Power transaction optimization method based on multi-dimensional features and intelligent prediction

Abstract

The invention relates to the technical field of power transaction, in particular to a power transaction optimization method based on multi-dimensional features and intelligent prediction, which comprises the following steps: S1, collecting and preprocessing multi-source data of power system operation, weather, economy, policies and the like; s2, classifying data according to constraint dimensions of time, users, power supplies and power grids, extracting feature vectors and splicing the feature vectors into multi-dimensional feature vectors so as to construct a historical database; s3, training and learning a mapping relation through a dual-output LSTM model, and outputting short, medium and long-term power load and market price prediction vectors; s4, risk indexes are obtained based on historical data statistical analysis, and transaction uncertainty is quantified; and S5, in combination with the prediction result and the risk correction item, establishing an objective function by taking the benefit maximization of the transaction subject as an optimization objective, setting constraint conditions and solving, and updating an optimal transaction scheme regularly. The prediction precision can be improved, the scheme stability is enhanced by embedding the risk index, and the market change is dynamically updated and adapted.

Description

Technical Field

[0001] This invention relates to the field of power trading technology, and specifically to a power trading optimization method based on multi-dimensional features and intelligent prediction. Background Technology

[0002] As the "dual carbon" goals are advanced, the electricity market trading model is evolving from traditional bilateral transactions to a spot market with multiple participants, high frequency, and strong volatility. This brings with it the problem of "unstable on-grid electricity volume and revenue that is greatly affected by market fluctuations." Power generation companies need to optimize their power generation and sales strategies by accurately predicting changes in electricity load to avoid over-generating electricity and selling it at low prices. Electricity purchasers, on the other hand, need to rely on load forecasts to rationally plan the timing and amount of electricity purchases to reduce electricity purchase costs.

[0003] In the prior art, such as the Chinese invention patent with authorization announcement number CN117764637B, a method and system for power trading based on multi-dimensional data analysis is disclosed. This method obtains multi-dimensional power trading data and power grid supply flow data, and generates an adaptive load optimization network structure and constructs a power trading strategy model through steps such as feature extraction, time series analysis, peak demand calculation, node network construction, and load analysis. However, it still has the following shortcomings: (1) It directly relies on neural networks for feature extraction and does not integrate meteorological, economic, and policy impact data, making it difficult for the model to capture the fluctuation pattern of power load in complex scenarios; (2) It focuses on load defect analysis and network structure optimization, ignoring core risks such as price and quantity mismatch in power trading, and does not effectively embed risk indicators into the trading optimization model, resulting in a weak risk resistance capability of the trading scheme. The forecasting system is inadequate: a multi-timescale power load forecasting system has not been established, and a dynamic correction mechanism for forecast results is lacking. This makes it difficult for trading entities to formulate dynamic optimization strategies based on load changes over different periods, and hinders the full utilization of peak-valley price differences to maximize profits. For example, the Chinese invention patent application with publication number CN120387537A discloses a method for coordinating the optimization of electricity volume and price in the power trading market. It proposes to achieve the balance of power plant revenue and improve the stability of transactions by summarizing the power generation plan, calculating historical profit and loss, bilateral adjustment of electricity price, predicting market electricity price and supply and demand matching through LSTM model. However, the existing technology still has the following shortcomings: (1) Single prediction dimension: It only relies on historical transaction data (power generation, electricity price) for prediction, without integrating multiple sources of influence factors such as meteorology (wind speed, temperature), economy (industrial operating rate), and policy (peak-shifting electricity consumption), resulting in insufficient prediction accuracy in scenarios with a high proportion of renewable energy access; (2) Lack of dynamic compensation mechanism: It cannot cope with real-time market fluctuations (such as sudden rise and fall of electricity price, sudden change of load) by balancing revenue through static constraints, and has weak risk resistance capability in transactions; (3) Lagging feedback adjustment: It has not established a real-time monitoring mechanism, and the prediction model and transaction strategy cannot dynamically adapt to changes in market rules (such as adjustment of deviation assessment coefficient). Long-term operation is prone to accuracy decay, making it difficult to continuously support optimization decision-making.

