Scheduling method for coupled operation of trading and spot trading in multi-energy systems

Abstract

The present invention discloses a scheduling method for coupled operation of transactions and spot transactions of a multi-energy system, which solves the problem that the existing method cannot predict the trend of electricity prices in the future, and further cannot plan the power generation plan of the electricity spot market in advance when the uncertainty and intermittent fluctuations of renewable energy appear. The method comprises: obtaining the day-ahead spot transaction related data, loading the day-ahead spot transaction related data in real time, the electricity price prediction model outputs the electricity price prediction result and the transaction risk value corresponding to the electricity price prediction result, and the comprehensive optimization scheduling model calculates the predicted output value of the thermal power unit and the predicted output value of the new energy based on the electricity price prediction result and the transaction risk value; the present invention can effectively cope with the uncertainty and intermittent fluctuation of renewable energy by constructing the electricity price prediction model and the comprehensive optimization scheduling model for coordinated cooperation, and ensure that the multi-energy system can operate stably under different energy supply conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy dispatching, and in particular relates to a dispatching method for coupled operation of transactions and spot transactions of a multi-energy system. Background Art

[0002] With the continuous changes in global energy demand and the increasing popularity of renewable energy, the traditional single energy market can no longer meet the needs of modern society. As a result, the coordinated market transaction coupling operation of multiple energy systems has emerged. It can better integrate different forms of energy, such as electricity, natural gas, thermal energy, etc., optimize resource allocation through market mechanisms, improve energy utilization efficiency, and promote the low-carbon transformation of energy consumption.

[0003] Chinese patent CN115481888A discloses a scheduling method for the coupled operation of medium- and long-term transactions and spot transactions of a multi-energy system, including: establishing a medium- and long-term contract electricity decomposition mechanism, determining the annual monthly electricity ratio and the monthly daily ratio according to the historical load data of the unified dispatch, and decomposing the annual electricity into monthly and daily electricity; determining the output ratio of each time period according to the average output of wind power and photovoltaic power in different time periods, and obtaining the time-divided power generation of contract electricity allocated to wind power and photovoltaic power, that is, the daily electricity of new energy electricity; decomposing the joint problem of hydropower and thermal power into two sub-problems of hydropower and thermal power through dynamic planning, using the load suppression method to initially allocate the hydropower and thermal power loads, and then solving the start-up and shutdown sequence and unit output of thermal power and hydropower units according to the allocated rated values, but the existing method cannot predict the trend of electricity prices in the future, and thus cannot plan the power generation plan of the electricity spot market in advance when there is uncertainty and intermittent fluctuations in renewable energy. In response to the above problems, we propose a scheduling method for the coupled operation of transactions and spot transactions of a multi-energy system. Summary of the invention

[0004] The purpose of the present invention is to address the shortcomings of the prior art and provide a scheduling method for coupling transactions and spot transactions of multiple energy systems, thereby solving the problem that the prior art methods are unable to predict the trend of electricity prices in the future, and thus are unable to plan power generation plans in advance in the electricity spot market when there are uncertainties and intermittent fluctuations in renewable energy.

[0005] Existing methods cannot predict the trend of electricity prices in the future, and therefore cannot plan the power generation plan of the electricity spot market in advance when there is uncertainty and intermittent fluctuations in renewable energy. To address the above problems, we propose a scheduling method for coupling transactions and spot transactions of multi-energy systems. In short, when the method is implemented, first obtain the day-ahead spot transaction-related data, and preprocess the day-ahead spot transaction-related data, then obtain the historical spot transaction-related data, divide the historical spot transaction-related data into a training set and a test set, train the electricity price prediction model based on the training set, and then execute the electricity price prediction model. The electricity price prediction model outputs the electricity price prediction results and the transaction risk value corresponding to the electricity price prediction results. Finally, the comprehensive optimization scheduling model calculates the predicted output value of the thermal power unit and the predicted output value of the new energy based on the electricity price prediction results and the transaction risk value, and generates power generation decision instructions based on the predicted output value of the thermal power unit and the predicted output value of the new energy. In the embodiment of the present invention, by constructing an electricity price prediction model and a comprehensive optimization scheduling model to work together, it is possible to effectively deal with the uncertainty and intermittent fluctuations of renewable energy, ensure that the multi-energy system can operate stably under different energy supply conditions, and by reasonably allocating energy storage facilities and dispatchable loads, it is possible to minimize operating costs while ensuring the stability of the power grid. This overcomes the problem that the existing methods cannot predict the trend of electricity prices in the future, and thus cannot plan the power generation plan of the electricity spot market in advance when there are uncertainties and intermittent fluctuations in renewable energy.

