Electricity quantity combination optimization and risk assessment method for electricity selling company
By combining deep learning and machine learning to optimize power allocation and assess risk, this method addresses the shortcomings of traditional methods in risk assessment under uncertainties in electricity prices and loads. It enables the optimization of power allocation and risk assessment, thereby improving the scientific nature of power sales companies' decision-making and their market responsiveness.
Patent Information
- Application Number
- CN202511094385.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-06
- Publication Date
- 2025-10-17
AI Technical Summary
Traditional power allocation management methods are unable to effectively capture and cope with the high nonlinearity, strong correlation and extreme uncertainty of electricity prices and loads. They lack systematic and coordinated assessment of multi-level risks, which makes it impossible for strategy formulation to find the global optimal balance between risk and return.
A fusion model of deep learning and machine learning is used for electricity demand forecasting, an ensemble learning model is used for market price forecasting, a multi-objective optimization model is constructed and an optimization algorithm is introduced, and Monte Carlo simulation and Bayesian network are combined for risk assessment. The optimal combination scheme and its profit and risk indicators are output to assist electricity sales companies in making scientific decisions.
It achieves overall optimization of power mix, provides quantifiable risk assessment, improves forecast accuracy and strategy robustness, and enhances the intelligent decision-making capabilities of power sales companies in dynamic markets.
Smart Images
Figure CN120806656A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of electricity market, more particularly, to a power purchase optimization and risk assessment method for a power retailer. BACKGROUND
[0002] On the one hand, the transaction variety is increasingly diversified and complex, covering long-term contracts (locking base price and quantity), annual agreements (providing certain flexibility), electricity spot market (reflecting real-time supply and demand, and price volatility), and various ancillary service markets such as frequency regulation and backup. On the other hand, the risk of electricity price fluctuation has been significantly exacerbated, driven by factors such as the intermittency of new energy generation, the uncertainty of fuel costs, the frequent occurrence of extreme weather events, and the real-time dynamic changes in market supply and demand. At the same time, the accuracy of user load forecasting is also challenged, further amplifying operational risks. Under this background, power retailers urgently need to develop a scientific, reasonable, and dynamic power portfolio optimization and risk assessment method. The core goal is to maximize overall revenue through fine and intelligent portfolio management strategies while effectively identifying, quantifying, and controlling multiple risks, especially the risk of electricity price fluctuations. However, traditional power portfolio management methods are mostly based on static or semi-static strategies (such as simple proportional allocation or fixed strategies based on historical data), which have increasingly prominent drawbacks: they are difficult to effectively capture and respond to the highly nonlinear, strong correlation, and extreme uncertainty of electricity prices and loads. More importantly, traditional methods generally lack a systematic and coordinated assessment mechanism for the overall portfolio's comprehensive benefits (such as revenue, cost, and market share) and multiple levels of risks (such as market risk, credit risk, and liquidity risk). This fragmented perspective often leads to a trade-off between risk and return, failing to find a global optimal balance point between the two. Therefore, breaking free from the constraints of traditional static frameworks and developing a collaborative decision support system that integrates advanced optimization algorithms, real-time data analysis, and dynamic risk quantification models has become a key to improving core competitiveness, achieving stable profits, and sustainable development for power retailers in the increasingly competitive electricity market. SUMMARY
[0003] The purpose of the present application is to provide an integrated power portfolio optimization and risk assessment method that models customer demand, electricity market quotes, prediction models, and risk assessment indicators to achieve optimal configuration of power purchase structure and quantify the risk level of portfolio schemes under different scenarios, thereby assisting power retailers in making scientific decisions.
[0004] The purpose of the present application can be achieved by the following technical solutions: A power purchase optimization and risk assessment method for a power retailer, comprising the following steps: S1, data collection, obtaining historical load, electricity price, weather, market mechanism and other multi-source data; S2, power demand prediction, using a deep learning and machine learning fusion model to predict future electricity load; S3, market price prediction, using an ensemble learning model to predict long-term correlation, spot, green and auxiliary service prices; S4, constructing a multi-objective optimization model, the objective function includes maximizing profit, minimizing risk, and green power proportion; S5, introducing an optimization algorithm for power combination decision, the constraint conditions include load demand, power upper limit and market mechanism restriction; S6, using Monte Carlo simulation and Bayesian network for scenario generation and risk assessment; S7, outputting the optimal combination scheme and its profit and risk indicators to assist the power selling company in formulating procurement strategies.