[0004] Therefore, there is an urgent need for a method that integrates multi-factor intelligent forecasting, dynamic trading operations, and adaptive compensation to solve the problems of low forecasting accuracy, weak risk resistance, and delayed feedback in existing technologies, so as to increase the amount of electricity traded to the grid and maximize the revenue of enterprises from power generation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an optimized power trading method based on multi-dimensional features and intelligent prediction, which can improve prediction accuracy, resist market fluctuations, ensure trading stability, and ultimately achieve a double increase in grid-connected electricity and power generation revenue.

[0006] The basic solution provided by this invention is a power trading optimization method based on multi-dimensional features and intelligent prediction, comprising the following steps: S1. Multi-source data acquisition and preprocessing: Acquire multi-source data, including power system operation data, core meteorological data, regional economic indicators and policy data, and preprocess the multi-source data to obtain multi-source standardized data. S2. Multi-dimensional feature vector extraction and historical database construction: Multi-source standardized data is classified from multiple dimensions, including time dimension, user dimension, power supply dimension, and grid constraint dimension. Feature vectors affecting power load and power market price under each dimension are extracted and concatenated to obtain multi-dimensional feature vectors. A historical database is constructed based on the multi-dimensional feature vectors. S3. Model Training and Prediction: Construct a dual-output LSTM prediction model, train it based on a historical database, learn the mapping relationship between multi-dimensional feature vectors and power load and power market price, and output power load prediction vector and power market price prediction vector for the corresponding time period based on the multi-dimensional feature vectors. The power load prediction vector and the power market price prediction vector both contain short-term, medium-term and long-term prediction results. S4. Risk Assessment: Based on statistical analysis of historical multi-source standardized data, risk indicators are obtained to quantify the impact of uncertainty in the trading process. These risk indicators include load deviation correction coefficient, price volatility correction coefficient, and policy impact correction coefficient. S5. Dynamic Trading Optimization Calculation: Based on the predicted power load, predicted power market price, and the quantitative correction terms corresponding to the risk indicators, a dynamic trading optimization objective function is established. The optimization objective is to maximize the interests of the trading entities. Constraints are set and solved to obtain the optimal trading scheme. Every preset update time interval, the objective function is re-solved based on the updated predicted values ​​and risk indicators to update the optimal trading scheme.

[0007] The principle of this invention lies in constructing a complete process of data collection, feature extraction, intelligent prediction, risk quantification, and transaction scheme optimization. Through multi-source data integration and intelligent algorithm collaboration, it achieves precision and dynamism in power trading. Specifically: By collecting multi-source data on power system operation, meteorology, economy, and policy, it covers the core factors affecting power load and market prices, solving the problem of low prediction accuracy in complex scenarios with a single data dimension; by preprocessing the data to eliminate outliers, missing values, and dimensional differences, it provides a data foundation for subsequent feature extraction and model training; by classifying data according to four dimensions—time, user, power source, and grid—and extracting feature vectors, it forms a structured multi-dimensional feature vector through vector concatenation and stores it in a historical database. This preserves the independent influence patterns of each dimension while integrating the correlation information between dimensions, enabling the model to learn a more comprehensive mapping relationship; by constructing a dual-output... The LSTM prediction model, trained on a historical database, leverages its ability to capture time-series data dependencies, simultaneously outputting load and price prediction vectors. It includes results across short, medium, and long-term timescales, providing longer-term decision-making reference data. Based on historical data statistical analysis, risk indicators are derived to quantify uncertainties such as market volatility and prediction bias, providing a quantitative basis for trading optimization. A dynamic trading optimization objective function is constructed by combining prediction results, investor demands, and risk correction terms, and the optimal trading solution is obtained through solving. Simultaneously, a preset update interval is set to achieve dynamic iteration of prediction values ​​and risk indicators, ensuring that the trading solution continuously adapts to market changes.

