Method for constructing, training and applying price difference direction prediction model

Abstract

The invention discloses a method for constructing, training and applying a price difference direction prediction model, and the method comprises the steps: building a multi-algorithm model in different scenes, and finally providing an optimal algorithm model for the prediction of the price difference direction through an automatic evaluation and screening mechanism between the models; in order to provide a more reliable price difference reference basis for spot declaration, a price difference prediction scheme is not adopted, but a more effective price difference direction prediction scheme is adopted on the premise of compatibly considering the price difference; and in combination with a loss function model constructed by the method, the reliability and accuracy of price difference direction prediction are greatly improved.

Description

Technical Field

[0001] The invention belongs to the technical field of power market data information processing, and specifically relates to a method for constructing, training and applying a price difference direction prediction model. Background Art

[0002] In the electricity spot market, the two market players, new energy power generation stations and power sales companies, will pay more attention to the price difference between the day-ahead clearing price and the real-time clearing price. For new energy power generation stations, station traders need to declare the amount of electricity based on the station's forecast of future short-term power generation, which will be used as a standard for assessment and calculation of settlement income; similarly, power sales companies also need to declare short-term electricity consumption based on their own users' electricity consumption forecasts, and use the declaration as a standard for assessment and calculation of settlement income. Because the accuracy of the price difference between the day-ahead clearing price and the real-time clearing price in the short term directly affects the final assessment and settlement income, the ability to accurately predict the size and direction of the price difference will naturally become the most critical factor.

[0003] The existing technology mainly uses the disclosed data provided by the spot market and artificially defines the data features based on the traders' own trading experience. Based on this data feature, a fixed algorithm model established in advance is trained to predict the price difference between the day-ahead clearing electricity price and the real-time clearing electricity price. This model is used to predict the short-term price difference in the future, and finally the spot electricity volume is declared based on the price difference prediction result.

[0004] The shortcomings of the existing technologies mainly exist in the following three aspects. First: In extracting effective feature data from massive disclosure data, clearing data and meteorological data, the existing technology still relies on the experience of traders themselves, so it is impossible to guarantee the long-term effectiveness and correctness of the features, and the artificially specified features cannot be guaranteed to be applicable to different seasons, different holidays, different wind seasons and different weather conditions. Second: Most of the algorithm models of the existing technologies rely on some pre-constructed algorithm models, so the generalization ability and universality of the model cannot be guaranteed. The most important performance is that a single model can only guarantee its effectiveness under certain climate conditions (such as summer and autumn climate conditions) and certain change rules (such as relatively stable load of the whole network and relatively moderate wind and solar output). Third: The existing technology mainly focuses on the prediction of the size of the spread, but the size of the spread changes strangely in most cases, the spread range is unstable, and there is no trace of the time period of the spread formation, which leads to a very low prediction accuracy of the size of the spread, and thus cannot provide a reliable, stable and effective reference basis for trading entities in the spot market.

[0005] To this end, we propose a method for constructing, training, and applying a price difference direction prediction model to solve the above problems. Summary of the invention

[0006] The technical problem to be solved by the present invention is: to provide a method for constructing, training and applying a price difference direction prediction model, with the aim of fully realizing the automatic data feature extraction function module, thereby ensuring that the extracted data features can meet the needs of different seasons, different holidays, different wind seasons and different weather conditions; in order to further improve the generalization ability and universality of the algorithm model, the present invention will establish a multi-algorithm model for use in different scenarios, and finally provide the best algorithm model for price difference direction prediction through an automatic evaluation and screening mechanism between models; in order to provide a more reliable price difference reference for spot declaration, the present invention does not adopt a price difference size prediction scheme, but adopts a more effective price difference direction prediction scheme on the premise of being compatible with the price difference size, and combines the loss function model constructed by the present invention to greatly improve the reliability and accuracy of price difference direction prediction.

