Multi-transaction variety optimization method based on clustering scene generation and sensitivity analysis and application thereof

Abstract

The invention discloses a multi-transaction variety optimization method based on clustering scene generation and sensitivity analysis, and the method comprises the following steps: a) collecting and cleaning historical transaction data of a new energy plant station, extracting related features through feature engineering, and generating multi-dimensional feature data; b) generating multiple groups of typical operation scenes by using a clustering analysis method; c) in each clustering scene, carrying out sensitivity analysis, and calculating the influence of each risk factor on the earnings; d) according to a trial result of the sensitivity analysis, carrying out risk sorting on each scene, and formulating a risk index according to the risk sorting; and e) carrying out feedback modification on sensitivity analysis and a clustering model through transaction redisk and error analysis, and carrying out rolling optimization on algorithm parameters according to actual transaction data and market conditions so as to realize dynamic updating and optimization of the model. And the competitiveness and adaptability of the new energy plant station in the electricity market are enhanced.

Description

Technical Field

[0001] The present invention relates to the fields of new energy power generation and electricity market trading, in particular to an optimization method for multiple trading varieties based on clustering scenario generation and sensitivity analysis and its application. Background Art

[0002] With the rapid development of new energy such as wind power and photovoltaic power, the dispatching of power grids faces increasing challenges. The volatility and uncertainty of new energy power generation bring complexity to power market price forecasting, trading strategies, and power grid dispatching. Most of the existing electricity market trading optimization methods rely on point forecasting and deterministic models, ignoring the complexity and variability of the market. Especially in the case of large uncertainty and volatility, the effects of traditional models are poor. Therefore, how to scientifically utilize historical trading data and market volatility characteristics, combine uncertainty modeling with clustering analysis, and optimize the trading strategies of multiple trading varieties for new energy power plants has become the core issue for optimizing new energy trading and improving power grid dispatching efficiency.

[0003] The present invention generates multiple typical market scenarios through clustering analysis, combines sensitivity analysis to rank and adjust key risk factors, and optimizes the trading strategies of new energy power plants. This method can effectively consider uncertainty factors such as market volatility and grid congestion, and provide more accurate medium- and long-term and spot trading strategy recommendations for new energy power plants. Summary of the Invention

[0004] The core objective of the present invention is to provide an optimization method for multiple trading varieties based on clustering scenario generation and sensitivity analysis and its application. This method analyzes the historical trading data of new energy power plants, generates multiple groups of typical operating scenarios, combines sensitivity analysis to evaluate key risk factors, optimizes the trading strategies of new energy power generation and energy storage systems, and improves the economy, robustness, and revenue maximization ability of electricity market trading.

[0005] The technical solution adopted by the present invention to solve the above technical problems is as follows:

[0006] An optimization method for multiple trading varieties based on clustering scenario generation and sensitivity analysis, the method comprising the following steps:

[0007] a) Collect and clean the historical trading data of new energy power plants, including power generation forecasts, historical power generation outputs, medium- and long-term trading volumes and prices, spot price forecasts, spot trading volumes and prices, and data on grid congestion conditions, and extract relevant features through feature engineering to generate multi-dimensional feature data;

[0008] b) Generate multiple groups of typical operation scenarios using clustering analysis methods based on historical data characteristics. The clustering analysis methods are K-means clustering and DBSCAN clustering, which are used to divide historical operation data into different market operation modes, and each market mode represents a typical operation scenario;

[0009] c) In each clustering scenario, conduct sensitivity analysis, considering risk factors in spot trading, including but not limited to spot price fluctuations, new energy power generation capacity fluctuations, and grid congestion. Use Monte Carlo simulation or Latin hypercube sampling method to simulate risk factors and calculate the impact of each risk factor on revenue;

[0010] d) According to the trial calculation results of sensitivity analysis, rank the risks of each scenario, formulate risk indicators based on the risk ranking, and adjust the levels of medium- and long-term and spot trading volume and price according to scenario probabilities and risk indicators to optimize trading decisions;

[0011] e) Through trading review, compare the actual settlement results with the predicted results, analyze the errors, and make feedback modifications to the sensitivity analysis and clustering model. Dynamically optimize the algorithm parameters based on actual trading data and market conditions to achieve dynamic update and optimization of the model.

