Energy storage system demand control method, device and medium based on maximum demand prediction

Abstract

The application relates to an energy storage system demand control method based on maximum demand prediction, equipment and medium, which comprises the following steps: S1, collecting data of a target scene to obtain a standardized data set; S2, constructing an improved attention mechanism LSTM prediction model based on the standardized data set; S3, establishing a multi-objective optimization function, and determining the weights of the objectives by using a dynamic weight distribution algorithm; S4, monitoring and analyzing the difference between the actual maximum demand and the monthly maximum demand prediction value by using a rolling window algorithm, and dynamically adjusting the charge-discharge power reference value and the EMS charging power limit value by combining a dynamic sliding error elimination method; and S5, controlling the energy storage inverter PCS to realize millisecond-level charge-discharge switching through an EMS-led hierarchical response control architecture according to the adjusted charge-discharge power reference value and the charging power limit value. Compared with the prior art, the application has the advantages of improving the accuracy, response speed and economy of energy storage demand control.

Description

Technical Field

[0001] This invention relates to the field of energy storage system control technology, and in particular to a method, device and medium for demand control of energy storage systems based on maximum demand prediction. Background Technology

[0002] With the deepening of the construction of new power systems, the fluctuation of electricity load in industrial and commercial settings, industrial park microgrids, and other scenarios is becoming increasingly severe. For users subject to two-part tariffs, the basic electricity charge is calculated based on the actual maximum demand for the month. If a sudden increase in electricity load or the superposition of energy storage charging causes the demand to break the monthly peak at a certain point in the month, it will directly increase the basic electricity cost for the entire month. Energy storage systems smooth the electricity load curve by "peak shaving and valley filling," and are the core equipment for reducing the maximum monthly demand and lowering the basic electricity cost. They have become an important means for industrial and commercial users to reduce electricity costs.

[0003] A search of Chinese patent publication CN116961042A reveals a method for controlling demand regulation in a commercial and industrial user-side energy storage system. This method involves real-time monitoring and calculation of user-side load, controlling and adjusting energy storage discharge power and duration by setting demand thresholds. When the threshold is exceeded, the system can self-learn and correct it. While this solution enables energy storage discharge control based on real-time demand monitoring, its demand prediction focuses on intraday real-time load monitoring and does not specifically predict the monthly maximum demand. Therefore, it cannot provide a core monthly threshold reference for full-month charging control and does not integrate historical monthly electricity consumption characteristics and coupled monthly energy storage charging data, resulting in a single predictive dimension.

[0004] Chinese Patent Publication No. CN119010145A discloses an energy storage charging and discharging control method. This method obtains the historical load demand curve of the energy storage system based on the historical demand of the grid's incoming and outgoing lines and the historical demand of the energy storage system's incoming and outgoing lines over a preset time period. It then predicts the user's maximum monthly demand based on this curve and adjusts the charging or discharging power of the energy storage system according to the maximum monthly demand and the real-time active power of the grid's incoming and outgoing lines during each charging and discharging time period. While this scheme involves predicting the maximum monthly demand, its prediction model is not optimized for the specific characteristics of the maximum monthly demand and does not employ a deep learning model to dynamically assign weights to input features across different dimensions, thus its prediction accuracy needs improvement.

[0005] Chinese patent publication CN117060424A discloses a demand control method that acquires the current actual load demand, current load power, and current operating status data of the energy storage system at the current moment. This data is then input into a target demand control model to predict the energy storage output value, obtaining the energy storage system's output value for the next moment. Based on this output value, the system's output power is controlled to regulate the load-side demand. While this scheme uses a predictive model for demand control, its prediction object is the current actual load demand, not the monthly maximum demand value, and it lacks a multi-objective optimization function with the primary hard constraint that the sum of charging power and load power does not exceed the predicted monthly maximum demand value.

[0006] The aforementioned existing technical solutions have the following main shortcomings in practical applications: First, they do not incorporate "monthly charging not exceeding the predicted maximum monthly demand" into the core control objective. Some methods blindly charge during peak electricity consumption periods in pursuit of peak-valley arbitrage profits, resulting in the charging power being superimposed with the real-time grid demand to refresh the maximum monthly demand, which in turn significantly increases the basic electricity cost for the entire month. Second, they lack a unified control platform centered on EMS for the entire month, leading to a disconnect between charging control and monthly demand control. There is no dynamic charging power hard-limiting mechanism based on the maximum monthly demand, making it impossible to achieve dynamic adjustment of the monthly charging strategy. Third, they suffer from slow response and insufficient robustness. Traditional static threshold triggering mechanisms are unable to quickly respond to sudden fluctuations in peak electricity consumption within the month, and there is no specific error correction mechanism for the maximum monthly demand. Once demand exceeds the limit within the month, it cannot be remedied in time. Fourth, they have poor coordination among multiple objectives, focusing solely on intraday peak-valley arbitrage or short-term demand control while ignoring the core objective of monthly demand control, charging demand constraints, and the long-term operational safety of energy storage equipment, thus failing to fully realize the comprehensive benefits.