[0006] The present invention is implemented as follows: a scheduling method for coupled operation of transactions and spot transactions of a multi-energy system, wherein the scheduling method for coupled operation of transactions and spot transactions of a multi-energy system comprises:

[0007] Acquire the day-ahead spot transaction-related data and pre-process the day-ahead spot transaction-related data, wherein the day-ahead spot transaction-related data includes historical electricity price data, historical cost data, supply and demand relationship information, fuel price information, day-ahead weather data, thermal power generation data, and new energy power generation data;

[0008] Obtain historical spot transaction-related data, divide the historical spot transaction-related data into a training set and a test set, train the electricity price prediction model based on the training set, and use the test set to evaluate the performance of the electricity price prediction model;

[0009] The day-ahead spot transaction related data is loaded in real time, the day-ahead spot transaction related data is normalized after preprocessing, and the electricity price prediction model is executed with the day-ahead spot transaction related data as input. The electricity price prediction model outputs the electricity price prediction result and the transaction risk value corresponding to the electricity price prediction result.

[0010] A comprehensive optimization scheduling model is constructed by combining the multi-objective moth-to-flame algorithm to obtain electricity price forecast results and transaction risk values. The comprehensive optimization scheduling model calculates the predicted output values ​​of thermal power units and new energy sources based on the electricity price forecast results and transaction risk values, and generates power generation decision instructions based on the predicted output values ​​of thermal power units and new energy sources.

[0011] Preferably, the method for training an electricity price prediction model based on a training set specifically comprises:

[0012] Construct the initial model of the electricity price prediction model. The initial model is a back-propagation neural network model. The discriminator and decoder are introduced into the initial model. The discriminator is based on the GAN structure, and the local sensitive attention mechanism is incorporated into the discriminator.

[0013] Among them, the discriminator quickly searches and discriminates neighboring samples based on the hash function, and the discriminant search calculation formula is:

[0014] (1)

[0015] in, for The input value of the training set at time instant, For prediction Time discriminator pair The predicted value of is the discriminant weight coefficient, is the acceleration discriminant factor, Random values ​​generated for the discriminator;

[0016] (2)

[0017] in, for The initial weight coefficient of

[0018] The hidden layer of the back-propagation neural network model is frozen and replaced by a decoder. The decoder simulates the prediction task based on the scheduled sampling algorithm combined with a two-pass decoding strategy, and processes the weighted sum of the input data by setting the decoder activation function to determine whether to use the predicted value or the fused value.

[0019] Preferably, the method for training an electricity price prediction model based on a training set further includes:

[0020] Get the decoder decoding result, adjust the weight and bias of the output layer according to the decoding result to minimize the loss function, calculate the error of the output layer, and then back-propagate the error to the decoder layer by layer, and gradually update the weight and bias to reduce the total error;

[0021] Among them, the loss function is the Center-Loss loss function, and the loss function expression is as follows:

[0022] (3)

[0023] in, is the true value of the sample, is the predicted value, is the number of samples, is the cosine similarity between the true value and the predicted value of the sample, is the softmax temperature;

[0024] Determine whether the total error is less than the preset error threshold. If it is less than the preset error threshold, stop model training;

[0025] A test set is obtained. The test set evaluates the performance of the electricity price prediction model through the evaluation indicators of accuracy and recall. If the accuracy and recall of the electricity price prediction model meet the preset accuracy, the trained electricity price prediction model is output.

[0026] Preferably, the execution of the electricity price prediction model, the electricity price prediction model outputting the electricity price prediction result and the transaction risk value corresponding to the electricity price prediction result, specifically includes:

[0027] Load the day-ahead spot transaction-related data, and normalize the pre-processed day-ahead spot transaction-related data;

[0028] The normalized day-ahead spot transaction correlation data is obtained, and the discriminator quickly searches and discriminates the sensitive attention points of the neighboring day-ahead spot transaction correlation data based on the hash function to generate an attention point set;

[0029] Based on the Scheduled Sampling algorithm, the attention point set is decomposed in parallel. The Scheduled Sampling algorithm unifies the price coupling constraint into the decoding function and constructs a set of prediction driving factors based on the price coupling constraint.

[0030] When the attention point set is decomposed in parallel, the decomposition function is expressed as:

[0031] (4)

[0032] in, For the The decomposition value of the group attention point, For the The input value of the group attention point, is the decomposition mixing coefficient, for Total power generated at any time, for The cost of generating electricity at any given moment.

[0033] Preferably, the execution of the electricity price prediction model, wherein the electricity price prediction model outputs an electricity price prediction result and a transaction risk value corresponding to the electricity price prediction result, specifically also includes:

[0034] Obtain a set of prediction driving factors, and the objective function of the output layer performs weighted summation on the prediction driving factors in the prediction driving factor set to output the electricity price prediction result;

[0035] (5)

[0036] in, for The electricity price forecast result at the moment, for The electricity price at the time, for The power generation at the time, To predict the number of driving factors, is the number of generating units, for Always make contributions, for The operating cost at the time;

[0037] Load the electricity price forecast results and calculate the transaction risk value corresponding to the electricity price forecast results based on the CVaR algorithm.

[0038] Preferably, the transaction risk value calculation formula corresponding to the electricity price prediction result calculated based on the CVaR algorithm is:

[0039] (6)

[0040] in, is the transaction risk value, is the unit load capacity, is the confidence level of the generator set, It is the proportion of the active power output of the generator set.