[0008] The beneficial effects of this invention are as follows: It combines multi-source data acquisition with multi-dimensional feature extraction, improving prediction accuracy through multi-dimensional feature training. The dual-output model simultaneously predicts load and price, covering short, medium, and long-term timescales, providing a more complete reference for trading decisions. Furthermore, it quantifies the impact of uncertainty through risk assessment and embeds risk indicators into the dynamic trading optimization objective function, enabling the trading plan to achieve profit maximization while also considering risk control requirements, thus correcting the impact of various risks on the dynamic trading optimization objective function results. Finally, based on the updated predicted values ​​and risk indicators, the trading plan is periodically and continuously updated and optimized to adapt to real-time market changes.

[0009] Furthermore, the preprocessing described in S1 includes data cleaning and data normalization. Data cleaning includes removing outliers and using linear interpolation to fill in missing data. The data normalization process uses the min-max standardization method, with the following formula:

[0010] in, The original data values, The minimum value in the dataset. For the maximum value of the dataset, These are the normalized data values.

[0011] Data cleaning removes outliers caused by sensor malfunctions and transmission anomalies, and fills in missing data using linear interpolation, ensuring data integrity and reliability and preventing outliers from interfering with subsequent model training. The min-max normalization method maps all data to the [0,1] interval, eliminating the dimensional differences between different types of data (such as the differences in units and numerical ranges of temperature, electricity price, and load data), improving training stability and convergence speed, and thus ensuring prediction accuracy.

[0012] Furthermore, S2 classifies multi-source standardized data and extracts feature vectors affecting electricity load and electricity market prices across various dimensions, specifically including: From the time dimension, it is divided into spring, summer, autumn and winter according to the season, weekdays, weekends and holidays according to the date, and peak period, ordinary period and valley period according to the time period. The seasonal factor feature vector, date weight feature vector and time period type feature vector are extracted. From the user perspective, by distinguishing between industrial users, commercial users, and residential users, feature vectors of user electricity load proportion characteristics and load fluctuation coefficients are extracted; From the power source perspective, based on the actual output data of thermal power, hydropower, wind power, photovoltaic power, and energy storage, we extract the characteristic variables of the proportion of new energy output, the amplitude of output fluctuation, and the characteristic vector of power generation cost coefficient. From the perspective of power grid constraints, based on transmission line capacity and substation load rate, feature vectors of power grid congestion early warning factor, incremental absorption space ratio, and transmission loss coefficient are extracted.

[0013] Feature vectors were extracted along four dimensions, enabling a structured decomposition of factors influencing electricity load and market prices: the time dimension captures the periodic fluctuations caused by seasons, dates, and time periods; the user dimension reflects the impact of different user types' electricity consumption behaviors on load and prices; the power source dimension adapts to scenarios with high proportions of renewable energy access, quantifying the impact of renewable energy output fluctuations and generation costs on the market; and the grid constraint dimension considers the constraints of grid physical limitations on power consumption and price formation. The precise extraction and concatenation of these feature vectors allows the model to focus on learning mapping relationships among core influencing factors, improving prediction accuracy.

[0014] Furthermore, the power load forecast vector and the power market price forecast vector described in S3 are composed in the same way, and are distributed in the order of short-term, medium-term and long-term: the short-term forecast part covers the next 24 hours and includes the forecast values ​​of each hour within the period, with a total of 24 dimensions; the medium-term forecast part covers the next 7 days and includes the forecast values ​​of each day within the period, with a total of 7 dimensions; the long-term forecast part covers the next 4 weeks and includes the forecast values ​​of each week within the period, with a total of 4 dimensions.

[0015] The output format is standardized and parsable. The short-term 24-hour hourly forecast is suitable for the high-frequency decision-making needs of intraday spot trading, the medium-term 7-day daily forecast supports weekly contract adjustments, and the long-term 4-week weekly forecast provides a reference for monthly and quarterly trading planning.