[0007] The objective of the present invention is achieved through the following technical solutions:

[0008] A price difference direction prediction model, comprising a feature screening algorithm model module, a prediction algorithm model module, and a model screening algorithm model module.

[0009] The feature screening algorithm model module includes an adaptive feature screening algorithm model;

[0010] The prediction algorithm model module includes a multi-scenario neural network algorithm model and a high-stability loss function algorithm model;

[0011] The model screening algorithm model module includes a strong intelligent model screening algorithm model and.

[0012] Another object of the present invention is to provide a method for training a price difference direction prediction model, wherein the feature screening algorithm model module obtains disclosure data, clearing data through the spot trading center platform and meteorological data through the European Meteorological Center, and then uses an adaptive feature screening algorithm model to extract the best feature data; the prediction algorithm model module is based on the screened data features, and establishes a multi-scenario neural network algorithm model that meets different scenarios, and combines a customized, highly stable loss function algorithm model suitable for spot declaration scenarios to form the core of the prediction algorithm model; the model screening algorithm model module mainly selects the best model from multiple trained models through an intelligent model screening algorithm model.

[0013] Further: the above training method specifically includes the following steps:

[0014] S1. In the feature screening algorithm model module, appropriate feature data is obtained from a large number of disclosure data, clearance data and meteorological data as input features of the neural network model. At the same time, the combination of feature fusion, feature dimension upgrading sub-algorithm model and Kendall correlation coefficient analysis sub-algorithm model can screen out more relevant adaptive data features;

[0015] S2. In the prediction algorithm model module, based on the data features screened in step S1, a multi-scenario neural network algorithm model is established, and then trained in combination with a loss function that is compatible with the price difference size and price difference direction, a highly stable loss function algorithm model can be obtained;

[0016] S3. In the model screening algorithm model module, since a variety of neural algorithm network models suitable for different scenarios are designed in the second step, a corresponding number of training models can be obtained by training and learning the screened data features. Recent feature data and clearing data are used as test sets, and all models are used cyclically to predict the test set. The designed price difference direction prediction accuracy algorithm is used to obtain the corresponding price difference direction accuracy. Finally, the algorithm model with the highest accuracy is selected, and this model is used as the best model for predicting the future short-term electricity price difference direction.

[0017] Furthermore, the feature fusion and dimension-raising sub-algorithm model is specifically as follows:

[0018] d t =n t -w t -p t -l n

[0019] in:

[0020] d t represents the non-renewable energy generation space at time t;

[0021] n t represents the load demand of the entire network at time t;

[0022] w t Indicates the wind power output of the entire grid at the tth moment;

[0023] p t Indicates the photovoltaic power output of the entire grid at the tth moment;

[0024] l t Indicates the transmission size of the interconnection line at the tth moment.

[0025] Furthermore, the Kendall correlation coefficient analysis sub-algorithm model is as follows:

[0026]

[0027]

[0028]

[0029]

[0030] in:

[0031] τ represents the size of the Kendall correlation coefficient, and its value range is [-1,1];

[0032] τ = -1 means that the two vectors are completely negatively correlated;

[0033] τ = 0 means that the two vectors have no correlation;

[0034] τ = 1 means that the two vectors are completely positively correlated;

[0035] n represents the sample size;

[0036] n a It indicates the number of concordant pairs in the comparison of two variables;

[0037] n b It indicates the number of discordant pairs in the comparison of two variables;

[0038] t i Indicates the number of clusters corresponding to the comparison vector when the price difference is the i-th clustering level; u j It indicates the number of clusters corresponding to the price difference when the comparison variable is the jth clustering level. The further high stability loss function algorithm model is as follows:

[0039]

[0040]

[0041] loss = diff_loss + log_loss

[0042] in:

[0043] diff_loss represents the difference size and direction loss function;

[0044] log_loss represents the logarithmic loss function in the price difference direction;

[0045] It represents the predicted value of the price difference direction at time t;

[0046] y t Indicates the true value of the price difference direction at time t;

[0047] priqian,t represents the day-ahead clearing electricity price at time t;

[0048] p shishi,t Represents the real-time clearing electricity price at time t.

