Energy storage system economy evaluation method and system and storage medium

Abstract

The invention discloses an energy storage system economy evaluation method and system and a storage medium, and belongs to the technical field of energy storage systems. Comprising the following steps: acquiring operation data, power market transaction data and equipment state data of an energy storage system in real time through a multi-source data acquisition module; a dynamic feature engineering module is used for screening key influence factors, and a comprehensive evaluation index system fusing market features, technical features and environment features is constructed; constructing an economy prediction model based on an LSTM neural network, inputting the data of the comprehensive evaluation index system into the economy prediction model, and outputting a multi-time scale economy evaluation result, including short-term income prediction, medium-term probabilistic income interval and long-term attenuation correction evaluation; and a revenue-risk balanced charging and discharging strategy is generated through a reinforcement learning algorithm, a strategy effect is rehearsed in a digital twin environment, and high-risk operation is dynamically adjusted. And the precision and adaptability of economic evaluation of the energy storage system can be remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of energy storage system technology, and in particular to an economic evaluation method, system, and storage medium for energy storage systems. Background Technology

[0002] With the expansion of renewable energy grid connection and the deepening of electricity market reforms, the economic evaluation of energy storage systems has become a core aspect of investment decisions. Traditional evaluation methods rely on static financial models (such as net present value and internal rate of return methods) and empirical assumptions, which have the following technical shortcomings:

[0003] The assessment suffers from several key shortcomings: First, it relies solely on historical electricity prices and fixed equipment parameters, lacking dynamic integration of real-time market data (such as demand response bidding), equipment status (such as battery health degradation), and environmental policies. This leads to significant evaluation biases. Second, it suffers from insufficient dynamic adaptability, employing a fixed-weight indicator system that fails to capture the coupling relationship between electricity market fluctuations and equipment performance degradation. Third, it lacks multi-objective synergy, focusing on isolated optimization of short-term returns and long-term asset value, and lacking a dynamic return-risk balancing mechanism. Fourth, it suffers from delayed strategy validation, relying on ex-post financial analysis and failing to anticipate systemic risks such as policy shifts, resulting in economic losses. Therefore, a dynamic, adaptive, and multi-dimensional evaluation method is urgently needed to improve the accuracy and reliability of energy storage system economic assessments. Summary of the Invention

[0004] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention proposes a method, system, and storage medium for economic evaluation of energy storage systems, which can significantly improve the accuracy and adaptability of economic evaluation of energy storage systems, thereby enhancing the economic benefits and investment reliability of energy storage systems.

[0005] An economic evaluation method for an energy storage system according to a first aspect of the present invention includes:

[0006] The system acquires real-time operational data, electricity market transaction data, and equipment status data of the energy storage system through a multi-source data acquisition module.

[0007] By utilizing the dynamic feature engineering module to screen key influencing factors, a comprehensive evaluation index system integrating market characteristics, technological characteristics, and environmental characteristics is constructed.

[0008] An economic forecasting model is constructed based on an LSTM neural network. The data of the comprehensive evaluation index system is input into the economic forecasting model, and the economic evaluation results at multiple time scales are output, including short-term return forecast, medium-term probabilistic return range and long-term decay correction evaluation.

[0009] A reward-risk balanced charging and discharging strategy is generated using a reinforcement learning algorithm, and the strategy effect is simulated in a digital twin environment to dynamically adjust high-risk operations.

[0010] According to some embodiments of the present invention, the dynamic feature engineering module further includes a feature weight adaptive unit. This feature weight adaptive unit can dynamically adjust the weight coefficients of market features, technical features, and environmental features based on the fluctuation cycle of the electricity market, and updates the weight coefficients of the market features, technical features, and environmental features daily through an online learning mechanism. The weight coefficients of the market features, technical features, and environmental features satisfy a normalization equation:

[0011] W t =α·W market +β·W thchnical +γ·W environment

[0012] in:

[0013] W t : The comprehensive feature weights within the time window t;

[0014] W market The basic weights of market characteristics;

[0015] W technical Cardinal weights of technical features;

[0016] W environment The basic weights of environmental characteristics;

[0017] α, β, γ: weighting coefficients for market characteristics, technological characteristics, and environmental characteristics, respectively.