[0016] Furthermore, the load deviation correction coefficient mentioned in S4 is calculated based on power system operation data from historically collected multi-source standardized data, and the calculation formula is as follows:

[0017] in, This is the load deviation correction factor. This represents the maximum difference between the actual and predicted power load values ​​within a time period T. Let T be the vector of actual power load values ​​at each time point within a period T. Let T be the vector of predicted electricity load values ​​at each time point within a period T. It is the average value of the actual power load at each time point within the period T; The price fluctuation correction coefficient is calculated based on electricity market price data from historically collected multi-source standardized data. The calculation formula is as follows:

[0018] in, This is a price fluctuation correction factor. This represents the maximum difference between the actual and predicted electricity market prices within a time period T. Let T be the vector of actual electricity market prices at each point in time within a period T. Let T be the vector of predicted electricity market prices at various points in time within a period T. This represents the average of the actual electricity market prices at each point in time within the period T. The policy impact correction coefficient is obtained based on policy data statistics from historically collected multi-source standardized data, denoted as... The value range is 0-0.15.

[0019] The system achieves precise quantification of trading risks through three types of risk indicators: the load deviation correction coefficient quantifies the risk of quantity-price mismatch caused by load forecast deviation; the price fluctuation correction coefficient quantifies the risk of electricity price forecast deviation and market fluctuation; and the policy impact correction coefficient quantifies the uncertainty brought about by policy adjustments. These three indicators cover the core risk types in electricity trading. The load deviation correction coefficient and price fluctuation correction coefficient are quantified based on historical data statistics, while the value of the policy impact correction coefficient is reasonably determined based on historical data, so that risks can be accurately embedded in the optimization objective function.

[0020] Furthermore, the trading entities mentioned in S5 include electricity sellers and electricity buyers, and maximizing the interests of the trading entities includes maximizing the profits of the electricity sellers or minimizing the costs of the electricity buyers. The dynamic trading optimization objective function includes: When the trading entity is an electricity seller, the dynamic trading optimization objective is to maximize electricity sales profit, as shown in the formula:

[0021] When the trading entity is an electricity purchaser, the dynamic trading optimization objective is to minimize the electricity purchase cost, as shown in the formula:

[0022] in, For the set of trading sessions, for Electricity market price forecast for the period, for Electricity sales during the period and meeting the requirements , To ensure that power generation companies operate at full capacity. To indicate A function of the total cost of electricity sales within a given time period. for Electricity purchase within the specified time period and meeting the requirements , To meet the minimum electricity demand of the electricity purchasing entity, To indicate A function of line loss cost within a time period.

[0023] The objective function is designed to address the differentiated needs of electricity sellers and buyers, matching the interests of different trading entities; this is achieved through risk indicator correction terms. By embedding a dynamic transaction optimization objective function, the goal of maximizing profits and mitigating risks is achieved. Electricity sellers reduce revenue volatility by deducting risk penalties, while electricity buyers avoid cost overruns by adding risk contingency items, thus improving the stability of the transaction scheme.

[0024] Furthermore, the constraints in S5 include generator set operating parameter constraints, environmental emission constraints, and power grid transmission capacity constraints; the preset update time interval is 1 hour.

[0025] Clear variable constraints ensure the feasibility of the trading scheme (such as generator output limits and grid transmission capacity limits), preventing optimization results from deviating from actual operating scenarios; the preset 1-hour update interval is adapted to the interval of short-term forecast results, ensuring that the trading scheme can respond quickly to market changes. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating an embodiment of an electricity trading optimization method based on multi-dimensional features and intelligent prediction according to the present invention. Detailed Implementation

