Energy scheduling method and device based on multi-target model solution, terminal equipment and storage medium

Abstract

The invention discloses an energy scheduling method and device based on multi-target model solution, terminal equipment and a storage medium, and can utilize a causal neural network to predict an output scheduling strategy at the next moment. In the training process of the causal neural network, the causal coefficient in the neural network is continuously updated based on the comparison result of the prediction result and the actual result, so that the causal neural network can learn and capture the causal relationship between the historical output scheduling strategy and the future output scheduling strategy; the method can better adapt to the uncertainty and dynamic nature of an energy supply and demand system, and improves the prediction reliability, so that when a multi-target scheduling optimization model is solved, multi-target solving can be carried out based on each accurate output scheduling strategy at the next moment, and the prediction efficiency is improved. Therefore, an accurate target output scheduling strategy capable of enabling the operation cost to be minimum is obtained, a reliable data basis is provided for subsequent scheduling optimization, and the situation of energy shortage or excess is avoided.

Description

Technical Field

[0001] The present invention relates to the field of energy scheduling technology, and in particular to an energy scheduling method, device, terminal equipment and storage medium based on multi-objective model solving. Background Art

[0002] With the continuous coupling between natural gas, heat-cooling networks and electrical systems, the utilization efficiency, economy, environmental protection and stability of each energy source need to be considered comprehensively in the planning and scheduling of each energy system. Different indicators are both interrelated and contradictory, so the problem of multiple indicators can be transformed into a multi-objective problem. In the integrated energy system, both economy and carbon emissions must be considered, while ensuring the optimal energy conversion cost. Therefore, there may be mutual constraints or conflicts between multiple objectives. Therefore, how to find a set of solutions that can simultaneously meet or balance these objectives is an urgent problem to be solved.

[0003] Existing technologies often solve multi-objective problems through mathematical analysis and intelligent optimization algorithms, that is, the optimal scheduling strategy can be solved from multiple scheduling strategies. However, due to the significant intermittency and volatility of new energy sources such as solar energy and wind energy, and the randomness of user behavior, existing technologies need to overcome uncertainty when using neural networks or models to predict future energy supply and demand to generate multiple scheduling strategies. However, existing models cannot fully capture the uncertainty and dynamics of energy supply and demand systems, and often find it difficult to accurately predict future energy scheduling strategies. Therefore, when solving the optimal scheduling strategy from multiple predicted scheduling strategies, the optimal scheduling strategy obtained is likely to have deviations due to multi-objective solutions based on inaccurate scheduling strategies. This deviation will be amplified in subsequent energy scheduling and output control, leading to energy shortages or surpluses. Summary of the invention

[0004] The embodiments of the present invention provide an energy scheduling method, apparatus, terminal device and storage medium based on solving a multi-objective model. By introducing a causal neural network, it can better adapt to the uncertainty and dynamics of the energy supply and demand system, improve the prediction stability and reliability of the optimal scheduling strategy, and effectively solve the problem of energy shortage or surplus caused by deviations in the optimal scheduling strategy in the prior art.

[0005] An embodiment of the present invention provides an energy scheduling method based on solving a multi-objective model, comprising:

[0006] A scheduling optimization model for minimizing operating costs is constructed based on first basic data of the energy conversion system, second basic data of the combined heat, cooling and power system, and third basic data of the power-to-gas system; wherein the first basic data includes first output data and first operating cost data of the energy conversion system; the second basic data includes second output data and second operating cost data of the combined heat, cooling and power system; and the third basic data includes third output data and third operating cost data of the power-to-gas system;

[0007] Inputting several historical output dispatch strategies corresponding to the current moments into a preset causal neural network, so that the causal neural network predicts the output dispatch strategy of the next moment corresponding to each current moment; wherein the historical output dispatch strategy includes the historical output dispatch values ​​corresponding to the energy conversion system, the combined heat and power system, and the power-to-gas system respectively;

[0008] Under the preset output constraints of the scheduling optimization model, based on a number of output scheduling strategies at the next moment, the scheduling optimization model is solved to generate a target output scheduling strategy when the operating cost is minimized;

[0009] According to the target output dispatching strategy, the output of the energy conversion system, the combined cooling, heating and power system and the power-to-gas system are respectively controlled;

[0010] Among them, the generation of causal neural network includes:

[0011] The sample output scheduling strategy corresponding to each sample moment is used as the input of the neural network, and the actual output scheduling strategy at the next moment corresponding to each sample moment is used as the output of the neural network. The neural network is iteratively trained until the neural network converges, and the trained neural network is used as the causal neural network;

[0012] During each training, the sample output scheduling strategy corresponding to each sample moment is input into the neural network, so that the neural network outputs the prediction result of the output scheduling strategy corresponding to each sample moment; the prediction result of the output scheduling strategy corresponding to each sample moment is compared with the actual output scheduling strategy of the next moment, and the causal coefficient in the neural network is adjusted according to the comparison result;

[0013] The causal coefficient is used to characterize the causal relationship between the sample output scheduling strategy corresponding to the sample moment and the actual output scheduling strategy corresponding to the next moment.