[0041] Preferably, the method of constructing a comprehensive optimization scheduling model in combination with a multi-objective moth-to-flame algorithm specifically includes:

[0042] Pre-constructing a comprehensive optimization scheduling model, wherein the comprehensive optimization scheduling model consists of an upper-level optimization module and a lower-level scheduling module;

[0043] The upper optimization module targets the electricity price forecast result and the transaction risk value to minimize the energy consumption cost of power generation. The upper optimization module objective function is:

[0044] (7)

[0045] (8)

[0046] in, To minimize the cost of power generation, For power generation mode The total energy consumption under For power generation mode The probability of the scenario, is the type of power generation mode, is the current power generation time, is the power generation cycle, To predict the output value, For power generation mode The unit price of energy, It is the energy consumption gain value that combines the electricity price forecast results and the transaction risk value.

[0047] The intraday rolling optimization model is used as the basic architecture of the lower-level scheduling module, and the multi-objective moth-to-fire algorithm is introduced to improve the basic architecture of the lower-level scheduling module;

[0048] Conduct ablation experiments on the upper-level optimization module and the lower-level scheduling module, and ablate and simplify the redundant modules of the model;

[0049] Use the preset model hyperparameters to iteratively train the comprehensive optimization scheduling model until convergence and save the model parameters.

[0050] Preferably, the method of introducing a multi-objective moth-to-fire algorithm to improve the infrastructure of the lower-level scheduling module specifically includes:

[0051] Based on the multi-objective moth-to-flame algorithm, at least one group of known individuals and one group of new individuals are randomly generated, and the group entropy value of the new individuals is calculated;

[0052] Determine whether the population entropy value is greater than the preset threshold. If it is greater than the preset threshold, add the new individual to the initial population to achieve population initialization;

[0053] The balanced particle swarm algorithm is used to perform chaotic perturbation on the optimal flame position, and the moth position is updated again based on the multi-objective algorithm to improve the infrastructure of the lower-level scheduling module.

[0054] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0055] In the embodiment of the present invention, by constructing an electricity price prediction model and a comprehensive optimization scheduling model to work together, it is possible to effectively deal with the uncertainty and intermittent fluctuations of renewable energy, ensure that the multi-energy system can operate stably under different energy supply conditions, and by reasonably allocating energy storage facilities and dispatchable loads, it is possible to minimize operating costs while ensuring the stability of the power grid. This overcomes the problem that the existing methods cannot predict the trend of electricity prices in the future, and thus cannot plan the power generation plan of the electricity spot market in advance when there are uncertainties and intermittent fluctuations in renewable energy.

[0056] In an embodiment of the present invention, an electricity price prediction model is provided. The electricity price prediction model is based on a back-propagation neural network model, so that the electricity price prediction model can realize effective information processing and error back-propagation. At the same time, a discriminator and a decoder are introduced into the model. The decoder simulates the prediction task based on the Scheduled Sampling algorithm combined with a two-pass decoding strategy, and simplifies the processing of high-dimensional data into the processing of data in a bucket, thereby reducing the amount of calculation, significantly reducing the deviation of the model during testing, and improving robustness.

[0057] In the embodiment of the present invention, the electricity price prediction model can output the electricity price prediction results and the transaction risk value corresponding to the electricity price prediction results by identifying and analyzing the day-ahead spot transaction related data, so as to facilitate the comprehensive optimization scheduling model to further solve the problem of uncertainty and intermittent fluctuations in evaluating new energy power generation units, so that the comprehensive optimization scheduling model can give full play to the regulation ability of new energy power generation units, complement new energy, and improve the stability and reliability of the overall power supply.

[0058] In an embodiment of the present invention, the comprehensive optimization scheduling model is composed of an upper-level optimization module and a lower-level scheduling module. Taking into account the influence of the electricity market electricity price forecast results and the transaction risk value, a comprehensive optimization scheduling model is constructed and trained in the electricity spot trading market environment, so as to compare the coupling operation effects under different proportions of new energy power generation scenarios, so that decision makers can generate power generation decision adjustment plans based on the predicted output values ​​of thermal power units and the predicted output values ​​of new energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the implementation flow of the scheduling method for coupled operation of multi-energy system transactions and spot transactions provided by the present invention.

[0060] Figure 2 It is a schematic diagram of the implementation flow of the method for training an electricity price prediction model based on a training set provided by the present invention.

[0061] Figure 3 It is a schematic diagram of the implementation process of the electricity price prediction model provided by the present invention outputting electricity price prediction results and transaction risk values ​​corresponding to the electricity price prediction results.

[0062] Figure 4 It is a schematic diagram of the implementation flow of the method for constructing a comprehensive optimization scheduling model in combination with a multi-objective moth-to-flame algorithm provided by the present invention.