[0012] Preferably, the clustering analysis method includes the K-means clustering or DBSCAN clustering algorithm. The K-means clustering algorithm is achieved by minimizing the objective function, and the objective function is:

[0013]

[0014] where y i is the clustering category to which sample i belongs, μ k is the center of clustering k, x i is the feature of sample i, and ||·|| is the Euclidean distance.

[0015] Preferably, the sensitivity analysis includes simulating risk factors through Monte Carlo simulation or Latin hypercube sampling method to obtain the probability distribution of risk factors, and calculating the impact of each risk factor change on revenue. The formula is:

[0016]

[0017] where ΔR is the change in total revenue, p i is the occurrence probability of risk factor i, and Δr i is the impact of the change in risk factor i on revenue.

[0018] Preferably, the risk ranking is carried out by calculating the impact of each risk factor on revenue. The formula is:

[0019]

[0020] Among them, RiskRank i is the ranking value of risk factor i, and ΔR j is the impact of risk factor i on the return at the j-th moment, and T is the time period.

[0021] Preferably, the calculation formula for the mid- and long-term and spot trading volume and price adjustment is:

[0022] P adjusted = P base ·(1 + α·ΔR)

[0023] Among them, P adjusted is the adjusted price, P base is the base price, α is the adjustment coefficient, and ΔR is the change in return.

[0024] Preferably, the trading review includes calculating the error between the actual settlement result and the predicted result, and the formula is:

[0025]

[0026] Among them, ∈ is the error, R actual (t) is the actual return at the t-th moment, and R forccast (t) is the predicted return at the t-th moment, and T is the length of the time period.

[0027] Preferably, the rolling optimization includes, after each round of trading, adjusting the parameters of the clustering model according to the actual trading data and the sensitivity analysis results, and using the backpropagation algorithm and the gradient descent method to optimize the parameters of the sensitivity analysis model to improve the prediction accuracy and risk control ability of the model.

[0028] A multi-trading variety optimization application based on clustering scenario generation and sensitivity analysis, characterized in that the steps include:

[0029] a) A data acquisition module for collecting historical trading data of new energy power plants, including data on power generation prediction, historical power generation output, mid- and long-term trading volume and price, spot price prediction, spot trading volume and price, and grid congestion;

[0030] b) A data cleaning module for cleaning the collected data, processing outliers and missing values, and performing standardization and normalization processing on the data;

[0031] c) A feature engineering module for extracting effective features from the cleaned data to generate a feature set for clustering analysis and risk sensitivity analysis;

[0032] d) A clustering analysis module, which is used to generate multiple groups of typical operation scenarios based on feature data and generate typical patterns of the market environment through clustering analysis methods;

[0033] e) A risk analysis module, which is used to conduct sensitivity analysis on risk factors under each clustering scenario, calculate the impact of changes in each risk factor on revenue, conduct risk ranking and formulate risk indicators;

[0034] f) A trading optimization module, which is used to adjust the volume and price of medium - and long - term and spot trading according to the risk analysis results, scenario probabilities and risk indicators, and provide trading decision - making suggestions;

[0035] g) A trading review module, which is used to compare the actual settlement results with the predicted results, calculate the error and feedback to optimize the model parameters to ensure the dynamic adjustment and rolling optimization of the model.

[0036] Preferably, the clustering analysis module adopts the K - means clustering or DBSCAN clustering algorithm, generates multiple market scenarios based on historical trading data, and divides the historical data into different categories by calculating the Euclidean distance or density algorithm. Each category represents a typical market environment.

[0037] Preferably, the risk analysis module simulates the changes of different risk factors through the Monte Carlo simulation method or the Latin hypercube sampling method, analyzes the impact of risk factors on revenue, and generates risk ranking and risk indicators.

[0038] Preferably, the trading optimization module adjusts the volume and price of medium - and long - term and spot trading based on scenario probabilities, risk ranking and risk thresholds, and dynamically adjusts trading strategies according to changes in the trading environment.

