where 1 means "completely retained" and 0 means "completely discarded". It should be noted that for the first neuron, since it does not have the output data of the previous neuron, the above formula cannot be used to determine its corresponding discarded data, so the discarded data corresponding to the first neuron can be directly set to 0 for complete discarding.

[0163] Step 505, using the first activation function of the input gate to update the output data and input data of the previous neuron to obtain first updated data, and using the second activation function of the input gate to perform alternative updates on the output data and input data of the previous neuron to obtain second updated data.

[0164] In this embodiment, after the discarded data corresponding to each neuron is determined, for any neuron, the output data of the previous neuron and the input data of the current neuron can be updated using the first activation function of the input gate in the neuron to obtain the first updated data, and the output data of the previous neuron and the input data of the current neuron can be alternatively updated using the second activation function of the input gate to obtain the second updated data. As a possible implementation, the calculation formula of the first activation function is as follows:

[0165] i n =σ(W i ·[h n-1 ,x n ]+b i )

[0166] Among them, i n Used to indicate the first update data; W i Used to represent weights, which can be determined by the LSTM network; h n-1 Used to represent the output data of the previous neuron; x n Used to represent the input data of the current neuron; b i It is used to represent the bias and can also be determined by the LSTM network; σ(x) is used to represent the sigmoid function, which is the first activation function.

[0167] The calculation formula of the second activation function is as follows:

[0168]

[0169] in, Used to indicate the second update data; W C Used to represent weights, which can be determined by the LSTM network; h n-1 Used to represent the output data of the previous neuron; x n Used to represent the input data of the current neuron; b C Used to represent bias, which can also be determined by the LSTM network; tanh(x) is used to represent the activation function of the tanh layer in the neuron, that is, the second activation function.

[0170] As a possible implementation method, the input gate of the neuron can be used to pass the output data h of the previous neuron to the n-1 and the input data x of the current neuron n Input to the sigmoid function for data update to generate i n At the same time, the output data h of the previous neuron n-1 and the input data x of the current neuron n Input to the tanh layer for an alternative update to generate

[0171] Step 506, determining third update data according to the first update data, the second update data, the output data of the previous neuron and the third update data of the previous neuron.

[0172] In this embodiment, after obtaining the first update data and the second update data, the input gate of the current neuron can combine the first update data of the current neuron, the second update data of the current neuron, the output data of the previous neuron, and the third update data of the previous neuron to update the input data of the current neuron and obtain the third update data of the current neuron. As a possible implementation, the calculation formula for determining the third update data of the current neuron is as follows:

[0173]

[0174] Among them, C n Used to represent the third updated data of the current neuron; C n-1 Used to represent the third updated data of the previous neuron; f n Used to represent the discarded data of the current neuron; i n Used to represent the first update data of the current neuron; Used to represent the second updated data of the current neuron.

[0175] Step 507: Use the output gate to process the third updated data to obtain corresponding output data.

[0176] In this embodiment, after the input gate updates the input data of the neuron to obtain the third updated data, the output gate of the neuron can be used to process the output data of the previous neuron, the input data of the current neuron and the third updated data to obtain the output data of the current neuron for input into the next neuron.

[0177] In order to clearly illustrate the relationship between neurons in the LSTM network, this embodiment provides the following Figure 7 The structural diagram of an LSTM network is shown in Figure 7 As shown, the input matrix of the LSTM network is H 7×nIn the case of , the thermal power unit data of the continuous set number of columns in the input matrix are input into each neuron in the LSTM network. Since the thermal power unit data includes seven kinds of data, namely, heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, AGC frequency modulation instruction and peak load instruction, the depth of the LSTM network can be seven layers, that is, there can be seven rows of neurons in the LSTM network. Since the thermal power unit data of the continuous set number of columns in the input matrix can be input into each neuron in the LSTM network, there can be a set number of columns of neurons in the LSTM network, therefore, the number of neurons in the LSTM network can be (7×set number of columns). It can be understood that the thermal power unit data of the continuous set number of columns in the input matrix are input into each neuron in the LSTM network, that is, the input data of each neuron is the thermal power unit data of the continuous set number of columns in the input matrix of the LSTM network, that is, the input data of a neuron is a corresponding element in any set number of columns in the input matrix of the LSTM network.