[0005] Preferably, the power demand prediction model is a hybrid prediction model based on LSTM and XGBoost.
[0006] Preferably, the market price prediction uses a Stacking ensemble method to integrate multiple basic prediction models, including random forest, XGBoost and time series model.
[0007] Preferably, the optimization algorithm is a multi-objective particle swarm optimization algorithm or a non-dominated sorting genetic algorithm.
[0008] Preferably, the risk assessment method includes conditional risk value, maximum loss value, profit volatility and Sharpe ratio.
[0009] Preferably, the output module can generate a report and link to the power trading platform for recommendation or execution.
[0010] Preferably, the method further comprises controlling the proportion of green power procurement to meet the requirements of the government or customers for the proportion of green power use.
[0011] Preferably, the method introduces user behavior labels as input variables of the prediction model to improve the personalized accuracy of load prediction.
[0012] Preferably, the system has self-learning ability and dynamically updates the prediction model and optimization parameters by collecting combination execution effect data.
[0013] Preferably, the system can provide simulation results of multiple combination strategies, including aggressive, conservative and balanced types for users to choose from, and support visual risk exposure chart and profit probability distribution chart.
[0014] The beneficial effects of the present application are: 1. Achieve overall optimization of electricity mix under multiple market mechanisms; 2. Provide quantifiable risk assessment indicators to effectively avoid operational risks caused by electricity price and load fluctuations; 3. Integrate multi-source data with advanced modeling algorithms to improve forecast accuracy and strategy robustness; 4. Build a decision support system that can be deployed on a cloud platform to enhance the intelligence level of power sales companies in responding to dynamic markets. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.
[0016] Figure 1 This is the system framework diagram of the method of the present invention, which includes six modules: data acquisition, load forecasting, electricity price forecasting, combination optimization, risk assessment, and result output. DETAILED DESCRIPTION
[0017] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0018] like Figure 1 As shown: This embodiment provides a method for optimizing and assessing the power portfolio of a power sales company. The method includes six modules: data collection, load forecasting, electricity price forecasting, portfolio optimization, risk assessment, and result output. The method includes the following steps: 1. Data acquisition and processing module 1.1. Collection dimensions include historical load, electricity price, weather, holidays, power structure, user tags and market mechanism parameters.
[0019] 1.2. Use the LOF algorithm to identify outliers, interpolation to repair missing values, Z-score standardization, and process the time window (hour / day / month) into the model input format.
[0020] 2. Power demand forecast module 2.1. Use a hybrid forecasting model (LSTM, GRU and XGBoost residual correction) for short-term and medium-term load forecasting.
[0021] 2.2, Introduce auxiliary variables (such as weather, holiday type) to improve model accuracy, use K-fold cross-validation and SMAPE indicators for evaluation.
[0022] 3, Electricity price prediction module 3.1, Long-term electricity price: regression model based on macro parameters.
[0023] 3.2, Spot electricity price: multivariate time series + machine learning model prediction (Stacking integration).
[0024] 3.3, Green electricity price: introduce carbon price factor linkage modeling.
[0025] 3.4, Auxiliary service price: model considers reserve capacity demand, response time, etc.
[0026] 4, Electricity combination optimization model 4.1, Multi-objective function construction: Maximize revenue: Max E[P] = Σ (Qi × (Si - Ci) - risk cost; Minimize risk: Min CVaRα(P) = E[P | P ≤ VaRα]; Meet green proportion, compliance stability, customer satisfaction and other goals.
[0027] 4.2, Decision variables: Electricity procurement proportion vector Q; Renewable priority and reserve capacity value.
[0028] 4.3, Constraint conditions: Total electricity demand balance: ΣQi = D; Electricity upper limit constraint: Qi ≤ Max_Qi; Market mechanism and transmission and distribution capacity limit, etc.
[0029] 4.4, Optimization algorithm: use multi-objective particle swarm optimization algorithm (MOPSO) or NSGA-II algorithm, and combine local search fine tuning.
