Electric power spot market settlement dynamic prediction method based on graph neural network

Abstract

The invention relates to an electric power spot market settlement dynamic prediction method based on a graph neural network, and the method comprises the steps: determining an evaluation index system of new energy participating in spot market settlement according to the construction principle of the evaluation index system; determining the weight of each index in the evaluation index system through an analytic hierarchy process; sorting the indexes according to the weights, selecting a high-weight index based on a sorting result, substituting the selected high-weight index into a graph neural network prediction model, and dynamically predicting new energy participating in electric power spot market settlement; and comparing the prediction result of the graph neural network prediction model with the prediction results of a plurality of different prediction models, and determining the prediction accuracy of the graph neural network prediction model according to the comparison result. According to the technical scheme, the settlement condition can be predicted when the new energy participates in settlement of the spot market, so that early warning is realized, the risk that the new energy participates in the electric power spot market is reduced, and the income of a new energy main body participating in the electric power spot market is guaranteed.

Description

Technical Field

[0001] The present application relates to the field of model prediction technology, and in particular to a method for dynamic prediction of electricity spot market settlement based on graph neural network. Background Art

[0002] Under the deployment of the "dual carbon" development strategy, the proportion of new energy connected to the power grid in northern Hebei has further increased. The random fluctuation characteristics and unpredictability of new energy represented by wind power and photovoltaic power have brought risks to the spot trading of power in northern Hebei. In addition, the spot market of power in northern Hebei lacks risk prediction methods for new energy entities participating in market settlement. Therefore, when some high-risk new energy entities participate in market settlement, the transaction risks they bring, such as failure to perform on schedule due to under-production of units, will undermine the smooth operation of the spot market of power in northern Hebei.

[0003] It is difficult to make a reasonable assessment of the settlement situation of new energy when participating in the electricity spot market in the existing technology, which leads to a series of problems, such as discouraging users from participating in the market and affecting the long-term development of the electricity spot market. Summary of the invention

[0004] In order to at least to some extent overcome the problem in the related art that it is difficult to make a reasonable assessment of the settlement situation when new energy sources participate in the electricity spot market, the present application provides a dynamic prediction method for electricity spot market settlement based on graph neural network.

[0005] The scheme of this application is as follows:

[0006] A method for dynamic prediction of electricity spot market settlement based on graph neural network, comprising:

[0007] According to the construction principles of the evaluation index system, determine the evaluation index system for new energy participating in spot market settlement;

[0008] Determine the weight of each indicator in the evaluation indicator system through the hierarchical analysis method;

[0009] The indicators are sorted according to their weights, and based on the sorting results, high-weight indicators are selected and substituted into the graph neural network prediction model to dynamically predict the participation of new energy in the electricity spot market settlement;

[0010] The prediction results of the graph neural network prediction model are compared with the prediction results of multiple different prediction models, and the prediction accuracy of the graph neural network prediction model is determined based on the comparison results.

[0011] Preferably, the method further comprises:

[0012] The graph neural network prediction model uses the substituted indicators as nodes of the graph neural network; each node of the graph neural network is naturally defined by its features and associated nodes;

[0013] The local transfer function is defined as a recursive function shared among all nodes, and the state of the current node is updated based on the input information from the domain when updating the vector of each node;

[0014] Each node uses the surrounding neighboring nodes and adjacent edges as source information to update its own expression h, and introduces the local output function g to obtain the node output o:

[0015] h n =f(x n ,x co[n] ,x ne[n] ,h ne[n] );

[0016] o n =g(h n ,x n );

[0017] Among them, x represents the current node, h represents the hidden state of the node, ne[n] represents the set of neighboring nodes of node n, and co[n] represents the set of adjacent edges of node n;

[0018] Stack all local transfer functions f:

[0019] H=F(H,X);

[0020] O=G(H,X n );

[0021] Among them, H represents the fixed point, F represents the global transfer function, X represents the global node, O represents the output of the global node, and G represents the global output function;

[0022] According to Banach's fixed point theorem, assuming that the global transfer function F is a compression mapping, the value of the fixed point H is iteratively solved:

[0023] H t+1 =F(H t ,X).

