Electric power marketing management information platform spot transaction prediction method

Abstract

The invention discloses a power marketing management information platform spot transaction prediction method, and relates to the technical field of power marketing spot transaction prediction, and the method comprises the steps: carrying out the semantic feature extraction of a power market related policy text through employing a text semantic modeling method, and obtaining a policy semantic vector representation; constructing a policy sensitivity function based on the policy semantic vector representation, and calculating a quantitative index of the influence of the policy on the market price to obtain a policy sensitivity value; performing feature fusion processing on the power market operation data, the meteorological data and the load demand data under multiple time scales by adopting a time sequence coding method to obtain a time sequence feature code; carrying out joint input on the policy sensitivity value and the time sequence feature code, carrying out training through a hybrid neural network model containing an attention mechanism, and outputting an electric power spot market price prediction value at a future moment; and constructing a quotation decision model by adopting a reinforcement learning strategy and taking the market price predicted value as state input.

Description

Technical Field

[0001] This invention relates to the field of electricity marketing spot trading forecasting technology, and in particular to a method for forecasting spot trading on an electricity marketing management information platform. Background Technology

[0002] Electricity marketing spot trading forecasting technology refers to the technique of predicting the price trend of the electricity spot market over a future period by analyzing historical data, combining it with current market conditions and external factors, and using mathematical models, machine learning algorithms, and other means. Therefore, how to utilize advanced technologies to improve the intelligence and security of electricity marketing spot trading forecasting has become one of the urgent problems to be solved.

[0003] In the field of electricity marketing spot trading forecasting, existing electricity spot trading forecasting methods ignore the impact of policy changes on market prices. Policy changes, such as new government regulations and subsidy policies, have a significant impact on market supply and demand, leading to price fluctuations. Furthermore, traditional time series forecasting models struggle to capture the nonlinear dynamic changes in electricity market prices, especially when faced with unforeseen circumstances or rapid changes in market conditions, where the model's forecasting accuracy drops significantly. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a spot trading forecasting method for the power marketing management information platform to address the problem that existing power spot trading forecasting methods ignore the impact of policy changes on market prices. Policy changes, such as new government regulations and subsidy policies, can have a significant impact on market supply and demand, leading to price fluctuations. Furthermore, traditional time series forecasting models struggle to capture the nonlinear dynamic changes in power market prices, especially when faced with unforeseen circumstances or rapid changes in market conditions, where the model's forecasting accuracy drops significantly.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a spot transaction forecasting method for an electricity marketing management information platform, comprising:

[0008] A text semantic modeling method is used to extract semantic features from electricity market-related policy texts to obtain policy semantic vector representations.

[0009] Based on the policy semantic vector representation, a policy sensitivity function is constructed to calculate a quantitative indicator of the policy's impact on market prices, thereby obtaining the policy sensitivity value.

[0010] Time series coding methods are used to perform feature fusion processing on electricity market operation data, meteorological data, and load demand data at multiple time scales to obtain time series feature codes.

[0011] The policy sensitivity value and the time series feature code are jointly input, trained through a hybrid neural network model containing an attention mechanism, and the predicted value of the electricity spot market price at future time is output.

[0012] Using a reinforcement learning strategy with the market price forecast as the state input, a pricing decision model is constructed to generate dynamic pricing suggestions for power generation companies or electricity sales companies.

[0013] Based on the rolling window update mechanism, the latest market feedback data is periodically acquired to perform online incremental updates on the hybrid neural network model and the pricing decision model.

[0014] As a preferred embodiment of the spot transaction forecasting method for the power marketing management information platform described in this invention, the step of extracting semantic features from power market-related policy texts using a text semantic modeling method to obtain a policy semantic vector representation includes the following steps:

[0015] The text of policy documents related to the electricity market is input into the pre-trained language model BERT. The contextual semantic information is extracted through a multi-layer Transformer structure, and a high-dimensional semantic vector of 768 dimensions is output. ;

[0016] Based on the high-dimensional semantic vector Principal component analysis (PCA) was used to reduce the dimensionality of the feature vector, reducing redundant dimensions while preserving the main semantic information, resulting in a low-dimensional policy semantic feature vector. ,in <768.

