Industrial and commercial energy storage energy management method based on multiple prediction models

Abstract

The invention discloses an industrial and commercial energy storage energy management method based on multiple prediction models, and the method comprises the steps: obtaining the power utilization data information of industrial and commercial users, predicting the future power utilization demands through multiple models, and correspondingly obtaining a prediction result; performing cooperative processing on the prediction result based on a multi-model integration method of a neural network to obtain a comprehensive prediction result; and formulating a charging and discharging scheduling strategy of the energy storage system according to the comprehensive prediction result, controlling the energy storage system to charge or discharge, continuously monitoring the change of the power demand and the market price, dynamically adjusting the weight of the prediction model, and optimizing the charging and discharging scheduling strategy of the energy storage system in real time according to the latest prediction result. By combining various prediction models and dynamic adjustment mechanisms, the scheduling precision of the energy storage system is remarkably improved, prediction deviation caused by a single model is avoided, price fluctuation of an electricity market can be flexibly coped with, the economic benefit is maximized, and the operation cost is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy storage energy management, and particularly relates to an industrial and commercial energy storage energy management method based on multiple prediction models. Background Art

[0002] Industrial and commercial energy storage energy management based on multiple prediction models refers to accurately predicting factors such as industrial and commercial energy demand, the charging and discharging processes of energy storage devices, and electricity market prices by using multiple prediction methods (such as machine learning, time series analysis, data mining, etc.), and then realizing the optimal scheduling and management of the energy storage system. Specifically, by predicting future power demand fluctuations, electricity price changes, and the remaining energy status of the energy storage system, it can provide the best energy storage strategies for industrial and commercial users, optimize energy use efficiency, reduce electricity costs, and improve system reliability. For example, the prediction model can help determine the best time for power charging (such as charging when the electricity price is low) and discharging (such as discharging during peak power demand or high electricity prices), so as to achieve cost savings and reasonable allocation of energy. This intelligent energy management can not only improve the economic benefits of the energy storage system but also promote the use of sustainable energy.

[0003] The existing technology has the following deficiencies: In the process of industrial and commercial energy storage energy management based on multiple prediction models, if different prediction models fail to work effectively together, it may lead to serious deviations in the scheduling strategy of the energy storage system. For example, some models may predict a decrease in power demand, while other models predict a sharp increase. This inconsistency in prediction results will cause the energy storage system to fail to respond in a timely manner at critical moments, resulting in overcharging or over-discharging. If overcharged, it may accelerate the aging or damage of the energy storage device; if it cannot discharge in time during the peak demand period, it may cause industrial and commercial users to face insufficient energy supply, and in severe cases, it may even affect the normal operation of production and operation. In addition, the incorrect scheduling of the energy storage system may also have an adverse impact on the stability of the power grid, resulting in unbalanced energy supply, thus triggering a series of chain reactions and bringing huge economic losses to the enterprise. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides an industrial and commercial energy storage energy management method based on multiple prediction models, including:

[0005] Obtain the historical electricity consumption data of industrial and commercial users, the current state of the energy storage device, and the electricity market price information;

[0006] Based on different types of prediction models, respectively predict the power demand, power price fluctuations, and charging and discharging demands of the energy storage device within a future period of time, and correspondingly obtain prediction results;

[0007] The collaborative processing of the prediction results is carried out by a multi-model integration method based on a neural network to obtain a comprehensive prediction result;

[0008] According to the comprehensive prediction result, considering the health status of the energy storage device, the user's power consumption demand, and the peak period of electricity price, a charging and discharging scheduling strategy for the energy storage system is formulated;

[0009] Control the energy storage system to charge or discharge according to the charging and discharging scheduling strategy, continuously monitor the changes in power demand and market price, dynamically adjust the weights of the prediction models, and optimize the charging and discharging scheduling strategy of the energy storage system in real time according to the latest prediction results.

[0010] Preferably, the prediction models include a time series analysis model and a machine learning model;

[0011] Among them, the time series analysis model uses the autoregressive integrated moving average algorithm for prediction;

[0012] The machine learning model uses the support vector machine algorithm for prediction.