[0018] According to some embodiments of the present invention, the online learning mechanism includes updating the weight coefficients of the market features, technical features, and environmental features based on a gradient descent algorithm, employing a sliding time window strategy to retain the latest data for a preset number of days, and introducing a forgetting factor to weaken the weight of historical data. The step of updating the weight coefficients of the market features, technical features, and environmental features based on a gradient descent algorithm includes the following steps:

[0019] Receive new daily operational data, electricity market transaction data, and equipment status data of the energy storage system.

[0020] The short-term earnings prediction error is used as the loss function.

[0021] The weighting coefficients of the market characteristics, technological characteristics, and environmental characteristics are adjusted through backpropagation.

[0022] According to some embodiments of the present invention, the dynamic feature engineering module further includes an abnormal data cleaning unit, which uses a generative adversarial network to simulate extreme environment data and includes the following steps:

[0023] Receive historical data and generate synthetic data containing extreme scenarios;

[0024] Distinguish between real and synthetic data, and improve the realism of synthetic data through adversarial training;

[0025] By adding synthetic data to the training set, the economic prediction model is exposed to more extreme scenarios during the training phase, thereby improving its ability to generalize to outliers.

[0026] According to some embodiments of the present invention, a real-time correction mechanism is also included, which triggers online fine-tuning of model parameters when the actual return deviates from the predicted value and exceeds a preset value. The online fine-tuning includes the following steps:

[0027] Update LSTM parameters: Learn the latest electricity market transaction data and equipment status data within the preset time range;

[0028] Corrected attenuation factor: Adjust long-term forecasts based on actual operating data of the energy storage system.

[0029] According to some embodiments of the present invention, a reward-risk balanced charging and discharging strategy is generated by a reinforcement learning algorithm, and the effect of the strategy is simulated in a digital twin environment. The high-risk operation is dynamically adjusted, including multi-objective game optimization and real-time strategy simulation.

[0030] The multi-objective game optimization is to balance returns and risks by establishing a return-risk balance equation, where the risk coefficient λ automatically matches conservative or aggressive strategies according to the investor type.

[0031] The real-time strategy simulation includes rehearsing the strategy effect in a digital twin environment, and prohibiting high-risk operations when the simulated net present value is lower than a threshold.

[0032] According to some embodiments of the present invention, the constructed benefit-risk balance equation is as follows:

[0033] U=E-λ·Var

[0034] in:

[0035] U is the utility function representing the overall economics of the energy storage system, E represents the expected return, Var represents the variance of risk, and λ represents the risk coefficient. Matching is performed according to preset conditions.

[0036] According to some embodiments of the present invention, the method of pre-simulating the effect of a strategy in a digital twin environment, prohibiting high-risk operations when the simulated net present value is lower than a threshold, includes:

[0037] The system acquires multi-source data, including at least real-time market data and equipment status data. It constructs a virtual model that is highly consistent with the real environment using the acquired multi-source data. The system then uses the virtual model to simulate a strategy and obtains the net present value after the strategy is executed. If the simulated net present value is lower than a preset threshold, the system automatically prohibits the execution of the strategy and triggers the generation of alternative solutions.

[0038] According to a second aspect of the present invention, an energy storage system economic evaluation system includes a memory and a processor. The memory stores a program for determining an energy storage system economic evaluation method, and the processor runs the program for determining the energy storage system economic evaluation method to enable the energy storage system economic evaluation system to perform the energy storage system economic evaluation method.

[0039] According to a third aspect of the present invention, a computer-readable storage medium includes: a program for determining an economic evaluation method for an energy storage system stored on the computer-readable storage medium, wherein when the program for determining the economic evaluation method for an energy storage system is executed by a processor, the economic evaluation method for an energy storage system is implemented.

[0040] According to an embodiment of the present invention, an economic evaluation method, system, and storage medium for energy storage systems have at least the following beneficial effects: A multi-source data acquisition module acquires real-time data on the operation, market, and equipment status of the energy storage system; a dynamic feature engineering module constructs a comprehensive index system integrating time-series, economic, and environmental characteristics; an economic prediction model uses an LSTM neural network to predict returns across multiple time scales; and a strategy optimization module combines reinforcement learning to generate a return-risk balance strategy and pre-validates it using digital twins. This solves the problems of traditional methods, such as single data dimension, insufficient dynamic adaptability, and lagging strategy validation, significantly improving the accuracy of economic evaluation of energy storage systems and the reliability of investment decisions.