[0027] The following detailed description illustrates the specific implementation method: The basic implementation examples are as follows: Figure 1 As shown: A power trading optimization method based on multi-dimensional features and intelligent prediction includes the following steps: S1. Multi-source data acquisition and preprocessing: Acquire multi-source data, including power system operation data, core meteorological data, regional economic indicators and policy data, and preprocess the multi-source data to obtain multi-source standardized data. S2. Multi-dimensional feature vector extraction and historical database construction: Multi-source standardized data is classified from multiple dimensions, including time dimension, user dimension, power supply dimension, and grid constraint dimension. Feature vectors affecting power load and power market price under each dimension are extracted and concatenated to obtain multi-dimensional feature vectors. A historical database is constructed based on the multi-dimensional feature vectors. S3. Model Training and Prediction: Construct a dual-output LSTM prediction model, train it based on a historical database, learn the mapping relationship between multi-dimensional feature vectors and power load and power market price, and output power load prediction vector and power market price prediction vector for the corresponding time period based on the multi-dimensional feature vectors. The power load prediction vector and the power market price prediction vector both contain short-term, medium-term and long-term prediction results. S4. Risk Assessment: Based on statistical analysis of historical multi-source standardized data, risk indicators are obtained to quantify the impact of uncertainty in the trading process. These risk indicators include load deviation correction coefficient, price volatility correction coefficient, and policy impact correction coefficient. S5. Dynamic Trading Optimization Calculation: Based on the predicted power load, predicted power market price, and the quantitative correction terms corresponding to the risk indicators, a dynamic trading optimization objective function is established. The optimization objective is to maximize the interests of the trading entities. Constraints are set and solved to obtain the optimal trading scheme. Every preset update time interval, the objective function is re-solved based on the updated predicted values ​​and risk indicators to update the optimal trading scheme.

[0028] Furthermore, the preprocessing described in S1 includes data cleaning and data normalization. Data cleaning includes removing outliers and using linear interpolation to fill in missing data. The data normalization process uses the min-max standardization method, with the following formula:

[0029] in, The original data values, The minimum value in the dataset. The maximum value in the dataset. These are the normalized data values.

[0030] In this embodiment, the power system operation data includes power load data, bus voltage data, transmission line load rate data, substation operation status data, planned output data of various power sources (thermal power, hydropower, wind power, photovoltaic, and energy storage), unit start-up and shutdown plan data, and electricity market price data (intraday spot price, peak-valley time-of-use price, and medium- and long-term contract price); core meteorological data includes daily average temperature data, hourly wind speed data, precipitation data, and sunshine duration data; regional economic indicators include industrial operating rate data, GDP growth rate data, and the proportion of the tertiary industry data; and policy data includes carbon emission control policy data, peak-shifting electricity consumption policy data, new energy subsidy policy data, and electricity market trading rule adjustment data.

[0031] Furthermore, S2 classifies multi-source standardized data and extracts feature vectors affecting electricity load and electricity market prices across various dimensions, specifically including: From the time dimension, it is divided into spring, summer, autumn and winter according to the season, weekdays, weekends and holidays according to the date, and peak period, ordinary period and valley period according to the time period. The seasonal factor feature vector, date weight feature vector and time period type feature vector are extracted. From the user perspective, by distinguishing between industrial users, commercial users, and residential users, feature vectors of user electricity load proportion characteristics and load fluctuation coefficients are extracted; From the power source perspective, based on the actual output data of thermal power, hydropower, wind power, photovoltaic power, and energy storage, we extract the characteristic variables of the proportion of new energy output, the amplitude of output fluctuation, and the characteristic vector of power generation cost coefficient. From the perspective of power grid constraints, based on transmission line capacity and substation load rate, feature vectors of power grid congestion early warning factor, incremental absorption space ratio, and transmission loss coefficient are extracted.