[0024] Preferably, the method further comprises:

[0025] In the graph neural network prediction model, the Pearson correlation coefficient between indicators is used to construct the graph structure;

[0026] When the Pearson correlation coefficient of the two indicators is greater than 0.8, it is determined that there is a correlation between the two indicators, otherwise it is determined that the two indicators are not correlated;

[0027] The gated recurrent unit in the graph neural network prediction model is used to predict the substituted indicators.

[0028] Preferably, the construction principles of the evaluation index system include:

[0029] The principles of relevance, coordination, comprehensiveness, practicality and specificity.

[0030] Preferably, the first-level indicators in the evaluation index system for new energy participating in spot market settlement include:

[0031] market structure, market supply and demand, market development, market behavior, market performance and market risks;

[0032] The secondary indicators in the market structure category include:

[0033] Market concentration, remaining supply rate and Top-m index;

[0034] The secondary indicators of market supply and demand indicators include:

[0035] Total monthly transaction volume of new energy, total monthly demand for new energy, and actual monthly settlement volume of new energy;

[0036] The secondary indicators of market development indicators include:

[0037] Monthly reporting volume growth rate, monthly transaction volume growth rate and transaction activity;

[0038] The secondary indicators in the market behavior indicators include:

[0039] Monthly increase in new energy unit price quotes, new energy unit price ceilings, quote markup index and settlement efficiency;

[0040] The secondary indicators in the market performance category include:

[0041] New energy clearing prices, additional issuance cost reduction rates and price convergence;

[0042] The secondary indicators of market risk indicators include:

[0043] Price volatility, electricity deviation rate and new energy benefits.

[0044] Preferably, the weight of each indicator in the evaluation index system is determined by the hierarchical analysis method, including:

[0045] Determine the first-level indicators in the evaluation index system for new energy participating in spot market settlement as the criterion layer in the hierarchical evaluation model;

[0046] A judgment matrix is ​​constructed to compare the first-level indicators of the criterion layer in pairs and determine the weight of each first-level indicator of the criterion layer to the target layer.

[0047] Preferably, a judgment matrix is ​​constructed to compare the first-level indicators of the criterion layer in pairs, and the weight of each first-level indicator of the criterion layer to the target layer is determined, including:

[0048] Construct a judgment matrix, calculate the product and 1 / n power of row elements, and obtain the m-dimensional vector of each row of the judgment matrix;

[0049] Normalize the m-dimensional vector to get the weight vector.

[0050] Preferably, the method further comprises:

[0051] Calculate the maximum eigenvalue of the judgment matrix;

[0052] Calculate the CI value of the judgment matrix based on the maximum eigenvalue of the judgment matrix;

[0053] Select the RI value according to the order of the judgment matrix, and calculate the CR value of the judgment matrix according to the CI value and RI value of the judgment matrix;

[0054] If the CR value of the judgment matrix is ​​less than 0.1, it is determined that the consistency of the judgment matrix is ​​within the allowable range;

[0055] If the CR value of the judgment matrix is ​​not less than 0.1, the judgment matrix is ​​adjusted until the consistency of the judgment matrix is ​​within the allowable range.

[0056] Preferably, the indicators are sorted according to their weights, and high-weight indicators are selected based on the sorting results and substituted into the graph neural network prediction model, including:

[0057] Based on the sorting results, the secondary indicators corresponding to the high-weight primary indicators are selected and substituted into the graph neural network prediction model.

[0058] Preferably, the prediction model different from the graph neural network prediction model includes at least:

[0059] Long short-term memory neural networks, random forests, and extreme gradient boosting trees;

[0060] Determining the prediction accuracy of the graph neural network prediction model according to the comparison result includes:

[0061] The average absolute error, absolute percentage error and root mean square error of the prediction results of the computational graph neural network prediction model and the prediction results of multiple different prediction models are used as comparison results.

[0062] The technical solution provided by this application may have the following beneficial effects:

[0063] The method for dynamic prediction of electricity spot market settlement based on graph neural network in the present application preferably includes: determining the evaluation index system for new energy participating in spot market settlement according to the construction principle of the evaluation index system; determining the weight of each indicator in the evaluation index system through hierarchical analysis; sorting each indicator according to the weight, selecting high-weight indicators based on the sorting results and substituting them into the graph neural network prediction model to dynamically predict the participation of new energy in electricity spot market settlement; comparing the prediction results of the graph neural network prediction model with the prediction results of multiple different prediction models, and determining the prediction accuracy of the graph neural network prediction model according to the comparison results.