[0049] Furthermore, the algorithm model formula with the highest accuracy is as follows:

[0050]

[0051] in:

[0052] Y t It represents the predicted value of the price difference direction at time t;

[0053] D t =1 means that the day-ahead clearing electricity price at time t is greater than the day-ahead real-time electricity price;

[0054] D t =0 means that the day-ahead clearing electricity price at time t is equal to the day-ahead real-time electricity price;

[0055] D t =-1 means that the day-ahead clearing electricity price at time t is less than the day-ahead real-time electricity price;

[0056] P riqian,t represents the day-ahead clearing electricity price at time t;

[0057] P shishi,t Represents the real-time clearing electricity price at time t.

[0058] The third object of the present invention is to provide a price difference direction prediction model training method for use in predicting the price difference direction in the short term in the future.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The present invention mainly predicts the direction of the price difference between the day-ahead clearing electricity price and the real-time clearing electricity price, and considers the influence of the price difference size in the design of the algorithm model, which is fundamentally different from the method of directly predicting the price difference size and this method is more effective. In the spot market, the starting points of the various participating entities are somewhat different, and the trading varieties are also different, but each entity will focus on the direction of the price difference between the day-ahead clearing electricity price and the real-time clearing electricity price, especially the time periods with a larger price difference, because the assessment and settlement income generated in these time periods accounts for a larger proportion. The loss function algorithm logic and model constructed in combination with the present invention can greatly improve the prediction accuracy in the time periods with a large price difference, which will largely solve the problem that the existing technology has a low accuracy rate in predicting the size of the price difference, especially inaccurate prediction or even reverse prediction in the time periods with a large price difference.

[0061] In order to ensure that the price difference direction prediction algorithm model has stable, reliable and accurate prediction effects in different seasons, holidays, wind seasons of different sizes and weather conditions, this patent constructs different algorithm network models, and designs reasonable model evaluation mechanisms and screening logic to realize autonomous evaluation and screening of algorithm models in different scenarios, thereby ensuring the freedom, stability and efficiency of the model.

[0062] At the same time, in order to improve the selection of data features that are more in line with the actual situation from the disclosed data and cleared data published in the spot market, the present invention realizes the automatic and intelligent selection of data features through algorithms such as feature fusion, feature dimension upgrade, and correlation analysis, thereby avoiding the problem that the selected features are limited, unstable, and not universal due to lack of manual experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is the overall flow chart of the present invention;

[0064] Figure 2 This is a flow chart of the feature screening algorithm model in the present invention;

[0065] Figure 3 This is a flow chart of the prediction algorithm model in the present invention;

[0066] Figure 4 This is a schematic diagram of the multi-scenario algorithm model in the present invention;

[0067] Figure 5 This is a flow chart of the model screening algorithm model in the present invention. DETAILED DESCRIPTION

[0068] The following will describe the implementation methods of the present application in detail with the help of accompanying drawings and examples, so that the implementation process of how the present application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0069] This invention proposes an adaptive, multi-scenario, highly stable, and highly intelligent algorithm logic model for predicting the price difference between the day-ahead clearing electricity price and the real-time clearing electricity price. The overall design idea and process of the algorithm model are as follows: Figure 1As shown, the feature screening algorithm model module obtains disclosure data and clearing data through the spot trading center platform and meteorological data through the European Meteorological Center, and then uses an adaptive feature screening algorithm model to extract the best feature data; the prediction algorithm model module is based on the screened data features, and establishes multiple neural network algorithm models that meet different scenarios, and combines a customized and highly stable loss function algorithm model suitable for spot declaration scenarios to form the core of the prediction algorithm model; the model screening algorithm model module mainly selects the best model from multiple trained models through an intelligent model screening algorithm model, and finally applies the model to the prediction of the price difference direction in the short term in the future.