[0178] For example, you can Figure 7 As shown in the figure, the thermal power plant data H1, H2, ..., H in the first column to the i-th column in the input matrix can be i-1 ,H i Input into each neuron in the LSTM network to obtain the output data P1, P2, ..., P corresponding to each column of data i-1 ,P i . It can be understood that, since the number of columns is set to i, there can be i columns of neurons in the LSTM network. Therefore, inputting the thermal power unit data from the 1st column to the i-th column in the input matrix into each neuron in the LSTM network is to input the seven types of thermal power unit data in the 1st column of the input matrix into the seven neurons in the 1st column of the LSTM network, and inputting the seven types of thermal power unit data in the 2nd column into the seven neurons in the 2nd column of the LSTM network, and so on, until the seven types of thermal power unit data in the i-th column of the input matrix are input into the seven neurons in the i-th column of the LSTM network. Specifically, it can be understood as inputting the P in the 1st column of the input matrix into each neuron in the LSTM network. D (1) ,P C (1) ,P Base (1) ,P TP (1) ,P TF (1) ,Δt D (1) ,Δt C (1) Input into A in LSTM network respectively 11 ,A12 ,A 13 ,A 14 ,A 15 ,A 16 ,A 17 , input the P in the second column of the matrix D (2) ,P C (2) ,P Base (2) ,P TP (2) ,P TF (2) ,Δt D (2) ,Δt C (2) Input into A in LSTM network respectively 21 ,A 22 ,A 23 ,A 24 ,A 25 ,A 26 ,A 27 , and so on, until the P in the i-th column of the input matrix is D (i) ,P C (i) ,P Base (i) ,P TP (i) ,P TF (i) ,Δt D (i) ,Δt C (i) Input into A in LSTM network respectively i1 ,A i2 ,A i3 ,A i4 ,A i5 ,A i6 ,A i7 Thus, it is possible to realize the transformation of the thermal power plant data H1 in the first column to the thermal power plant data H in the i-th column in the input matrix. i Input into each neuron in the LSTM network to obtain the output data P1, P2, ..., P corresponding to each column of data i-1 ,P i .

[0179] It can be understood that each time a round of iterative training is performed, the LSTM network can use the above method to determine the output data of each neuron, and then predict the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration at subsequent moments after the input thermal power unit data of at least one historical moment, and obtain the corresponding predicted values.

[0180] Step 508: training the LSTM network based on the difference between each predicted value and the actual value at a subsequent moment after the corresponding historical moment.

[0181] Step 509: for at least one combination of the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration, a trained LSTM network is used to predict the heat storage power generation, heat release power generation, heat storage duration and heat release duration at subsequent moments after the corresponding historical moment for the input thermal power unit data, so as to obtain corresponding predicted values.

[0182] Step 510, for any combination, determine the value of the objective function according to each predicted value, and determine the target combination according to the value of the objective function.

[0183] Step 511, adjusting the molten salt heat storage system according to the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration in the target combination.

[0184] It should be noted that the execution process of steps 508-511 can refer to the execution process of steps 304-307 in the above embodiment, and the principles are the same, which will not be repeated here.

[0185] The molten salt heat storage system adjustment method provided by the embodiment of the present invention, by performing a round of iterative training, for each neuron, using at least one thermal power unit data of a historical moment as the input data corresponding to each neuron, to achieve for any neuron, using a forget gate to process the output data and input data of the previous neuron, and determining the discarded data corresponding to the input data, so as to update the output data and input data of the previous neuron using the first activation function of the input gate to obtain the first updated data, and using the second activation function of the input gate to perform an alternative update on the output data and input data of the previous neuron to obtain the second updated data, after determining the third updated data according to the first updated data, the second updated data, and the output data and input data of the previous neuron, the output gate is used to process the third updated data to obtain the corresponding output data. Thus, it can be achieved that each time a round of iterative training is performed, the LSTM network is used to determine the corresponding predicted value.