[0017] As a preferred embodiment of the spot transaction forecasting method for the power marketing management information platform described in this invention, the step of constructing a policy sensitivity function based on the policy semantic vector representation, calculating a quantitative indicator of the policy's impact on market prices, and obtaining a policy sensitivity value includes the following steps:

[0018] Nonlinear mapping functions are used to transform policy semantic features into numerical indicators that reflect the degree of policy impact on market prices;

[0019] The nonlinear mapping function is:

[0020] ;

[0021] in, Indicates the first The policy sensitivity value at any given time is used to measure the direction and intensity of the impact of current policies on fluctuations in electricity spot market prices.

[0022] As a preferred embodiment of the spot transaction forecasting method for the power marketing management information platform described in this invention, the step of using a time series coding method to perform feature fusion processing on power market operation data, meteorological data, and load demand data at multiple time scales to obtain time series feature codes includes the following steps:

[0023] We collected and organized electricity market operation data, meteorological data, and load demand data, and divided the data into three granularities: hourly, daily, and weekly, forming three types of raw datasets. ;

[0024] Standardize each dataset separately to eliminate the differences in the units of measurement between different variables;

[0025] Extract statistical and trend features at each time granularity;

[0026] The above features are fused using a time series encoder. The time series encoder adopts a multi-head self-attention mechanism model based on the Transformer structure, and sequentially performs position encoding addition, multi-head attention calculation, feedforward neural network processing and residual connection operation.

[0027] Output time series feature encoding This serves as one of the inputs for subsequent prediction models.

[0028] As a preferred embodiment of the spot market transaction forecasting method for the power marketing management information platform described in this invention, the step of jointly inputting the policy sensitivity value and the time series feature code, training the model through a hybrid neural network model containing an attention mechanism, and outputting the predicted value of the future power spot market price includes the following steps:

[0029] Encode the time series features Policy sensitivity value Perform a concatenation operation to generate a joint input feature vector. ;

[0030] The joint input feature vector is input into a hybrid neural network model containing convolutional layers, long short-term memory layers, and attention modules for modeling processing;

[0031] The modeling process includes:

[0032] The trend pattern within a local time window is extracted using a one-dimensional convolutional layer, and a local feature representation is output. ;

[0033] Local features are input into the LSTM layer to model long-term dependencies, and the output is a sequence of hidden states. ;

[0034] The importance weights of each time step are calculated using an attention mechanism. After weighted fusion, context-aware features are obtained. ;

[0035] Finally, a fully connected layer is used to perform a nonlinear transformation on the extracted features, outputting the predicted electricity spot market price at future time t+1. ;

[0036] The output calculation expression of the hybrid neural network model is as follows:

[0037] ;

[0038] In the formula, and These are the output layer parameters.

[0039] As a preferred embodiment of the spot transaction forecasting method for the power marketing management information platform described in this invention, the step of employing a reinforcement learning strategy to construct a pricing decision model using the predicted market price as state input, and generating dynamic pricing suggestions for power generation companies or power sales companies, includes the following steps:

[0040] The market price forecast As the state input for reinforcement learning, the action space is defined as the set of pricing strategies that power generation companies or electricity sales companies can choose from. Each action This corresponds to a specific price level;

[0041] The reward function is designed to maximize the expected return per unit time, while taking into account the probability of a transaction and risk control constraints.

[0042] The reward function is defined as follows:

[0043] ;

[0044] In the formula, The quote value corresponding to the quote action. For marginal cost, This is the expected transaction volume for this price, fitted based on historical data. This is the probability factor for a transaction. Price deviation penalty coefficient;

[0045] The strategy is trained using either Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) algorithm to generate optimal bid suggestions and output dynamic bid decision results. This refers to the optimal bidding action under the current circumstances.

[0046] As a preferred embodiment of the spot transaction forecasting method for the power marketing management information platform described in this invention, the step of periodically acquiring the latest market feedback data based on a rolling window update mechanism and incrementally updating the hybrid neural network model and the pricing decision model online includes the following steps:

[0047] Set a fixed time interval Get the latest market feedback data, including actual transaction prices, transaction volume, and policy updates;

[0048] Based on the latest market feedback data, the parameters of the hybrid neural network model are fine-tuned and updated using an incremental learning approach.

[0049] The update formula is as follows:

[0050] ;

[0051] In the formula, These are the parameters of the original model. For learning rate, For loss function, This is a newly added sample;

[0052] Simultaneously, the historical experience pool of the reinforcement learning policy model is updated by adding new state-action-reward triples. This enhances the generalization ability of the pricing decision model under new market conditions.