[0013] Preferably, the process of the collaborative processing of the prediction results by the multi-model integration method based on a neural network to obtain a comprehensive prediction result includes:

[0014] Perform reliability analysis on the prediction results of different types of prediction models, and dynamically adjust the weights of the prediction models according to the reliability analysis results.

[0015] Preferably, the process of performing reliability analysis on the prediction results of different types of prediction models includes:

[0016] Normalize the prediction results of each prediction model, and perform collaborative processing using the weighted average method to determine the weight of each model;

[0017] Weightedly average the prediction results of all prediction models according to the corresponding weights to generate a comprehensive prediction result.

[0018] Preferably, the process of normalizing the prediction results of each prediction model and performing collaborative processing using the weighted average method to determine the weight of each model includes:

[0019] Define the output of the prediction model as P i (t), where i is the number of the prediction model and t is the prediction time. Standardize the prediction results P i (t) of each model to unify the scale;

[0020] Based on the historical prediction accuracy of each prediction model and the current model prediction error, determine the weight of each model, and dynamically adjust the weight value by comparing the prediction errors; the weight formula expression is:

[0021]

[0022] Wherein, w i is the dynamic weight of model i at time t, MAE(Pi(t)) is the mean absolute error of model i at time t, RMSE(P i (t)) is the root mean square error of model i at time t, and both α and β are adjustment factors. α is used to control the influence degree of MAE in calculating the weight, and β is used to control the influence degree of RMSE in calculating the weight.

[0023] Preferably, the predicted results of all prediction models are weighted and averaged according to the corresponding weights, and the formula expression for generating the comprehensive prediction result is:

[0024] The predicted results of all models are weighted and averaged according to the weight w i to generate the comprehensive prediction result, and the calculation expression is as follows:

[0025]

[0026] Wherein, w i is the model weight, P com (t) is the comprehensive prediction result, and n is the number of prediction models.

[0027] Preferably, the process of dynamically adjusting the weights of the prediction models includes:

[0028] At each time period t, calculate the prediction error of each model, and the formula expression is:

[0029] e i (t) = |P true (t) - P i (t)|

[0030] Wherein, e i (t) is the model prediction error, and P true (t) is the actual observed value;

[0031] Based on the prediction error e i (t) and the error variance calculate the dynamic weight of each model, and the formula expression is:

[0032]

[0033] Wherein, w i (t) is the weight after dynamic adjustment for model i at time t, and both γ and δ are adjustment factors. γ is used to control the influence degree of the prediction error in weight calculation, and δ is used to control the influence of the model error variance in weight calculation;

[0034] By normalizing the weights of all prediction models, the normalized weights are obtained, and the formula expression is:

[0035]

[0036] In the formula, w i '(t) is the normalized weight;

[0037] The final comprehensive prediction result is generated by the weighted average method, and the formula expression is:

[0038]

[0039] In the formula, P com (t)' is the final comprehensive prediction result.

[0040] Preferably, the charge-discharge scheduling strategy takes into account the remaining service life of the energy storage device, and automatically limits the charging power when overcharging is predicted.

[0041] Preferably, the process of continuously monitoring the changes in power demand and market price and dynamically adjusting the weights of the prediction model includes:

[0042] Real-time obtain key data such as the charge-discharge state, temperature, and health state of the energy storage device through intelligent metering devices, and feedback them into the prediction model to adjust the prediction range and accuracy of the prediction model.

[0043] Preferably, the process of optimizing the charge-discharge scheduling strategy of the energy storage system in real time according to the latest prediction results includes:

[0044] According to the charge-discharge efficiency and loss of the battery of the energy storage system, use a dynamic optimization algorithm to optimize the charge-discharge strategy.