[0064] In this technical solution, we first construct an evaluation index system for new energy participating in spot market settlement, and then substitute the high-weighted indicators into the graph neural network prediction model to dynamically predict the participation of new energy in electricity spot market settlement. At the same time, we compare and evaluate the prediction results of the graph neural network prediction model through multiple different prediction models to obtain accurate prediction results. This technical solution can predict the settlement situation of new energy when it participates in spot market settlement, thereby achieving early warning, reducing the risk of new energy participating in the electricity spot market, and protecting the benefits of new energy entities participating in the electricity spot market.

[0065] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0067] Figure 1 It is a flow chart of a method for dynamic prediction of electricity spot market settlement based on graph neural network provided by an embodiment of the present application;

[0068] Figure 2 It is a schematic diagram of a prediction process of a graph neural network prediction model provided by an embodiment of the present application;

[0069] Figure 3 This is a schematic diagram of a consistency detection process of a judgment matrix provided by an embodiment of the present application;

[0070] Figure 4 It is a judgment matrix constructed based on the scale opinions of experts provided in an embodiment of the present application. DETAILED DESCRIPTION

[0071] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0072] Figure 1 This is a flow chart of a method for dynamic prediction of electricity spot market settlement based on graph neural network provided by an embodiment of the present application, with reference to Figure 1 , a dynamic prediction method for electricity spot market settlement based on graph neural network, comprising:

[0073] S11: According to the construction principles of the evaluation index system, determine the evaluation index system for new energy participating in spot market settlement;

[0074] It should be noted that the construction principles of the evaluation index system include:

[0075] The principles of relevance, coordination, comprehensiveness, practicality and specificity.

[0076] The selection of evaluation indicators must first clarify the fundamental purpose, that is, to evaluate the effectiveness and rationality of market operation and market rules and mechanisms, and to regulate market behavior; at the same time, attention should be paid to the comprehensiveness and practicality of the evaluation indicators, that is, the availability of the required market data; then, it is necessary to consider whether the market has a benign development trend, set corresponding indicators or evaluation themes, and guide the healthy and sustainable development of the market. Based on the current status of the electricity spot market with a proportion of new energy and the problems and challenges that may be faced in the future, the construction principles of the electricity spot market are determined as follows:

[0077] Relevance principle: Relevance means that the selection of each indicator should be consistent with and reflect the development goals of the electricity spot market;

[0078] Coordination principle: Coordination means that the indicators should be organically combined to form a hierarchical system that conforms to the market characteristics. When selecting specific indicators for comprehensive evaluation, duplication or contradiction should be avoided;

[0079] Comprehensiveness principle: The index system should be able to systematically, comprehensively and coherently reflect the market operation effect and the effectiveness and rationality of the market operation mechanism rules, and should cover all aspects of the power spot market while grasping the main factors;

[0080] Principle of practicality: The principle of practicality has two meanings. First, the availability of market data required to form indicators should not be of theoretical significance but not practical application. Second, the indicator system should not be too complex and large, making it difficult to operate. Instead, it should highlight the key points, reflect the main factors of the main aspects, and select simple, easy-to-understand and easy-to-implement indicators to construct the indicator system.

[0081] Principle of specificity: The evaluation index system is designed according to the specific evaluation object, namely the electricity spot market, and the current stage of market development. It is neither universal nor static.

[0082] On the premise of following the above-mentioned construction principles, a series of specific indicators are selected and formulated from six aspects: market structure, market supply and demand, market development, market behavior, market performance and market risk to evaluate the market.

[0083] That is, the first-level indicators in the evaluation index system for new energy participating in spot market settlement include:

[0084] Market structure, market supply and demand, market development, market behavior, market performance and market risks.

[0085] (1) Secondary indicators in market structure indicators include:

[0086] 1) Market concentration index (HHI)

[0087]

[0088] Among them, O i represents the market share of the ith supplier participating in the Jibei new energy power spot market, and n represents the total number of suppliers participating in the new energy power spot market.