[0007] Therefore, the existing technologies have problems such as low accuracy of monthly maximum demand prediction, failure to include "monthly charging not exceeding the predicted monthly maximum demand" as a core control objective, lack of a unified monthly control carrier with EMS as the core, insufficient robustness due to response lag, poor coordination of multiple objectives, and the fact that the energy storage charging process drives up the grid's monthly maximum demand, resulting in a significant increase in the basic electricity cost for the whole month. These are the technical problems that need to be solved. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art by providing a demand control method, device and medium for energy storage systems based on maximum demand prediction.

[0009] The objective of this invention can be achieved through the following technical solutions: According to a first aspect of the present invention, a method for demand control of an energy storage system based on maximum demand prediction is provided, the method comprising the following steps: Step S1: Collect historical monthly electricity consumption data, real-time electricity consumption data, environmental data, and energy storage system operation data of the target scenario, and clean, normalize, and extract features from the data to obtain a standardized dataset. Step S2: Construct an improved attention mechanism LSTM prediction model based on a standardized dataset to perform multi-dimensional monthly maximum demand prediction, and output the predicted value of the monthly maximum demand of the power grid for the target month and the prediction error range. Step S3: With the objectives of ensuring that the sum of the energy storage charging power and the load power does not exceed the predicted monthly maximum demand, achieving demand control targets, maximizing peak-valley arbitrage profits, and minimizing energy storage system losses, a multi-objective optimization function is established. A dynamic weight allocation algorithm is used to determine the weights of each objective, and the benchmark value of the energy storage system's charging and discharging power is obtained. At the same time, the energy storage energy management system (EMS) generates a dynamically adjustable charging power limit value. Step S4: The rolling window algorithm is used to monitor and analyze the difference between the actual maximum demand and the predicted monthly maximum demand. Combined with the dynamic slip error elimination method, the charging and discharging power reference value and the EMS charging power limit value are dynamically adjusted. Step S5: Based on the adjusted charging and discharging power reference value and charging power limit value, the energy storage inverter PCS is controlled by the EMS-led hierarchical response control architecture to achieve millisecond-level charging and discharging switching, complete the monthly demand control, and provide real-time feedback of the operating status to the EMS prediction model and optimization module, forming a full closed-loop control of the EMS.

[0010] As a preferred technical solution, the historical monthly electricity consumption data in step S1 includes the maximum monthly demand value for each month over the past 1 to 3 years, the distribution of peak electricity consumption periods within the month, and coupled data on grid demand changes during monthly energy storage charging periods; the real-time electricity consumption data includes active power, reactive power, voltage, current, and real-time grid demand data for the current moment of the day; the environmental data includes monthly meteorological trends, daily ambient temperature, humidity, light intensity, and seasonal characteristics; and the energy storage system operation data includes battery SOC, battery temperature, internal resistance, charging and discharging efficiency, inverter operating parameters, and real-time charging power. The feature extraction employs wavelet transform combined with random forest algorithm to extract the trend features of monthly electricity load, peak load fluctuation features within the month, correlation features, and coupling features between energy storage charging and the grid's maximum monthly demand. Abnormal data and redundant features are eliminated, and key time periods within the month that are prone to triggering maximum demand are identified.

[0011] As a preferred technical solution, the improved attention mechanism LSTM prediction model in step S2 is to add a multi-scale attention module to the input layer of the traditional LSTM network, assign dynamic attention weights to input features of different dimensions, and focus on features that are strongly correlated with the maximum monthly demand of the power grid, including monthly production plans, monthly weather trends, monthly start-up and shutdown plans of large equipment, and monthly charging plans for energy storage. During the training process of the improved attention mechanism LSTM prediction model, an adaptive learning rate adjustment algorithm is introduced, combined with an early stopping strategy to avoid overfitting. The output prediction error range is generated based on the Monte Carlo simulation method and is used for the threshold determination of subsequent charging demand constraints. The improved attention mechanism LSTM prediction model is a single-value prediction model for monthly maximum demand. It outputs a unique monthly maximum demand prediction value for the target month. The prediction period is either a calendar month or a billing month, and it is adapted to the billing rules of the power grid that charge basic electricity fees on a monthly basis.

[0012] As a preferred technical solution, the expression of the multi-objective optimization function in step S3 is: in, The charging power for energy storage; Power consumption of the load; This is the maximum demand forecast; The preset demand control threshold and ≤ ; R C represents peak-valley arbitrage profits; C represents the operating losses of the energy storage system. .

[0013] As a preferred technical solution, the dynamic weight allocation algorithm in step S3 specifically involves dynamically adjusting the values ​​of each weight based on the real-time operating conditions within the month: When the actual maximum demand of the power grid Pgrid ≥ P _ predmax When the charge reaches 90%, adjust ω1 to 0.5-0.7, and simultaneously reduce ω3 to limit or even stop charging. During peak hours of the month and Pgrid ≤ P _ predmax When the value is 70%, decrease ω1 to 0.4 and increase ω3 to improve the peak-valley arbitrage profit; When the SOC of the energy storage battery is ≤10% or ≥90%, increase ω4 to 0.4-0.5 to prioritize equipment safety; When the month is nearing the end of the billing cycle and the actual maximum demand has already reached its limit. P _ predmaxAt this time, ω1 is adjusted to 0.7-0.8 to forcibly lock the charging power limit; Peak-valley arbitrage profits R Based on real-time peak-valley electricity prices, system losses are calculated. C This includes battery charging and discharging losses, inverter losses, and line losses.