[0045] Compared with the prior art, the present invention has the following advantages and technical effects:

[0046] By combining multiple prediction models and a dynamic adjustment mechanism, the present invention significantly improves the scheduling accuracy of the energy storage system, avoids the prediction deviation caused by a single model, ensures an accurate charge-discharge strategy, and improves the system stability. Through multi-model integration and optimized scheduling strategies, it can flexibly respond to power market price fluctuations, maximize economic benefits, and reduce operating costs. Combining real-time monitoring and intelligent scheduling, the present invention also effectively avoids overcharging and over-discharging of energy storage devices, extends the device life, improves the long-term stability and reliability of the system, and ensures the competitive advantage of enterprises in cost control and energy management.

[0047] By combining various different types of prediction models (such as time series analysis, machine learning, and deep learning models), the present invention can comprehensively capture power demand fluctuations and market price changes from multiple dimensions, thereby improving the accuracy of prediction. In addition, by dynamically adjusting the weights of the models, the models that perform well at a specific moment can account for a larger proportion in the final decision-making, further enhancing the prediction accuracy of the energy storage system. This accurate prediction can provide a more scientific charge-discharge scheduling strategy for the energy storage system, thus optimizing the allocation of power resources, avoiding overcharging or over-discharging problems caused by prediction errors, and ensuring that the energy storage system always operates efficiently and stably.

[0048] Through the integration of multiple prediction models and dynamic optimization scheduling strategies, the present invention can effectively reduce the operating costs of the energy storage system. Especially in the case of large fluctuations in electricity market prices, by adopting the multi-model integration method of the present invention, the system can judge the best charge-discharge timing according to real-time prediction results, charge when the electricity price is low, and discharge when the electricity price peaks, thereby maximizing the economic benefits of the energy storage system. In addition, by further adjusting the scheduling strategy of the energy storage system through dynamic optimization algorithms (such as genetic algorithms or particle swarm optimization algorithms), it is possible to minimize energy waste while meeting power demand, improving the overall economic benefits of the energy storage system. In this way, not only can the electricity costs of industrial and commercial users be reduced, but also the energy use efficiency of enterprises can be improved, helping them gain a greater competitive advantage in cost control and energy management.

[0049] By introducing a health monitoring and real-time feedback mechanism for energy storage devices, the present invention effectively avoids overcharging and over-discharging of the energy storage system, thereby extending the service life of the devices.

[0050] The present invention combines real-time sensor data (such as device temperature, battery status, etc.) to dynamically adjust the charge-discharge strategy to ensure that the device operates in the best state. Specifically, when the system monitors that the health status of the energy storage device is abnormal (such as overheating, over-discharging, etc.), it will automatically adjust the charge-discharge strategy to avoid damaging the device. In addition, by optimizing the charge-discharge cycle and power, the system can achieve intelligent management of the energy storage device, slow down the device aging process, and ensure that the device maintains stable performance for a long time. This not only improves the long-term operation stability of the energy storage system, but also effectively avoids operation interruptions and economic losses caused by device failures, ensuring the reliability of the enterprise's energy supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The drawings forming a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0052] Figure 1 It is a schematic flowchart of the method according to an embodiment of the present invention. Detailed implementation manners

[0053] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0054] It should be noted that the steps shown in the flowchart of the drawings may be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than here.

[0055] As Figure 1 shown, in this embodiment, a method for industrial and commercial energy storage energy management based on multiple prediction models is provided, including:

[0056] Obtain the historical electricity consumption data of industrial and commercial users, the current state of the energy storage device, and the electricity market price information;

[0057] Use at least two different types of prediction models to respectively predict the electricity demand, electricity price fluctuations, and charging and discharging demands of the energy storage device in a future period of time;

[0058] Perform collaborative processing on the results of the prediction models to generate a comprehensive prediction result. The collaborative processing is based on a multi-model integration method of a neural network to ensure the collaborative consistency between each prediction model;

[0059] Based on the comprehensive prediction result, formulate a charging and discharging scheduling strategy for the energy storage system, and consider the health state of the energy storage device, the user's electricity consumption demand, and the peak period of electricity price;

[0060] Control the energy storage system to charge or discharge according to the formulated scheduling strategy to optimize the usage efficiency and lifespan of the energy storage device;

[0061] Continuously monitor the changes in electricity demand and market price, dynamically adjust the weights of the prediction models, and optimize the scheduling strategy of the energy storage system in real time according to the latest prediction result;

[0062] Further, different types of prediction models include but are not limited to time series analysis models and machine learning models. Among them, the time series analysis model in this embodiment adopts the ARIMA (Autoregressive Integrated Moving Average) algorithm, and the machine learning model adopts the support vector machine (SVM) algorithm.