[0014] Preferably, the energy conversion system is a gas turbine that realizes the conversion between power supply, heat supply and gas supply; the combined heat and power system is a refrigerator that can realize the integration of refrigeration, heat supply and power generation; the power-to-gas system is a power-to-gas device for converting wind power into natural gas;

[0015] The first operating cost data includes power generation cost data of the gas turbine, carbon emission cost data of the gas turbine and energy conversion cost data of the gas turbine; the second operating cost data includes conversion cost data of the refrigeration equipment, carbon emission cost data of the refrigeration equipment and energy generation cost data of the refrigeration equipment; the third operating cost data includes cost data of the power-to-gas equipment converted to natural gas, carbon emission cost data of the power-to-gas equipment and energy conversion cost data of the power-to-gas equipment;

[0016] A scheduling optimization model to minimize operating costs is constructed according to the following formula:

[0017] F = argmin(F1, F2, F3);

[0018]

[0019]

[0020] Among them, F is used to represent the scheduling optimization model that minimizes the total operating cost of gas turbines, refrigeration equipment and power-to-gas equipment, F1 is used to represent the total startup cost of gas turbines, refrigeration equipment and power-to-gas equipment, F2 is used to represent the total carbon emission cost of gas turbines, refrigeration equipment and power-to-gas equipment, and F3 is used to represent the total energy conversion cost of gas turbines, refrigeration equipment and power-to-gas equipment; The cost of electricity generation from gas turbines; The conversion cost of the refrigeration equipment; Energy cost for refrigeration equipment; The cost of conversion to natural gas for power-to-gas equipment; is the carbon emission cost of gas turbines; Carbon emission costs for chiller equipment; The carbon emission cost of the power-to-gas equipment; Energy conversion costs for gas turbine operation; Energy conversion costs for combined cooling, heating and power systems; Energy conversion costs for power-to-gas equipment;

[0021] a, b and c are different efficiency coefficients of micro gas turbines; P mt is the start and stop state of the gas turbine; H ac is the input thermal power of the refrigeration equipment; COP ac D is the refrigeration coefficient of the refrigeration equipment; ac P is the conversion coefficient of the refrigerator through electrical energy cooling; ac is the operating energy consumption of the refrigeration equipment; η p2g is the conversion efficiency of the power-to-gas equipment; Pp2g is the power of the power-to-gas technology; α, β, and γ are the quadratic factor, linear factor, and constant factor of carbon emissions from gas turbines, respectively; r ac is the carbon emission factor for refrigeration equipment; Energy conversion factor for refrigeration equipment; is the energy conversion factor of the power-to-gas equipment.

[0022] Preferably, the preset output constraints of the scheduling optimization model include:

[0023]

[0024] in, is the preset upper limit of the gas turbine output power, is the preset upper limit of the input thermal power of the refrigeration equipment, It is the preset upper limit of the power-to-gas equipment power.

[0025] Preferably, the neural network comprises: a causal selection network and a causal prediction network;

[0026] The step of inputting the sample output scheduling strategy corresponding to each sample moment into the neural network so that the neural network outputs the prediction result of the output scheduling strategy corresponding to each sample moment includes:

[0027] The sample output scheduling strategies corresponding to each sample time are input into the neural network so that the causal selection network extracts the strategy features corresponding to each sample output scheduling strategy;

[0028] The causal prediction network predicts the output scheduling strategy prediction result corresponding to each sample moment according to the strategy characteristics.

[0029] Preferably, comparing the output dispatch strategy prediction result corresponding to each sample moment with the corresponding actual output dispatch strategy at the next moment, and adjusting the causal coefficient in the neural network according to the comparison result, includes:

[0030] The output scheduling strategy prediction result corresponding to each sample moment is compared with the actual output scheduling strategy corresponding to the next moment, and the loss function corresponding to the neural network is updated according to the comparison result; wherein the loss function is:

[0031]

[0032] Among them, L loss represents the loss value, S' is the prediction result of the output scheduling strategy corresponding to the sample time; S is the actual output scheduling strategy at the next moment; m is the data size of the small batch sampling, and i is the sample index of the small batch sampling;

[0033] The gradient value of the output dispatch strategy prediction result corresponding to each sample moment is calculated according to the updated loss function; wherein the gradient value is used to indicate the adjustment direction and adjustment size corresponding to the parameters in the neural network;

[0034] The causal coefficient in the neural network is adjusted according to the gradient value corresponding to the prediction result of each output scheduling strategy.

[0035] Preferably, the gradient value of the output scheduling strategy prediction result corresponding to each sample moment is calculated according to the updated loss function, including:

[0036] The loss value of each output dispatch strategy prediction result is calculated according to the loss function;

[0037] The loss value of each output scheduling strategy prediction result is multiplied by the activation function of the causal selection network and the causal prediction network respectively to obtain the gradient value corresponding to each output scheduling strategy prediction result in the causal selection network and the causal prediction network respectively.

[0038] Preferably, adjusting the causal coefficient in the neural network according to the gradient value corresponding to each output scheduling strategy prediction result includes:

[0039] According to the gradient values ​​of the prediction results of each output dispatch strategy in the causal selection network and the causal prediction network, the causal coefficients corresponding to the causal selection network and the causal prediction network are adjusted respectively.

[0040] Based on the above method embodiments, the present invention provides corresponding device embodiments.