[0014] As a preferred technical solution, in step S4, the window size of the rolling window algorithm is set to 15-60 minutes based on the monthly electricity consumption characteristics, and the rolling step size is 5-10 minutes. Each time the window is rolled, the superimposed deviation value of the charging demand within the current window is calculated. : Simultaneously monitor the actual maximum demand within the month. P_month_actual Compared with the predicted value P_pred_max The deviation; when Δ P_charge >0 or P_month_actual ≥ P_month_pred When the charge power reaches 95%, a dynamic slip error elimination process is initiated. The charging power limit value for subsequent periods is adjusted by linear interpolation, while the charging and discharging power reference value is corrected to forcibly reduce the charging power. when P_month_actual far below P_pred_max Furthermore, when there is no risk of peak electricity consumption during the day, the charging power limit will be relaxed.

[0015] As a preferred technical solution, the EMS-led hierarchical response control architecture in step S5 is specifically as follows: The main control layer is the core computing module of EMS. It is used to receive monthly maximum demand forecast data and intraday real-time error correction information, output charging and discharging control commands and charging power hard limit commands, and dynamically update and send them to the execution layer throughout the month. The execution layer includes the energy storage inverter PCS and the battery management system BMS. The PCS performs charging operations according to the charging power limit value issued by the EMS, refuses to receive any charging commands that exceed the limit, achieves millisecond-level charging and discharging power switching, and has a response time of no more than 50ms. The BMS monitors the battery status in real time and feeds the data back to the EMS to meet the long-term charging regulation requirements within the month. The monitoring layer includes smart meters, energy storage status acquisition terminals, and grid demand monitoring terminals. It collects real-time grid demand, actual maximum demand within the month, electricity load, and energy storage system operation data in real time and uploads them to the EMS in milliseconds, providing real-time data support for the EMS to adjust its monthly demand control strategy. During the control process, the EMS simultaneously incorporates hard limits on charging power, upper and lower bounds on power, battery SOC threshold, and inverter rated power constraints. This achieves dynamic hard limits on the monthly charging power at the EMS level, ensuring that the sum of charging and electricity load at any time during the month does not exceed the predicted maximum monthly demand, while also preventing the energy storage system from overcharging, over-discharging, or overloading.

[0016] As a preferred technical solution, this method is suitable for two monthly billing modes: contract demand billing and actual maximum demand billing, and the billing mode can be adaptively switched through EMS. When using the contract demand billing model, EMS uses the monthly contract demand value as the core threshold. If the monthly contract demand is ≤ P_month_pred Then replace the contract month demand with P_month_pred As a threshold for charging demand constraints, the weight of ω1 is increased; if the contract monthly demand > P_month_pred, the contract demand control strategy is adjusted based on P_month_pred, and the incremental basic electricity charge for the portion exceeding the contract monthly demand is calculated and included in the optimization objective. When using the actual maximum demand billing model, EMS focuses on optimizing the difference between the actual maximum demand and the predicted maximum demand for the month, ensuring that the actual maximum demand for the month is ≤ P_month_pred The incremental basic electricity fee will be reduced, and the charging power limit will be dynamically adjusted based on real-time electricity consumption data within the month.

[0017] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.