[0063] Figure 5 It is a schematic diagram of the implementation flow of the method for improving the infrastructure of the lower-level scheduling module by introducing a multi-objective moth-to-fire algorithm provided by the present invention. DETAILED DESCRIPTION

[0064] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of this application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0065] Existing methods cannot predict the trend of electricity prices in the future, and therefore cannot plan the power generation plan of the electricity spot market in advance when there is uncertainty and intermittent fluctuations in renewable energy. To address the above problems, we propose a scheduling method for coupling transactions and spot transactions of multi-energy systems. In short, when the method is implemented, first obtain the day-ahead spot transaction-related data, and preprocess the day-ahead spot transaction-related data, then obtain the historical spot transaction-related data, divide the historical spot transaction-related data into a training set and a test set, train the electricity price prediction model based on the training set, and then execute the electricity price prediction model. The electricity price prediction model outputs the electricity price prediction results and the transaction risk value corresponding to the electricity price prediction results. Finally, the comprehensive optimization scheduling model calculates the predicted output value of the thermal power unit and the predicted output value of the new energy based on the electricity price prediction results and the transaction risk value, and generates power generation decision instructions based on the predicted output value of the thermal power unit and the predicted output value of the new energy. In the embodiment of the present invention, by constructing an electricity price prediction model and a comprehensive optimization scheduling model to work together, it is possible to effectively deal with the uncertainty and intermittent fluctuations of renewable energy, ensure that the multi-energy system can operate stably under different energy supply conditions, and by reasonably allocating energy storage facilities and dispatchable loads, it is possible to minimize operating costs while ensuring the stability of the power grid. This overcomes the problem that the existing methods cannot predict the trend of electricity prices in the future, and thus cannot plan the power generation plan of the electricity spot market in advance when there are uncertainties and intermittent fluctuations in renewable energy.

[0066] The embodiment of the present invention provides a scheduling method for coupled operation of transactions and spot transactions in a multi-energy system. Figure 1The schematic diagram of the implementation process of the scheduling method for coupled operation of transactions and spot transactions of the multi-energy system is shown. The scheduling method for coupled operation of transactions and spot transactions of the multi-energy system specifically includes:

[0067] Step S10, obtaining the day-ahead spot transaction related data, and preprocessing the day-ahead spot transaction related data.

[0068] It should be noted that the day-ahead spot transaction-related data include but are not limited to historical electricity price data, historical cost data, supply and demand relationship information, fuel price information, day-ahead weather data, thermal power generation data, and new energy power generation data;

[0069] In this embodiment, the preprocessing method for the day-ahead spot transaction correlation data may be wavelet transformation, normalization and then normalization.

[0070] Among them, Box-Cox transformation can be used for normalization of the day-ahead spot transaction correlation data, and the data acquisition period for the day-ahead spot transaction correlation data is 7 days, 15 days or 30 days.

[0071] Step S20, obtaining historical spot transaction related data, dividing the historical spot transaction related data into a training set and a test set, training an electricity price prediction model based on the training set, and using the test set to evaluate the performance of the electricity price prediction model;

[0072] In this embodiment, the historical spot transaction related data can be obtained from the power trading network and the trading database, and the allocation ratio of the training set and the test set can be 80% and 20%.

[0073] Step S30, loading the day-ahead spot transaction related data in real time, normalizing the pre-processed day-ahead spot transaction related data, executing the electricity price prediction model with the day-ahead spot transaction related data as input, and the electricity price prediction model outputs the electricity price prediction result and the transaction risk value corresponding to the electricity price prediction result.

[0074] Step S40, combining the multi-objective moth-to-flame algorithm to build a comprehensive optimization scheduling model, obtain electricity price forecast results and transaction risk values, the comprehensive optimization scheduling model calculates the predicted output value of the thermal power unit and the predicted output value of the new energy based on the electricity price forecast results and the transaction risk value, and generates a power generation decision instruction based on the predicted output value of the thermal power unit and the predicted output value of the new energy.

[0075] In the embodiment of the present invention, by constructing an electricity price prediction model and a comprehensive optimization scheduling model to work together, it is possible to effectively deal with the uncertainty and intermittent fluctuations of renewable energy, ensure that the multi-energy system can operate stably under different energy supply conditions, and by reasonably allocating energy storage facilities and dispatchable loads, it is possible to minimize operating costs while ensuring the stability of the power grid. This overcomes the problem that the existing methods cannot predict the trend of electricity prices in the future, and thus cannot plan the power generation plan of the electricity spot market in advance when there are uncertainties and intermittent fluctuations in renewable energy.

[0076] The embodiment of the present invention provides a method for training an electricity price prediction model based on a training set. Figure 2 The schematic diagram of the implementation process of the method for training an electricity price prediction model based on a training set is shown. The method for training an electricity price prediction model based on a training set specifically includes:

[0077] Step S101, constructing an initial model of the electricity price prediction model, the initial model is a back propagation neural network model, a discriminator and a decoder are introduced into the initial model, the discriminator is based on a GAN structure, and a local sensitive attention mechanism is incorporated into the discriminator;

[0078] It should be noted that the GAN structure also includes a generator, which is a deep neural network structure. The generator can receive random noise as input and convert it into predicted data similar to the real data distribution. During the training process, the discriminator first fixes the parameters of the generator, then updates the parameters of the discriminator, and then uses back propagation to optimize the discriminator and generator.