[0053] As a preferred embodiment of the spot transaction forecasting method for the power marketing management information platform described in this invention, the process of updating records and uploading performance evaluation results to the power marketing management information platform database includes:

[0054] After each model update, calculate and record the changes in the model's performance metrics before and after the update, including:

[0055] Mean absolute percentage error ;

[0056] Root mean square error ;

[0057] Predicted direction accuracy ;

[0058] The performance evaluation results, along with the model update logs and parameter change records, are uploaded to the power marketing management information platform database so that system administrators can view historical version comparisons and model evolution trends.

[0059] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the spot transaction prediction method for the power marketing management information platform as described in the first aspect of the present invention.

[0060] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the spot transaction prediction method for the power marketing management information platform as described in the first aspect of the present invention.

[0061] The beneficial effects of this invention are as follows: By introducing policy semantic modeling, policy sensitivity measurement, multi-timescale feature fusion, and a hybrid neural network prediction model based on attention mechanism, a set of intelligent prediction systems for electricity spot trading with external environment perception and dynamic adaptability is constructed. Policy text is incorporated as a structured input into the prediction system, which significantly improves the model's responsiveness to policy changes. At the same time, by combining multi-granularity market data and reinforcement learning pricing strategies, the prediction accuracy, decision-making intelligence level, and market adaptability are enhanced, effectively supporting power companies in making scientific decisions and controlling risks in complex spot markets. Attached Figure Description

[0062] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a flowchart of the spot transaction forecasting method of the power marketing management information platform in Example 1. Detailed Implementation

[0064] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0065] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0066] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0067] Example, refer to Figure 1 This embodiment of the invention provides a spot trading forecasting method for an electricity marketing management information platform, comprising the following steps:

[0068] S1. Use text semantic modeling methods to extract semantic features from electricity market-related policy texts to obtain policy semantic vector representations;

[0069] Furthermore, the text of policy documents related to the electricity market is input into the pre-trained language model BERT, and contextual semantic information is extracted through a multi-layer Transformer structure, outputting a 768-dimensional high-dimensional semantic vector. ;

[0070] Based on high-dimensional semantic vectors Principal component analysis (PCA) was used to reduce the dimensionality of the feature vector, reducing redundant dimensions while preserving the main semantic information, resulting in a low-dimensional policy semantic feature vector. ,in <768;

[0071] It should be noted that this embodiment uses the BERT model to perform semantic modeling of policy texts. Based on its powerful contextual understanding capabilities in the field of natural language processing, compared with traditional keyword extraction or TF-IDF methods, BERT can more accurately capture the deep semantic information in policy documents, laying the foundation for subsequent quantitative analysis. At the same time, the high-dimensional semantic vectors are reduced through principal component analysis (PCA), which not only reduces the dimension of the model input and improves computational efficiency, but also retains the main semantic features, avoids information redundancy from interfering with the prediction model, and ensures that the policy sensitivity function constructed subsequently has good generalization ability and stability.

[0072] S2. Based on the policy semantic vector representation, construct a policy sensitivity function, calculate the quantitative indicator of the policy's impact on market prices, and obtain the policy sensitivity value;

[0073] Furthermore, nonlinear mapping functions are used to transform policy semantic features into numerical indicators that reflect the degree of policy impact on market prices;

[0074] The nonlinear mapping function is:

[0075] ;

[0076] in, Indicates the first The policy sensitivity value at any given time is used to measure the direction and intensity of the impact of current policies on fluctuations in electricity spot market prices;

[0077] It should be noted that the nonlinear mapping function constructed in this embodiment is used to transform policy semantic features into numerical indicators that reflect the degree of policy impact. Its core purpose is to establish a quantifiable correlation between policy variables and market prices. By introducing learnable weight parameters and bias terms, the function enables the model to adaptively identify the impact weights of different policy elements on market fluctuations. The hyperbolic tangent function is used to limit the output to the interval [-1, 1], which facilitates model understanding and comparison. The design breaks through the limitations of traditional prediction models that ignore policy factors or only perform qualitative processing, and significantly enhances the model's ability to perceive and respond to changes in the external environment.

[0078] S3. Time series coding method is used to perform feature fusion processing on power market operation data, meteorological data and load demand data under multiple time scales to obtain time series feature codes;

[0079] Furthermore, electricity market operation data, meteorological data, and load demand data are collected and organized, and the data windows are divided into three granularities: hourly, daily, and weekly, forming three types of raw datasets. ;

[0080] Standardize each dataset separately to eliminate the differences in the units of measurement between different variables;

[0081] Extract statistical and trend features at each time granularity;

[0082] The above features are fused using a time series encoder. The time series encoder adopts a multi-head self-attention mechanism model based on the Transformer structure, which sequentially performs position encoding addition, multi-head attention calculation, feedforward neural network processing and residual connection operation.