[0018] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0019] Compared with the prior art, the present invention has the following advantages: 1) This invention accurately controls the maximum monthly demand, ensuring that the monthly charging does not exceed the predicted value from the root cause: By accurately predicting the maximum monthly demand in multiple dimensions, combined with the soft optimization constraints of the monthly charging demand at the EMS level and the hard limit constraints of the charging power at the execution layer, the dynamic and accurate matching of charging power and real-time grid demand is achieved, ensuring that the sum of the two does not exceed the predicted maximum monthly demand, thus completely solving the industry pain point of energy storage charging driving up the maximum monthly demand of the grid; it can stably control the actual maximum monthly demand of the grid within the predicted value, with a maximum demand reduction of 10%-30%, significantly reducing users' monthly basic electricity expenses; 2) This invention achieves multi-objective collaborative optimization throughout the month, resulting in significant comprehensive economic benefits: By taking full-month charging compliance as the primary hard constraint objective, it achieves four-fold collaborative optimization: charging does not exceed the predicted maximum monthly demand, monthly demand control meets the target, peak-valley arbitrage profits are maximized within the month, and energy storage system losses are minimized. Under the premise of ensuring charging compliance and long-term equipment safety, it improves peak-valley arbitrage compliance profits and reduces monthly operating losses of the energy storage system. Combined with monthly demand optimization and compliance arbitrage, it can increase the internal rate of return (IRR) of energy storage projects by 5%-10%, fully leveraging the multiple values ​​of the energy storage system and avoiding the problem of "pushing up basic electricity prices for arbitrage". 3) This invention features fast response speed and strong monthly control robustness: Through a hierarchical response control architecture led by EMS, it achieves millisecond-level charging and discharging switching and charging command issuance, enabling rapid response to sudden fluctuations in peak electricity consumption within the month and timely avoidance of demand exceeding limits; at the same time, the rolling window monitoring mechanism for charging demand superposition deviation and the EMS full closed-loop optimization enable the control strategy to adapt to changes in operating conditions within the month, effectively addressing issues such as data acquisition delays, equipment execution deviations, and adjustments to monthly production plans, thereby improving monthly control robustness and preventing a single exceedance from causing an increase in basic electricity costs for the entire month; 4) This invention is based on EMS, is highly versatile and easy to implement: relying on the existing EMS of the energy storage system as the core control carrier, it does not require large-scale modification of existing energy storage equipment, and can be realized only through algorithm upgrades and function expansion; it can flexibly adjust prediction parameters, charging safety margin, weight parameters, etc. according to the monthly electricity consumption characteristics of different scenarios, adapting to various scenarios such as industrial and commercial, park microgrids and other scenarios implementing two-part electricity pricing, and supports two monthly billing modes: contract demand and actual maximum demand. It has the ability to dynamically adapt energy storage capacity, is highly practical, and is easy to promote and apply in the industry; 5) This invention extends the service life of equipment and reduces the total life cycle operating cost: Through EMS's precise SOC control, power constraint, loss optimization, and full-month charging compliance control, it avoids battery overcharging, over-discharging, and overload operation, maintains the consistency of battery cell voltage, and can extend the service life of energy storage batteries by 1-2 years; at the same time, it avoids the incremental basic electricity charges and grid penalties caused by charging exceeding the predicted maximum monthly demand, reduces the total life cycle operating cost of the energy storage system, and improves the overall benefits of the project. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the specific process of the method of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] To address the shortcomings of existing technologies, such as low accuracy in monthly maximum demand prediction, slow response, poor multi-objective coordination, and insufficient robustness, as well as the core technical problems of energy storage charging driving up the grid's predicted monthly maximum demand, this invention provides an energy storage system EMS demand control method based on maximum demand prediction. Using the energy storage management system (EMS) as the core control carrier, it achieves accurate single-value prediction of monthly maximum demand across multiple dimensions, multi-objective optimization with hard constraints on charging demand, real-time error correction within the month, and EMS-led hierarchical rapid response. This ensures, from both optimization and execution perspectives, that the sum of charging power and real-time grid demand at any point in the month does not exceed the predicted monthly maximum demand value. This improves the accuracy and economy of monthly demand control for energy storage, extends the lifespan of energy storage equipment, and is compatible with two billing models where the grid charges basic electricity fees monthly, meeting the practical needs of industrial and commercial applications, industrial park microgrids, and other scenarios.

[0023] like Figure 1 As shown, an energy storage system demand control method based on monthly maximum demand forecasting uses an energy storage management system (EMS) as the core control carrier to achieve a closed-loop process encompassing data acquisition, monthly maximum demand forecasting, full-month optimization, real-time control, and feedback. The core focus is on conducting full-month demand control and charging management around the single value of monthly maximum demand, including the following steps: Step S1: Data Acquisition and Preprocessing The EMS (Energy Management System) coordinates and schedules various data acquisition terminals to collect multi-dimensional data from the target scenario, focusing on monthly electricity consumption characteristics and coupled data of energy storage charging. This provides foundational data for predicting the maximum monthly demand and controlling charging throughout the month. Historical monthly electricity consumption data covers the maximum monthly demand for the past 1-3 years, the distribution of peak electricity consumption periods within the month, and coupled data on grid demand changes during monthly energy storage charging periods. This data is used to uncover historical patterns in monthly electricity load and the triggering characteristics of maximum monthly demand. Real-time electricity consumption data includes active power, reactive power, voltage, current, and real-time grid demand values ​​at the current time of day, collected at a frequency of no less than 10Hz to ensure accurate capture of real-time operating conditions and demand within the month. Environmental data includes monthly weather trends, ambient temperature, humidity, light intensity, and seasonal characteristics, used to analyze the impact of meteorological factors on monthly electricity load and maximum monthly demand. Energy storage system operation data includes battery SOC, battery temperature, internal resistance, charge / discharge efficiency, inverter operating parameters, and real-time charging power, used to ensure the long-term safe operation of the energy storage system and the regulation of charging power within the month.