[0089] 2) Remaining Supply Rate (RSI)

[0090]

[0091] Where N represents the total number of suppliers participating in the market, q j represents the supply provided by the jth supplier participating in the market, Q D Represents the total market demand.

[0092] 3) Top-m Index

[0093] The Top-m share index refers to the market share of the largest m power generation groups in the market. The Top-4 index is commonly used in the industry. The larger the index, the higher the market concentration.

[0094] (2) Secondary indicators in market supply and demand indicators include:

[0095] 1) Total monthly electricity volume of new energy transactions

[0096] It means the total amount of electricity traded by Jibei New Energy in that month.

[0097] 2) Monthly total electricity demand of new energy

[0098] It means the total electricity demand for new energy units in northern Hebei that month.

[0099] 3) Actual monthly settlement electricity consumption of new energy

[0100] It means the total amount of electricity actually settled in the Jibei Electricity Spot Market that month.

[0101] (3) Secondary indicators in the market development category include:

[0102] 1) Monthly reporting volume growth rate

[0103] 2) Monthly transaction volume growth rate

[0104] 3) Trading activity

[0105] (4) Secondary indicators in market behavior indicators include:

[0106] 1) Monthly price increase of new energy units

[0107] Meaning: the month-on-month increase in the average price of new energy units

[0108] 2) Price ceiling for new energy units

[0109] It means the quotation cap set by the trading center for new energy units in each trading cycle.

[0110] 3) Quote markup index

[0111] The bidding behavior index of power generation groups or units is one of the most intuitive measures to reflect whether they exercise market power. Before making pricing decisions, power generation groups participating in market bidding may make long-term bidding behaviors far higher than marginal costs based on their judgment of market demand, supply capacity and their own geographical advantages. In this embodiment, the bidding markup index is used to describe the bidding behavior of power generation groups or units participating in the market, and its definition is as follows:

[0112]

[0113] in, and C ac They represent the weighted average bid and the approved power generation cost of the ith power generation group or unit respectively. They are defined as follows:

[0114]

[0115] C ac=C fuel +C trans (6)

[0116] In the formula and Q i,k are the quotation and capacity of the kth section of the i-th unit respectively; K is the number of quotation sections; C fuel is the average coal consumption cost per kilowatt-hour based on government release; C trans is the average transportation cost of fuel.

[0117] 4) Settlement efficiency

[0118] It means the ratio of the transaction volume actually settled in each trading cycle to the total transaction volume in that cycle.

[0119] (5) Secondary indicators in market performance indicators include:

[0120] 1) New energy clearing price

[0121] It means that in spot electricity trading, new energy units trade electricity one by one in the order of the quotations from the power generation side from low to high, so that the traded electricity meets the quotation of the last new energy unit that meets the load demand.

[0122] 2) Reduction rate of additional issuance cost

[0123] The cost reduction rate of new energy generation brought about by calling on inter-provincial power generation resources is lower than that brought about by power generation within the province.

[0124] 3) Price Convergence

[0125]

[0126] Among them, I PCI represents the price consistency index; rt represents the real-time market clearing price; da represents the day-ahead market clearing price.

[0127] (6) Secondary indicators in market risk indicators include:

[0128] 1) Price volatility

[0129] It means the ratio of the difference between the highest monthly clearing price and the lowest monthly clearing price in the market to the monthly average price.

[0130] I FR =(P H -P L ) / P;

[0131] Among them, P H represents the highest marginal electricity price in the new energy power spot market; P L represents the minimum marginal electricity price; P represents the monthly average electricity price.

[0132] 2) Electricity deviation rate

[0133] The ratio of monthly deviation electricity to transaction clearing electricity is the electricity deviation rate.

[0134] 3) New energy revenue

[0135] The grid-connected electricity price of new energy minus the cost of power generation and peak-shaving costs describes the benefits of new energy in obtaining power generation through purchasing peak-shaving services.