[0083] Output time series feature encoding , as one of the inputs to the subsequent prediction model;

[0084] It should be noted that this embodiment adopts a multi-timescale fusion strategy to address the periodicity and uncertainty of price fluctuations in the electricity spot market. By extracting statistical and trend features from hourly, daily, and weekly data respectively, it can comprehensively depict the short-term fluctuation patterns and long-term operating trends of the market. Furthermore, a time series encoder based on the Transformer structure is used for unified encoding to effectively capture cross-cycle time dependencies and improve the model's ability to understand complex market behaviors. Compared with single-granularity modeling, this method significantly improves the robustness and generalization ability of the prediction model, and is particularly suitable for accurate prediction needs in high-frequency trading and dynamic scheduling scenarios.

[0085] S4. The policy sensitivity value and time series feature code are jointly input, trained through a hybrid neural network model with attention mechanism, and the predicted value of the electricity spot market price at future time is output.

[0086] Furthermore, time series features are encoded. Policy sensitivity value Perform a concatenation operation to generate a joint input feature vector. ;

[0087] The joint input feature vector is input into a hybrid neural network model containing convolutional layers, long short-term memory layers, and attention modules for modeling processing;

[0088] Modeling processes include:

[0089] The trend pattern within a local time window is extracted using a one-dimensional convolutional layer, and a local feature representation is output. ;

[0090] Local features are input into the LSTM layer to model long-term dependencies, and the output is a sequence of hidden states. ;

[0091] The importance weights of each time step are calculated using an attention mechanism. After weighted fusion, context-aware features are obtained. ;

[0092] Finally, a fully connected layer is used to perform a nonlinear transformation on the extracted features, outputting the predicted electricity spot market price at future time t+1. ;

[0093] The output calculation expression of the hybrid neural network model is as follows:

[0094] ;

[0095] In the formula, and For output layer parameters;

[0096] It should be noted that the CNN-LSTM-Attention hybrid neural network model constructed in this embodiment fully integrates three major capabilities: local trend extraction, time series modeling, and key feature focusing. Among them, the convolutional layer is used to capture short-term price fluctuation patterns, the LSTM layer is used to model long-term time dependencies, and the attention mechanism gives the model the ability to selectively focus on key historical information, thereby improving prediction accuracy. In addition, the policy sensitivity value is introduced as an external variable into the model input, realizing effective modeling of policy-driven price fluctuations. The model structure not only improves prediction performance but also enhances the model's interpretability and adaptability to market changes, providing high-quality data support for subsequent pricing decisions.

[0097] S5. Using a reinforcement learning strategy with market price forecasts as state input, construct a pricing decision model to generate dynamic pricing suggestions for power generation companies or electricity sales companies.

[0098] Furthermore, the market price forecast As the state input for reinforcement learning, the action space is defined as the set of pricing strategies that power generation companies or electricity sales companies can choose from. Each action This corresponds to a specific price level;

[0099] The reward function is designed to maximize the expected return per unit time, while taking into account the probability of a transaction and risk control constraints.

[0100] The reward function is defined as follows:

[0101] ;

[0102] In the formula, The quote value corresponding to the quote action. For marginal cost, This is the expected transaction volume for this price, fitted based on historical data. This is the probability factor for a transaction. Price deviation penalty coefficient;

[0103] The strategy is trained using either Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) algorithm to generate optimal bid suggestions and output dynamic bid decision results. This refers to the optimal bidding action under the current circumstances;

[0104] It should be noted that the reinforcement learning strategy used in this embodiment to construct the pricing decision model is to achieve intelligent closed-loop control from price prediction to pricing recommendation. By using the market price prediction value as the state input and defining a reasonable action space and reward function, the model can find the optimal balance between maximizing revenue and minimizing risk. The application of Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) algorithm ensures the stability and convergence of the strategy training process. At the same time, it supports dynamic adjustment of the pricing strategy to adapt to the ever-changing market environment. The method changes the inefficient mode of traditional human experience-driven pricing, realizes the automation and intelligence of pricing decision-making, and helps power generation companies or electricity sales companies gain stronger competitiveness in the spot market.