[0041] An embodiment of the present invention provides an energy scheduling device based on multi-objective model solving, including: a model building module, a scheduling strategy prediction module, a multi-objective solving module and an output control module;

[0042] The model building module is used to build a scheduling optimization model for minimizing operating costs based on first basic data of the energy conversion system, second basic data of the combined cooling, heat and power system, and third basic data of the power-to-gas system; wherein the first basic data includes first output data and first operating cost data of the energy conversion system; the second basic data includes second output data and second operating cost data of the combined cooling, heat and power system; and the third basic data includes third output data and third operating cost data of the power-to-gas system;

[0043] The scheduling strategy prediction module is used to input several historical output scheduling strategies corresponding to the current moment into a preset causal neural network, so that the causal neural network predicts the output scheduling strategy of the next moment corresponding to each current moment; wherein the historical output scheduling strategy includes the historical output scheduling values ​​corresponding to the energy conversion system, the combined heat and power system, and the power-to-gas system respectively;

[0044] The multi-objective solving module is used to solve the scheduling optimization model based on a number of output scheduling strategies at the next moment under the preset output constraints of the scheduling optimization model, and generate a target output scheduling strategy when the operating cost is minimized;

[0045] The output control module is used to control the output of the energy conversion system, the combined cooling, heating and power system and the power-to-gas system respectively according to the target output scheduling strategy;

[0046] Among them, the generation of causal neural network includes:

[0047] The sample output scheduling strategy corresponding to each sample moment is used as the input of the neural network, and the actual output scheduling strategy at the next moment corresponding to each sample moment is used as the output of the neural network. The neural network is iteratively trained until the neural network converges, and the trained neural network is used as the causal neural network;

[0048] During each training, the sample output scheduling strategy corresponding to each sample moment is input into the neural network, so that the neural network outputs the prediction result of the output scheduling strategy corresponding to each sample moment; the prediction result of the output scheduling strategy corresponding to each sample moment is compared with the actual output scheduling strategy of the next moment, and the causal coefficient in the neural network is adjusted according to the comparison result;

[0049] The causal coefficient is used to characterize the causal relationship between the sample output scheduling strategy corresponding to the sample moment and the actual output scheduling strategy corresponding to the next moment.

[0050] Based on the above method embodiments, the present invention provides corresponding terminal device embodiments.

[0051] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements an energy scheduling method based on multi-objective model solving as described in the above-mentioned embodiment of the invention.

[0052] Based on the above method embodiments, the present invention provides corresponding storage medium item embodiments.

[0053] Another embodiment of the present invention provides a storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute an energy scheduling method based on multi-objective model solving as described in the above-mentioned embodiment of the invention.

[0054] The following beneficial effects are achieved by implementing the present invention:

[0055] The embodiment of the present invention provides an energy scheduling method, device, terminal device and storage medium based on solving a multi-objective model. The energy scheduling method of the present invention can predict the output scheduling strategy of the next moment by using a causal neural network, and in the training process of the causal neural network, the causal coefficient in the neural network is continuously updated based on the comparison result of the predicted result and the actual result, so that the causal neural network can learn and capture the causal relationship between the historical output scheduling strategy and the future output scheduling strategy, and can more accurately predict the output scheduling strategy of the future (i.e., the next moment) when applied. Therefore, the present invention can better adapt to the uncertainty and dynamics of the energy supply and demand system by introducing the causal neural network, and improve the stability and reliability of the prediction. Compared with the prior art, when solving the multi-objective scheduling optimization model, the present invention can perform multi-objective solutions based on each accurate output scheduling strategy of the next moment to obtain an accurate target output scheduling strategy that can minimize the operating cost, and provide a reliable data basis for subsequent scheduling optimization, so that the output can be adjusted based on the target output scheduling strategy with a small deviation from the actual situation, thereby avoiding the occurrence of energy shortages or surpluses. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flowchart of an energy scheduling method based on solving a multi-objective model provided by an embodiment of the present invention.

[0057] Figure 2 It is a schematic diagram of a causal neural network structure provided by an embodiment of the present invention.

[0058] Figure 3 It is a schematic diagram of parameter self-update based on meta-learning provided by an embodiment of the present invention.

[0059] Figure 4 It is a structural schematic diagram of an energy scheduling device based on multi-objective model solving provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] like Figure 1 FIG. 1 is a flow chart of an energy scheduling method based on solving a multi-objective model provided by an embodiment of the present invention. The energy scheduling method based on solving a multi-objective model includes:

[0062] Step S1: constructing a scheduling optimization model for minimizing operating costs based on first basic data of the energy conversion system, second basic data of the combined cooling, heat and power system, and third basic data of the power-to-gas system; wherein the first basic data includes first output data and first operating cost data of the energy conversion system; the second basic data includes second output data and second operating cost data of the combined cooling, heat and power system; and the third basic data includes third output data and third operating cost data of the power-to-gas system;

[0063] Step S2: inputting several historical output dispatch strategies corresponding to the current moments into a preset causal neural network, so that the causal neural network predicts the output dispatch strategy of the next moment corresponding to each current moment; wherein the historical output dispatch strategy includes the historical output dispatch values ​​corresponding to the energy conversion system, the combined heat and power system, and the power-to-gas system respectively;

[0064] Wherein, the generation of the causal neural network includes:

[0065] The sample output scheduling strategy corresponding to each sample moment is used as the input of the neural network, and the actual output scheduling strategy at the next moment corresponding to each sample moment is used as the output of the neural network. The neural network is iteratively trained until the neural network converges, and the trained neural network is used as the causal neural network;

[0066] During each training, the sample output scheduling strategy corresponding to each sample moment is input into the neural network, so that the neural network outputs the prediction result of the output scheduling strategy corresponding to each sample moment; the prediction result of the output scheduling strategy corresponding to each sample moment is compared with the actual output scheduling strategy of the next moment, and the causal coefficient in the neural network is adjusted according to the comparison result;

[0067] The causal coefficient is used to characterize the causal relationship between the sample output scheduling strategy corresponding to the sample moment and the actual output scheduling strategy corresponding to the next moment;

[0068] Step S3: Under the preset output constraints of the scheduling optimization model, based on a number of output scheduling strategies at the next moment, solving the scheduling optimization model to generate a target output scheduling strategy when the operating cost is minimized;

[0069] Step S4: controlling the output of the energy conversion system, the combined cooling, heating and power system and the power-to-gas system respectively according to the target output dispatching strategy.