[0024] The collected data is preprocessed uniformly by EMS: abnormal monthly data and intraday outliers are removed using the 3σ criterion, and missing data are filled in using linear interpolation; all data are normalized and mapped to the [0,1] interval to eliminate the influence of dimensions; wavelet transform combined with random forest algorithm is used for feature extraction, focusing on extracting the trend characteristics of monthly electricity load, the fluctuation characteristics of monthly peak load, and the coupling characteristics of energy storage charging and the grid's monthly maximum demand. At the same time, the characteristics of key periods that are likely to trigger the maximum demand in the month (such as the start-up and shutdown of large equipment and peak production periods) are also explored, and redundant features are removed to obtain a standardized dataset, which is stored by EMS and provides high-quality data support for subsequent single-value prediction of monthly maximum demand.

[0025] Step S2: Multi-dimensional maximum demand unit value prediction Based on the standardized dataset obtained in step S1, EMS calls its built-in dedicated algorithm module to construct an improved attention mechanism LSTM monthly maximum demand prediction model. This differs from traditional intraday / hourly load forecasting, achieving precise prediction of a single maximum demand value for the target month. The model adds a multi-scale attention module to the input layer of the traditional LSTM network, dynamically assigning attention weights to input features of different dimensions (such as historical monthly maximum demand values, monthly production plans, monthly weather trends, monthly large equipment start-up and shutdown plans, and monthly energy storage charging plans). It focuses on features strongly correlated with the grid's monthly maximum demand, effectively capturing the cumulative impact of energy storage charging on the monthly maximum demand. During model training, an adaptive learning rate adjustment algorithm is introduced, dynamically adjusting the learning rate based on the training error. An early stopping strategy is also employed, stopping training when the validation set error no longer decreases for several consecutive epochs to avoid overfitting.

[0026] The model outputs the single-value forecast of the maximum monthly demand of the power grid for the target month from the EMS. P_pred_max And the prediction error range, which is generated based on the Monte Carlo simulation method and is used to determine the threshold of the monthly charging demand constraint. The prediction period is the natural month or billing month specified by the power grid, which is fully compatible with the core rule of charging basic electricity fees by month. P_pred_max This serves as the core threshold for full-month demand control and charging management, providing the sole basis for the EMS to subsequently generate dynamic monthly charging power limits, ensuring a deep match between charging control and monthly demand control. Compared to traditional prediction models, the improved attention mechanism LSTM model in this step is optimized for single-value monthly maximum demand, improving prediction accuracy by more than 15%. It can effectively capture the long-term trend of monthly electricity load, the characteristics of peak surges within the month, and the coupled impact of energy storage charging on monthly maximum demand.

[0027] Step S3: Generation of multi-objective optimization control strategy with charging demand constraints With EMS as the core computing unit, the sum of the energy storage charging power and the real-time grid demand at any time during the month shall not exceed the single value of the predicted maximum monthly demand. P_pred_max With the primary hard constraint objective as the objective, and combining the achievement of monthly demand control targets, maximization of monthly peak-valley arbitrage profits, and minimization of monthly energy storage system losses, a multi-objective optimization function is established. This breaks through the limitations of existing technologies that focus solely on intraday arbitrage and ignore the core monthly demand objective, and achieves coordinated optimization of multiple objectives throughout the month.

[0028] The expression for the multi-objective optimization function is: in, The charging power for energy storage; Power consumption of the load; This is the maximum demand forecast value, which is the core hard threshold for demand control throughout the month; The preset demand control threshold and ≤ This is used to adjust daily demand in advance and avoid the risk of triggering the maximum monthly demand. R C represents peak-valley arbitrage profits; C represents the operating losses of the energy storage system. ,and It always has the highest priority weight, with a value no lower than 0.4, to ensure the core position of charging demand constraints.

[0029] The dynamic weight allocation algorithm dynamically adjusts the values ​​of each weight based on real-time operating conditions within the month: when the actual maximum demand of the power grid... Pgrid ≥ P _ predmax When the usage rate reaches 90%, adjust ω1 to 0.5-0.7, simultaneously reduce ω3, and limit or even suspend charging; during peak usage periods within the month and Pgrid ≤ P _ predmax When the SOC is 70%, appropriately reduce ω1 to 0.4 and increase ω3 to improve peak-valley arbitrage profits; when the SOC of the energy storage battery is close to the threshold (≤10% or ≥90%), increase ω4 to 0.4-0.5 to prioritize equipment safety; when the billing cycle is nearing its end and the actual maximum demand is close to the threshold... P _ predmax At this time, ω1 is adjusted to 0.7-0.8 to forcibly lock the charging power limit and ensure that the demand does not exceed the limit.

[0030] For the two monthly billing modes, contract demand billing and actual maximum demand billing, EMS implements adaptive switching between the billing modes: Under contract demand billing mode, if the contract monthly demand ≤ PpredmaxEMS will replace the contracted monthly demand. Ppredmax As the core threshold for charging demand constraints, the weight of ω1 is increased to above 0.5; if the contracted monthly demand > Ppredmax EMS Ppredmax Based on the adjustment of contract demand control strategies, the incremental basic electricity charges for the portion exceeding the contracted monthly demand will be included in the optimization targets, balancing monthly demand control and contract compliance. Under the actual maximum demand billing model, EMS will focus on optimizing the difference between the actual maximum demand and the predicted maximum demand within the month, with the core objective being to ensure that the actual maximum demand is ≤ P_month_ pred This reduces the incremental basic electricity cost and dynamically adjusts the charging power limit based on real-time electricity consumption data within the month. Finally, the EMS uses a genetic algorithm to solve a multi-objective optimization function to obtain the baseline value of the energy storage system's charging and discharging power, while simultaneously sending the dynamic charging power limit value to the PCS for execution.