[0041] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0042] The present invention will be further described below with reference to the accompanying drawings and embodiments, wherein:

[0043] Figure 1 This is a flowchart illustrating an embodiment of the energy storage system economic evaluation method of the present invention;

[0044] Figure 2 This is a schematic diagram of the multi-source data acquisition module of the present invention;

[0045] Figure 3 This is a schematic diagram of the dynamic feature engineering module of the present invention;

[0046] Figure 4This is a schematic diagram illustrating the economic prediction and strategy optimization of the storage system according to the present invention.

[0047] Figure 5 This is a schematic diagram of the energy storage system economic evaluation system of the present invention. Detailed Implementation

[0048] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0049] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0050] In the description of this invention, "multiple" refers to two or more. The use of "first" and "second" is for distinguishing technical features only and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features or their sequential relationship.

[0051] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0052] Reference Figures 1 to 4 As shown, this invention discloses an economic evaluation method for energy storage systems, comprising:

[0053] Step S100: Real-time acquisition of energy storage system operation data, electricity market transaction data, and equipment status data through multi-source data acquisition module;

[0054] Step S200: Use the dynamic feature engineering module to screen key influencing factors, construct a comprehensive evaluation index system that integrates market characteristics, technological characteristics, and environmental characteristics, and use adversarial generative networks to enhance the robustness of the model data;

[0055] Step S300: Construct an economic forecasting model based on an LSTM neural network, input the data from the comprehensive evaluation index system into the economic forecasting model, and output economic evaluation results at multiple time scales, including short-term return forecasts, medium-term probabilistic return ranges, and long-term decay correction assessments.

[0056] Step S400: Generate a reward-risk balanced charging and discharging strategy using a reinforcement learning algorithm, and preview the strategy's effect in a digital twin environment to dynamically adjust high-risk operations.

[0057] In this embodiment, the operating data of the energy storage system is collected in real time from the BMS (Battery Management System) and SCADA (Supervisory and Data Acquisition System) of the energy storage system, including charging and discharging power, SOC (State of Charge), efficiency curve, battery health (SOH), cycle life degradation rate, etc.

[0058] Electricity market transaction data is obtained through the electricity trading platform API, including real-time electricity prices, demand response subsidies, carbon emission rights prices, and electricity spot market bidding data.

[0059] Device status data is monitored via sensors, including battery temperature, charge / discharge cycles, and capacity decay rate.

[0060] The random forest algorithm is used to rank the features of multi-source data by importance and screen out key influencing factors. These include market features such as real-time electricity price volatility and demand response subsidy intensity; technical features such as dynamic trends in State of Charge (SOC) and the slope of State of Health (SOH) decay; and environmental features such as carbon emission rights price trends and the probability of extreme weather events. Redundant features (such as outdated parameters from historical subsidy policies) are removed to reduce model complexity.

[0061] The selected features are used to construct a comprehensive evaluation index system, which includes market features (such as 24-hour electricity price cycle), technological features (such as peak-valley price difference benefits), and environmental features (such as policy compliance costs).

[0062] The system employs an LSTM (Long Short-Term Memory) neural network, with the input being a comprehensive evaluation index system output by dynamic feature engineering. It outputs economic evaluation results across multiple time scales, specifically including: short-term forecasts (e.g., 24 hours), providing a high-precision peak-valley arbitrage profit curve (error rate <5%); long-term forecasts (e.g., 30 days), generating probabilistic profit ranges (e.g., a profit range with 90% confidence); and long-term forecasts (e.g., 5 years), combined with a battery capacity degradation model, outputting a corrected Lifetime Cost (LCOE) value.

[0063] Based on the DQN (Deep Q-Network) algorithm, a charge / discharge strategy is generated with the goal of maximizing net present value (NPV). It also requires that the battery health is not lower than 70% and that the number of charge / discharge cycles per day does not exceed 3.

[0064] The strategy execution effect is simulated in a digital twin environment. Real-time market data and equipment status are input, and simulated NPV is output. If the simulated NPV is lower than a safety threshold (e.g., investment payback period > 8 years), high-risk operations are automatically prohibited.

[0065] High-risk operations specifically refer to actions that may cause the economic performance of the energy storage system to deviate significantly from expectations or lead to irreversible losses. These mainly include two categories:

[0066] 1) Market risk operations: For example, when electricity market prices fluctuate wildly, aggressive charging and discharging strategies are used to pursue high short-term returns without fully assessing the risk of electricity price reversal.