[0136] S12: Determine the weight of each indicator in the evaluation index system through the hierarchical analysis method;

[0137] The principle of the hierarchical analysis method is that before analyzing a phenomenon or problem, the phenomenon or problem is first decomposed into relevant factors according to their nature, and a multi-level structural model is formed according to the classification of the relationship between them. Then, through experience or experts, the relative importance of low-level factors to high-level factors is judged and measured, and the weight ranking is obtained according to the degree of importance, and then quantitative analysis and comparison can be performed. The core of the hierarchical analysis method is to hierarchize and digitize the influencing factors. It decomposes an abstract phenomenon or problem from difficult to easy, and it is easy to intuitively judge complex problems and make decisions. The hierarchical analysis method has the advantages of simplifying complex problems and simple calculations. It is widely used, such as in personnel quality assessment, multi-scheme comparison, scientific and technological achievement evaluation, and work performance evaluation. In short, the hierarchical analysis method is to decompose a decision event into a target layer, a criterion layer (factors affecting decision-making, in this invention, it refers to the indicators in the evaluation index system of Jibei New Energy's participation in spot market settlement) and a scheme layer.

[0138] It should be noted that the weight of each indicator in the evaluation index system is determined through the hierarchical analysis method, including:

[0139] Determine the first-level indicators in the evaluation index system for new energy participating in spot market settlement as the criterion layer in the hierarchical evaluation model;

[0140] A judgment matrix is ​​constructed to compare the first-level indicators of the criterion layer in pairs and determine the weight of each first-level indicator of the criterion layer to the target layer.

[0141] Based on the constructed evaluation system, market structure indicators, market supply and demand indicators, market development indicators, market behavior indicators, market performance indicators and market risk indicators can be determined as the criterion layers in the hierarchical evaluation model.

[0142] The hierarchical evaluation model is an evaluation model constructed by the hierarchical analysis method for the evaluation index system.

[0143] Specifically, a judgment matrix is ​​constructed to compare the first-level indicators of the criterion layer in pairs, and the weight of each first-level indicator of the criterion layer to the target layer is determined, including:

[0144] Construct a judgment matrix, calculate the product and 1 / n power of row elements, and obtain the m-dimensional vector of each row of the judgment matrix;

[0145] Normalize the m-dimensional vector to get the weight vector.

[0146] For the criterion layer A, the following matrix can be constructed:

[0147]

[0148] The elements in A satisfy:

[0149]

[0150] Furthermore, based on the expert's scaling opinion, we construct Figure 4 The judgment matrix shown.

[0151] The weight vector is calculated using the square root method. First, the product of the row elements is calculated, and then the 1 / n power is calculated. The specific formula is as follows:

[0152]

[0153] in, represents an m-dimensional vector in each row, m represents the dimension of the vector (the value is 6 in this embodiment), i represents the number of rows of the vector, and j represents the number of columns.

[0154] Then the m-dimensional vector is standardized to obtain the weight vector, that is, the weight is obtained. The specific formula is as follows:

[0155]

[0156] in, Represents the sum of the m-dimensional vectors in each row.

[0157] Reference Figure 3 , the method further comprises:

[0158] S31: Calculate the maximum eigenvalue of the judgment matrix;

[0159] S32: Calculate the CI value of the judgment matrix according to the maximum characteristic root of the judgment matrix;

[0160] S33: selecting the RI value according to the order of the judgment matrix, and calculating the CR value of the judgment matrix according to the CI value and the RI value of the judgment matrix;

[0161] S34: If the CR value of the judgment matrix is ​​less than 0.1, it is determined that the consistency of the judgment matrix is ​​within the allowable range;

[0162] S35: If the CR value of the judgment matrix is ​​not less than 0.1, the judgment matrix is ​​adjusted until the consistency of the judgment matrix reaches an allowable range.

[0163] Calculate the maximum eigenvalue of the judgment matrix. The specific formula is as follows:

[0164]

[0165] Among them, λ max represents the maximum eigenvalue of the matrix, AW represents the weight of the judgment matrix after standardization, and then the accumulated value is calculated by row.