[0079] Among them, the discriminator quickly searches and discriminates neighboring samples based on the hash function, and the discriminant search calculation formula is:

[0080] (1)

[0081] in, for The input value of the training set at time instant, For prediction Time discriminator pair The predicted value of is the discriminant weight coefficient, is the acceleration discrimination factor. In this embodiment, the acceleration discrimination factor may be 0.1-0.3. The random value generated by the discriminator is in the range of (0-1);

[0082] (2)

[0083] in, for The initial weight coefficient can be set by initializing with Gaussian distribution.

[0084] Step S102, freeze the hidden layer of the back propagation neural network model, and replace the hidden layer with a decoder. The decoder simulates the prediction task based on the scheduled sampling algorithm combined with the two-pass decoding strategy, and processes the weighted sum of the input data by setting the decoder activation function, thereby determining whether to use the predicted value or the fused value.

[0085] Step S103, obtaining the decoding result of the decoder, adjusting the weight and bias of the output layer according to the decoding result to minimize the loss function, calculating the error of the output layer, and then back-propagating the error to the decoder layer by layer, and gradually updating the weight and bias to reduce the total error. In this embodiment, the error of the output layer can be 0.05-0.08;

[0086] Among them, the loss function is the Center-Loss loss function, and the loss function expression is as follows:

[0087] (3)

[0088] in, is the true value of the sample, is the predicted value, is the number of samples, is the cosine similarity between the true value and the predicted value of the sample, is the softmax temperature;

[0089] Step S104, determining whether the total error is less than a preset error threshold;

[0090] Step S105, if the error is less than the preset error threshold, stop model training;

[0091] If it is greater than the preset error threshold, return to step S102.

[0092] Step S106, obtaining a test set, the test set evaluates the performance of the electricity price prediction model by evaluating the accuracy and recall of the evaluation indicators;

[0093] Step S107: if the accuracy and recall rate of the electricity price prediction model meet the preset accuracy rate, the trained electricity price prediction model is output.

[0094] If the accuracy and recall rate of the electricity price prediction model do not meet the preset accuracy rates, return to step S102.

[0095] In an embodiment of the present invention, an electricity price prediction model is provided. The electricity price prediction model is based on a back-propagation neural network model, so that the electricity price prediction model can realize effective information processing and error back-propagation. At the same time, a discriminator and a decoder are introduced into the model. The decoder simulates the prediction task based on the Scheduled Sampling algorithm combined with a two-pass decoding strategy, and simplifies the processing of high-dimensional data into the processing of data in a bucket, thereby reducing the amount of calculation, significantly reducing the deviation of the model during testing, and improving robustness.

[0096] The embodiment of the present invention provides an execution electricity price prediction model, the electricity price prediction model outputs an electricity price prediction result and a transaction risk value corresponding to the electricity price prediction result, Figure 3 The schematic diagram of the implementation process of executing the electricity price prediction model, the electricity price prediction model outputting the electricity price prediction result and the transaction risk value corresponding to the electricity price prediction result is shown, and the execution of the electricity price prediction model, the electricity price prediction model outputting the electricity price prediction result and the transaction risk value corresponding to the electricity price prediction result specifically includes:

[0097] Step S201, loading the day-ahead spot transaction related data, and normalizing the pre-processed day-ahead spot transaction related data;

[0098] It should be noted that the normalization processing of the pre-processed day-ahead spot transaction correlation data can be performed by min-max normalization or Z-score normalization.

[0099] Step S202, obtaining the normalized day-ahead spot transaction related data, and the discriminator quickly searches and discriminates the sensitive attention points of the neighbor day-ahead spot transaction related data based on the hash function to generate an attention point set;

[0100] It should be noted that the hash function is a technique for quickly finding approximate nearest neighbors for massive high-dimensional data. The basic idea is that after adjacent data points in the original data space are transformed by mapping or projection, the probability that they are still adjacent in the new data space is high, while the probability that non-adjacent data points are mapped to the same bucket is low. In this embodiment, the hash function can be used to quickly find and identify sensitive attention points, thereby improving the computing power and prediction accuracy of the model.

[0101] Step S203, based on the Scheduled Sampling algorithm, the attention point set is decomposed in parallel, the Scheduled Sampling algorithm unifies the price coupling constraint into the decoding function, and constructs a prediction driving factor set based on the price coupling constraint;

[0102] It should be noted that the predicted driving factors include but are not limited to risk factors, cost factors, benefit factors, weight factors, economic activity factors, climate factors, fuel costs and other driving factors, and the decoding function can be an inverse sigmoid decay function. The Scheduled Sampling algorithm judges and extracts the predicted driving factors based on the increase and decrease rate of the inverse sigmoid decay function.

[0103] When the attention point set is decomposed in parallel, the decomposition function is expressed as:

[0104] (4)

[0105] in, For the The decomposition value of the group attention point, For the The input value of the group attention point, is the decomposition mixing coefficient, for Total power generated at any time, for The cost of generating electricity at any given moment.