[0186] From the above analysis, it can be known that in the embodiment of the present invention, for any combination, the value of the objective function can be determined according to each predicted value, so as to determine the target combination according to the value of the objective function. In order to clearly explain how the present invention determines the value of the objective function for any combination according to each predicted value, so as to determine the target combination according to the value of the objective function, the present invention also provides a molten salt heat storage system adjustment method.

[0187] Figure 8 A schematic diagram of another molten salt heat storage system adjustment method provided in an embodiment of the present invention. Figure 8 As shown, the molten salt heat storage system adjustment method may include the following steps:

[0188] Step 801, obtaining the thermal power unit data of the thermal power unit at multiple historical moments.

[0189] Step 802: training a long short-term memory (LSTM) network based on the thermal power unit data at each historical moment.

[0190] Step 803: for at least one combination of the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration, a trained LSTM network is used to predict the heat storage power generation, heat release power generation, heat storage duration and heat release duration at subsequent moments after the corresponding historical moment for the input thermal power unit data, so as to obtain corresponding predicted values.

[0191] It should be noted that the execution process of steps 801-803 can refer to the execution process of steps 101-103 in the above embodiment, and the principles are the same, which will not be repeated here.

[0192] Step 804, for any combination, determine the value of the objective function according to the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration.

[0193] In this embodiment, for any combination, the value of the objective function can be determined according to the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration predicted by the trained LSTM network. Optionally, the objective function can be the cycle electrical efficiency η TES As a possible implementation method, the cycle electrical efficiency η TES The calculation formula can be:

[0194]

[0195] Therefore, the objective function, i.e., the cycle electrical efficiency η, can be calculated based on the predicted values ​​of heat storage power generation, heat release power generation, heat storage duration, and heat release duration corresponding to each combination. TES The value of .

[0196] Step 805: In response to the value of the objective function being greater than or equal to a set threshold, the corresponding combination is determined as a reference combination.

[0197] In this embodiment, for any combination, when the value of the objective function of the combination is greater than or equal to the set threshold, the combination can be determined as a reference combination to determine the target combination from the reference combination. The specific value of the set threshold is not limited in this embodiment, and can be set according to manual experience, for example, the set threshold can be set to 90%, or it can be dynamically adjusted according to actual application requirements, which is not limited in this embodiment.

[0198] It can be understood that the number of the determined reference combinations may be one or more, which is not limited in this embodiment.

[0199] Step 806: In response to there being multiple reference combinations, determine a target combination according to the values ​​of the target functions corresponding to the reference combinations.

[0200] In this embodiment, when there are multiple reference combinations to be determined, the target combination can be determined according to the values ​​of the objective functions corresponding to the reference combinations. Optionally, the reference combination with the largest value of the objective functions corresponding to the reference combinations can be determined as the target combination.

[0201] Step 807: In response to the reference combination being one, determine the target combination from the reference combination.

[0202] In this embodiment, when the number of determined reference combinations is one, the reference combination may be directly determined as the target combination.

[0203] Step 808, adjusting the molten salt heat storage system according to the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration in the target combination.

[0204] It should be noted that the execution process of this step can refer to the execution process of step 105 in the above embodiment, and the principles are the same, which will not be repeated here.

[0205] The molten salt heat storage system adjustment method provided in the embodiment of the present invention determines the value of the objective function according to the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration for any combination, so that in response to the value of the objective function being greater than or equal to the set threshold, the corresponding combination is determined as a reference combination, and then in response to the reference combination being multiple, the target combination is determined according to the size of the value of the objective function corresponding to each reference combination, or, in response to the reference combination being one, the reference combination is determined as the target combination. Thus, it is possible to determine the value of the objective function for any combination according to the predicted values ​​of the combination, so as to determine the target combination according to the value of the objective function.

[0206] In order to explain the above embodiment more clearly, an example is now given for illustration.

[0207] like Fig. 9 As shown, the molten salt heat storage system adjustment method may include the following steps:

[0208] Step 901, taking a day as a cycle, obtaining the thermal power group data of n days in a historical period of the thermal power group, so as to generate an input matrix according to the thermal power group data.