[0166] Calculate the CI value of the judgment matrix based on the maximum eigenvalue of the judgment matrix:

[0167]

[0168] Select the RI value according to the order of the judgment matrix, and calculate the CR value of the judgment matrix according to the CI value and RI value of the judgment matrix:

[0169]

[0170] The RI value table is as follows:

[0171]

[0172] Furthermore, based on the above calculation method, the AHP hierarchical analysis results are as follows:

[0173]

[0174] The consistency test results are as follows:

[0175]

[0176] S13: Sort each indicator according to its weight, select high-weight indicators based on the sorting results and substitute them into the graph neural network prediction model to dynamically predict the participation of new energy in the electricity spot market settlement;

[0177] In this embodiment, the secondary indicators corresponding to the high-weighted primary indicators are selected based on the sorting results and substituted into the graph neural network prediction model. The three market indicators with the highest weights in the AHP hierarchical analysis results are selected, which are: market structure indicators, market development indicators and market risk indicators.

[0178] Furthermore, the detailed secondary indicators of the three indicators are: market concentration, surplus supply rate, Top-m index, monthly declaration volume growth rate, monthly transaction volume growth rate, transaction activity, price volatility, electricity deviation rate, and new energy income. All the above secondary indicators are substituted into the graph neural network prediction model.

[0179] Reference Figure 2 , the method further comprises:

[0180] S21: The graph neural network prediction model uses the substituted indicators as nodes of the graph neural network; each node of the graph neural network is naturally defined by its features and associated nodes;

[0181] S22: define the local transfer function as a loop recursive function to be shared among all nodes, and update the state of the current node based on the input information from the domain when updating the vector of each node;

[0182] S23: Each node uses the surrounding neighboring nodes and adjacent edges as source information to update its own expression h, and introduces the local output function g to obtain the node output o:

[0183] h n =f(x n ,x co[n] ,x ne[n] ,h ne[n] );

[0184] o n =g(h n ,x n );

[0185] Among them, x represents the current node, h represents the hidden state of the node, ne[n] represents the set of neighboring nodes of node n, and co[n] represents the set of adjacent edges of node n;

[0186] S24: Stack all local transfer functions f:

[0187] H=F(H,X);

[0188] O=G(H,X n );

[0189] Among them, H represents the fixed point, F represents the global transfer function, X represents the global node, O represents the output of the global node, and G represents the global output function;

[0190] Assuming that F is a compression mapping, the value of the fixed point H can be uniquely determined, and step S25 is executed.

[0191] S25: According to Banach's fixed point theorem, assuming that the global transfer function F is a compression mapping, the value of the fixed point H is iteratively solved:

[0192] H t+1 =F(H t ,X).

[0193] Furthermore, in order to train the parameters of function f and equation g, the target information (t n ) for supervision, the loss function can be expressed as:

[0194]

[0195] Here, p represents the number of nodes involved in the calculation, and the training algorithm can be based on various types of gradient descent algorithms.

[0196] It should be noted that the method also includes:

[0197] In the graph neural network prediction model, the Pearson correlation coefficient between indicators is used to construct the graph structure;

[0198] When the Pearson correlation coefficient of the two indicators is greater than 0.8, it is determined that there is a correlation between the two indicators, otherwise it is determined that the two indicators are not correlated;

[0199] The gated recurrent unit in the graph neural network prediction model is used to predict the substituted indicators.

[0200] Since the prediction by the graph neural network prediction model requires more sample data, in this embodiment, the closing volume of these 9 secondary indicators (market concentration, surplus supply rate, Top-m index, monthly declaration volume growth rate, monthly transaction volume growth rate, transaction activity, price volatility, electricity deviation rate, and new energy income) on all settlement days from 2015 to 2023 is selected as the known data for prediction. At the same time, the data corresponding to these 9 secondary indicators are processed by the upper and lower mean filling method and the descriptive statistical risk method, and the GRU units in the graph convolutional neural network and the recursive neural network are used to predict these 9 secondary indicators. In the graph neural network prediction model, the Pearson correlation coefficient between indicators is used to construct the graph structure. When the Pearson correlation coefficient of the two indicators is greater than 0.8, it is determined that there is a correlation between the two indicators, and the element value of the corresponding position of the adjacent matrix is ​​1. When the Pearson coefficient is less than 0.8, it is determined that the two indicators are unrelated, and the value of the corresponding element of the adjacent matrix is ​​0.