[0106] Step S204, obtaining a set of prediction driving factors, performing a weighted summation of the prediction driving factors in the set of prediction driving factors by the objective function of the output layer, and outputting a prediction result of the electricity price;

[0107] (5)

[0108] in, for The electricity price forecast result at the moment, for The electricity price at the time, for The power generation at the time, is the number of predicted driving factors. In this embodiment, the number of predicted driving factors is 3-15. is the number of generating units, for Always make contributions, for The operating cost at the moment;

[0109] Step S205, loading the electricity price forecast result, and calculating the transaction risk value corresponding to the electricity price forecast result based on the CVaR algorithm.

[0110] In this embodiment, the transaction risk value calculation formula corresponding to the electricity price prediction result calculated based on the CVaR algorithm is:

[0111] (6)

[0112] in, is the transaction risk value, is the unit load capacity, is the confidence of the generator set. It should be noted that the confidence of the generator set can be calculated using the sequential Monte Carlo method and the chord intercept method. The confidence of the generator set can be 0.1-1. It is the proportion of the active power output of the generator set.

[0113] In the embodiment of the present invention, the electricity price prediction model can output the electricity price prediction results and the transaction risk value corresponding to the electricity price prediction results by identifying and analyzing the day-ahead spot transaction related data, so as to facilitate the comprehensive optimization scheduling model to further solve the problem of uncertainty and intermittent fluctuations in evaluating new energy power generation units, so that the comprehensive optimization scheduling model can give full play to the regulation ability of new energy power generation units, complement new energy, and improve the stability and reliability of the overall power supply.

[0114] The embodiment of the present invention provides a method for constructing a comprehensive optimization scheduling model in combination with a multi-objective moth-to-fire algorithm. Figure 4 The schematic diagram of the implementation process of the method for constructing a comprehensive optimization scheduling model by combining the multi-objective moth-to-fire algorithm is shown. The method for constructing a comprehensive optimization scheduling model by combining the multi-objective moth-to-fire algorithm specifically includes:

[0115] Step S301, pre-constructing a comprehensive optimization scheduling model, wherein the comprehensive optimization scheduling model is composed of an upper-layer optimization module and a lower-layer scheduling module;

[0116] It should be noted that the upper-level optimization module is used to optimize the energy consumption cost of power generation. The upper-level optimization module is connected to the lower-level scheduling module. The upper-level optimization module is based on the ytorch deep learning framework, and a self-attention mechanism is introduced into the upper-level optimization module. The activation function of the upper-level optimization module is the Sigmoid function, and the self-attention mechanism can be the S2-MLPv2 attention mechanism.

[0117] Step S302, the upper optimization module targets the electricity price prediction result and the transaction risk value to minimize the energy consumption cost of participating in power generation. The upper optimization module objective function is:

[0118] (7)

[0119] (8)

[0120] in, To minimize the cost of power generation, For power generation mode The total energy consumption under For power generation mode The probability of the scenario, is the type of power generation mode, is the current power generation time, is the power generation cycle, To predict the output value, For power generation mode The unit price of energy, In order to combine the electricity price prediction result and the energy consumption gain value of the transaction risk value, in this embodiment, the predicted output value includes the predicted output value of the thermal power unit and the predicted output value of the new energy, and the type of power generation mode can be 2-5.

[0121] Step S303, using the intraday rolling optimization model as the infrastructure of the lower-level scheduling module, and introducing a multi-objective moth-to-fire algorithm to improve the infrastructure of the lower-level scheduling module;

[0122] Step S304, performing ablation experiments on the upper layer optimization module and the lower layer scheduling module, and ablating and simplifying the model redundant modules;

[0123] Step S305, iteratively train the comprehensive optimization scheduling model using preset model hyperparameters until convergence, and save the model parameters.

[0124] In an embodiment of the present invention, the comprehensive optimization scheduling model is composed of an upper-level optimization module and a lower-level scheduling module. Taking into account the influence of the electricity market electricity price forecast results and the transaction risk value, a comprehensive optimization scheduling model is constructed and trained in the electricity spot trading market environment, so as to compare the coupling operation effects under different proportions of new energy power generation scenarios, so that decision makers can generate power generation decision adjustment plans based on the predicted output values ​​of thermal power units and the predicted output values ​​of new energy.

[0125] The embodiment of the present invention provides a method for improving the infrastructure of the lower-level scheduling module by introducing a multi-objective moth-to-fire algorithm. Figure 5 The schematic diagram of the implementation process of the method for improving the infrastructure of the lower scheduling module by introducing the multi-objective moth-to-fire algorithm is shown. The method for improving the infrastructure of the lower scheduling module by introducing the multi-objective moth-to-fire algorithm specifically includes:

[0126] Step S3031, randomly generating at least one group of known individuals and one group of new individuals based on the multi-target moth-to-flame algorithm, and calculating the group entropy value of the new individuals;

[0127] Step S3032, determining whether the population entropy value is greater than a preset threshold. In this embodiment, the preset threshold can be set to 0.096-0.098;

[0128] Step S3033, if it is greater than the preset threshold, the new individual is added to the initial population to initialize the population; if it is less than the preset threshold, the current new individual is discarded.

[0129] Step S3034, using the balanced particle swarm algorithm to perform chaotic perturbation on the optimal flame position, and updating the moth position again based on the multi-objective algorithm to improve the infrastructure of the lower-level scheduling module.