[0070] Step 1: In the feature screening algorithm model module, it is necessary to obtain appropriate feature data from a large number of disclosed data, clearing data and meteorological data as the input features of the neural network model. Reasonable and effective feature selection directly affects the learning effect of the model and the prediction results of the final model. The disclosed data that can generally be obtained in the spot market trading center include network-wide load forecast data, network-wide wind power output forecast data, network-wide photovoltaic output forecast data, interconnection line forecast data and other data; the clearing data that can be obtained include day-ahead clearing electricity price data and real-time clearing electricity price data. At the same time, more accurate meteorological data can be obtained through the European Meteorological Center (ECMWF). However, considering that the data type is still small, the present invention proposes a sub-algorithm model of feature fusion and feature dimensionality increase, and in order to extract more effective and accurate data features, taking into account the existence of data mutations (fluctuating) and outliers in the price difference data, the present invention adopts the sub-algorithm model of Kendall correlation coefficient analysis. Finally, the combination of the two algorithm sub-modules can screen out more relevant adaptive data features. The detailed design ideas and processes of this algorithm module are as follows. Figure 2 shown.

[0071] (1) Feature fusion and dimension-upgrading sub-algorithm model

[0072] According to the spot market rules and the universal supply and demand relationship in nature, it is easy to conclude that when power generation is greater than power consumption, the price of electricity will inevitably fall; similarly, when power generation is less than power consumption, the price of electricity will inevitably rise. Taking this obvious natural law into consideration, the present invention designs a feature fusion sub-algorithm model to reflect the reasonable degree of supply and demand relationship, and uses this data as one of the data features. The specific algorithm model is as follows:

[0073] d t =n t -w t -p t -l n

[0074] in:

[0075] d t represents the non-renewable energy generation space at time t;

[0076] n t represents the load demand of the entire network at time t;

[0077] w t Indicates the wind power output of the entire grid at the tth moment;

[0078] p t Indicates the photovoltaic power output of the entire grid at the tth moment;

[0079] l t It represents the transmission size of the tie line at the tth moment;

[0080] (2) Kendall correlation coefficient analysis sub-algorithm model

[0081] Since there must be a difference between the predicted disclosure data and the actual operating data, there is generally a certain price difference between the day-ahead clearing electricity price and the real-time clearing electricity price. However, due to many situations such as sudden changes in weather, sudden changes in load demand, abnormal unit output, etc., the electricity price difference may fluctuate abnormally and change frequently. Because abnormal differences can also reflect the laws and conditions of the market, in order to better utilize these abnormal data, the present invention first uses the k-nearest neighbor algorithm to perform 10 types of clustering operations on the electricity price difference data at each moment and the disclosed data and meteorological data of each variable at the corresponding moment, and then obtains the corresponding clustering level, and finally uses the Kendall correlation coefficient analysis sub-algorithm model to obtain characteristic variable data with a correlation greater than 0.6. The specific algorithm model is as follows

[0082]

[0083]

[0084]

[0085]

[0086] in:

[0087] τ represents the size of the Kendall correlation coefficient, and its value range is [-1,1];

[0088] τ = -1 means that the two vectors are completely negatively correlated;

[0089] τ = 0 means that the two vectors have no correlation;

[0090] τ = 1 means that the two vectors are completely positively correlated;

[0091] n represents the sample size;

[0092] n a It indicates the number of concordant pairs in the comparison of two variables;

[0093] n b It indicates the number of discordant pairs in the comparison of two variables;

[0094] t i Indicates the number of clusters corresponding to the comparison vector when the price difference is the i-th clustering level;

[0095] u j It indicates the number of clusters corresponding to the price difference when the contrast variable is at the jth clustering level;

[0096] Step 2: In the prediction algorithm model module, based on the data features screened in the first step, a neural algorithm network model is established for use in different scenarios, such as algorithm models suitable for different wind seasons, algorithm models suitable for different levels of load demand, and algorithm models suitable for different weather conditions. Then, the loss function designed by this patent is combined with the loss function that considers the size and direction of the price difference to train, and the training model for different scenarios can be obtained. The detailed design ideas and processes of this algorithm module are as follows: Figure 3 shown.