[0201] In specific practice, 320 hidden units are used in the graph neural network prediction model constructed in this embodiment, and the size of each batch is 40. The learning rate is 0.005, and the number of training times is 600.

[0202] S14: Compare the prediction results of the graph neural network prediction model with the prediction results of multiple different prediction models, and determine the prediction accuracy of the graph neural network prediction model based on the comparison results.

[0203] Preferably, the prediction model different from the graph neural network prediction model includes at least:

[0204] Long short-term memory neural networks, random forests, and extreme gradient boosting trees;

[0205] The prediction accuracy of the graph neural network prediction model is determined based on the comparison results, including:

[0206] The average absolute error, absolute percentage error and root mean square error of the prediction results of the computational graph neural network prediction model and the prediction results of multiple different prediction models are used as comparison results.

[0207] In this embodiment, the prediction results of the long short-term memory neural network (LSTM), random forest and extreme gradient boosting tree (XGBoost) are compared with the graph neural network prediction model to determine the prediction accuracy of the graph neural network prediction model.

[0208] Among them, LSTM is a special recurrent neural network that can control the output state through the gate state to process long sequence data. The LSTM used is a two-layer LSTM with a Dropout before the output layer to prevent overfitting. The Dropout Rate is set to 0.5.

[0209] Random forest is an ensemble algorithm consisting of multiple decision trees, and there is no correlation between failed decision trees. Random forest is a commonly used machine learning model for classification and regression problems. The number of random forest trees used in this article is 24.

[0210] XGBoost is also known as Extreme Gradient Boosting Trees. It is an implementation of a boosting algorithm. The number of trees is 300 and the maximum depth is 6.

[0211] Furthermore, in this embodiment, the mean absolute error (MAE), the mean absolute percentage error (MAPE), and the root mean square error (RMSE) are selected as statistical indicators of the above prediction model. In this embodiment, 98% of the data is used for training, and the remaining samples are 56 as a test set. The prediction errors of several methods for 8 market indicators are as follows:

[0212]

[0213]

[0214] It can be seen that among several machine learning models, the graph neural network prediction model has the highest prediction accuracy for the nine market indicators. At the same time, in specific practice, the parameters of various prediction methods were selected and replaced, and it was found that none of them exceeded the prediction accuracy of the graph neural network. Therefore, among all models, the graph neural network prediction model has the highest prediction accuracy.

[0215] In this technical solution, we first construct an evaluation index system for new energy participating in spot market settlement, and then substitute the high-weighted indicators into the graph neural network prediction model to dynamically predict the participation of new energy in electricity spot market settlement. At the same time, we compare and evaluate the prediction results of the graph neural network prediction model through multiple different prediction models to obtain accurate prediction results. This technical solution can predict the settlement situation of new energy when it participates in spot market settlement, thereby achieving early warning, reducing the risk of new energy participating in the electricity spot market, and protecting the benefits of new energy entities participating in the electricity spot market.

[0216] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0217] It should be noted that, in the description of this application, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0218] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0219] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0220] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0221] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0222] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0223] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0224] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A dynamic prediction method for electricity spot market settlement based on graph neural network, characterized in that: include: According to the construction principles of the evaluation index system, determine the evaluation index system for new energy participating in spot market settlement; Determine the weight of each indicator in the evaluation indicator system through the hierarchical analysis method; The indicators are sorted according to their weights, and based on the sorting results, high-weight indicators are selected and substituted into the graph neural network prediction model to dynamically predict the participation of new energy in the electricity spot market settlement; The prediction results of the graph neural network prediction model are compared with the prediction results of multiple different prediction models, and the prediction accuracy of the graph neural network prediction model is determined based on the comparison results.