[0031] Step S4: Real-time error correction and strategy adjustment within the month The EMS employs a rolling window algorithm to monitor and analyze the difference between the real-time maximum demand of the power grid and the predicted maximum monthly demand throughout the month. It focuses on calculating the cumulative deviation of charging demand to mitigate the risk of exceeding the maximum monthly demand due to sudden fluctuations in peak electricity consumption and charging deviations, ensuring that charging remains uninterrupted throughout the month. P_month_pred Threshold. The size of the rolling window is set to 15-60 minutes based on the monthly electricity consumption characteristics, and the rolling step size is 5-10 minutes. Each time the window is rolled, the EMS calculates the cumulative deviation value of the charging demand within the current window. Simultaneously monitor the actual maximum demand within the month in real time. P_month_actual Compared with the predicted value P_month_pred The deviation.

[0032] When Δ P_charge >0 (i.e., the sum of charging and load is close to 0) P_month_pred )or P_month_actual ≥ P_month_pred When the charge rate reaches 95%, the EMS immediately initiates a dynamic slip error elimination process, significantly reducing the charging power limit for subsequent periods using linear interpolation, while simultaneously correcting the charging and discharging power reference values, forcibly reducing the charging power, and, if necessary, pausing charging and initiating energy storage discharge peak shaving; when P_month_actual far below P_month_pred Furthermore, when there is no peak electricity demand risk during the day, the EMS can appropriately relax the charging power limit to increase peak-valley arbitrage profits, balancing economic efficiency and control safety; when Δ P_charge ≤0 and P_month_actualWhen within a safe range, the EMS maintains the current baseline values ​​for charging and discharging power and the charging power limit to avoid excessive correction that could reduce energy storage utilization. Furthermore, the EMS feeds back real-time deviation data (including the cumulative deviation of charging demand and the deviation of the monthly actual maximum demand) to the monthly maximum demand prediction model, fine-tuning the model parameters online. This achieves dynamic adaptation of the prediction model to the actual operating conditions within the month, further improving the prediction accuracy for subsequent months and forming a closed-loop mechanism led by the EMS: "monthly prediction—full-month control—intraday correction—model iteration".

[0033] Step S5: Execution of EMS-led hierarchical response control Based on the charging and discharging power benchmark value and the dynamic limit value of charging power adjusted by EMS in step S4, the energy storage system achieves millisecond-level response and stable regulation of charging and discharging through the hierarchical response control architecture of "main control layer - execution layer - monitoring layer" coordinated by EMS. From the hardware execution level, it forcibly ensures that the monthly charging does not exceed the predicted maximum monthly demand value, solving the problems of response lag and disconnect between charging and monthly demand control in traditional control methods.

[0034] This hierarchical architecture is centered on the EMS (Electronic Management System), with all functions at each level uniformly scheduled by the EMS. Signal transmission and command issuance achieve millisecond-level response, and the core focuses on full-month control based on the maximum monthly demand value. 1. Main Control Layer: This is the core computing module of the EMS and the decision center for monthly demand control and charging management. It is responsible for receiving monthly maximum demand forecast data and real-time error correction information within the month. Combined with the monthly operating status of the energy storage system, it outputs precise charging and discharging control commands and charging power hard limit commands. The commands are dynamically updated according to the daily operating conditions and are continuously issued throughout the month. All commands follow the principle of "charging does not exceed the predicted monthly maximum demand". 2. Execution Layer: Composed of a power storage inverter (PCS) and a battery management system (BMS), the PCS acts as the execution terminal, strictly adhering to the charging power limits issued by the EMS to perform charging operations, rejecting any charging commands exceeding the limits, achieving millisecond-level charging and discharging power switching, with a response time of no more than 50ms, enabling rapid response to sudden fluctuations in peak electricity consumption within the month; the BMS monitors the battery's SOC, temperature, and other statuses in real time, feeding the data back to the EMS in milliseconds, allowing the EMS to dynamically adjust the charging strategy based on the battery status, balancing charging compliance with the long-term operational safety of the battery within the month; 3. Monitoring Layer: Composed of smart meters, energy storage status acquisition terminals, and grid demand monitoring terminals, this layer collects real-time grid demand, monthly maximum actual demand, electricity load, and energy storage system operation data (battery SOC, charging power, inverter status, etc.) in real time and uploads them to the EMS at a frequency of no less than 10Hz. This provides accurate and real-time data support for the EMS's real-time decision-making and strategy adjustments throughout the month, ensuring that the EMS can promptly grasp the monthly demand change trend.