[0070] For step S1, in a preferred embodiment, the energy conversion system may be a gas turbine that realizes conversion between power supply, heat supply and gas supply; the cogeneration system may be a refrigerator device that can realize integrated refrigeration, heating and power generation; the power-to-gas system may be a power-to-gas device for converting wind power into natural gas.

[0071] Furthermore, the first operating cost data mainly includes the power generation cost data of the gas turbine, the carbon emission cost data of the gas turbine and the energy conversion cost data of the gas turbine; the second operating cost data mainly includes the conversion cost data of the refrigeration equipment, the carbon emission cost data of the refrigeration equipment and the energy generation cost data of the refrigeration equipment; the third operating cost data mainly includes the cost data of the power-to-gas equipment converted to natural gas, the carbon emission cost data of the power-to-gas equipment and the energy conversion cost data of the power-to-gas equipment.

[0072] The micro gas turbine is the main equipment for realizing the conversion between power supply, heat supply and gas supply. Its mathematical model can be expressed as follows:

[0073]

[0074] in, is the power generation cost of the gas turbine; P mt is the start and stop status of the gas turbine; a, b and c are different efficiency coefficients of the micro gas turbine.

[0075] The combined cooling, heating and power system is the main equipment for realizing the integration of refrigeration, heating and power generation. The mathematical model of its absorption chiller equipment is as follows:

[0076]

[0077] in, is the conversion cost of the refrigeration equipment; H ac is the input thermal power of the refrigeration equipment; COP ac is the refrigeration coefficient of the refrigeration equipment, is the energy cost of the refrigeration equipment; P mt is the start and stop state of the gas turbine, D ac is the conversion factor of the refrigerator through electrical energy cooling; is the cost of conversion from power to gas equipment to natural gas; p2 g is the conversion efficiency of the power-to-gas equipment; α, β, and γ are the quadratic factor, linear factor, and constant factor of carbon emissions from the gas turbine, respectively; P ac is the operating energy consumption of the refrigeration equipment; ac is the carbon emission factor of the refrigeration equipment.

[0078] Power-to-gas technology converts wind power into natural gas technology. The mathematical model of the corresponding power-to-gas equipment is as follows:

[0079]

[0080] in, Energy conversion costs for combined cooling, heating and power systems; Energy conversion costs for power-to-gas equipment; Energy conversion factor for refrigeration equipment; is the energy conversion factor of the power-to-gas device. Schematically, the energy conversion cost of the combined cooling, heating and power system is the carbon emission cost data of the power-to-gas device in the third operating cost data.

[0081] Through the above-mentioned mathematical models of electricity, gas, and cold (hot) gas systems, the present invention can provide a scheduling optimization model that takes into account economy, reduction of carbon emissions, and energy conversion costs as the goal as shown below:

[0082] F = argmin(F1, F2, F3);

[0083]

[0084] Among them, F is used to represent the scheduling optimization model that minimizes the total operating cost of gas turbines, refrigeration equipment and power-to-gas equipment, F1 is used to represent the total startup cost of gas turbines, refrigeration equipment and power-to-gas equipment, F2 is used to represent the total carbon emission cost of gas turbines, refrigeration equipment and power-to-gas equipment, and F3 is used to represent the total energy conversion cost of gas turbines, refrigeration equipment and power-to-gas equipment; The cost of electricity generation from gas turbines; The conversion cost of the refrigeration equipment; Energy cost for refrigeration equipment; The cost of conversion to natural gas for power-to-gas equipment; is the carbon emission cost of gas turbines; Carbon emission costs for chiller equipment; The carbon emission cost of the power-to-gas equipment; Energy conversion costs for gas turbine operation; Energy conversion costs for combined cooling, heating and power systems; is the energy conversion cost of the power-to-gas equipment.

[0085] Furthermore, the preset output constraints of the scheduling optimization model are the processing constraints corresponding to the energy conversion system, the combined cooling, heating and power system and the power-to-gas system respectively, and then:

[0086]

[0087] in, is the preset upper limit of the gas turbine output power, is the preset upper limit of the input thermal power of the refrigeration equipment, It is the preset upper limit of the power-to-gas equipment power.

[0088] For step S2, in a preferred embodiment, existing technologies such as mathematical analysis and intelligent optimization algorithms have achieved certain results in solving multi-objective problems, but the accuracy of their prediction and scheduling strategies is challenged when faced with the intermittent and volatile nature of new energy and the randomness of user behavior. Moreover, due to these characteristics of new energy and the randomness of user behavior, the uncertainty and dynamics of the energy supply and demand system are increased, making it difficult for existing models to fully capture these changes, thereby affecting the accuracy of prediction and scheduling.

[0089] When solving multiple objectives based on inaccurate scheduling strategies, the optimal scheduling strategy obtained is likely to be biased, which will lead to energy shortages or surpluses due to biased output supply in subsequent energy scheduling and output control. Energy shortages may lead to insufficient power supply, heating or cooling in some areas, affecting residents' lives and industrial production; while energy surpluses will cause resource waste and environmental pollution.

[0090] In order to overcome these problems, before solving the above-mentioned multi-objective model (i.e., the scheduling optimization model), the present invention can predict accurate scheduling strategies through a preset causal neural network, and then solve the multi-objective model based on these predicted scheduling strategies to ensure that the target output scheduling strategies solved subsequently are more accurate and reduce the deviation from the actual output scheduling.