[0209] Here, the thermal power unit data includes heat storage power generation, heat release power generation, base load power generation, heat storage duration, heat release duration, AGC frequency regulation instruction and peak load regulation instruction. There is no restriction on the specific value of the historical period. The generated input matrix can be:

[0210]

[0211] Among them, P D (1) , P D (2) ,…,P D (n) It is used to indicate the heat release power generation on the first day, the second day, ..., the nth day; P C (1) , P C (2) ,…,P C (n) It is used to indicate the thermal storage power generation on the first day, the second day, ..., and the nth day; P Base (1) , P Base (2) ,…,P Base (n) It is used to represent the basic load power generation on the first day, the second day, ..., and the nth day; P TP (1) , P TP (2),…,P TP (n) It is used to indicate the AGC frequency modulation instructions for the first day, the second day, ..., and the nth day; P TF (1) , P TF (2) ,…,P TF (n) It is used to indicate the peak load instructions for the first day, the second day, ..., and the nth day; Δt D (1) , Δt D (2) ,…,Δt D (n) Used to indicate the duration of heat release on the first day, the second day, ..., the nth day; Δt C (1) , Δt C (2) ,…,Δt C (n) Used to indicate the heat storage duration on the first day, the second day, ..., the nth day.

[0212] Step 902, input the input matrix into the LSTM network to obtain the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration on the n+mth day (m=1, 2, 3, 4...) output by the LSTM network.

[0213] Here, the input matrix H 7×n The data are input into the LSTM network to obtain the predicted values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration on the n+1th day output by the LSTM network.

[0214] Step 903, calculating the cycle electrical efficiency on the n+m day according to the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration on the n+m day.

[0215] Here, the cycle electrical efficiency on the n+mth day is calculated based on the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration, and heat release duration on the n+mth day. TES The calculation formula can be:

[0216]

[0217] Step 904: compare the calculated cycle electrical efficiency on the (n+m)th day with a set threshold.

[0218] Here, the calculated cycle electrical efficiency on the n+mth day is compared with a set threshold value, so that different comparison results are processed differently.

[0219] Step 905, in response to the cycle electrical efficiency on the n+mth day being less than the set threshold, the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration on the n+mth day are changed, and the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration and the input matrix on the n+mth day are used as inputs for a new round of iterative training.

[0220] Here, when the cycle electrical efficiency on the n+mth day is less than the set threshold, the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration on the n+mth day can be changed, and the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration on the n+mth day and the input matrix are used as the input of a new round of iterative training. As a possible implementation method, when the cycle electrical efficiency on the n+mth day is less than the set threshold, the predicted value of the heat storage duration on the n+mth day can be increased, the predicted value of the heat release duration on the n+mth day can be reduced, or the predicted value of the heat storage power generation and the predicted value of the heat release power generation on the n+mth day can be increased. Among them, the specific value of the set threshold is not limited in this embodiment. Optionally, it can be set according to manual experience. For example, the set threshold can be set to 90%, or it can be dynamically adjusted according to actual application requirements. This is not limited in this embodiment.

[0221] Step 906, in response to the cycle electrical efficiency on the n+mth day being greater than or equal to the set threshold, the molten salt heat storage system is adjusted according to the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration and heat release duration on the n+mth day.

[0222] Here, when the cycle electrical efficiency on the n+mth day is greater than or equal to the set threshold, the operation mode of the molten salt heat storage system can be adjusted according to the predicted values ​​of the heat storage power generation, heat release power generation, heat storage duration, and heat release duration on the n+mth day. As a possible implementation method, the operation mode of the molten salt heat storage system can be adjusted according to the prediction curves of each predicted value and the prediction curve of the cycle electrical efficiency.

[0223] It should be noted that when the LSTM network is overfitting, the iteration can be terminated and the network parameters, such as the depth of the LSTM network, can be adjusted; when the LSTM network is fitting normally, the next round of iteration can be continued.