[0063] Further, the process of performing collaborative processing on the results of the prediction models to generate a comprehensive prediction result includes:

[0064] Perform reliability analysis on the results of each prediction model, and dynamically adjust the weights of the prediction models according to the results of the reliability analysis to ensure that the proportion of the more accurate prediction models in the final results is higher.

[0065] Furthermore, the specific steps for the reliability analysis of the results of each prediction model are as follows:

[0066] Normalize the results of each prediction model, and define the output of the prediction model as P i (t), where i is the number of the prediction model and t is the prediction time. Standardize the prediction results P i (t) of each model to make it conform to a unified scale;

[0067] Use the weighted average method for collaborative processing. Based on the historical prediction accuracy of each prediction model and the current model prediction error, determine the weight of each model, and dynamically adjust the weight value by comparing the prediction errors. The weight calculation formula is as follows:

[0068]

[0069] In the formula, w i is the dynamic weight of model i at time t, MAE(P i (t)) is the mean absolute error of model i at time t, RMSE(P i (t)) is the root mean square error of model i at time t, and both α and β are adjustment factors. α is used to control the influence degree of MAE (mean absolute error) when calculating the weight, and β is used to control the influence degree of RMSE (root mean square error) when calculating the weight;

[0070] Weight the prediction results of all models according to the weight w i and perform weighted averaging to generate a comprehensive prediction result. The calculation expression is as follows:

[0071]

[0072] In the formula, P com (t) is the comprehensive prediction result, indicating the final predicted value obtained by weighted summation of the results of each prediction model at time t, and n is the number of prediction models.

[0073] Furthermore, the specific steps for dynamically adjusting the weights of the prediction models are as follows:

[0074] At each time period t, calculate the prediction error of each model. The calculation expression is as follows:

[0075] e i (t) = |P true (t) - P i (t)|

[0076] In the formula, e i (t) is the model prediction error, and P true (t) is the actual observed value;

[0077] Based on the prediction error e i (t) and the error variance Calculate the dynamic weight of each model, and the calculation expression is as follows:

[0078]

[0079] In the formula, w i (t) is the weight after dynamic adjustment for model i at time t. Both γ and δ are adjustment factors. γ is used to control the influence degree of the prediction error in the weight calculation, and δ is used to control the influence of the model error variance in the weight calculation;

[0080] By normalizing the weights of all prediction models, the normalized weights are obtained so that the sum of all weights is 1, and the calculation expression is as follows:

[0081]

[0082] In the formula, w i '(t) is the normalized weight;

[0083] The final comprehensive prediction result is generated by the weighted average method, and the generation formula is as follows:

[0084]

[0085] In the formula, P com (t)' is the final comprehensive prediction result

[0086] Furthermore, based on the comprehensive prediction result, a charge-discharge scheduling strategy for the energy storage system is formulated, and the health state of the energy storage device, the user's electricity demand, and the peak period of electricity price are considered;

[0087] The charge-discharge scheduling strategy designed in this embodiment also considers the remaining service life of the energy storage device. When it is predicted that the charging is excessive, the scheduling strategy will take measures to limit the charging power to avoid shortening the service life due to overuse of the device.

[0088] Furthermore, the charge-discharge scheduling strategy designed in this embodiment also includes further optimizing the charge-discharge strategy using a dynamic optimization algorithm (such as a genetic algorithm or a particle swarm optimization algorithm) according to the charge-discharge efficiency and loss of the battery of the energy storage system, reducing energy loss and extending the service life of the energy storage device.