[0130] In this embodiment, the generation distance, dispersion, and hypervolume ratio can be used to evaluate the performance of the lower-level scheduling module. The generation distance is used to evaluate the closeness of the crowding entropy of the multi-objective algorithm to the true frontier. The smaller the generation distance, the closer the surface crowding entropy is to the true frontier, and the smaller the dispersion, the better the crowding entropy diversity. The hypervolume is used to evaluate the volume covered by the crowding entropy in the target space.

[0131] In the embodiment of the present invention, the infrastructure of the lower-level scheduling module is improved through a multi-objective moth-to-flame algorithm, so as to improve the global search capability of the comprehensive optimization scheduling model, optimize and adjust the search range, increase the convergence speed, enhance the robustness of the algorithm, and improve the multi-objective optimization performance, thereby providing decision makers with a variety of options and helping to make more reasonable decisions in practical problems.

[0132] On the other hand, an embodiment of the present invention further provides a computer device, which includes a display screen, a memory, a processor and a computer program, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the scheduling method for coupling transactions and spot transactions of the multi-energy system.

[0133] It should be noted that the computer device may also be a notebook computer, a personal digital assistant (PDA), a mobile phone or other device capable of communication.

[0134] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer program instructions, and the computer program instructions can be executed by a processor. When the computer program instructions are executed, the method of any one of the above embodiments is implemented.

[0135] 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 user terminal.

[0136] The computer-readable storage medium (e.g., memory) herein may be a volatile memory or a nonvolatile memory, or may include both a volatile memory and a nonvolatile memory. As an example and not by way of limitation, a nonvolatile memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. A volatile memory may include a random access memory (RAM), which may act as an external cache memory. As an example and not by way of limitation, RAM may be obtained in a variety of forms, such as synchronous RAM (DRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The storage devices of the disclosed aspects are intended to include, but are not limited to, these and other suitable types of memory.

[0137] In summary, the present invention provides a scheduling method for coupled operation of transactions and spot transactions of a multi-energy system. In the embodiment of the present invention, by constructing an electricity price prediction model and a comprehensive optimization scheduling model to coordinate, it is possible to effectively cope with the uncertainty and intermittent fluctuations of renewable energy, and ensure that the multi-energy system can operate stably under different energy supply conditions. By rationally deploying energy storage facilities and dispatchable loads, it is possible to minimize operating costs while ensuring the stability of the power grid. This overcomes the problem that the existing methods cannot predict the trend of electricity prices in the future, and thus cannot plan the power generation plan of the electricity spot market in advance when there are uncertainties and intermittent fluctuations in renewable energy.

[0138] It should be noted that, for the above-mentioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0139] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also belong to the scope of protection of the present invention.

Claims

1. A scheduling method for coupling trading and spot trading of multi-energy systems, characterized in that: The scheduling method for coupling the transaction and spot transaction of the multi-energy system includes: Acquire the day-ahead spot transaction-related data and pre-process the day-ahead spot transaction-related data, wherein the day-ahead spot transaction-related data includes historical electricity price data, historical cost data, supply and demand relationship information, fuel price information, day-ahead weather data, thermal power generation data, and new energy power generation data; Obtain historical spot transaction-related data, divide the historical spot transaction-related data into a training set and a test set, train the electricity price prediction model based on the training set, and use the test set to evaluate the performance of the electricity price prediction model; Load the day-ahead spot transaction-related data in real time, normalize the pre-processed day-ahead spot transaction-related data, use the day-ahead spot transaction-related data as input, execute the electricity price prediction model, and the electricity price prediction model outputs the electricity price prediction result and the transaction risk value corresponding to the electricity price prediction result; A comprehensive optimization dispatching model is constructed by combining the multi-objective moth-to-flame algorithm to obtain the electricity price forecast results and transaction risk values. The comprehensive optimization dispatching model calculates the predicted output values ​​of thermal power units and the predicted output values ​​of new energy based on the electricity price forecast results and transaction risk values, and generates power generation decision instructions based on the predicted output values ​​of thermal power units and the predicted output values ​​of new energy. The method of constructing a comprehensive optimization scheduling model by combining the multi-objective moth-to-flame algorithm specifically includes: Pre-constructing a comprehensive optimization scheduling model, wherein the comprehensive optimization scheduling model consists of an upper-level optimization module and a lower-level scheduling module; The upper optimization module targets the electricity price forecast result and the transaction risk value to minimize the energy consumption cost of power generation. The upper optimization module objective function is: （7） （8） in, To minimize the cost of power generation, For power generation mode The total energy consumption under For power generation mode The probability of the scenario, is the type of power generation mode, is the current power generation time, is the power generation cycle, To predict the output value, For power generation mode The unit price of energy, In order to combine the electricity price forecast results and the energy consumption gain value of the transaction risk value, for The electricity price forecast result at the moment, is the transaction risk value; The intraday rolling optimization model is used as the basic architecture of the lower-level scheduling module, and the multi-objective moth-to-fire algorithm is introduced to improve the basic architecture of the lower-level scheduling module; Conduct ablation experiments on the upper-level optimization module and the lower-level scheduling module, and ablate and simplify the redundant modules of the model; Use the preset model hyperparameters to iteratively train the comprehensive optimization scheduling model until convergence and save the model parameters; The method of introducing the multi-objective moth-to-fire algorithm to improve the infrastructure of the lower-level scheduling module includes: Based on the multi-objective moth-to-flame algorithm, at least one group of known individuals and one group of new individuals are randomly generated, and the group entropy value of the new individuals is calculated; Determine whether the population entropy value is greater than the preset threshold. If it is greater than the preset threshold, add the new individual to the initial population to achieve population initialization; The balanced particle swarm algorithm is used to perform chaotic perturbation on the optimal flame position, and the moth position is updated again based on the multi-objective algorithm to improve the infrastructure of the lower-level scheduling module.