[0224] In summary, by taking a day as a cycle, the data of thermal power units for n days in the historical period of the thermal power units are obtained, so as to generate an input matrix according to the data of the thermal power units, and input the input matrix into the LSTM network to obtain the predicted values ​​of the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration of the n+1th day output by the LSTM network. According to the predicted values ​​of the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration of the n+1th day, the cycle electrical efficiency of the n+1th day is calculated, and the calculated cycle electrical efficiency of the n+1th day is compared with the set threshold, so as to respond to If the cycle electrical efficiency on the n+1th day is less than the set threshold, the predicted values ​​of the thermal storage power generation, thermal release power generation, thermal storage duration and thermal release duration on the n+1th day are changed, and the predicted values ​​of the thermal storage power generation, thermal release power generation, thermal storage duration and thermal release duration on the n+mth day are added to the input matrix for iterative training, and then in response to the cycle electrical efficiency on the n+mth day (m=1,2,3,4……) being greater than or equal to the set threshold, the molten salt thermal storage system is adjusted according to the predicted values ​​of the thermal storage power generation, thermal release power generation, thermal storage duration and thermal release duration on the n+mth day. In this way, the operation mode of the molten salt thermal storage system can be optimized to maximize the system benefits.

[0225] In order to implement the above-mentioned embodiment, the present invention also proposes a molten salt heat storage system adjustment device.

[0226] Fig.10 A schematic structural diagram of a molten salt heat storage system adjustment device provided in an embodiment of the present invention.

[0227] like Fig.10 As shown, the molten salt heat storage system adjustment device includes: an acquisition module 11, a training module 12, a prediction module 13, a determination module 14 and an adjustment module 15.

[0228] The acquisition module 11 is used to acquire the thermal power unit data at multiple historical moments of the thermal power unit; wherein the thermal power unit data includes heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, automatic generation control AGC frequency modulation instruction and peak load regulation instruction;

[0229] A training module 12 is used to train a long short-term memory (LSTM) network based on the thermal power unit data at each historical moment;

[0230] The prediction module 13 is used to predict the heat storage power generation power, heat release power generation power, heat storage duration and heat release duration at subsequent moments after the corresponding historical moment for the input thermal power unit data, using a trained LSTM network for at least one combination of the values ​​of the heat storage power generation power, the heat release power generation power, the heat storage duration and the heat release duration, so as to obtain the corresponding prediction value;

[0231] The determination module 14 is used to determine the value of the objective function according to each predicted value for any combination, so as to determine the target combination according to the value of the objective function; wherein the objective function is used to indicate the cycle electrical efficiency of the thermal power unit;

[0232] The adjustment module 15 is used to adjust the molten salt heat storage system according to the values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration in the target combination.

[0233] Furthermore, in a possible implementation of the embodiment of the present invention, the training module 12 includes:

[0234] An input unit 1201 is used to input the thermal power unit data at each historical moment as training data into the LSTM network to perform multiple rounds of iterative training;

[0235] The prediction unit 1202 is used to predict the thermal storage power generation, thermal release power generation, thermal storage duration and thermal release duration at subsequent moments after the corresponding historical moment by using the LSTM network for each round of iterative training, based on the thermal power unit data at at least one historical moment input, so as to obtain corresponding prediction values;

[0236] The training unit 1203 is used to train the LSTM network according to the difference between each predicted value and the actual value at a subsequent moment after the corresponding historical moment.

[0237] Furthermore, in a possible implementation of the embodiment of the present invention, the input unit 1201 is further configured to:

[0238] The normalization method is used to normalize the thermal power unit data at historical moments to obtain training data;

[0239] Generate an input matrix of the LSTM network based on the training data; wherein each column in the input matrix corresponds to a historical moment, and each element in the same column of the input matrix is ​​used to indicate the thermal power unit data collected at the historical moment corresponding to the column;

[0240] In the order of historical moments, the thermal power unit data of at least one column in the input matrix is ​​input into the LSTM network to determine the predicted values ​​of heat storage power generation, heat release power generation, heat storage duration and heat release duration according to the output of the LSTM network.