[0032] In this embodiment, each feature vector is quantified based on the correlation analysis between multi-source standardized data and the corresponding dimension's influence, as detailed below: Seasonal factor feature vector: Quantified based on the correlation analysis of historical seasonal related power load, power market price data and meteorological data (temperature, sunshine, etc.), reflecting the intensity of the impact of different seasons on load and price, with a value range of 0.1-0.4; Date-weighted feature vector: It is quantified by statistically analyzing the differences in peak electricity load and average electricity market price for different date types (weekdays, weekends, and holidays), reflecting the impact of date attributes on electricity consumption behavior and market prices, with a value range of 0.8-1.2; Time Period Type Feature Vector: Based on load demand intensity and electricity price difference data during peak, flat, and valley periods, it distinguishes the market supply and demand characteristics of different time periods, with a value range of 1-3; User power load proportion feature vector: It is obtained by statistically analyzing the historical load data of industrial, commercial and residential users and their changing trends, reflecting the impact of user structure on total power load; Load fluctuation coefficient feature vector: calculated by the ratio of the standard deviation to the mean of the load data of various users over the past 30 days, quantifying the power load stability of different user groups; New energy output proportion feature vector: quantified based on the ratio of actual wind power and photovoltaic power output data to total power output data and historical fluctuation patterns; Power output fluctuation amplitude characteristic vector: calculated as the ratio of the difference between the daily maximum and minimum values ​​of new energy power output to the mean, reflecting the uncertainty of new energy power output; Power generation cost coefficient feature vector: The unit power generation cost of various power sources is quantified by combining data such as thermal power fuel cost, hydropower operation and maintenance cost, and new energy subsidy policies; The feature vector of the power grid congestion early warning factor is derived by quantifying the historical load rate of transmission lines, the frequency of congestion occurrence, and the scope of impact, with a value range of 0-1. The incremental space ratio characteristic vector is calculated as the ratio of the rated capacity of the transmission line minus the historical maximum load to the rated capacity of the transmission line, reflecting the regional power absorption potential. Transmission loss coefficient feature vector: It is quantified based on data such as transmission line length, material, and load rate, combined with the transmission loss formula, and reflects the loss level in the power transmission process.

[0033] Furthermore, the power load forecast vector and the power market price forecast vector described in S3 are composed in the same way, and are distributed in the order of short-term, medium-term and long-term: the short-term forecast part covers the next 24 hours and includes the forecast values ​​of each hour within the period, with a total of 24 dimensions; the medium-term forecast part covers the next 7 days and includes the forecast values ​​of each day within the period, with a total of 7 dimensions; the long-term forecast part covers the next 4 weeks and includes the forecast values ​​of each week within the period, with a total of 4 dimensions.

[0034] In this embodiment, the dual-output LSTM prediction model is structured as follows: the input layer receives multi-dimensional feature vectors, the hidden layer uses a 3-layer LSTM network (128 neurons per layer), and the output layer consists of two parallel output heads, outputting a 35-dimensional power load prediction vector and a 35-dimensional power market price prediction vector, respectively. During model training, 80% of the data in the historical database is used as the training set and 20% as the test set. The Adam optimizer (learning rate 0.001) is employed, with mean squared error (MSE) as the loss function. The model is iteratively trained for 500 rounds until the MSE loss value is below 0.001. Example prediction outputs: In the short-term forecast, the predicted load for the third hour is 13200MW, and the predicted power market price is 0.65 yuan / kWh; in the medium-term forecast, the predicted load for the third day (working day) is 178000MW, and the predicted power market price is 0.58 yuan / kWh; in the long-term forecast, the average load for the second week is 1232000MW, and the average weekly price is 0.55 yuan / kWh.

[0035] Furthermore, the load deviation correction coefficient mentioned in S4 is calculated based on power system operation data from historically collected multi-source standardized data, and the calculation formula is as follows:

[0036] in, This is the load deviation correction factor. This represents the maximum difference between the actual and predicted power load values ​​within a time period T. Let T be the vector of actual power load values ​​at each time point within a period T. Let T be the vector of predicted electricity load values ​​at each time point within a period T. It is the average value of the actual power load at each time point within the period T; The price fluctuation correction coefficient is calculated based on electricity market price data from historically collected multi-source standardized data. The calculation formula is as follows:

[0037] in, This is a price fluctuation correction factor. This represents the maximum difference between the actual and predicted electricity market prices within a time period T. Let T be the vector of actual electricity market prices at each point in time within a period T. Let T be the vector of predicted electricity market prices at various points in time within a period T. This represents the average of the actual electricity market prices at each point in time within the period T. The policy impact correction coefficient is obtained based on policy data statistics from historically collected multi-source standardized data, denoted as... The value range is 0-0.15.

[0038] In this embodiment, the period T is set to 30 days, and the policy impact correction coefficient is calculated based on the recent policy data and set to 0.05 (corresponding to the cost fluctuation impact brought about by the adjustment of carbon emission control policies).