[0105] S6. Based on the rolling window update mechanism, the latest market feedback data is periodically obtained to perform online incremental updates on the hybrid neural network model and the pricing decision model;

[0106] Furthermore, set a fixed time interval. Get the latest market feedback data, including actual transaction prices, transaction volume, and policy updates;

[0107] Based on the latest market feedback data, incremental learning is used to fine-tune and update the parameters of the hybrid neural network model;

[0108] The update formula is as follows:

[0109] ;

[0110] In the formula, These are the parameters of the original model. For learning rate, For loss function, This is a newly added sample;

[0111] At the same time, the historical experience pool of the reinforcement learning policy model is updated by adding new state-action-reward triples. This enhances the generalization ability of the pricing decision model under new market conditions;

[0112] After each model update, calculate and record the changes in the model's performance metrics before and after the update, including:

[0113] Mean absolute percentage error ;

[0114] Root mean square error ;

[0115] Predicted direction accuracy ;

[0116] The performance evaluation results, along with the model update log and parameter change records, are uploaded to the power marketing management information platform database so that system administrators can view historical version comparisons and model evolution trends.

[0117] It should be noted that the rolling window update mechanism introduced in this embodiment is to address the model degradation problem caused by frequent updates to electricity spot market data and continuous policy changes. By setting a fixed time interval to obtain the latest market feedback data and perform incremental learning, model biases can be corrected in a timely manner, maintaining the timeliness and accuracy of prediction results. At the same time, the historical experience pool of the reinforcement learning strategy model is updated synchronously, enhancing its adaptability under new market conditions. Combined with the model performance evaluation system, MAPE, RMSE, and directional accuracy indicators, the model evolution process can be effectively monitored, assisting system administrators in version management and strategy optimization, and ensuring the long-term stable operation of the entire prediction and decision-making system.

[0118] This embodiment also provides a computer device applicable to the spot trading forecasting method of the power marketing management information platform, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the spot trading forecasting method of the power marketing management information platform as proposed in the above embodiment.

[0119] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0120] This embodiment also provides a storage medium on which a computer program is stored. When executed by a processor, the program implements the spot trading prediction method for the power marketing management information platform as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0121] In summary, this invention constructs an intelligent prediction system for electricity spot trading with external environment perception and dynamic adaptability by introducing policy semantic modeling, policy sensitivity measurement, multi-timescale feature fusion, and a hybrid neural network prediction model based on attention mechanisms. By incorporating policy text as structured input into the prediction system, the system significantly improves the model's responsiveness to policy changes. Furthermore, by combining multi-granularity market data and reinforcement learning pricing strategies, the system enhances prediction accuracy, decision-making intelligence, and market adaptability, effectively supporting power companies in making scientific decisions and controlling risks in complex spot markets.

[0122] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A spot transaction forecasting method for an electricity marketing management information platform, characterized by: include: A text semantic modeling method is used to extract semantic features from electricity market-related policy texts to obtain policy semantic vector representations. Based on the policy semantic vector representation, a policy sensitivity function is constructed to calculate a quantitative indicator of the policy's impact on market prices, thereby obtaining the policy sensitivity value. Time series coding methods are used to perform feature fusion processing on electricity market operation data, meteorological data, and load demand data at multiple time scales to obtain time series feature codes. The policy sensitivity value and the time series feature code are jointly input, trained through a hybrid neural network model containing an attention mechanism, and the predicted value of the electricity spot market price at future time is output. Using a reinforcement learning strategy with the market price forecast as the state input, a pricing decision model is constructed to generate dynamic pricing suggestions for power generation companies or electricity sales companies. Based on the rolling window update mechanism, the latest market feedback data is periodically acquired to perform online incremental updates on the hybrid neural network model and the pricing decision model.

2. The spot transaction forecasting method for the power marketing management information platform as described in claim 1, characterized in that: The method of extracting semantic features from electricity market-related policy texts using text semantic modeling to obtain policy semantic vector representations includes the following steps: The text of policy documents related to the electricity market is input into the pre-trained language model BERT. The contextual semantic information is extracted through a multi-layer Transformer structure, and a high-dimensional semantic vector of 768 dimensions is output. ; Based on the high-dimensional semantic vector Principal component analysis (PCA) was used to reduce the dimensionality of the feature vector, reducing redundant dimensions while preserving the main semantic information, resulting in a low-dimensional policy semantic feature vector. ,in <768.

3. The spot transaction forecasting method for the power marketing management information platform as described in claim 2, characterized in that: The process of constructing a policy sensitivity function based on the policy semantic vector representation, calculating a quantitative indicator of the policy's impact on market prices, and obtaining a policy sensitivity value includes the following steps: Nonlinear mapping functions are used to transform policy semantic features into numerical indicators that reflect the degree of policy impact on market prices; The nonlinear mapping function is: ； in, Indicates the first The policy sensitivity value at any given time is used to measure the direction and intensity of the impact of current policies on fluctuations in electricity spot market prices.