[0089] Furthermore, the process of continuously monitoring the changes in power demand and market price and dynamically adjusting the weights of the prediction model includes: obtaining key data such as the charge and discharge status, temperature, and health status of the energy storage device in real time through intelligent metering devices, and feeding them back into the prediction model to adjust the prediction range and accuracy of the model to ensure that the energy storage system is always in the best working state.

[0090] Embodiment 1: The core of this embodiment is to utilize the historical electricity consumption data of industrial and commercial users, the price fluctuations in the electricity market, and the operating status of energy storage devices to construct multiple prediction models, and optimize the energy management and scheduling of the energy storage system by dynamically optimizing the weight allocation of these models. The specific implementation process is divided into the following steps:

[0091] First, the historical electricity consumption data of industrial and commercial users will be analyzed as a basic data source. The time series analysis method (such as the ARIMA model) is used to predict the power demand fluctuations in the next period of time. ARIMA (AutoRegressive Integrated Moving Average) is a linear model based on historical data, which can well capture the trends, seasonality, and cyclic changes in the data. Therefore, based on the historical electricity consumption data, the ARIMA model can effectively predict the future demand changes and provide a decision-making basis for the charge and discharge scheduling of the energy storage system.

[0092] Secondly, combined with the price fluctuations in the electricity market, a machine learning model (such as the support vector machine SVM or decision tree model) is used to predict the change trend of the electricity market price. The price fluctuations in the electricity market are usually affected by various factors, including power supply, demand, policies, etc. Therefore, a single linear model (such as ARIMA) may not be able to fully reflect the dynamic changes in the electricity market, and the machine learning model can better handle non-linear and complex relationships, thus making a more accurate prediction of the price fluctuations in the electricity market.

[0093] Then, the prediction results of all models (including machine learning models such as ARIMA and SVM) will be comprehensively processed. The comprehensive prediction results are calculated using the weighted average method, and the weight allocation is based on the prediction accuracy of the models. To ensure the collaborative work among different models, the weight adjustment is not fixed but dynamic. The weights of the models are adjusted according to factors such as the historical prediction accuracy and real-time error. If the prediction error of a certain model is small, its weight will increase accordingly, thus occupying a larger proportion in the comprehensive prediction results. Specifically, we can dynamically adjust the weights of the models according to the MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) of each model. For example, when the error of a certain model is small, it indicates that the model performs excellently in the current prediction task, and then its weight will be increased, otherwise decreased. In this way, the system can automatically "learn" which models are more reliable in different situations, thereby improving the overall prediction accuracy.

[0094] In the charge and discharge scheduling of the energy storage system, the comprehensive prediction results will be used as the decision-making basis to optimize the use of the energy storage system. Specifically, when the power demand is low, the energy storage device will charge according to the predicted power market price, while when the power demand is high or the power market price is high, the energy storage device will discharge. Through this dynamic scheduling strategy, not only can the grid load be effectively balanced, but also the electricity cost of enterprises can be significantly reduced. In addition, regarding the aging problem of the energy storage device, this method will also consider the health status of the device and avoid overusing the energy storage device during the charge and discharge process to extend the service life of the device.

[0095] This method of optimizing weight allocation based on historical data can give full play to the advantages of each model in the collaborative work of multiple models and avoid the problem that a single model overly relies on a certain specific prediction direction. It can effectively cope with the volatility of the power market and ensure that the energy storage system can make accurate and economic scheduling decisions under different time periods and demand conditions, thereby achieving the optimization of industrial and commercial energy storage energy management.

[0096] Embodiment 2: This embodiment optimizes the weights of the prediction models of the energy storage system by introducing a real-time data feedback mechanism, thereby improving the response ability and prediction accuracy of the system. The core idea of the dynamic feedback mechanism is to use the actual operation data of the energy storage system to continuously correct and adjust the weights of each prediction model, so that the energy storage system can make real-time optimal scheduling in the face of the changing power market and user demands. The specific implementation steps are as follows:

[0097] First, the system will collect key data such as the charge and discharge status, battery health status, and temperature of the energy storage system in real time through sensors and intelligent metering devices. In addition, the real-time electricity market price and the real-time electricity demand of industrial and commercial users will also be monitored by the system. All of this real-time data will be used as input signals and fed back into the prediction model to help the system evaluate and correct the prediction accuracy of each model.