[0091] In an embodiment of the present invention, several historical output scheduling strategies corresponding to the current moment can be input into a preset causal neural network, so that the causal neural network predicts the output scheduling strategy of the next moment corresponding to each current moment;

[0092] The present invention provides a real-time scheduling strategy distribution prediction scheme for comprehensive energy based on causal deep learning, such as Figure 2The structural diagram of the causal neural network shown in the figure, the neural network (i.e., the causal neural network) includes: a causal selection network and a causal prediction network; wherein, the causal selection network and the causal prediction network are both composed of multiple hidden layers.

[0093] In order to improve the prediction accuracy of the causal neural network of the present invention, during the training process of the causal neural network, when the sample output scheduling strategy corresponding to each sample moment is input into the neural network so that the neural network outputs the prediction result of the output scheduling strategy corresponding to each sample moment, it specifically includes:

[0094] The sample output scheduling strategies corresponding to each sample time are input into the neural network so that the causal selection network extracts the strategy features corresponding to each sample output scheduling strategy;

[0095] The causal prediction network predicts the output scheduling strategy prediction result corresponding to each sample moment according to the strategy characteristics.

[0096] Specifically, the causal selection network can extract historical time series features. Assuming the prediction time is T = (t1, t2, ..., tn), the sample historical output dispatch strategy of the energy supplier can be expressed as:

[0097] {S={(F1,t1),(F2,t2),…,(F n ,t n}};

[0098] Then the output dispatch strategy prediction result corresponding to each sample time predicted by the causal prediction network can be expressed as:

[0099] {S'={(F1',t1'),(F2',t'2),…,(F n ',t' n )}}.

[0100] After obtaining the output dispatch strategy prediction result corresponding to each sample moment, the output dispatch strategy prediction result corresponding to each sample moment can be compared with the actual output dispatch strategy of the next corresponding moment;

[0101] First, the loss function corresponding to the neural network is updated according to the comparison result; wherein the loss function is:

[0102]

[0103] Among them, L loss represents the loss value, S' is the prediction result of the output scheduling strategy corresponding to the sample time; S is the actual output scheduling strategy at the next moment; m is the data size of the small batch sampling, and i is the sample index of the small batch sampling;

[0104] Then, back propagation is performed based on the updated loss function, and the ReLU function is used to update the parameters of the network to accurately predict the output scheduling strategy at the next moment. Specifically:

[0105] The loss value of each output dispatch strategy prediction result is calculated according to the loss function;

[0106] The loss value of each output scheduling strategy prediction result is multiplied by the activation function of the causal selection network and the causal prediction network respectively, to obtain the gradient value corresponding to each output scheduling strategy prediction result in the causal selection network and the causal prediction network respectively;

[0107] The gradient value is used to indicate the adjustment direction and size corresponding to the parameters in the neural network; the gradient value reflects the influence of the difference between the predicted result and the actual result on each parameter in the network. In the back propagation process, these gradient values ​​are used to guide the update of network parameters. For the causal selection network and the causal prediction network of the present invention, their respective gradient values ​​reflect their contribution to the final prediction result and the direction that needs to be adjusted.

[0108] Furthermore, the causal coefficients corresponding to the causal selection network and the causal prediction network can be adjusted respectively according to the gradient values ​​of the prediction results of each output scheduling strategy in the causal selection network and the causal prediction network. Schematically, the parameters in the network, including the causal coefficients and other network weights, can be updated according to the calculated gradient, and the direction of the update is the opposite direction of the gradient because the goal is to minimize the loss function. Through multiple iterative training, the network parameters are continuously optimized until the preset number of training rounds is reached or other stopping conditions are met.

[0109] It is understandable that causal neural networks can capture causal relationships in data. By updating the parameters of the network (including but not limited to causal coefficients), the causal neural network can better learn and understand the causal relationship between the input data (sample output scheduling strategy) and the output data (output scheduling strategy prediction results), so that the network can be closer to the actual results in subsequent predictions. Schematically, compared to traditional Sigmoid or Tanh activation functions, the ReLU function has a constant gradient (ie 1) when the input is positive, which helps alleviate the common gradient vanishing problem in deep neural networks.

[0110] In the causal neural network, the causal selection network is responsible for extracting historical time series features, while the causal prediction network makes predictions based on these features. By adjusting the causal coefficient, the impact of these two parts on the prediction results can be balanced to ensure that they can fully play their respective roles.

[0111] When adjusting the causal coefficient, the present invention iteratively updates various parameters in the network through an optimization algorithm (such as a gradient descent algorithm). Specifically, based on the gradient value and the learning rate of the network (a hyperparameter that controls the parameter update step), the update amount of the causal coefficient can be calculated and applied to the network. This process will be repeated until the network reaches a convergence state or meets other stopping conditions.

[0112] By adjusting the causal coefficient, the network can be made more adaptable to specific data distribution and task requirements, thereby improving its performance in practical applications. In the training process of the causal neural network, the present invention continuously updates the causal coefficient in the neural network based on the comparison results of the predicted results and the actual results, so that the causal neural network can learn and capture the causal relationship between the historical output scheduling strategy and the future output scheduling strategy. When applied, the output scheduling strategy of the future (i.e., the next moment) can be predicted more accurately. Therefore, by introducing the causal neural network, the present invention can better adapt to the uncertainty and dynamics of the energy supply and demand system, and improve the stability and reliability of the prediction.