[0035] Meanwhile, during the control process, the EMS incorporates four constraints to comprehensively ensure stable system operation and charging compliance throughout the month: First, a hard limit constraint on charging power, generated based on the predicted maximum monthly demand and dynamically updated daily, preventing charging from pushing up the maximum monthly demand from the execution level; second, upper and lower bound constraints on power, combined with the inverter's rated power, to prevent overload operation; third, a battery SOC threshold constraint (typically 10%-90%), implemented by the EMS in conjunction with the BMS, to prevent battery overcharging and over-discharging, ensuring long-term operational safety throughout the month; and fourth, monthly demand regulation constraints, implemented through preset... P_limit ≤ P_month_pred Adjust daily demand in advance to avoid the risk of triggering the maximum monthly demand.

[0036] Finally, the monitoring layer feeds back all real-time operating data for the month to the EMS main control layer and the prediction model. The EMS then aggregates, analyzes, and iterates the data, forming a closed-loop control system for the entire EMS process: "data acquisition—monthly prediction—full-month optimization—intra-day correction—real-time control—monthly feedback." This ensures that the entire system always achieves coordinated optimization of monthly demand control, intra-monthly peak-valley arbitrage, and energy storage safety, under the premise that "the monthly charging does not exceed the predicted maximum monthly demand."

[0037] The above is an introduction to the method embodiments. The following embodiments using electronic devices and storage media will further illustrate the solution of the present invention.

[0038] This invention also provides an electronic device including a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0039] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0040] The processing unit executes the various methods and processes described above, such as methods S1 to S5. For example, in some embodiments, methods S1 to S5 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of methods S1 to S5 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S5 by any other suitable means (e.g., by means of firmware).

[0041] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0042] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0043] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0044] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A demand control method for an energy storage system based on maximum demand prediction, characterized in that, The method includes the following steps: Step S1: Collect historical monthly electricity consumption data, real-time electricity consumption data, environmental data, and energy storage system operation data of the target scenario, and clean, normalize, and extract features from the data to obtain a standardized dataset. Step S2: Construct an improved attention mechanism LSTM prediction model based on a standardized dataset to perform multi-dimensional monthly maximum demand prediction, and output the predicted value of the monthly maximum demand of the power grid for the target month and the prediction error range. Step S3: With the objectives of ensuring that the sum of the energy storage charging power and the load power does not exceed the predicted monthly maximum demand, achieving demand control targets, maximizing peak-valley arbitrage profits, and minimizing energy storage system losses, a multi-objective optimization function is established. A dynamic weight allocation algorithm is used to determine the weights of each objective, and the benchmark value of the energy storage system's charging and discharging power is obtained. At the same time, the energy storage energy management system (EMS) generates a dynamically adjustable charging power limit value. Step S4: The rolling window algorithm is used to monitor and analyze the difference between the actual maximum demand and the predicted monthly maximum demand. Combined with the dynamic slip error elimination method, the charging and discharging power reference value and the EMS charging power limit value are dynamically adjusted. Step S5: Based on the adjusted charging and discharging power reference value and charging power limit value, the energy storage inverter PCS is controlled by the EMS-led hierarchical response control architecture to achieve millisecond-level charging and discharging switching, complete the monthly demand control, and provide real-time feedback of the operating status to the EMS prediction model and optimization module, forming a full closed-loop control of the EMS.

2. The demand control method for an energy storage system based on maximum demand prediction according to claim 1, characterized in that, The historical monthly electricity consumption data in step S1 includes the maximum monthly demand value for each month over the past 1 to 3 years, the distribution of peak electricity consumption periods within the month, and coupled data on grid demand changes during monthly energy storage charging periods; the real-time electricity consumption data includes active power, reactive power, voltage, current, and real-time grid demand data for the current moment of the day; the environmental data includes monthly meteorological trends, daily ambient temperature, humidity, light intensity, and seasonal characteristics; the energy storage system operation data includes battery SOC, battery temperature, internal resistance, charge and discharge efficiency, inverter operating parameters, and real-time charging power; The feature extraction employs wavelet transform combined with random forest algorithm to extract the trend features of monthly electricity load, peak load fluctuation features within the month, correlation features, and coupling features between energy storage charging and the grid's maximum monthly demand. Abnormal data and redundant features are eliminated, and key time periods within the month that are prone to triggering maximum demand are identified.

3. The demand control method for an energy storage system based on maximum demand prediction according to claim 1, characterized in that, In step S2, the improved attention mechanism LSTM prediction model adds a multi-scale attention module to the input layer of the traditional LSTM network, assigns dynamic attention weights to input features of different dimensions, and focuses on features that are strongly correlated with the maximum monthly demand of the power grid, including monthly production plans, monthly weather trends, monthly start-up and shutdown plans of large equipment, and monthly charging plans for energy storage. During the training process of the improved attention mechanism LSTM prediction model, an adaptive learning rate adjustment algorithm is introduced, combined with an early stopping strategy to avoid overfitting. The output prediction error range is generated based on the Monte Carlo simulation method and is used for the threshold determination of subsequent charging demand constraints. The improved attention mechanism LSTM prediction model is a single-value prediction model for monthly maximum demand. It outputs a unique monthly maximum demand prediction value for the target month. The prediction period is either a calendar month or a billing month, and it is adapted to the billing rules of the power grid that charge basic electricity fees on a monthly basis.