[0067] 2) Risky operations: For example, frequent deep charging and discharging when the battery is in a high health state may increase short-term gains, but it will accelerate the degradation of battery life and lead to a significant increase in the total life cycle cost.

[0068] In the strategy optimization module, the impact of such operations on net present value is simulated in a digital twin environment. If the simulation results show that the NPV is lower than the safety threshold, the system will automatically prohibit the execution.

[0069] In this embodiment, the limitations of traditional methods that rely on single historical data are overcome by integrating market, equipment, and environmental data in real time. For example, by incorporating real-time electricity price fluctuations into the assessment, prediction errors can be significantly reduced.

[0070] By using the dynamic feature engineering module to screen key influencing factors, eliminate redundant features, reduce model complexity, and improve training efficiency.

[0071] Short-term forecasts provide minute-level strategies for intraday arbitrage (such as buying low and selling high); long-term forecasts provide a decay correction reference for investment planning. Digital twin simulations intercept high-risk operations, avoiding equipment damage or profit loss due to strategy failure, and reducing investment risk.

[0072] In some embodiments of the present invention, the dynamic feature engineering module further includes a feature weight adaptive unit. This unit can dynamically adjust the weight coefficients of market features, technical features, and environmental features according to the fluctuation cycle of the electricity market, and updates these weight coefficients daily through an online learning mechanism. The weight coefficients of the market features, technical features, and environmental features satisfy a normalization equation.

[0073] W t =α·W market +β·W thchnical +γ·W environment

[0074] in:

[0075] W t : The comprehensive feature weights within the time window t;

[0076] W market The basic weights of market characteristics;

[0077] W technical Cardinal weights of technical features;

[0078] W environment The basic weights of environmental characteristics;

[0079] α, β, γ: weighting coefficients for market characteristics, technological characteristics, and environmental characteristics, respectively.

[0080] Input daily updated market data (such as real-time electricity prices and demand response subsidies), technical data (such as battery health status (SOH)), and environmental data (such as carbon emission rights prices) into the economic forecasting model. Set weighting coefficients for market characteristics, technical characteristics, and environmental characteristics, such as α = 0.4, β = 0.3, and γ = 0.3.

[0081] The weight coefficients of market characteristics, technical characteristics, and environmental characteristics are updated daily through an online learning mechanism, and the softmax function ensures that the sum of the weights is 1.

[0082] Example 1:

[0083] If the daily price volatility exceeds 20%, the market characteristic weight α will automatically increase (e.g., from 0.4 to 0.6) to enhance the capture of short-term arbitrage opportunities.

[0084] Example 2:

[0085] If the battery health (SOH) drops by more than 5%, the technical feature weight β is increased (e.g., adjusted from 0.3 to 0.4) to prioritize device life protection.

[0086] By adjusting the weighting coefficients (α, β, γ) in real time based on daily data, the model maintains high accuracy under scenarios such as electricity price fluctuation cycles and policy adjustments. For example:

[0087] In the early stages of electricity market reform, frequent policy changes led to an increase in the environmental characteristic weight γ from 0.2 to 0.35, with the model focusing more on the impact of carbon emission costs on long-term economics.

[0088] During the battery aging stage, the technical feature weight β continues to increase, and the strategy optimization module automatically reduces the depth of charge and discharge to extend the equipment life.

[0089] In some embodiments of the present invention, the online learning mechanism includes updating the weight coefficients of market features, technical features, and environmental features based on a gradient descent algorithm, employing a sliding time window strategy to retain data for the latest preset number of days, and introducing a forgetting factor to weaken the weight of historical data. Updating the weight coefficients of market features, technical features, and environmental features based on a gradient descent algorithm includes the following steps:

[0090] It receives new energy storage system operation data, electricity market transaction data, and equipment status data daily.

[0091] The short-term earnings prediction error is used as the loss function.

[0092] The weighting coefficients of market characteristics, technological characteristics, and environmental characteristics are adjusted through backpropagation.

[0093] In this embodiment, daily updated market data (such as real-time electricity prices and demand response subsidies), technical data (such as battery health status (SOH)) and environmental data (such as carbon emission rights prices) are input into the economic forecasting model.

[0094] Only retain data for the latest preset number of days (e.g., 30 days), and automatically remove the oldest historical data after each new day's data is added. This limits the data size, reduces computational complexity, and focuses on recent dynamic changes (e.g., electricity price fluctuation cycles, equipment performance degradation trends).