[0113] In a preferred embodiment, in order to make the prediction accuracy of the scheduling strategy higher and speed up the training of the neural network, the present invention introduces the idea of ​​meta-learning to compare the similarity between the historical scheduling strategy and the real-time predicted scheduling strategy, so as to better optimize the output scheduling strategy predicted at the next moment. Figure 3 The schematic diagram of parameter self-update based on meta-learning is shown. In meta-learning, the historical time series features of energy suppliers extracted by the causal selection network are first pre-trained;

[0114] The historical time series characteristics can be: D tr ={(F1,t1),(F2,t2),...(F n ,t n )};

[0115] Then the training set is set to D tr ={(F1,t1),(F2,t2),...(F n ,t n )}; then the causal prediction network is based on the test set, takes the test set as input, and finally outputs a probability value between the test set and the training value. The larger the probability, the more similar the test set is to the training value. The test set updates the parameter φ in the network in an adaptive manner until the network converges. Thus, the present invention improves the extraction ability of the causal selection network and the prediction ability of the causal prediction network through adaptive parameter updating, and enables the causal neural network to accurately predict the output scheduling strategy of the energy supply even with a small number of samples.

[0116] For step S3 and step S4, in a preferred embodiment, when solving a multi-objective scheduling optimization model, the present invention can perform multi-objective solutions based on each accurate output scheduling strategy for the next moment. Then, when solving the scheduling optimization model, an accurate target output scheduling strategy that can minimize the operating cost can be generated, providing a reliable data basis for subsequent scheduling optimization.

[0117] Therefore, the output of the comprehensive multi-energy supply system can be adjusted based on a target output dispatching strategy that deviates less from the actual situation, thereby avoiding energy shortages or surpluses.

[0118] Specifically, by adjusting the power (i.e., output) of the equipment, refined management of energy use can be achieved, thereby further optimizing the energy use strategy. For example, during peak power consumption periods, the power of non-critical equipment can be reduced to reduce the load on the power grid and improve the stability of the power grid; during low power consumption periods, the power of the equipment can be increased to fully utilize the remaining capacity of the power grid. The present invention can predict the output scheduling strategy at the next moment in real time based on the historical output scheduling strategy at the current moment, which enables the energy scheduling system to respond to changes in real time and quickly adjust the scheduling strategy to adapt to the changing energy demand, which can help to achieve staggered use of energy and improve energy utilization efficiency.

[0119] In a preferred embodiment, in a comprehensive multi-energy supply system, the scheduling strategy may also contain plans and decisions for energy production, conversion, distribution and use made by each energy supplier based on factors such as energy demand, supply conditions, equipment status and market conditions. The scheduling strategy is thus intended to optimize the operation of the energy system to meet energy demand while achieving goals such as economy, environmental protection and stability. When the historical scheduling strategy of an energy supplier is associated with output data, this generally means that the actual effects of the historical scheduling strategy and the corresponding output data are taken into account during the formulation and execution of the scheduling strategy. In an embodiment of the present invention, output data refers to the amount of energy or power output actually generated by an energy supply device (such as a gas turbine, a wind turbine, a solar photovoltaic panel, etc.) under a specific scheduling strategy.

[0120] The present invention realizes the coordinated optimization and scheduling of multiple energy systems by combining the scheduling optimization model and causal neural network prediction. It can not only consider the operating characteristics and costs of each system, but also use historical data to predict future output scheduling strategies, thereby improving energy utilization efficiency and reducing operating costs.

[0121] In summary, the present invention realizes the real-time scheduling strategy distribution prediction and multi-objective solution method of comprehensive energy through causal deep learning, which solves the problem that the prior art cannot cope with complex systems and cannot dynamically maintain the optimization of multiple objectives. It improves the accuracy of prediction, optimizes the scheduling strategy, and enables the comprehensive energy system to minimize multiple objectives such as economy, carbon emissions and energy conversion costs while ensuring safe and stable operation.

[0122] like Figure 3 As shown, based on the above-mentioned various embodiments of the energy scheduling method based on solving the multi-objective model, the present invention provides a corresponding device item embodiment;

[0123] An embodiment of the present invention provides an energy scheduling device based on multi-objective model solving, including: a model building module, a scheduling strategy prediction module, a multi-objective solving module and an output control module;

[0124] The model building module is used to build a scheduling optimization model for minimizing operating costs based on first basic data of the energy conversion system, second basic data of the combined cooling, heat and power system, and third basic data of the power-to-gas system; wherein the first basic data includes first output data and first operating cost data of the energy conversion system; the second basic data includes second output data and second operating cost data of the combined cooling, heat and power system; and the third basic data includes third output data and third operating cost data of the power-to-gas system;

[0125] The scheduling strategy prediction module is used to input several historical output scheduling strategies corresponding to the current moment into a preset causal neural network, so that the causal neural network predicts the output scheduling strategy of the next moment corresponding to each current moment; wherein the historical output scheduling strategy includes the historical output scheduling values ​​corresponding to the energy conversion system, the combined heat and power system, and the power-to-gas system respectively;

[0126] The multi-objective solving module is used to solve the scheduling optimization model based on a number of output scheduling strategies at the next moment under the preset output constraints of the scheduling optimization model, and generate a target output scheduling strategy when the operating cost is minimized;

[0127] The output control module is used to control the output of the energy conversion system, the combined cooling, heating and power system and the power-to-gas system respectively according to the target output scheduling strategy;

[0128] Among them, the generation of causal neural network includes:

[0129] The sample output scheduling strategy corresponding to each sample moment is used as the input of the neural network, and the actual output scheduling strategy at the next moment corresponding to each sample moment is used as the output of the neural network. The neural network is iteratively trained until the neural network converges, and the trained neural network is used as the causal neural network;

[0130] During each training, the sample output scheduling strategy corresponding to each sample moment is input into the neural network, so that the neural network outputs the prediction result of the output scheduling strategy corresponding to each sample moment; the prediction result of the output scheduling strategy corresponding to each sample moment is compared with the actual output scheduling strategy of the next moment, and the causal coefficient in the neural network is adjusted according to the comparison result;

[0131] The causal coefficient is used to characterize the causal relationship between the sample output scheduling strategy corresponding to the sample moment and the actual output scheduling strategy corresponding to the next moment.