[0241] Further, in a possible implementation of the embodiment of the present invention, the LSTM network includes a plurality of neurons, the neurons include a forget gate, an input gate, and an output gate, and the prediction unit 1202 is further used for:

[0242] Each time a round of iterative training is performed, for each neuron, the thermal power unit data at at least one historical moment is used as the input data corresponding to each neuron;

[0243] For any neuron, use the forget gate to compare the output data and input data of the previous neuron to determine the discarded data corresponding to the input data;

[0244] The output data and input data of the previous neuron are updated by using a first activation function of the input gate to obtain first updated data, and the output data and input data of the previous neuron are alternatively updated by using a second activation function of the input gate to obtain second updated data;

[0245] Determine third update data according to the first update data, the second update data, and the output data and input data of the previous neuron;

[0246] The third update data is processed using an output gate to obtain corresponding output data.

[0247] Furthermore, in a possible implementation of the embodiment of the present invention, the prediction unit 1202 is further configured to:

[0248] The output data and input data of the previous neuron are filtered using the third activation function of the output gate to obtain filtered data;

[0249] The third updated data and the filtered data are processed using the second activation function of the output gate to obtain corresponding output data.

[0250] Furthermore, in a possible implementation of the embodiment of the present invention, the determining module 14 includes:

[0251] The first determining unit 1401 is used to determine the value of the objective function according to the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration for any combination;

[0252] A second determining unit 1402 is configured to determine the corresponding combination as a reference combination in response to the value of the objective function being greater than or equal to a set threshold;

[0253] The third determining unit 1403 is configured to determine the target combination according to the magnitude of the value of the target function corresponding to each reference combination in response to the number of reference combinations being multiple;

[0254] The fourth determining unit 1404 is configured to determine a target combination from the reference combination in response to the reference combination being one.

[0255] It should be noted that the aforementioned explanation of the embodiment of the molten salt heat storage system adjustment method is also applicable to the molten salt heat storage system adjustment device of this embodiment, and will not be repeated here.

[0256] The molten salt heat storage system adjustment device provided by the embodiment of the present invention obtains the thermal power unit data of multiple historical moments of the thermal power unit, and realizes training the long short-term memory LSTM network according to the thermal power unit data at each historical moment, so as to predict the thermal power generation power, heat release power generation power, heat storage duration and heat release duration of the subsequent moment after the corresponding historical moment for the input thermal power unit data by using the trained LSTM network, so as to obtain the corresponding predicted value, and for any combination, according to each predicted value, determine the value of the objective function, and after determining the target combination according to the value of the objective function, adjust the molten salt heat storage system according to the values ​​of the thermal power generation power, heat release power generation power, heat storage duration and heat release duration in the target combination. Therefore, by adjusting the molten salt heat storage system, the operation mode of the molten salt heat storage system is optimized, and the overall benefits of the molten salt heat storage system and the thermal power unit are improved.

[0257] In order to implement the above embodiments, the present invention also proposes an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the molten salt heat storage system adjustment method proposed in any of the aforementioned embodiments of the present invention.

[0258] Fig.11 A schematic diagram of the structure of an electronic device provided in an embodiment of the present invention can implement the present invention Figure 1-10 The process of the embodiment shown is as follows: Fig.11 As shown, the electronic device may include: a shell 1101, a processor 1102, a memory 1103, a circuit board 1104 and a power supply circuit 1105, wherein the circuit board 1104 is arranged inside the space enclosed by the shell 1101, and the processor 1102 and the memory 1103 are arranged on the circuit board 1104; the power supply circuit 1105 is used to supply power to various circuits or devices of the above-mentioned electronic device; the memory 1103 is used to store executable program codes; the processor 1102 runs a program corresponding to the executable program code by reading the executable program code stored in the memory 1103, so as to execute the molten salt heat storage system adjustment method described in any of the aforementioned embodiments.

[0259] For details on the specific execution process of the above steps by the processor 1102 and the steps further executed by the processor 1102 by running the executable program code, please refer to the present invention. Figure 1-10 The description of the illustrated embodiment will not be repeated here.

[0260] In order to implement the above embodiments, the present invention further proposes a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the molten salt heat storage system adjustment method proposed in any of the aforementioned embodiments of the present invention.

[0261] In order to implement the above embodiments, the present invention further proposes a computer program product, including a computer program, which, when executed by a processor, implements the molten salt heat storage system adjustment method proposed in any of the above embodiments of the present invention.