[0039] Furthermore, the trading entities mentioned in S5 include electricity sellers and electricity buyers, and maximizing the interests of the trading entities includes maximizing the profits of the electricity sellers or minimizing the costs of the electricity buyers. The dynamic trading optimization objective function includes: When the trading entity is an electricity seller, the dynamic trading optimization objective is to maximize electricity sales profit, as shown in the formula:

[0040] When the trading entity is an electricity purchaser, the dynamic trading optimization objective is to minimize the electricity purchase cost, as shown in the formula:

[0041] in, For the set of trading sessions, for Electricity market price forecast for the period, for Electricity sales during the period and meeting the requirements , To ensure that power generation companies operate at full capacity. To indicate A function of the total cost of electricity sales within a given time period. for Electricity purchase within the specified time period and meeting the requirements , The minimum electricity demand of the electricity purchasing entity. To indicate A function of line loss cost within a time period.

[0042] Furthermore, the constraints in S5 include generator set operating parameter constraints, environmental emission constraints, and power grid transmission capacity constraints; the preset update time interval is 1 hour.

[0043] In this embodiment, the constraints on generator unit operating parameters include minimum output (not less than 30% of rated capacity), maximum output (not exceeding 105% of rated capacity), ramp rate constraints (ramp rate of thermal power not exceeding 15% of rated capacity per hour, ramp rate of new energy units not exceeding 30% of rated capacity per hour), and continuous operating time constraints (single continuous operating time of thermal power not less than 72 hours); environmental emission constraints include carbon emission concentration of thermal power plants not exceeding 800 mg / m³. 3 Nitrogen oxide emission concentration not exceeding 100 mg / m³ 3 To meet regional environmental protection policy requirements; power grid transmission capacity constraints include ensuring that the transmission power of each transmission line does not exceed 90% of its rated capacity, so as to avoid safety accidents caused by line overload and ensure the stable operation of the power grid.

[0044] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A power trading optimization method based on multi-dimensional features and intelligent prediction, characterized in that, Includes the following steps: S1. Multi-source data acquisition and preprocessing: Acquire multi-source data, including power system operation data, core meteorological data, regional economic indicators and policy data, and preprocess the multi-source data to obtain multi-source standardized data. S2. Multi-dimensional feature vector extraction and historical database construction: Multi-source standardized data is classified from multiple dimensions, including time dimension, user dimension, power supply dimension, and grid constraint dimension. Feature vectors affecting power load and power market price under each dimension are extracted and concatenated to obtain multi-dimensional feature vectors. A historical database is constructed based on the multi-dimensional feature vectors. S3. Model Training and Prediction: Construct a dual-output LSTM prediction model, train it based on a historical database, learn the mapping relationship between multi-dimensional feature vectors and power load and power market price, and output power load prediction vector and power market price prediction vector for the corresponding time period based on the multi-dimensional feature vectors. The power load prediction vector and the power market price prediction vector both contain short-term, medium-term and long-term prediction results. S4. Risk Assessment: Based on statistical analysis of historical multi-source standardized data, risk indicators are obtained to quantify the impact of uncertainty in the trading process. These risk indicators include load deviation correction coefficient, price volatility correction coefficient, and policy impact correction coefficient. S5. Dynamic Trading Optimization Calculation: Based on the predicted power load, predicted power market price, and the quantitative correction terms corresponding to the risk indicators, a dynamic trading optimization objective function is established. The optimization objective is to maximize the interests of the trading entities. Constraints are set and solved to obtain the optimal trading scheme. Every preset update time interval, the objective function is re-solved based on the updated predicted values ​​and risk indicators to update the optimal trading scheme.

2. The power trading optimization method based on multi-dimensional features and intelligent prediction according to claim 1, characterized in that, The preprocessing described in S1 includes data cleaning and data normalization. Data cleaning includes removing outliers and using linear interpolation to fill in missing data. The data normalization process uses the min-max standardization method, with the following formula: in, The original data values, The minimum value in the dataset. For the maximum value of the dataset, These are the normalized data values.