[0098] Then, the system will calculate the prediction error of each model and compare it with the historical prediction error. Based on the real-time data, the system can evaluate the performance of each model in the current situation in real time. If the prediction error of a certain model is large at the current moment, the system will reduce the weight of this model, thereby reducing the impact of this model on the final comprehensive prediction result. On the contrary, if the prediction result of a certain model is very close to the actual situation, the system will automatically increase the weight of this model, thereby strengthening the role of this model in the comprehensive prediction.

[0099] Finally, the introduction of the dynamic feedback mechanism enables the energy storage system to perform optimized scheduling based on real-time data. For example, if the electricity market price rises sharply and a certain model predicts an increase in electricity demand, the system will preferentially execute the prediction result of this model and adjust the discharge amount of the energy storage device to the optimal level; if the system monitors situations such as overcharging of the energy storage device or abnormal battery temperature, the system will also adjust the charging strategy according to the real-time feedback to avoid damage to the energy storage device.

[0100] Through real-time feedback adjustment, this embodiment enables the energy storage system to always maintain the optimal charge and discharge strategy in a complex and changing power environment, thereby achieving the stable operation, cost optimization, and equipment protection of the industrial and commercial energy storage system.

[0101] Embodiment 3: This embodiment combines multiple prediction models and intelligent optimization algorithms to achieve intelligent management of the charge and discharge scheduling of the energy storage system. Through the combination of multi-model integration and optimization algorithms, the system can improve the efficiency of charge and discharge scheduling and the long-term stability of energy storage devices while ensuring high-precision prediction. The specific implementation steps are as follows:

[0102] First, the system uses multiple prediction models to predict electricity demand, electricity market price, and the status of energy storage devices. These include the ARIMA model based on time series, the support vector machine (SVM) model based on machine learning, and deep learning models, etc. Each model has different advantages for different data features. The time series model is good at capturing long-term trends and seasonal changes, the machine learning model can handle complex non-linear relationships, and the deep learning model can process high-dimensional and complex power system data. By integrating these models, the system can make full use of the advantages of different models to provide more accurate and comprehensive prediction results.

[0103] Based on the prediction results, the system adopts intelligent optimization algorithms (such as genetic algorithms or particle swarm optimization algorithms) to further optimize the charge-discharge strategy of the energy storage system. Genetic algorithms can find the global optimal solution among multiple candidate solutions by simulating the natural selection process and are suitable for complex non-linear optimization problems. Particle swarm optimization algorithms simulate the foraging behavior of bird flocks and can quickly find the optimal solution through information sharing among particles. In the scheduling of energy storage systems, these optimization algorithms can adjust the charge-discharge strategy of energy storage devices in real time according to changes in power demand and market prices to minimize energy waste and maximize economic benefits.

[0104] Specifically, the optimization algorithm will intelligently adjust the charge-discharge cycle of the energy storage device according to multiple factors such as the charging efficiency, discharging efficiency, battery health status of the energy storage device, and electricity market price during power demand fluctuations. For example, when the electricity market price is low, the system will guide the energy storage device to charge through the optimization algorithm; during the peak power demand period, the system will discharge to meet user needs and reduce costs. The optimization algorithm will also monitor the device status in real time to avoid battery damage caused by overcharging and over-discharging and ensure the service life of the energy storage device.

[0105] Through the combination of multi-model integration and intelligent optimization algorithms, this embodiment can effectively improve the charge-discharge scheduling ability of the energy storage system and achieve efficient, intelligent, and stable energy management.

[0106] This method can not only cope with the uncertainty of the electricity market but also perform dynamic scheduling according to the actual needs of industrial and commercial users and the health status of energy storage devices to ensure that the system always operates in the best state.