[0097] (1) Multi-scenario algorithm model

[0098] In order to adapt to different wind seasons, different levels of load demand, different weather conditions and other scenarios, the present invention proposes to design a neural network algorithm model that can give full play to its own network advantages in a variety of different scenarios and dimensions. The specific implementation mechanism and algorithm principle of the algorithm model are as follows Figure 4 shown.

[0099] (2) Highly stable loss function algorithm model

[0100] A reasonable, efficient, universal and stable loss function is the key to the excellent learning effect of the algorithm model. In the spot scenario, each participant pays more attention to the price difference between the day-ahead clearing price and the real-time clearing price. However, the existing technology has poor prediction effect and low accuracy on the price difference. The present invention proposes a loss function that considers both the price difference size and the price difference direction. By designing a reasonable compatible method, the two different loss functions are integrated to obtain a highly stable loss function model. The specific implementation mechanism and algorithm principle of the algorithm model are as follows

[0101]

[0102]

[0103] loss = diff_loss + log_loss

[0104] in:

[0105] diff_loss represents the difference size and direction loss function;

[0106] log_loss represents the logarithmic loss function in the price difference direction;

[0107] It represents the predicted value of the price difference direction at time t;

[0108] y t Indicates the true value of the price difference direction at time t;

[0109] p riqian,t represents the day-ahead clearing electricity price at time t;

[0110] p shishi,t represents the real-time clearing electricity price at time t;

[0111] Step 3: In the model screening algorithm model module, since a variety of neural algorithm network models suitable for different scenarios are designed in the second step, a corresponding number of training models can be obtained by training and learning the screened data features. In order to select the training model with the best prediction effect for future short-term prediction from these models, the present invention designs a model screening algorithm that meets the spot declaration scenario. The specific algorithm model is as follows:

[0112]

[0113] in:

[0114] Y t It represents the predicted value of the price difference direction at time t;

[0115] D t =1 means that the day-ahead clearing electricity price at time t is greater than the day-ahead real-time electricity price;

[0116] D t =0 means that the day-ahead clearing electricity price at time t is equal to the day-ahead real-time electricity price;

[0117] D t =-1 means that the day-ahead clearing electricity price at time t is less than the day-ahead real-time electricity price;

[0118] P riqian,t represents the day-ahead clearing electricity price at time t;

[0119] P shishi,t represents the real-time clearing electricity price at time t;

[0120] The specific implementation principle of this algorithm module is: take the characteristic data and clearing data of the last few days as the test set, use all models cyclically to predict the test set, and use the designed price difference direction prediction accuracy algorithm to obtain the corresponding price difference direction accuracy, and finally select the algorithm model with the highest accuracy, and use this model as the best model for predicting the future short-term electricity price difference direction. The detailed design ideas and processes of this algorithm module are as follows: Figure 5 shown.

[0121] In summary, the above are only preferred embodiments of the present invention and are not intended to limit the present invention in any form. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention fall within the protection scope of the present invention.

[0122] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are within the scope of the technical solution of the present invention.

Claims

1. A price difference direction prediction model is constructed, characterized by: Including feature screening algorithm model module, prediction algorithm model module, model screening algorithm model module, The feature screening algorithm model module includes an adaptive feature screening algorithm model; The prediction algorithm model module includes a multi-scenario neural network algorithm model and a high-stability loss function algorithm model; The model screening algorithm model module includes a strong intelligent model screening algorithm model and.