[0095] The optimization objective is the difference between the predicted and actual revenue of the energy storage system over the next 24 hours. The loss function is:

[0096]

[0097] The loss function uses mean squared error (MSE), and the weighting coefficients α, β, and γ are dynamically optimized by minimizing the squared difference between the predicted value and the actual return. pred To predict the rate of return; y true This represents the actual rate of return. Backpropagation is a commonly used algorithm for training neural networks. Its core principle is to calculate the gradient of the loss function with respect to the network parameters, and then use gradient descent to update the parameters, thereby reducing prediction errors.

[0098] Processing flow:

[0099] Forward calculation: Input the latest 24-hour data (such as electricity price and battery status) into the model to obtain the revenue forecast.

[0100] Error calculation: Compare the predicted value with the actual profit and calculate the error (e.g., using mean squared error).

[0101] Reverse parameter tuning: Working backward from the output layer, calculate the degree of influence (gradient) of each parameter (such as LSTM weights) on the error, and then make small adjustments to the parameters to reduce the error.

[0102] The direction of parameter adjustment is the direction in which the error decreases the fastest. By repeatedly fine-tuning the parameters with new data, the model becomes increasingly closer to reality.

[0103] By introducing a forgetting factor, a decay weight is assigned to each day's data in the sliding window data. For example, with a decay factor of 0.95, the data from 30 days ago has a weight of 0.95. 30 The value is approximately 0.21, significantly reducing its influence. This gradually weakens the impact of historical data on current weights, preventing outdated data (such as obsolete policies) from interfering with the model.

[0104] By introducing a forgetting factor, in scenarios of sudden policy changes (such as the cancellation of carbon emission rights pricing), the influence of old policy data decays to less than 5% after 30 days, allowing the model to quickly adapt to the new rules. This balances the impact of historical data with real-time dynamics. A sliding window physically removes old data, while the forgetting factor logically weakens the influence of the remaining old data; both together enhance the model's sensitivity to recent dynamics.

[0105] refer to Figure 3 As shown, in some embodiments of the present invention, the dynamic feature engineering module further includes an abnormal data cleaning unit. The abnormal data cleaning unit uses a generative adversarial network to simulate extreme environment data and includes the following steps:

[0106] Receive historical data and generate synthetic data containing extreme scenarios;

[0107] Distinguish between real and synthetic data, and improve the realism of synthetic data through adversarial training;

[0108] Adding synthetic data to the training set allows the economic forecasting model to be exposed to more extreme scenarios during the training phase, improving its ability to generalize to outliers.

[0109] In this embodiment, a generator receives historical data (such as electricity price and equipment status) and generates synthetic data that includes extreme scenarios (such as a 50% drop in electricity price or a sudden battery failure).

[0110] The discriminator receives real and synthetic data and outputs the probability of data authenticity. Through adversarial training, the discriminator continues to be trained until it can no longer distinguish between real and generated data (Nash equilibrium).

[0111] The generator produces synthetic data that simulates extreme fluctuations in electricity prices (e.g., ±50%) and extreme scenarios such as equipment failure (sudden 30% drop in power). Synthetic and real data are mixed at a certain ratio (e.g., 1:9) for training simulations. The model's prediction errors are validated on a test set (containing real outlier data) to ensure that the generated data does not introduce bias.

[0112] The outlier cleaning unit reduces prediction errors in extreme scenarios. For example, in special cases where electricity prices are low for an extended period, the model can identify and adjust its strategy in advance. In the event of a sudden equipment failure, such as a sudden drop in simulated battery capacitance, the strategy module automatically triggers a backup power switch to prevent revenue loss. Cleaned data reduces the risk of overfitting, and the model's stability is improved with noisy data (such as sensor errors).

[0113] By exposing the model to extreme but reasonable data distributions, the model is forced to learn the economic evaluation capabilities under unconventional conditions, avoiding strategy failures caused by overfitting to historical data and enhancing the model's data robustness.

[0114] In some embodiments of the present invention, a real-time correction mechanism is also included, which triggers online fine-tuning of model parameters when the actual return deviates from the predicted value and exceeds a preset value. The online fine-tuning includes the following steps:

[0115] Update LSTM parameters: Learn the latest electricity market transaction data and equipment status data within the preset time range;

[0116] Corrected attenuation factor: Adjust long-term forecasts based on actual energy storage system operating data.