[0107] In summary, through the combination of multiple prediction models and intelligent optimization algorithms, this embodiment further improves the management accuracy and scheduling efficiency of the energy storage system in a complex power environment, providing an efficient and intelligent solution for industrial and commercial energy storage management.

[0108] By introducing a variety of prediction models and a dynamic adjustment mechanism, the present invention can significantly improve the scheduling accuracy of the energy storage system. In traditional energy storage management systems, a single model is often relied on to predict power demand and electricity market prices, resulting in low prediction accuracy or model bias. However, by combining multiple different types of prediction models (such as time series analysis, machine learning, and deep learning models), the present invention can comprehensively capture power demand fluctuations and market price changes from multiple dimensions, thereby improving the accuracy of prediction. In addition, by dynamically adjusting the weights of the models, the models that perform well at a specific moment can account for a larger proportion in the final decision, thus further improving the prediction accuracy of the energy storage system. This accurate prediction can provide a more scientific charge-discharge scheduling strategy for the energy storage system, thereby optimizing the allocation of power resources, avoiding overcharging or over-discharging problems caused by prediction errors, and ensuring that the energy storage system always operates efficiently and stably.

[0109] By integrating multiple prediction models and dynamically optimizing the scheduling strategy, the present invention can effectively reduce the operating costs of the energy storage system, especially in the case of large fluctuations in electricity market prices. Traditional energy storage systems often cannot flexibly respond to fluctuations in electricity market prices, so they fail to charge in time when electricity prices are low or discharge in time during peak price periods, thus missing opportunities to save costs. Using the multi-model integration method of the present invention, the system can determine the best charge-discharge timing based on real-time prediction results, charge when electricity prices are low, and discharge during peak electricity price periods, thereby maximizing the economic benefits of the energy storage system. In addition, by further adjusting the scheduling strategy of the energy storage system through dynamic optimization algorithms (such as genetic algorithms or particle swarm optimization algorithms), it is possible to minimize energy waste while meeting power demand, and improve the overall economic benefits of the energy storage system. In this way, not only can the electricity costs of industrial and commercial users be reduced, but also the energy use efficiency of enterprises can be improved, helping them gain a greater competitive advantage in cost control and energy management.

[0110] By introducing a health monitoring and real-time feedback mechanism for energy storage devices, the present invention effectively avoids overcharging and over-discharging of energy storage systems, thereby extending the service life of the devices. Energy storage devices, especially battery energy storage systems, are negatively affected by overcharging or over-discharging during long-term charging and discharging processes, which in turn affects the performance and lifespan of the devices. The present invention combines real-time sensor data (such as device temperature, battery status, etc.) to dynamically adjust the charging and discharging strategies to ensure that the devices operate in the best state. Specifically, when the system monitors an abnormal health state of the energy storage device (such as overheating, over-discharging, etc.), it will automatically adjust the charging and discharging strategies to avoid damaging the device. In addition, by optimizing the charging and discharging cycles and power, the system can achieve intelligent management of energy storage devices, slow down the device aging process, and ensure stable performance of the devices over a long period of time. This not only improves the long-term operational stability of the energy storage system, but also effectively avoids operational interruptions and economic losses caused by device failures, ensuring the reliability of the enterprise's energy supply.

[0111] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An industrial and commercial energy storage energy management method based on multiple prediction models, characterized in that, Including: Obtain the historical electricity consumption data of industrial and commercial users, the current state of energy storage devices, and electricity market price information; Based on different types of prediction models, respectively predict the electricity demand, electricity price fluctuations, and charge and discharge demands of energy storage devices within a future period of time, and obtain corresponding prediction results; Perform collaborative processing on the prediction results by a multi-model integration method based on a neural network to obtain a comprehensive prediction result; According to the comprehensive prediction result, considering the health status of the energy storage device, the user's electricity demand, and the peak electricity price period, formulate a charge and discharge scheduling strategy for the energy storage system; Control the energy storage system to charge or discharge according to the charge and discharge scheduling strategy, and at the same time continuously monitor the changes in electricity demand and market price, dynamically adjust the weights of the prediction models, and optimize the charge and discharge scheduling strategy of the energy storage system in real time according to the latest prediction results.