[0132] It should be noted that the device embodiments described above are merely schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, and may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art can understand and implement it without paying creative labor.

[0133] Those skilled in the art can clearly understand that, for the sake of convenience and simplicity, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0134] Based on the above-mentioned various embodiments of the energy scheduling method based on solving the multi-objective model, the present invention provides corresponding terminal equipment item embodiments.

[0135] An embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements an energy scheduling method based on multi-objective model solving as described in any method embodiment of the present invention.

[0136] The terminal device may be a computing terminal device such as a desktop computer, a notebook, a palm computer, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0137] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0138] The memory can be used to store the computer program, and the processor realizes various functions of the terminal device by running or executing the computer program stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device or other volatile solid-state storage device.

[0139] Based on the above-mentioned various embodiments of the energy scheduling method based on solving the multi-objective model, the present invention provides a corresponding storage medium item embodiment.

[0140] An embodiment of the present invention provides a storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute an energy scheduling method based on multi-objective model solving as described in any method embodiment of the present invention.

[0141] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by the processor, the steps of each method embodiment described above can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0142] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An energy scheduling method based on solving a multi-objective model, characterized in that: include: A scheduling optimization model for minimizing operating costs is constructed based on first basic data of the energy conversion system, second basic data of the combined heat, cooling and power system, and third basic data of the power-to-gas system; wherein the first basic data includes first output data and first operating cost data of the energy conversion system; the second basic data includes second output data and second operating cost data of the combined heat, cooling and power system; and the third basic data includes third output data and third operating cost data of the power-to-gas system; Inputting several historical output dispatch strategies corresponding to the current moments into a preset causal neural network, so that the causal neural network predicts the output dispatch strategy of the next moment corresponding to each current moment; wherein the historical output dispatch strategy includes the historical output dispatch values ​​corresponding to the energy conversion system, the combined heat and power system, and the power-to-gas system respectively; Under the preset output constraints of the scheduling optimization model, based on a number of output scheduling strategies at the next moment, the scheduling optimization model is solved to generate a target output scheduling strategy when the operating cost is minimized; According to the target output dispatching strategy, the output of the energy conversion system, the combined cooling, heating and power system and the power-to-gas system are respectively controlled; Among them, the generation of causal neural network includes: The sample output scheduling strategy corresponding to each sample moment is used as the input of the neural network, and the actual output scheduling strategy at the next moment corresponding to each sample moment is used as the output of the neural network. The neural network is iteratively trained until the neural network converges, and the trained neural network is used as the causal neural network; During each training, the sample output scheduling strategy corresponding to each sample moment is input into the neural network, so that the neural network outputs the prediction result of the output scheduling strategy corresponding to each sample moment; the prediction result of the output scheduling strategy corresponding to each sample moment is compared with the actual output scheduling strategy of the next moment, and the causal coefficient in the neural network is adjusted according to the comparison result; The causal coefficient is used to characterize the causal relationship between the sample output scheduling strategy corresponding to the sample moment and the actual output scheduling strategy corresponding to the next moment.

2. The energy scheduling method based on solving a multi-objective model according to claim 1, characterized in that: The energy conversion system is a gas turbine that realizes the conversion between power supply, heat supply and gas supply; the combined heat and power system is a refrigerator that can realize the integration of refrigeration, heat supply and power generation; the power-to-gas system is a power-to-gas device used to convert wind power into natural gas; The first operating cost data includes power generation cost data of the gas turbine, carbon emission cost data of the gas turbine and energy conversion cost data of the gas turbine; the second operating cost data includes conversion cost data of the refrigeration equipment, carbon emission cost data of the refrigeration equipment and energy generation cost data of the refrigeration equipment; the third operating cost data includes cost data of the power-to-gas equipment converted to natural gas, carbon emission cost data of the power-to-gas equipment and energy conversion cost data of the power-to-gas equipment; A scheduling optimization model to minimize operating costs is constructed according to the following formula: F = argmin(F1, F2, F3); Among them, F is used to represent the scheduling optimization model that minimizes the total operating cost of gas turbines, refrigeration equipment and power-to-gas equipment, F1 is used to represent the total startup cost of gas turbines, refrigeration equipment and power-to-gas equipment, F2 is used to represent the total carbon emission cost of gas turbines, refrigeration equipment and power-to-gas equipment, and F3 is used to represent the total energy conversion cost of gas turbines, refrigeration equipment and power-to-gas equipment; The cost of electricity generation from gas turbines; The conversion cost of the refrigeration equipment; Energy cost for refrigeration equipment; The cost of conversion to natural gas for power-to-gas equipment; is the carbon emission cost of gas turbines; Carbon emission costs for chiller equipment; The carbon emission cost of the power-to-gas equipment; Energy conversion costs for gas turbine operation; Energy conversion costs for combined cooling, heating and power systems; Energy conversion costs for power-to-gas equipment; a, b and c are different efficiency coefficients of micro gas turbines; P mt is the start and stop state of the gas turbine; H ac is the input thermal power of the refrigeration equipment; COP ac D is the refrigeration coefficient of the refrigeration equipment; ac P is the conversion coefficient of the refrigerator through electrical energy cooling; ac is the operating energy consumption of the refrigeration equipment; η p2 g is the conversion efficiency of the power-to-gas equipment; P p2g is the power of the power-to-gas technology; α, β, and γ are the quadratic factor, linear factor, and constant factor of carbon emissions from gas turbines, respectively; r ac is the carbon emission factor for refrigeration equipment; Energy conversion factor for refrigeration equipment; is the energy conversion factor of the power-to-gas equipment.