[0030] 5, Risk assessment and scenario simulation module 5.1, Use double-layer Monte Carlo simulation: the first layer simulates load changes, and the second layer simulates electricity price fluctuations, forming 1000+ groups of scenario data.
[0031] 5.2, Risk index evaluation includes: expected revenue, revenue volatility σ, maximum loss, VaR, CVaR, Sharpe ratio.
[0032] 5.3, Correlation modeling between risk factors: use Bayesian network to learn the dependency structure between variables.
[0033] 6、Results output and intelligent auxiliary decision-making system 6.1、The output content includes the optimal power combination scheme, the income distribution diagram under each scenario, the risk heat map, and the comparison of user customized target strategies.
[0034] 6.2、The system has online adjustment and API docking capabilities, and can automatically generate reports or link to execute system recommendations.
[0035] The method constructs a complete closed-loop system from data input to decision output, which covers key links such as load prediction, price prediction, combination optimization, and risk simulation, ensuring the scientificity and robustness of the decision. By introducing multi-source data, multi-model collaborative prediction, and risk control, the method has good scalability and cross-regional adaptability, significantly improving the profit stability and risk resistance of the power selling company.
[0036] Further, the power demand prediction model is a hybrid prediction model based on LSTM and XGBoost. In this embodiment, the LSTM model can effectively capture the long-term dependence of power load and is suitable for complex time series feature learning; while the XGBoost model is good at handling nonlinear relationships and abnormal data. After combining the two, not only the prediction accuracy is improved, but also the generalization ability of the model in special scenarios such as holidays and extreme weather is enhanced, reducing the interference of prediction bias on the downstream optimization module.
[0037] Further, the market price prediction adopts a Stacking ensemble method to integrate multiple base prediction models, including random forests, XGBoost, and time series models. In this embodiment, the Stacking ensemble method can utilize the advantages of multiple models in different data feature dimensions, implement output fusion through a secondary learner, and improve prediction stability. For the differences between different price varieties, the ensemble model avoids the local optimal risk caused by the prediction bias of a single model, thereby better serving the revenue estimation of the combination optimization module.
[0038] Further, the optimization algorithm is a multi-objective particle swarm optimization algorithm (MOPSO) or a non-dominated sorting genetic algorithm (NSGA-II). In this embodiment, MOPSO and NSGA-II, as mature multi-objective optimization algorithms, can maintain the diversity and convergence speed of solutions in solving non-convex and nonlinear problems, and are suitable for multi-objective and multi-constrained power procurement combination scenarios. This type of algorithm not only generates a set of Pareto optimal solutions, but also facilitates subsequent strategy screening and matching for different risk preferences.
[0039] Further, the risk assessment method includes conditional value at risk (CVaR), maximum loss value, return volatility, and Sharpe ratio. In this embodiment, the introduction of CVaR, maximum loss, volatility, and other commonly used indicators in the financial field for systematic evaluation of the return curve helps the electricity sales company to build a comprehensive risk identification and early warning mechanism. At the same time, the introduction of the Sharpe ratio provides a quantitative reference for the trade-off between returns and risks, helping managers to make more robust strategy choices.
[0040] Further, the output module can generate a report of the results and link to the electricity trading platform for execution or recommendation. In this embodiment, the result output supports a configurable format, supports the generation of standardized reports that can be directly used for internal audit or regulatory disclosure, and can interface with the electricity trading platform API to realize the automation of transaction instructions, reducing the time and operational risk of human intervention, and improving response efficiency and execution accuracy.
[0041] Further, the method further includes controlling the proportion of green electricity procurement to meet the requirements of the government or customers for the proportion of green electricity use. In this embodiment, the method considers the adjustability and compliance requirements of the green electricity procurement strategy, sets green proportion constraints or objective function weights in the optimization model, and guides the system to seek a balance between the yield target and environmental compliance.
[0042] Further, the method introduces user behavior labels as input variables of the prediction model to improve the personalized accuracy of load prediction. In this embodiment, after the user behavior labels such as industry category, production shift, historical response elasticity, and other information are introduced into the prediction model, the understanding of the load time series distribution and fluctuation characteristics can be enhanced, and fine-grained load management can be realized. At the same time, the prediction resolution in the diversified user structure scenario can be improved, and the local adaptability of the strategy can be enhanced.