4. The demand control method for an energy storage system based on maximum demand prediction according to claim 1, characterized in that, The expression for the multi-objective optimization function in step S3 is: in, The charging power for energy storage; Power consumption of the load; This is the maximum demand forecast; The preset demand control threshold and ≤ ; R C represents peak-valley arbitrage profits; C represents the operating losses of the energy storage system. .

5. The demand control method for an energy storage system based on maximum demand prediction according to claim 4, characterized in that, The dynamic weight allocation algorithm in step S3 specifically involves dynamically adjusting the values ​​of each weight based on real-time operating conditions within the month. When the actual maximum demand of the power grid Pgrid ≥ P _ predmax When the charge reaches 90%, adjust ω1 to 0.5-0.7, and simultaneously reduce ω3 to limit or even stop charging. During peak hours of the month and Pgrid ≤ P _ predmax When the value is 70%, decrease ω1 to 0.4 and increase ω3 to improve the peak-valley arbitrage profit; When the SOC of the energy storage battery is ≤10% or ≥90%, increase ω4 to 0.4-0.5 to prioritize equipment safety; When the month is nearing the end of the billing cycle and the actual maximum demand has already reached its limit. P _ predmax At this time, ω1 is adjusted to 0.7-0.8 to forcibly lock the charging power limit; Peak-valley arbitrage profits R Based on real-time peak-valley electricity prices, system losses are calculated. C This includes battery charging and discharging losses, inverter losses, and line losses.

6. The demand control method for an energy storage system based on maximum demand prediction according to claim 4, characterized in that, In step S4, the window size of the rolling window algorithm is set to 15-60 minutes based on the monthly electricity consumption characteristics, and the rolling step size is 5-10 minutes. Each time the window is rolled, the cumulative deviation value of the charging demand within the current window is calculated. : Simultaneously monitor the actual maximum demand within the month. P_month_actual Compared with the predicted value P_pred_max The deviation; when Δ P_ charge >0 or P_month_actual ≥ P_month_pred When the charge power reaches 95%, a dynamic slip error elimination process is initiated. The charging power limit value for subsequent periods is adjusted by linear interpolation, while the charging and discharging power reference value is corrected to forcibly reduce the charging power. when P_month_actual far below P_pred_max Furthermore, when there is no risk of peak electricity consumption during the day, the charging power limit will be relaxed.

7. The demand control method for an energy storage system based on maximum demand prediction according to claim 1, characterized in that, The EMS-led hierarchical response control architecture in step S5 is specifically as follows: The main control layer is the core computing module of EMS. It is used to receive monthly maximum demand forecast data and intraday real-time error correction information, output charging and discharging control commands and charging power hard limit commands, and dynamically update and send them to the execution layer throughout the month. The execution layer includes the energy storage inverter PCS and the battery management system BMS. The PCS performs charging operations according to the charging power limit value issued by the EMS, refuses to receive any charging commands that exceed the limit, achieves millisecond-level charging and discharging power switching, and has a response time of no more than 50ms. The BMS monitors the battery status in real time and feeds the data back to the EMS to meet the long-term charging regulation requirements within the month. The monitoring layer includes smart meters, energy storage status acquisition terminals, and grid demand monitoring terminals. It collects real-time grid demand, actual maximum demand within the month, electricity load, and energy storage system operation data in real time and uploads them to the EMS in milliseconds, providing real-time data support for the EMS to adjust its monthly demand control strategy. During the control process, the EMS simultaneously incorporates hard limits on charging power, upper and lower bounds on power, battery SOC threshold, and inverter rated power constraints. This achieves dynamic hard limits on the monthly charging power at the EMS level, ensuring that the sum of charging and electricity load at any time during the month does not exceed the predicted maximum monthly demand, while also preventing the energy storage system from overcharging, over-discharging, or overloading.

8. The demand control method for an energy storage system based on maximum demand prediction according to claim 1, characterized in that, This method is suitable for two monthly billing modes: contract demand billing and actual maximum demand billing, and the billing mode can be automatically switched through EMS. When using the contract demand billing model, EMS uses the monthly contract demand value as the core threshold. If the monthly contract demand is ≤ P_ month_pred Then replace the contract month demand with P_month_pred As a threshold for charging demand constraints, the weight of ω1 is increased; if the contract monthly demand > P_month_pred, the contract demand control strategy is adjusted based on P_month_pred, and the incremental basic electricity charge for the portion exceeding the contract monthly demand is calculated and included in the optimization objective. When using the actual maximum demand billing model, EMS focuses on optimizing the difference between the actual maximum demand and the predicted maximum demand for the month, ensuring that the actual maximum demand for the month is ≤ P_month_pred The incremental basic electricity fee will be reduced, and the charging power limit will be dynamically adjusted based on real-time electricity consumption data within the month.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images