[0039] The beneficial effects that can be achieved by the present invention adopting the above - mentioned technical solution are as follows:

[0040] Through clustering analysis and sensitivity analysis, the present invention can accurately identify the key risk factors affecting revenue and provide data - driven decision - making support for trading strategies. According to the probability distribution and risk ranking of market scenarios, it can adjust medium - and long - term and spot trading strategies in real time to improve the revenue and robustness of electricity market trading. Through trading review and rolling optimization, it can continuously improve the accuracy of strategies and adapt to the rapidly changing market environment. Through multi - scenario generation and dynamic decision - making adjustment, it enhances the competitiveness and adaptability of new - energy power plants in the electricity market. The present invention provides a multi - trading - variety optimization method based on clustering scenario generation and sensitivity analysis. Through the combination of in - depth analysis of historical data, clustering to generate typical scenarios, sensitivity analysis, risk ranking and trading strategy optimization, it can more accurately provide optimized medium - and long - term and spot trading decisions for new - energy power plants and improve the economy and robustness of electricity market trading. Detailed implementation manners

[0041] 1. Data cleaning and feature engineering:

[0042] For effective prediction and optimization, it is first necessary to clean and process the historical transaction data of new energy power plants. The data includes the following main parts:

[0043] Power generation prediction data: including the predicted power generation of new energy sources such as photovoltaic and wind power;

[0044] Historical power generation output data: recording the historical actual power generation;

[0045] Medium- and long-term trading volume and price data: including the trading volume and price of medium- and long-term contracts;

[0046] Spot price prediction data: price prediction data for the electricity spot market;

[0047] Spot trading volume and price data: trading volume and transaction price in the electricity spot market;

[0048] Grid congestion data: recording the operating conditions of the grid and grid congestion situations.

[0049] First, data cleaning is carried out, including removing outliers and noise and handling missing data. Then, through feature engineering methods, features related to market fluctuations, power generation capacity, grid congestion, etc. are extracted. These features include:

[0050] Power generation capacity fluctuation: the deviation between historical power generation and predicted power generation;

[0051] Electricity price fluctuation: the fluctuation of the historical electricity spot market price;

[0052] Grid load: the change of grid load in different time periods;

[0053] Price difference: the price difference between the spot market and the medium- and long-term contract market, etc.

[0054] Highly correlated features are selected through methods such as principal component analysis (PCA) and recursive feature elimination (RFE).

[0055] 2. Cluster analysis to generate typical scenarios:

[0056] After processing and cleaning the data, the clustering analysis method is used to generate multiple groups of typical market scenarios. Each cluster represents a market operation mode and can reflect the electricity trading behaviors under different market conditions. The clustering analysis methods include:

[0057] K-means clustering method: This method divides the historical operation day data into multiple clusters according to similarity, and each cluster represents a typical market operation scenario. Its objective function is as follows:

[0058]

[0059] where y i is the clustering category to which sample i belongs, and μ k is the center of cluster k, and x i is the feature of sample i, and ||·|| is the Euclidean distance.

[0060] Through cluster analysis, multiple typical operation scenarios are generated, and each scenario has different market prices, power generation fluctuations, grid loads, and congestion situations.

[0061] 3. Risk Sensitivity Analysis and Risk Ranking

[0062] In electricity market trading, risk factors such as spot price fluctuations, grid congestion, and new energy generation capacity fluctuations have an important impact on revenue. Therefore, the present invention uses sensitivity analysis to evaluate these risk factors. Sensitivity analysis method: We use Monte Carlo simulation or Latin hypercube sampling method to simulate the risk factors, and by changing each risk factor, observe its impact on the total revenue. For example, in the case of electricity price fluctuations, calculate the impact of spot market price fluctuations on the revenue of new energy power plants. Risk factor trial calculation and ranking: Conduct trial calculations on the risk factors (such as electricity price fluctuations, new energy generation capacity fluctuations, etc.) under each clustering scenario, and calculate the impact of changes in risk factors on revenue. Through the trial calculation results, rank them according to the degree of impact of risk factors on revenue. The calculation formula for risk impact is:

[0063]

[0064] where ΔR is the change in total revenue, and p i is the occurrence probability of risk factor i, and Δr i is the impact of the change in risk factor i on revenue.