[0117] In this embodiment, the daily comparison model predicts the 24-hour return (y). pred ) and actual returns (y) true ), calculate the absolute percentage error (APE):

[0118]

[0119] If the daily APE exceeds the preset value (e.g., exceeding 15% in a single day or accumulating an error of more than 30% over three consecutive days), the system will automatically trigger a parameter fine-tuning process.

[0120] The parameters to be adjusted specifically include:

[0121] 1) LSTM hidden layer weights: used to optimize the time series modeling capability for short-term return prediction.

[0122] 2) Battery capacity degradation coefficient: used to correct the impact of battery aging on the total cost of ownership (LCOE).

[0123] 3) Gaussian distribution parameters (μ,σ) for medium-term return intervals: used to adjust the confidence level of probabilistic return intervals.

[0124] Online fine-tuning includes the following steps:

[0125] Learn the latest electricity market transaction data and equipment status data within a preset time frame (e.g., the latest 24 hours). Adjust long-term forecasts based on actual energy storage system operating data (e.g., battery degradation). For example, if equipment failure causes a sharp drop in revenue, adjust the model using data from the period of failure to reduce subsequent forecast errors from 20% to 8%.

[0126] After parameter adjustments, data from the past seven days is used to verify whether the adjusted model's predicted values ​​are lower than the preset values. Simultaneously, in a virtual environment, the simulation demonstrates that the adjusted strategy achieves the expected net present value (NPV) maximization. If verification passes, the new parameters take effect immediately; otherwise, the system rolls back to the previous version and triggers a manual intervention alert.

[0127] By adjusting LSTM weights in real time, the prediction error for short-term (24-hour) returns is reduced. In scenarios with sudden electricity price fluctuations (such as daily fluctuations of ±200%), the model updates its parameters within one hour, restoring the prediction error to within the threshold range. Dynamic calibration of the attenuation coefficient reduces the prediction error for the 5-year LCOE, preventing inflated returns on investment. Adjusting the Gaussian distribution parameters (μ, σ) improves the confidence coverage of the medium-term return range. Furthermore, only the latest 24-hour data needs to be processed, shortening the LSTM weight update time. The fine-tuning algorithm can run on the local server of the energy storage system, reducing cloud dependence and latency.

[0128] In some embodiments of the present invention, the strategy optimization module generates a payoff-risk balanced charging and discharging strategy through a reinforcement learning algorithm, and pre-simulates the effect of the strategy in a digital twin environment, dynamically adjusting high-risk operations including multi-objective game optimization and real-time strategy simulation.

[0129] Multi-objective game optimization balances returns and risks by establishing a return-risk balance equation, where the risk coefficient λ automatically matches conservative or aggressive strategies based on investor type.

[0130] Real-time strategy simulation includes rehearsing the effects of a strategy in a digital twin environment, and prohibiting high-risk operations when the simulated net present value falls below a threshold.

[0131] In this embodiment, a multi-objective optimization model is established to maximize benefits while minimizing risks. The PPO (Proximal Policy Optimization) algorithm from Deep Reinforcement Learning (DRL) is employed, with the input being a comprehensive evaluation index from the dynamic feature engineering module, and the output being the optimal charge / discharge strategy. Strategy constraints include: battery health (SOH) not lower than 70%; daily charge / discharge cycles not exceeding 3 times; and compliance with carbon emission standards.

[0132] Digital twin environment construction: Real-time access to energy storage system operation data (SOC, SOH), electricity market data (electricity prices, subsidies), and environmental data (policy changes). A high-fidelity virtual energy storage system is constructed based on physical laws (such as battery degradation models and electricity market settlement rules). Discrete event simulation (DES) technology is used to simulate the economic impact of the strategy execution from 24 hours to 5 years later. The net present value (NPV) after strategy execution is simulated in the digital twin. If the simulated NPV is lower than a preset threshold (e.g., payback period > 7 years), the system automatically prohibits the execution of the strategy and triggers the generation of alternative solutions.

[0133] By establishing a multi-objective optimization model, different investment types can be satisfied. For example, the strategy for conservative investors (λ=1.5) limits the charge-discharge depth to within 50%, reducing NPV volatility by 40%. The strategy for aggressive investors (λ=0.5) allows for a charge-discharge depth of up to 80%, increasing daily returns by 25%.