3. The power trading optimization method based on multi-dimensional features and intelligent prediction according to claim 2, characterized in that, S2 classifies multi-source standardized data and extracts feature vectors that affect electricity load and electricity market prices under various dimensions, specifically including: From the time dimension, it is divided into spring, summer, autumn and winter according to the season, weekdays, weekends and holidays according to the date, and peak period, ordinary period and valley period according to the time period. The seasonal factor feature vector, date weight feature vector and time period type feature vector are extracted. From the user perspective, by distinguishing between industrial users, commercial users, and residential users, feature vectors of user electricity load proportion characteristics and load fluctuation coefficients are extracted; From the power source perspective, based on the actual output data of thermal power, hydropower, wind power, photovoltaic power, and energy storage, we extract the characteristic variables of the proportion of new energy output, the amplitude of output fluctuation, and the characteristic vector of power generation cost coefficient. From the perspective of power grid constraints, based on transmission line capacity and substation load rate, feature vectors of power grid congestion early warning factor, incremental absorption space ratio, and transmission loss coefficient are extracted.

4. The power trading optimization method based on multi-dimensional features and intelligent prediction according to claim 3, characterized in that, The power load forecast vector and the power market price forecast vector described in S3 have the same composition, both distributed in the order of short-term, medium-term, and long-term: the short-term forecast part covers the next 24 hours, including the forecast values ​​of each hour within that period, with a total of 24 dimensions; the medium-term forecast part covers the next 7 days, including the forecast values ​​of each day within that period, with a total of 7 dimensions; and the long-term forecast part covers the next 4 weeks, including the forecast values ​​of each week within that period, with a total of 4 dimensions.

5. The power trading optimization method based on multi-dimensional features and intelligent prediction according to claim 4, characterized in that, The load deviation correction factor mentioned in S4 is calculated based on power system operation data from historically collected multi-source standardized data. The calculation formula is as follows: in, This is the load deviation correction factor. This represents the maximum difference between the actual and predicted power load values ​​within a time period T. Let T be the vector of actual power load values ​​at each time point within a period T. Let T be the vector of predicted electricity load values ​​at each time point within a period T. It is the average value of the actual power load at each time point within the period T; The price fluctuation correction coefficient is calculated based on electricity market price data from historically collected multi-source standardized data. The calculation formula is as follows: in, This is a price fluctuation correction factor. This represents the maximum difference between the actual and predicted electricity market prices within a time period T. Let T be the vector of actual electricity market prices at each point in time within a period T. Let T be the vector of predicted electricity market prices at various points in time within a period T. This represents the average of the actual electricity market prices at each point in time within the period T. The policy impact correction coefficient is obtained based on policy data statistics from historically collected multi-source standardized data, denoted as... The value range is 0-0.

15.

6. The power trading optimization method based on multi-dimensional features and intelligent prediction according to claim 5, characterized in that, The trading entities mentioned in S5 include electricity sellers and electricity buyers. Maximizing the interests of the trading entities includes maximizing the profits of the electricity sellers or minimizing the costs of the electricity buyers. The dynamic trading optimization objective function includes: When the trading entity is an electricity seller, the dynamic trading optimization objective is to maximize electricity sales profit, as shown in the formula: When the trading entity is an electricity purchaser, the dynamic trading optimization objective is to minimize the electricity purchase cost, as shown in the formula: in, For the set of trading sessions, for Electricity market price forecast for the period, for Electricity sales during the period and meeting the requirements , To ensure that power generation companies operate at full capacity. To indicate A function of the total cost of electricity sales within a given time period. for Electricity purchase within the specified time period and meeting the requirements , To meet the minimum electricity demand of the electricity purchasing entity, To indicate A function of line loss cost within a time period.

7. The power trading optimization method based on multi-dimensional features and intelligent prediction according to claim 6, characterized in that, The constraints in S5 include generator set operating parameter constraints, environmental emission constraints, and power grid transmission capacity constraints; the preset update time interval is 1 hour.

Images