[0065] Conduct trial calculations on the risk factors (such as electricity price fluctuations, new energy generation capacity fluctuations, etc.) under each clustering scenario, and calculate the impact of changes in risk factors on revenue. Through the trial calculation results, rank them according to the degree of impact of risk factors on revenue. The calculation formula for risk ranking is:

[0066]

[0067] where RiskRank i is the ranking value of risk factor i, ΔR j is the impact of risk factor i on revenue at the jth moment, and T is the time period.

[0068] Based on the risk ranking, different risk indicators are formulated to further determine the risk level of each scenario, and the trading strategies for different scenarios are adjusted through these risk indicators. According to the risk ranking and the trial calculation results, the calculation formula for the risk indicator is as follows:

[0069]

[0070] Among them, R indicator is the risk indicator, w i is the weight of risk factor i, and p i is the probability of risk factor i.

[0071] 4. Optimization of trading strategies based on scenario probability and risk indicators:

[0072] According to the probability distribution of scenarios and the risk ranking results, combined with the actual trading days and the risk preferences of power plants and substations, adjust the trading volume and price levels in the medium- and long-term and spot markets.

[0073] Scenario probability and risk threshold: According to the scenario probability distribution generated from historical data, combined with the risk ranking and the set risk threshold, adjust the trading ratio of medium- and long-term contracts. For high-risk scenarios, reduce the participation in the spot market and increase the proportion of medium- and long-term contracts; for low-risk scenarios, increase the trading volume in the spot market. The adjustment formulas for the trading volume and price of medium- and long-term and spot are as follows:

[0074] P adjusted = P base ·(1 + α·ΔR)

[0075] Among them, P adjusted is the adjusted price, P base is the base price, α is the adjustment coefficient, and ΔR is the change in return.

[0076] 5. Trading review and rolling optimization:

[0077] After the trading is executed, compare the actual settlement with the trial calculation results through trading review, analyze the errors and conduct feedback adjustments to optimize the algorithm model. By calculating the errors between the actual settlement results and the trial calculation results, analyze the actual impact of changes in risk factors on the return, and adjust the parameters of the sensitivity analysis model. The error calculation formula is as follows:

[0078]

[0079] Among them, ∈ is the error, R actual (t) is the actual return at the t-th moment, R forccast (t) is the predicted return at the t-th moment, and T is the time period length.

[0080] Rolling optimization: Continuously optimize the parameters of the clustering algorithm and sensitivity analysis model through real-time data feedback and error adjustment to ensure that in future transactions, new energy power plants can make optimal decisions based on the latest market information and risk situations.

Claims

1. A multi-trading product optimization method based on clustering scenario generation and sensitivity analysis, characterized in that: The method comprises the following steps: a) Collect and clean historical transaction data of new energy plants and stations, including power generation forecasts, historical power generation output, medium- and long-term transaction volume and price, spot price forecasts, spot transaction volume and price, and grid congestion data, and extract relevant features through feature engineering to generate multi-dimensional feature data; b) Based on the characteristics of historical data, a cluster analysis method is used to generate multiple groups of typical operation scenarios, where the cluster analysis method is K-means clustering and DBSCAN clustering, which is used to divide the historical operation data into different market operation modes, each of which represents a typical operation scenario; c) Conduct sensitivity analysis in each clustering scenario, consider the risk factors in spot trading, including but not limited to spot price fluctuations, fluctuations in renewable energy generation capacity and grid congestion, simulate risk factors using Monte Carlo simulation or Latin hypercube sampling, and calculate the impact of each risk factor on returns; d) According to the results of sensitivity analysis, the risk of each scenario is ranked, and risk indicators are formulated according to the risk ranking. The level of medium- and long-term and spot transaction volume and price is adjusted according to the scenario probability and risk indicators to optimize trading decisions; e) Through transaction review, the actual settlement results are compared with the predicted results, the errors are analyzed, and the sensitivity analysis and clustering models are modified in a feedback manner. The algorithm parameters are optimized on a rolling basis according to the actual transaction data and market conditions to achieve dynamic updating and optimization of the model.

2. The multi-trading product optimization method based on clustering scenario generation and sensitivity analysis according to claim 1 is characterized in that: The cluster analysis method includes K-means clustering or DBSCAN clustering algorithm, wherein the K-means clustering algorithm is implemented by minimizing the objective function, and the objective function is: Among them, y i is the cluster category to which sample i belongs, μ k is the center of cluster k, x i is the feature of sample i, and ||·|| is the Euclidean distance.