2. The method according to claim 1, characterized in that The prediction models include a time series analysis model and a machine learning model; Among them, the time series analysis model uses the autoregressive integrated moving average algorithm for prediction; The machine learning model uses the support vector machine algorithm for prediction.

3. The method according to claim 1, characterized in that The process of performing collaborative processing on the prediction results by a multi-model integration method based on a neural network to obtain a comprehensive prediction result includes: Perform reliability analysis on the prediction results of different types of prediction models, and dynamically adjust the weights of the prediction models according to the reliability analysis results.

4. The method according to claim 3, characterized in that The process of performing reliability analysis on the prediction results of different types of prediction models includes: Perform normalization processing on the prediction results of each prediction model, and use the weighted average method for collaborative processing to determine the weight of each model; Perform weighted averaging on the prediction results of all prediction models according to the corresponding weights to generate a comprehensive prediction result.

5. The method according to claim 4, characterized in that The process of performing normalization processing on the prediction results of each prediction model and using the weighted average method for collaborative processing to determine the weight of each model includes: Define the output of the prediction model as P i (t), where i is the number of the prediction model and t is the prediction time. Standardize the prediction results P i (t) of each model to unify the scale; Based on the historical prediction accuracy and current model prediction error of each prediction model, determine the weight of each model, and dynamically adjust the weight value by comparing the prediction errors; the weight formula expression is: where w i is the dynamic weight of model i at time t, MAE(P i (t)) is the mean absolute error of model i at time t, RMSE(P i (t)) is the root mean square error of model i at time t, and both α and β are adjustment factors. α is used to control the influence degree of MAE when calculating the weight, and β is used to control the influence degree of RMSE when calculating the weight.

6. The method according to claim 4, characterized in that The formula expression for performing weighted averaging on the prediction results of all prediction models according to the corresponding weights to generate a comprehensive prediction result is: Weight the prediction results of all models according to the weight w i Perform weighted averaging to generate a comprehensive prediction result. The calculation formula is as follows: where w i is the model weight, P com (t) is the comprehensive prediction result, and n is the number of prediction models.

7. The method according to claim 1, characterized in that The process of dynamically adjusting the weights of the prediction models includes: At each time period t, calculate the prediction error of each model, and the formula expression is: e i (t) = |P ture (t) - P i (t)| where, e i (t) is the model prediction error, and P true (t) is the actual observed value; Based on the prediction error e i (t) and the error variance Calculate the dynamic weight of each model, and the formula expression is: where w i (t) is the weight after dynamic adjustment for model i at time t, γ and δ are both adjustment factors, γ is used to control the influence degree of prediction error in weight calculation, and δ is used to control the influence of model error variance in weight calculation; Obtain the normalized weights by performing normalization processing on the weights of all prediction models, and the formula expression is: where w i '(t) is the normalized weight; Generate the final comprehensive prediction result by the weighted average method, and the formula expression is: where P com (t)' is the final comprehensive prediction result.

8. The method according to claim 1, characterized in that The charge and discharge scheduling strategy considers the remaining service life of the energy storage device, and automatically limits the charging power when it is predicted that the charging is excessive.

9. The method according to claim 1, wherein: The process of continuously monitoring changes in power demand and market prices and dynamically adjusting the weights of the prediction model includes: Obtaining key data such as the charge and discharge status, temperature, and health status of the energy storage device in real time through intelligent metering devices, and feeding it back into the prediction model to adjust the prediction range and accuracy of the prediction model.

10. The method according to claim 1, wherein: The process of real-time optimizing the charge and discharge scheduling strategy of the energy storage system according to the latest prediction results includes: Optimizing the charge and discharge strategy using a dynamic optimization algorithm based on the charge and discharge efficiency and losses of the battery of the energy storage system.

Images