2. The method according to claim 1, characterized in that The method further comprises: The graph neural network prediction model uses the substituted indicators as nodes of the graph neural network; each node of the graph neural network is naturally defined by its features and associated nodes; The local transfer function is defined as a recursive function shared among all nodes, and the state of the current node is updated based on the input information from the domain when updating the vector of each node; Each node uses the surrounding neighboring nodes and adjacent edges as source information to update its own expression h, and introduces the local output function g to obtain the node output o: h n =f(x n ,x co[n] ,x ne[n] ,h ne[n] ); o n =g(h n ,x n ); Among them, x represents the current node, h represents the hidden state of the node, ne[n] represents the set of neighboring nodes of node n, and co[n] represents the set of adjacent edges of node n; Stack all local transfer functions f: H=F(H,X); O=G(H,X n ); Among them, H represents the fixed point, F represents the global transfer function, X represents the global node, O represents the output of the global node, and G represents the global output function; According to Banach's fixed point theorem, assuming that the global transfer function F is a compression mapping, the value of the fixed point H is iteratively solved: H t+1 =F(H t ,X)。 3. The method according to claim 1, characterized in that The method further comprises: In the graph neural network prediction model, the Pearson correlation coefficient between indicators is used to construct the graph structure; When the Pearson correlation coefficient of the two indicators is greater than 0.8, it is determined that there is a correlation between the two indicators, otherwise it is determined that the two indicators are not correlated; The gated recurrent unit in the graph neural network prediction model is used to predict the substituted indicators.

4. The method according to claim 1, characterized in that: The construction principles of the evaluation index system include: The principles of relevance, coordination, comprehensiveness, practicality and specificity.

5. The method according to claim 1, characterized in that The first-level indicators in the evaluation index system for new energy participating in spot market settlement include: market structure, market supply and demand, market development, market behavior, market performance and market risks; The secondary indicators in the market structure category include: Market concentration, remaining supply rate and Top-m index; The secondary indicators of market supply and demand indicators include: Total monthly transaction volume of new energy, total monthly demand for new energy, and actual monthly settlement volume of new energy; The secondary indicators in the market development category include: Monthly reporting volume growth rate, monthly transaction volume growth rate and transaction activity; The secondary indicators in the market behavior indicators include: Monthly increase in new energy unit price quotes, new energy unit price ceilings, quote markup index and settlement efficiency; The secondary indicators in the market performance category include: New energy clearing prices, additional issuance cost reduction rates and price convergence; The secondary indicators of market risk indicators include: Price volatility, electricity deviation rate and new energy benefits.

6. The method according to claim 5, characterized in that The weight of each indicator in the evaluation index system is determined by the hierarchical analysis method, including: Determine the first-level indicators in the evaluation index system for new energy participating in spot market settlement as the criterion layer in the hierarchical evaluation model; A judgment matrix is ​​constructed to compare the first-level indicators of the criterion layer in pairs and determine the weight of each first-level indicator of the criterion layer to the target layer.

7. The method according to claim 6, characterized in that Construct a judgment matrix to compare the first-level indicators of the criterion layer in pairs and determine the weight of each first-level indicator of the criterion layer to the target layer, including: Construct a judgment matrix, calculate the product and 1 / n power of row elements, and obtain the m-dimensional vector of each row of the judgment matrix; Normalize the m-dimensional vector to get the weight vector.

8. The method according to claim 7, characterized in that The method further comprises: Calculate the maximum eigenvalue of the judgment matrix; Calculate the CI value of the judgment matrix based on the maximum eigenvalue of the judgment matrix; Select the RI value according to the order of the judgment matrix, and calculate the CR value of the judgment matrix according to the CI value and RI value of the judgment matrix; If the CR value of the judgment matrix is ​​less than 0.1, it is determined that the consistency of the judgment matrix is ​​within the allowable range; If the CR value of the judgment matrix is ​​not less than 0.1, the judgment matrix is ​​adjusted until the consistency of the judgment matrix is ​​within the allowable range.

9. The method according to claim 5, characterized in that The indicators are sorted according to their weights, and high-weight indicators are selected based on the sorting results and substituted into the graph neural network prediction model, including: Based on the sorting results, the secondary indicators corresponding to the high-weight primary indicators are selected and substituted into the graph neural network prediction model.

10. The method according to claim 1, characterized in that The prediction model different from the graph neural network prediction model includes at least: Long short-term memory neural networks, random forests, and extreme gradient boosting trees; Determining the prediction accuracy of the graph neural network prediction model according to the comparison result includes: The average absolute error, absolute percentage error and root mean square error of the prediction results of the computational graph neural network prediction model and the prediction results of multiple different prediction models are used as comparison results.

Images