3. The multi-trading product optimization method based on clustering scenario generation and sensitivity analysis according to claim 1, characterized in that: The sensitivity analysis includes simulating risk factors through Monte Carlo simulation or Latin hypercube sampling method to obtain the probability distribution of risk factors and calculate the impact of each risk factor change on returns. The formula is: Among them, ΔR is the change in total revenue, p i is the probability of occurrence of risk factor i, Δr i is the impact of changes in risk factor i on returns.

4. The multi-trading product optimization method based on clustering scenario generation and sensitivity analysis according to claim 1, characterized in that: The risk ranking is performed by calculating the impact of each risk factor on the return, and the formula is: Among them, RiskRank i is the ranking value of risk factor i, ΔR j is the impact of risk factor i on returns at the jth moment, and T is the time period.

5. The multi-trading product optimization method based on clustering scenario generation and sensitivity analysis according to claim 1 is characterized in that: The calculation formula for the mid- to long-term and spot transaction volume and price adjustments is: P adjusted =P base ·(1+α·ΔR) Among them, P adjusted is the adjusted price, P base is the base price, α is the adjustment coefficient, and ΔR is the change in revenue.

6. The multi-trading product optimization method based on clustering scenario generation and sensitivity analysis according to claim 1, characterized in that: The transaction review includes calculating the error between the actual settlement result and the predicted result. The formula is: in, ∈ is the error, R actual (t) is the actual return at time t, R forccast (t) is the predicted return at time t, and T is the length of the time period.

7. The multi-trading product optimization method based on clustering scenario generation and sensitivity analysis according to claim 1, characterized in that: The rolling optimization includes adjusting the parameters of the clustering model according to the actual transaction data and sensitivity analysis results after each round of transactions, and optimizing the parameters of the sensitivity analysis model using the back propagation algorithm and the gradient descent method to improve the prediction accuracy and risk control ability of the model.

8. The multi-trading product optimization application based on clustering scenario generation and sensitivity analysis according to claim 1, 2, 3, 4, 5, 6 or 7, characterized in that the step include: a) Data collection module, used to collect historical transaction data of new energy plants and stations, including power generation forecasts, historical power generation output, medium- and long-term transaction volume and price, spot price forecasts, spot transaction volume and price, and grid congestion data; b) Data cleaning module, used to clean the collected data, process outliers and missing values, and standardize and normalize the data; c) Feature engineering module, used to extract effective features from the cleaned data and generate feature sets for cluster analysis and risk sensitivity analysis; d) a cluster analysis module, which is used to generate multiple groups of typical operation scenarios based on feature data, and to generate typical patterns of the market environment through cluster analysis methods; e) Risk analysis module, which is used to conduct sensitivity analysis on the risk factors in each clustering scenario, calculate the impact of changes in each risk factor on returns, rank risks and develop risk indicators; f) Transaction optimization module, which is used to adjust the volume and price of medium- and long-term and spot transactions based on risk analysis results, scenario probability and risk indicators, and provide transaction decision suggestions; g) The transaction review module is used to compare the actual settlement results with the predicted results, calculate the error and provide feedback to optimize the model parameters to ensure dynamic adjustment and rolling optimization of the model.

9. The multi-trading product optimization application based on clustering scenario generation and sensitivity analysis according to claim 8, characterized in that: The cluster analysis module adopts K-means clustering or DBSCAN clustering algorithm, generates multiple market scenarios based on historical transaction data, and divides historical data into different categories by calculating Euclidean distance or density algorithm, each category representing a typical market environment.

10. The multi-trading product optimization application based on clustering scenario generation and sensitivity analysis according to claim 8, characterized in that: The risk analysis module simulates the changes of different risk factors through Monte Carlo simulation method or Latin hypercube sampling method, analyzes the impact of risk factors on returns, and generates risk rankings and risk indicators.

11. The multi-trading product optimization application based on clustering scenario generation and sensitivity analysis according to claim 8, characterized in that: The transaction optimization module adjusts the volume and price of medium- and long-term and spot transactions based on scenario probability, risk ranking and risk threshold, and dynamically adjusts the trading strategy according to changes in the trading environment.

Images