[0134] By constructing a digital twin environment, strategy failures caused by electricity price reversals are identified in the simulation. When the battery health (SOH) falls below 70%, it automatically switches to protection mode to extend battery life cycles.

[0135] In some embodiments of the present invention, the constructed benefit-risk balance equation is as follows:

[0136] U=E-λ·Var

[0137] in:

[0138] U is the utility function, representing the overall economic efficiency of the energy storage system;

[0139] E represents expected returns, including short-term arbitrage profits, demand response subsidies, etc.

[0140] Var represents the variance of risk, which covers market volatility risk (such as electricity price reversals) and equipment performance risk (such as battery life degradation).

[0141] λ represents the risk coefficient, which is matched according to preset conditions. Specifically, the preset conditions include two strategies: conservative or aggressive. Conservative strategies (such as pension funds) prioritize reducing risk, so λ has a larger value, for example, λ = 1.5; aggressive strategies (such as hedge funds) pursue high returns, so λ has a smaller value, for example, λ = 0.5. Of course, λ can be adjusted according to actual circumstances.

[0142] In some embodiments of the present invention, the risk coefficient λ is used to simulate the effect of a strategy in a digital twin environment. When the simulated net present value is lower than a threshold, prohibiting high-risk operations includes:

[0143] Acquire multi-source data, including at least real-time market data and equipment status data. Construct a virtual model that is highly consistent with the real environment using the acquired multi-source data. Pre-simulate the strategy using the virtual model to obtain the net present value after the simulated strategy is executed. If the simulated net present value is lower than a preset threshold, the system automatically prohibits the execution of the strategy and triggers the generation of alternative solutions.

[0144] The present invention also discloses an energy storage system economic evaluation system based on machine learning, including a memory and a processor. The memory stores a program for determining the energy storage system economic evaluation method, and the processor runs the program for determining the energy storage system economic evaluation method so that the energy storage system economic evaluation system executes the energy storage system economic evaluation method.

[0145] refer to Figure 5 As shown, the present invention also discloses a computer-readable storage medium, comprising: a program for determining an economic evaluation method for an energy storage system stored on the computer-readable storage medium, wherein when the program for determining the economic evaluation method for an energy storage system is executed by a processor, the economic evaluation method for an energy storage system is implemented. The medium includes a memory and a processor; the memory stores the program for determining the economic evaluation method for an energy storage system, and the processor runs the program for determining the economic evaluation method for an energy storage system to cause the economic evaluation system for an energy storage system to execute the economic evaluation method for an energy storage system.

[0146] According to a third aspect of the present invention, a computer-readable storage medium includes: a program for determining an economic evaluation method for an energy storage system stored on the computer-readable storage medium, wherein when the program for determining the economic evaluation method for an energy storage system is executed by a processor, the economic evaluation method for an energy storage system is implemented.

[0147] This invention discloses an economic evaluation method, system, and storage medium for energy storage systems. It utilizes a multi-source data acquisition module to acquire real-time data on the operation, market, and equipment status of the energy storage system. A dynamic feature engineering module constructs a comprehensive indicator system integrating time-series, economic, and environmental characteristics. An economic prediction model based on an LSTM neural network achieves multi-timescale revenue prediction. A strategy optimization module combines reinforcement learning to generate a revenue-risk balance strategy and pre-validates it using digital twins. This invention solves the problems of traditional methods, such as limited data dimensions, insufficient dynamic adaptability, and lagging strategy validation, significantly improving the accuracy of economic evaluation of energy storage systems and the reliability of investment decisions.

[0148] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for evaluating the economic viability of an energy storage system, characterized in that, include: The system acquires real-time operational data, electricity market transaction data, and equipment status data of the energy storage system through a multi-source data acquisition module. By utilizing the dynamic feature engineering module to screen key influencing factors, a comprehensive evaluation index system integrating market characteristics, technological characteristics, and environmental characteristics is constructed. An economic forecasting model is constructed based on an LSTM neural network. The data of the comprehensive evaluation index system is input into the economic forecasting model, and the economic evaluation results at multiple time scales are output, including short-term return forecast, medium-term probabilistic return range and long-term decay correction evaluation. A reward-risk balanced charging and discharging strategy is generated using a reinforcement learning algorithm, and the strategy effect is simulated in a digital twin environment to dynamically adjust high-risk operations.