4. The spot transaction forecasting method for the power marketing management information platform as described in claim 3, characterized in that: The method employs time series coding to perform feature fusion processing on electricity market operation data, meteorological data, and load demand data across multiple time scales to obtain time series feature codes, including the following steps: We collected and organized electricity market operation data, meteorological data, and load demand data, and divided the data into three granularities: hourly, daily, and weekly, forming three types of raw datasets. ; Standardize each dataset separately to eliminate the differences in the units of measurement between different variables; Extract statistical and trend features at each time granularity; The above features are fused using a time series encoder. The time series encoder adopts a multi-head self-attention mechanism model based on the Transformer structure, and sequentially performs position encoding addition, multi-head attention calculation, feedforward neural network processing and residual connection operation. Output time series feature encoding This serves as one of the inputs for subsequent prediction models.

5. The spot transaction forecasting method for the power marketing management information platform as described in claim 4, characterized in that: The step of jointly inputting the policy sensitivity value and the time series feature code, training the model through a hybrid neural network including an attention mechanism, and outputting a predicted value for the future electricity spot market price includes the following steps: Encode the time series features Policy sensitivity value Perform a concatenation operation to generate a joint input feature vector. ; The joint input feature vector is input into a hybrid neural network model containing convolutional layers, long short-term memory layers, and attention modules for modeling processing; The modeling process includes: The trend pattern within a local time window is extracted using a one-dimensional convolutional layer, and a local feature representation is output. ; Local features are input into the LSTM layer to model long-term dependencies, and the output is a sequence of hidden states. ; The importance weights of each time step are calculated using an attention mechanism. After weighted fusion, context-aware features are obtained. ; Finally, a fully connected layer is used to perform a nonlinear transformation on the extracted features, outputting the predicted electricity spot market price at future time t+1. ; The output calculation expression of the hybrid neural network model is as follows: ； In the formula, and These are the output layer parameters.

6. The spot transaction forecasting method for the power marketing management information platform as described in claim 5, characterized in that: The process of employing a reinforcement learning strategy, using the market price forecast as the state input, to construct a pricing decision model and generate dynamic pricing recommendations for power generation companies or electricity sales companies includes the following steps: The market price forecast As the state input for reinforcement learning, the action space is defined as the set of pricing strategies that power generation companies or electricity sales companies can choose from. Each action This corresponds to a specific price level; The reward function is designed to maximize the expected return per unit time, while taking into account the probability of a transaction and risk control constraints. The reward function is defined as follows: ； In the formula, The quote value corresponding to the quote action. For marginal cost, This is the expected transaction volume for this price, fitted based on historical data. This is the probability factor for a transaction. Price deviation penalty coefficient; The strategy is trained using either Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) algorithm to generate optimal bid suggestions and output dynamic bid decision results. This refers to the optimal bidding action under the current circumstances.

7. The spot transaction forecasting method for the power marketing management information platform as described in claim 6, characterized in that: The method of periodically acquiring the latest market feedback data based on the rolling window update mechanism and performing online incremental updates to the hybrid neural network model and the pricing decision model includes the following steps: Set a fixed time interval Get the latest market feedback data, including actual transaction prices, transaction volume, and policy updates; Based on the latest market feedback data, the parameters of the hybrid neural network model are fine-tuned and updated using an incremental learning approach. The update formula is as follows: ； In the formula, These are the parameters of the original model. For learning rate, For loss function, This is a newly added sample; Simultaneously, the historical experience pool of the reinforcement learning policy model is updated by adding new state-action-reward triples. This enhances the generalization ability of the pricing decision model under new market conditions.

8. The spot transaction forecasting method for the power marketing management information platform as described in claim 7, characterized in that: The record update process and performance evaluation results are uploaded to the power marketing management information platform database, including: After each model update, calculate and record the changes in the model's performance metrics before and after the update, including: Mean absolute percentage error ; Root mean square error ; Predicted direction accuracy ; The performance evaluation results, along with the model update logs and parameter change records, are uploaded to the power marketing management information platform database so that system administrators can view historical version comparisons and model evolution trends.

9. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the spot transaction prediction method for the power marketing management information platform as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the spot transaction prediction method for the power marketing management information platform as described in any one of claims 1 to 8.

Images