2. The scheduling method for coupled operation of trading and spot trading of a multi-energy system as claimed in claim 1, characterized in that: The method for training an electricity price prediction model based on a training set specifically includes: Construct the initial model of the electricity price prediction model. The initial model is a back-propagation neural network model. The discriminator and decoder are introduced into the initial model. The discriminator is based on the GAN structure, and the local sensitive attention mechanism is incorporated into the discriminator. Among them, the discriminator quickly searches and discriminates neighboring samples based on the hash function, and the discriminant search calculation formula is: （1） in, for The input value of the training set at time instant, For prediction Time discriminator pair The predicted value of is the discriminant weight coefficient, is the acceleration discriminant factor, Random values ​​generated for the discriminator; （2） in, for The initial weight coefficient of The hidden layer of the back-propagation neural network model is frozen and replaced by a decoder. The decoder simulates the prediction task based on the ScheduledSampling algorithm combined with a two-pass decoding strategy, and processes the weighted sum of the input data by setting the decoder activation function to determine whether to use the predicted value or the fused value.

3. The scheduling method for coupled operation of trading and spot trading of a multi-energy system as claimed in claim 2 is characterized by: The method for training the electricity price prediction model based on the training set specifically includes: Get the decoder decoding result, adjust the weight and bias of the output layer according to the decoding result to minimize the loss function, calculate the error of the output layer, and then back-propagate the error to the decoder layer by layer, and gradually update the weight and bias to reduce the total error; Among them, the loss function is the Center-Loss loss function, and the loss function expression is as follows: （3） in, is the true value of the sample, is the predicted value, is the number of samples, is the cosine similarity between the true value and the predicted value of the sample, is the softmax temperature; Determine whether the total error is less than the preset error threshold. If it is less than the preset error threshold, stop model training; A test set is obtained. The test set evaluates the performance of the electricity price prediction model through the evaluation indicators of accuracy and recall. If the accuracy and recall of the electricity price prediction model meet the preset accuracy, the trained electricity price prediction model is output.

4. The scheduling method for coupled operation of trading and spot trading of a multi-energy system as claimed in claim 1, characterized in that: The executing electricity price prediction model, wherein the electricity price prediction model outputs an electricity price prediction result and a transaction risk value corresponding to the electricity price prediction result, specifically includes: Load the day-ahead spot transaction-related data, and normalize the pre-processed day-ahead spot transaction-related data; The normalized day-ahead spot transaction correlation data is obtained, and the discriminator quickly searches and discriminates the sensitive attention points of the neighboring day-ahead spot transaction correlation data based on the hash function to generate an attention point set; Based on the Scheduled Sampling algorithm, the attention point set is decomposed in parallel. The Scheduled Sampling algorithm unifies the price coupling constraint into the decoding function and constructs a set of prediction driving factors based on the price coupling constraint. When the attention point set is decomposed in parallel, the decomposition function is expressed as: （4） in, For the The decomposition value of the group attention point, For the The input value of the group attention point, is the decomposition mixing coefficient, for Total power generated at any time, for The cost of generating electricity at any given moment.

5. The scheduling method for coupled operation of trading and spot trading of a multi-energy system as claimed in claim 4, characterized in that: The execution of the electricity price prediction model, wherein the electricity price prediction model outputs the electricity price prediction result and the transaction risk value corresponding to the electricity price prediction result, specifically includes: Obtain a set of prediction driving factors, and the objective function of the output layer performs weighted summation on the prediction driving factors in the prediction driving factor set to output the electricity price prediction result; （5） in, for The electricity price forecast result at the moment, for The electricity price at the time, for The power generation at the time, To predict the number of driving factors, is the number of generating units, for Always make contributions, for The operating cost at the time; Load the electricity price forecast results and calculate the transaction risk value corresponding to the electricity price forecast results based on the CVaR algorithm.

6. The scheduling method for coupled operation of trading and spot trading of a multi-energy system as claimed in claim 5, characterized in that: The calculation formula of the transaction risk value corresponding to the electricity price forecast result calculated based on the CVaR algorithm is: （6） in, is the transaction risk value, is the unit load capacity, is the confidence level of the generator set, It is the proportion of the working power output of the generator set.

Images