2. The economic evaluation method for energy storage systems according to claim 1, characterized in that: The dynamic feature engineering module also includes a feature weight adaptive unit. This unit can dynamically adjust the weight coefficients of market features, technical features, and environmental features according to the fluctuation cycle of the electricity market, and update these weight coefficients daily through an online learning mechanism. The weight coefficients of the market features, technical features, and environmental features satisfy a normalization equation: W t =α·W market +β·W technical +γ·W environment in: W t : The comprehensive feature weights within the time window t; W market The basic weights of market characteristics; W technical Cardinal weights of technical features; W environment The basic weights of environmental characteristics; α, β, γ: weighting coefficients for market characteristics, technological characteristics, and environmental characteristics, respectively.

3. The economic evaluation method for energy storage systems according to claim 2, characterized in that: The online learning mechanism includes updating the weight coefficients of the market features, technical features, and environmental features based on a gradient descent algorithm, employing a sliding time window strategy to retain data for the latest preset number of days, and introducing a forgetting factor to weaken the weight of historical data. The step of updating the weight coefficients of the market features, technical features, and environmental features based on the gradient descent algorithm includes the following steps: Receive new daily operational data, electricity market transaction data, and equipment status data of the energy storage system. The short-term earnings prediction error is used as the loss function. The weighting coefficients of the market characteristics, technological characteristics, and environmental characteristics are adjusted through backpropagation.

4. The economic evaluation method for energy storage systems according to claim 2, characterized in that: The dynamic feature engineering module also includes an abnormal data cleaning unit, which uses a generative adversarial network to simulate extreme environment data, and includes the following steps: Receive historical data and generate synthetic data containing extreme scenarios; Distinguish between real and synthetic data, and improve the realism of synthetic data through adversarial training; By adding synthetic data to the training set, the economic prediction model is exposed to more extreme scenarios during the training phase, thereby improving its ability to generalize to outliers.

5. The economic evaluation method for energy storage systems according to claim 1, characterized in that: It also includes a real-time correction mechanism that triggers online fine-tuning of model parameters when the actual return deviates from the predicted value and exceeds a preset value. The online fine-tuning includes the following steps: Update LSTM parameters: Learn the latest electricity market transaction data and equipment status data within the preset time range; Corrected attenuation factor: Adjust long-term forecasts based on actual operating data of the energy storage system.

6. The economic evaluation method for energy storage systems according to claim 1, characterized in that: A reward-risk balanced charging and discharging strategy is generated through reinforcement learning algorithms, and the effect of the strategy is simulated in a digital twin environment. High-risk operations are dynamically adjusted, including multi-objective game optimization and real-time strategy simulation. The multi-objective game optimization is to balance returns and risks by establishing a return-risk balance equation, where the risk coefficient λ automatically matches conservative or aggressive strategies according to the investor type. The real-time strategy simulation includes rehearsing the strategy effect in a digital twin environment, and prohibiting high-risk operations when the simulated net present value is lower than a threshold.

7. The economic evaluation method for energy storage systems according to claim 6, characterized in that: The constructed payoff-risk balance equation is as follows: U=E-λ·Var in: U is the utility function representing the overall economics of the energy storage system, E represents the expected return, Var represents the variance of risk, and λ represents the risk coefficient.

8. The method for evaluating the economic efficiency of an energy storage system according to claim 6, characterized in that: The provision regarding pre-simulating the strategy's effectiveness in a digital twin environment, prohibiting high-risk operations when the simulated net present value falls below a threshold, includes: The system acquires multi-source data, including at least real-time market data and equipment status data. It constructs a virtual model that is highly consistent with the real environment using the acquired multi-source data. The system then uses the virtual model to simulate a strategy and obtains the net present value after the strategy is executed. If the simulated net present value is lower than a preset threshold, the system automatically prohibits the execution of the strategy and triggers the generation of alternative solutions.

9. An economic evaluation system for energy storage systems, characterized in that: The system includes a memory and a processor. The memory stores a program for determining the economic evaluation method of an energy storage system. The processor runs the program for determining the economic evaluation method of an energy storage system to enable the energy storage system economic evaluation system to perform the energy storage system economic evaluation method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, include: The computer-readable storage medium stores a program for determining the economic evaluation method of an energy storage system. When the program for determining the economic evaluation method of an energy storage system is executed by a processor, it implements the economic evaluation method of an energy storage system as described in any one of claims 1 to 8.

Images