2. The price difference direction prediction model training method according to claim 1, characterized in that: The feature screening algorithm model module obtains disclosure data, clearing data through the spot trading center platform and meteorological data through the European Meteorological Center, and then uses an adaptive feature screening algorithm model to extract the best feature data; the prediction algorithm model module is based on the screened data features, by establishing a multi-scenario neural network algorithm model that meets different scenarios, and combines it with a customized, highly stable loss function algorithm model suitable for spot declaration scenarios to form the core of the prediction algorithm model; the model screening algorithm model module mainly selects the best model from multiple trained models through an intelligent model screening algorithm model.

3. The price difference direction prediction model training method according to claim 2, characterized in that: The specific steps include: S1. In the feature screening algorithm model module, appropriate feature data is obtained from a large number of disclosure data, clearance data and meteorological data as input features of the neural network model. At the same time, the combination of feature fusion, feature dimension upgrading sub-algorithm model and Kendall correlation coefficient analysis sub-algorithm model can screen out more relevant adaptive data features; S2. In the prediction algorithm model module, based on the data features screened in step S1, a multi-scenario neural network algorithm model is established, and then trained in combination with a loss function that is compatible with the price difference size and price difference direction, a highly stable loss function algorithm model can be obtained; S3. In the model screening algorithm model module, since a variety of neural algorithm network models suitable for different scenarios are designed in the second step, a corresponding number of training models can be obtained by training and learning the screened data features. Recent feature data and clearing data are used as test sets, and all models are used cyclically to predict the test set. The designed price difference direction prediction accuracy algorithm is used to obtain the corresponding price difference direction accuracy. Finally, the algorithm model with the highest accuracy is selected, and this model is used as the best model for predicting the future short-term electricity price difference direction.

4. The price difference direction prediction model training method according to claim 3 is characterized in that: The feature fusion and dimension-raising sub-algorithm models are specifically as follows: d t =n t -w t -p t -l n in: d t represents the non-renewable energy generation space at time t; n t represents the load demand of the entire network at time t; w t Indicates the wind power output of the entire grid at the tth moment; p t Indicates the photovoltaic power output of the entire grid at the tth moment; l t Indicates the transmission size of the interconnection line at the tth moment.

5. The price difference direction prediction model training method according to claim 3, characterized in that: The Kendall correlation coefficient analysis sub-algorithm model is as follows: in: τ represents the size of the Kendall correlation coefficient, and its value range is [-1,1]; τ = -1 means that the two vectors are completely negatively correlated; τ = 0 means that the two vectors have no correlation; τ = 1 means that the two vectors are completely positively correlated; n represents the sample size; n a It indicates the number of concordant pairs in the comparison of two variables; n b It indicates the number of discordant pairs in the comparison of two variables; t i Indicates the number of clusters corresponding to the comparison vector when the price difference is the i-th clustering level; u j It indicates the number of clusters corresponding to the price difference when the contrast variable is at the jth clustering level.

6. The price difference direction prediction model training method according to claim 3, characterized in that: The highly stable loss function algorithm model is specifically as follows: loss = diff_loss + log_loss in: diff_loss represents the difference size and direction loss function; log_loss represents the logarithmic loss function in the price difference direction; It represents the predicted value of the price difference direction at time t; y t Indicates the true value of the price difference direction at time t; p riqian,t represents the day-ahead clearing electricity price at time t; p shishi,t Represents the real-time clearing electricity price at time t.

7. The price difference direction prediction model training method according to claim 3, characterized in that: The algorithm model formula with the highest accuracy is as follows: in: Y t It represents the predicted value of the price difference direction at time t; D t =1 means that the day-ahead clearing electricity price at time t is greater than the day-ahead real-time electricity price; D t =0 means that the day-ahead clearing electricity price at time t is equal to the day-ahead real-time electricity price; D t =-1 means that the day-ahead clearing electricity price at time t is less than the day-ahead real-time electricity price; P riqian,t represents the day-ahead clearing electricity price at time t; P shishi,t Represents the real-time clearing electricity price at time t.

8. A price difference direction prediction model training method as described in any one of claims 1 to 7 is applied to predict the price difference direction in the short term in the future.

Images