3. The energy scheduling method based on multi-objective model solving as claimed in claim 2 is characterized in that: The preset output constraints of the scheduling optimization model include: in, is the preset upper limit of the gas turbine output power, is the preset upper limit of the input thermal power of the refrigeration equipment, It is the preset upper limit of the power-to-gas equipment power.

4. The energy scheduling method based on multi-objective model solving as claimed in claim 3 is characterized in that: The neural network includes: a causal selection network and a causal prediction network; The step of inputting the sample output scheduling strategy corresponding to each sample moment into the neural network so that the neural network outputs the prediction result of the output scheduling strategy corresponding to each sample moment includes: The sample output scheduling strategies corresponding to each sample time are input into the neural network so that the causal selection network extracts the strategy features corresponding to each sample output scheduling strategy; The causal prediction network predicts the output scheduling strategy prediction result corresponding to each sample moment according to the strategy characteristics.

5. The energy scheduling method based on multi-objective model solving as claimed in claim 4 is characterized in that: The method of comparing the output dispatch strategy prediction result corresponding to each sample moment with the corresponding actual output dispatch strategy at the next moment, and adjusting the causal coefficient in the neural network according to the comparison result, includes: The output scheduling strategy prediction result corresponding to each sample moment is compared with the actual output scheduling strategy corresponding to the next moment, and the loss function corresponding to the neural network is updated according to the comparison result; wherein the loss function is: Among them, L loss represents the loss value, S' is the prediction result of the output scheduling strategy corresponding to the sample time; S is the actual output scheduling strategy at the next moment; m is the data size of the small batch sampling, and i is the sample index of the small batch sampling; The gradient value of the output dispatch strategy prediction result corresponding to each sample moment is calculated according to the updated loss function; wherein the gradient value is used to indicate the adjustment direction and adjustment size corresponding to the parameters in the neural network; The causal coefficient in the neural network is adjusted according to the gradient value corresponding to the prediction result of each output scheduling strategy.

6. The energy scheduling method based on multi-objective model solving according to claim 5 is characterized in that: The gradient value of the output scheduling strategy prediction result corresponding to each sample time is calculated according to the updated loss function, including: The loss value of each output dispatch strategy prediction result is calculated according to the loss function; The loss value of each output scheduling strategy prediction result is multiplied by the activation function of the causal selection network and the causal prediction network respectively to obtain the gradient value corresponding to each output scheduling strategy prediction result in the causal selection network and the causal prediction network respectively.

7. The energy scheduling method based on multi-objective model solving according to claim 6 is characterized in that: The step of adjusting the causal coefficient in the neural network according to the gradient value corresponding to each output scheduling strategy prediction result includes: According to the gradient values ​​of the prediction results of each output dispatch strategy in the causal selection network and the causal prediction network, the causal coefficients corresponding to the causal selection network and the causal prediction network are adjusted respectively.

8. An energy dispatching device based on solving a multi-objective model, characterized in that: include: Model building module, scheduling strategy prediction module, multi-objective solution module and output control module; The model building module is used to build a scheduling optimization model for minimizing operating costs based on first basic data of the energy conversion system, second basic data of the combined cooling, heat and power system, and third basic data of the power-to-gas system; wherein the first basic data includes first output data and first operating cost data of the energy conversion system; the second basic data includes second output data and second operating cost data of the combined cooling, heat and power system; and the third basic data includes third output data and third operating cost data of the power-to-gas system; The scheduling strategy prediction module is used to input several historical output scheduling strategies corresponding to the current moment into a preset causal neural network, so that the causal neural network predicts the output scheduling strategy of the next moment corresponding to each current moment; wherein the historical output scheduling strategy includes the historical output scheduling values ​​corresponding to the energy conversion system, the combined heat and power system, and the power-to-gas system respectively; The multi-objective solving module is used to solve the scheduling optimization model based on a number of output scheduling strategies at the next moment under the preset output constraints of the scheduling optimization model, and generate a target output scheduling strategy when the operating cost is minimized; The output control module is used to control the output of the energy conversion system, the combined cooling, heating and power system and the power-to-gas system respectively according to the target output scheduling strategy; Among them, the generation of causal neural network includes: The sample output scheduling strategy corresponding to each sample moment is used as the input of the neural network, and the actual output scheduling strategy at the next moment corresponding to each sample moment is used as the output of the neural network. The neural network is iteratively trained until the neural network converges, and the trained neural network is used as the causal neural network; During each training, the sample output scheduling strategy corresponding to each sample moment is input into the neural network, so that the neural network outputs the prediction result of the output scheduling strategy corresponding to each sample moment; the prediction result of the output scheduling strategy corresponding to each sample moment is compared with the actual output scheduling strategy of the next moment, and the causal coefficient in the neural network is adjusted according to the comparison result; The causal coefficient is used to characterize the causal relationship between the sample output scheduling strategy corresponding to the sample moment and the actual output scheduling strategy corresponding to the next moment.

9. A terminal device, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements an energy scheduling method based on multi-objective model solution as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium includes a stored computer program, wherein when the computer program is running, the device where the storage medium is located is controlled to execute an energy scheduling method based on multi-objective model solving as described in any one of claims 1 to 7.

Images