[0043] Further, the system has self-learning ability, and dynamically updates the prediction model and optimization parameters by collecting combined execution effect data. In this embodiment, the system introduces an online learning mechanism to collect and model update the actual execution data, including model weight adjustment, abnormal error correction, risk parameter re-estimation, etc., so that the system has the ability to dynamically adapt to market changes, and builds a truly "closed-loop intelligent optimization system".
[0044] Further, the system can provide simulation results of multiple combination strategies, including aggressive, conservative, and balanced, and support visual risk exposure graphs and return probability distribution graphs. In this embodiment, the multi-strategy simulation and visual output meet the decision-making preferences of different management levels, and the strategy covers a complete range from high yield and high risk to low risk and low yield. The risk exposure graph and the return distribution graph help users intuitively determine the possible result interval, improve the transparency and credibility of decision-making, and are suitable for electricity sales subjects of various operating styles.
[0045] In the description of the application, unless otherwise specified, the meaning of "a plurality of" is two or more; it should be understood that the terms "opening", "upper", "lower", "thickness", "top", "middle", "length", "inner", "periphery" and the like indicate the orientation or positional relationship, only for the purpose of facilitating the description of the application and simplifying the description, and do not indicate or imply that the components or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application.
[0046] Finally, it should be noted that: the above only for the preferred embodiments of the present application, and not for the limitation of the present application, although the application is described in detail with reference to the foregoing embodiments, for those skilled in the art, it still can be modified to the technical scheme recorded in the foregoing embodiments, or equivalent replacement of some of the technical features, any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application, should be included in the protection scope of the present application.
Claims
1. A method for optimizing and assessing the power mix of a power sales company, characterized in that: The steps include: S1. Data collection: obtaining multi-source data such as historical load, electricity price, weather, and market mechanism; S2, electricity demand forecasting, uses a deep learning and machine learning fusion model to predict future electricity load; S3, market quotation forecast, using an integrated learning model to predict long-term contract, spot, green and ancillary service electricity prices; S4. Build a multi-objective optimization model with objective functions including maximizing revenue, minimizing risk, and determining the proportion of green electricity; S5. Introduce an optimization algorithm to make power combination decisions, with constraints including load demand, power limit, and market mechanism restrictions; S6. Scenario generation and risk assessment using Monte Carlo simulation and Bayesian networks; S7. Output the optimal combination plan and its profit-risk indicators to assist power sales companies in formulating procurement strategies.
2. The method for optimizing and assessing the power sales company's power mix according to claim 1 is characterized by: The power demand forecasting model is a hybrid forecasting model based on LSTM and XGBoost.
3. The method for optimizing and assessing the power sales company's power mix according to claim 1 is characterized by: The market quotation forecast adopts the Stacking integration method to integrate multiple basic forecasting models, including random forest, XGBoost and time series model.
4. The method for optimizing and assessing the power sales company's power mix according to claim 1 is characterized by: The optimization algorithm is a multi-objective particle swarm optimization algorithm or a non-dominated sorting genetic algorithm.
5. The method for optimizing and assessing the power sales company's power mix according to claim 1 is characterized by: The risk assessment methods include conditional risk value, maximum loss value, return volatility and Sharpe ratio.
6. The method for optimizing and assessing the power sales company's power mix according to claim 1 is characterized by: The output module can generate a report based on the results and link it to the power trading platform for execution or recommendation.
7. The method for optimizing and assessing the power sales company's power mix according to claim 1 is characterized by: The method also includes controlling the proportion of green electricity purchases to meet the government or customer's requirements on the proportion of green electricity usage.
8. The method for optimizing and assessing the power sales company's power mix according to claim 1 is characterized by: The method introduces user behavior tags as prediction model input variables to improve the personalized accuracy of load forecasting.
9. The method for optimizing and assessing the power sales company's power mix according to claim 1 is characterized by: The system has self-learning capabilities and dynamically updates the prediction model and optimization parameters by collecting combined execution effect data.
10. The method for optimizing and assessing the power sales company's power mix according to claim 1, characterized in that: The system can provide a variety of combination strategy simulation results, including aggressive, conservative and balanced types for users to choose from, and supports visualization of risk exposure graphs and return probability distribution graphs.