[0262] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0263] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0264] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present invention belong.

[0265] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0266] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0267] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0268] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0269] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present invention. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for adjusting a molten salt heat storage system, characterized in that: The following steps are involved: Obtaining thermal power unit data at multiple historical moments of the thermal power unit; wherein the thermal power unit data includes heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, automatic generation control AGC frequency modulation instruction and peak load regulation instruction; According to the thermal power unit data at each of the historical moments, a long short-term memory LSTM network is trained, including: The data of the thermal power unit at each of the historical moments are used as training data and input into the LSTM network to perform multiple rounds of iterative training; each time a round of iterative training is performed, the LSTM network is used to predict the thermal storage power generation, the thermal release power generation, the thermal storage duration and the thermal release duration at the subsequent moment after the corresponding historical moment for the input thermal power unit data at least one of the historical moments, so as to obtain the corresponding predicted values; the LSTM network is trained according to the difference between each of the predicted values ​​and the actual value at the subsequent moment after the corresponding historical moment; wherein the LSTM network includes a plurality of neurons, and the neurons include a forgetting gate, an input gate and an output gate. Each time a round of iterative training is performed, for each of the neurons, the data of the thermal power unit at at least one of the historical moments is used as the input data corresponding to each of the neurons; for any one of the neurons element, using the forget gate to process the output data and the input data of the previous neuron to determine the discarded data corresponding to the input data; using the first activation function of the input gate to update the output data and the input data of the previous neuron to obtain the first updated data, and using the second activation function of the input gate to perform an alternative update on the output data and the input data of the previous neuron to obtain the second updated data; determining the third updated data according to the first updated data, the second updated data, the output data of the previous neuron and the third updated data of the previous neuron; using the output gate to process the output data, the input data and the third updated data of the previous neuron to obtain the corresponding output data; the calculation formula of the first activation function is as follows: in, represents the first update data; represents weight; Represents the output data of the previous neuron; represents the input data; Indicates bias; represents the sigmoid function, i.e., the first activation function; The calculation formula of the second activation function is as follows: in, represents the second update data; represents weight; Represents the output data of the previous neuron; represents the input data; Indicates bias; Indicates that the neuron The activation function of the layer, i.e., the second activation function; For at least one combination of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration, a trained LSTM network is used to predict the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration at subsequent moments after the corresponding historical moment based on the input thermal power unit data, so as to obtain corresponding predicted values; For any of the combinations, the value of the objective function is determined according to the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration; in response to the value of the objective function being greater than or equal to the set threshold, the corresponding combination is determined as a reference combination; in response to the reference combination being multiple, the target combination is determined according to the value of the objective function corresponding to each reference combination; in response to the reference combination being one, the reference combination is determined as the target combination; wherein the objective function is used to indicate the cycle electrical efficiency of the thermal power unit; the calculation formula for the cycle electrical efficiency is: in, represents the predicted value of the cycle electrical efficiency, represents the predicted value of the thermal storage power generation power, represents the predicted value of heat storage duration, represents the predicted value of the exothermic power generation, represents the predicted value of the heat release duration, represents the base load power generation; The molten salt heat storage system is adjusted according to the values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration in the target combination.

2. The method according to claim 1, characterized in that: The thermal power unit data at each of the historical moments is used as training data and input into the LSTM network to perform multiple rounds of iterative training, including: Normalizing the thermal power unit data at each of the historical moments using a normalization method to obtain the training data; Generate an input matrix of the LSTM network according to the training data; wherein each column in the input matrix corresponds to a historical moment, and each element in the same column in the input matrix is ​​used to indicate the thermal power unit data collected at the historical moment corresponding to the column; In the order of each of the historical moments, the thermal power unit data of at least one column in the input matrix is ​​input into the LSTM network to determine the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration according to the output of the LSTM network.

3. The method according to claim 1, characterized in that The using the output gate to process the output data of the previous neuron, the input data and the third update data to obtain corresponding output data includes: Using the third activation function of the output gate to perform screening processing on the output data and the input data of the previous neuron to obtain screened data; The third updated data and the screened data are processed using a fourth activation function of the output gate to obtain corresponding output data.

4. A molten salt heat storage system adjustment device, characterized in that: include: An acquisition module is used to acquire the thermal power unit data at multiple historical moments of the thermal power unit; wherein the thermal power unit data includes heat storage power generation, heat release power generation, basic load power generation, heat storage duration, heat release duration, automatic generation control AGC frequency modulation instruction and peak load regulation instruction; A training module is used to use the thermal power unit data of each of the historical moments as training data, and input it into the LSTM network to perform multiple rounds of iterative training; each time a round of iterative training is performed, the LSTM network is used to predict the thermal storage power generation power, the thermal release power generation power, the thermal storage duration and the thermal release duration at the subsequent moment after the corresponding historical moment for the input thermal power unit data of at least one of the historical moments, so as to obtain the corresponding predicted values; the LSTM network is trained according to the difference between each of the predicted values ​​and the actual value at the subsequent moment after the corresponding historical moment; wherein the LSTM network includes a plurality of neurons, and the neurons include a forgetting gate, an input gate and an output gate. Each time a round of iterative training is performed, for each of the neurons, the thermal power unit data of at least one of the historical moments is used as the input data corresponding to each of the neurons; for any one The neuron uses the forget gate to process the output data and the input data of the previous neuron to determine the discarded data corresponding to the input data; uses the first activation function of the input gate to update the output data and the input data of the previous neuron to obtain the first updated data, and uses the second activation function of the input gate to perform alternative updates on the output data and the input data of the previous neuron to obtain the second updated data; determines the third updated data according to the first updated data, the second updated data, the output data of the previous neuron and the third updated data of the previous neuron; uses the output gate to process the output data, the input data and the third updated data of the previous neuron to obtain the corresponding output data; the calculation formula of the first activation function is as follows: in, represents the first update data; represents weight; Represents the output data of the previous neuron; represents the input data; Indicates bias; represents the sigmoid function, i.e., the first activation function; The calculation formula of the second activation function is as follows: in, represents the second update data; represents weight; Represents the output data of the previous neuron; represents the input data; Indicates bias; Indicates that the neuron The activation function of the layer, i.e., the second activation function; A prediction module, for predicting, for at least one combination of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration, the heat storage power generation, the heat release duration and the heat release duration at a subsequent moment after a corresponding historical moment for the input thermal power unit data, using a trained LSTM network to obtain a corresponding predicted value; A determination module is used to determine the value of the objective function for any of the combinations according to the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration; in response to the value of the objective function being greater than or equal to a set threshold, the corresponding combination is determined as a reference combination; in response to the reference combination being multiple, the target combination is determined according to the value of the objective function corresponding to each reference combination; in response to the reference combination being one, the reference combination is determined as the target combination; wherein the objective function is used to indicate the cycle electrical efficiency of the thermal power unit; the calculation formula for the cycle electrical efficiency is: in, represents the predicted value of the cycle electrical efficiency, represents the predicted value of the thermal storage power generation power, represents the predicted value of heat storage duration, represents the predicted value of the exothermic power generation, represents the predicted value of the heat release duration, represents the base load power generation; An adjustment module is used to adjust the molten salt heat storage system according to the values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration in the target combination.

5. The device according to claim 4, characterized in that The training module is also used for: Normalizing the thermal power unit data at each of the historical moments using a normalization method to obtain the training data; Generate an input matrix of the LSTM network according to the training data; wherein each column in the input matrix corresponds to a historical moment, and each element in the same column in the input matrix is ​​used to indicate the thermal power unit data collected at the historical moment corresponding to the column; In the order of each of the historical moments, the thermal power unit data of at least one column in the input matrix is ​​input into the LSTM network to determine the predicted values ​​of the heat storage power generation, the heat release power generation, the heat storage duration and the heat release duration according to the output of the LSTM network.

6. The device according to claim 4, characterized in that The training module is also used for: Using the third activation function of the output gate to perform screening processing on the output data and the input data of the previous neuron to obtain screened data; The third updated data and the screened data are processed using a fourth activation function of the output gate to obtain corresponding output data.

7. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 3.

8. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 3.

9. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method according to any one of claims 1 to 3.

Images