Industrial and commercial energy storage dynamic control method based on user power consumption behavior prediction

Abstract

The invention discloses an industrial and commercial energy storage dynamic control method based on user power consumption behavior prediction, and relates to the technical field of energy storage system control, and the method comprises the steps: S1, edge-cloud cooperative control system deployment, S2, multi-source data and feature collection, S3, load prediction model construction, S4, dynamic optimization control, and S5, countercurrent prevention and demand control. According to the method, intelligent dynamic scheduling of the energy storage system is realized by constructing an edge-cloud collaborative hierarchical control architecture; the data acquisition contains multi-dimensional characteristics such as a load curve, an equipment start-stop event, environmental parameters, photovoltaic output and electricity price information, and provides comprehensive input for the prediction model; the dynamic optimization control is based on a model prediction control framework, and the purpose is to minimize the comprehensive power consumption cost; according to the method, the photovoltaic consumption efficiency and the demand electric charge optimization effect are remarkably improved, the problem of adjustment lag caused by an overlong period of a traditional control strategy is solved through cooperation of edge layer millisecond-level response and cloud intelligent decision, and an economical and efficient energy storage system solution is provided for industrial and commercial users.

Description

Technical Field

[0001] This invention relates to the field of energy storage system control technology, and specifically to a dynamic scheduling method for industrial and commercial energy storage based on machine learning prediction. Background Technology

[0002] With the upgrading of energy consumption in the industrial and commercial sectors and the advancement of dual-carbon goals, energy storage systems, as core equipment for optimizing energy allocation and improving energy efficiency, have attracted much attention for their dynamic control technology.

[0003] Currently, commercial and industrial energy storage systems face three major technical bottlenecks in dynamic control, severely restricting their economic efficiency and energy utilization efficiency: First, rigid and ineffective control strategies: Traditional EMS systems use preset peak-valley charging and discharging periods, failing to respond to dynamic load changes. For example, actual data from a manufacturing plant in Jiangsu shows that under a fixed-period charging and discharging strategy, the maximum demand exceedance rate of the energy storage system reaches 27%, with demand electricity costs only reduced by 9%–12% (far below the theoretical optimization space of 20%–25%), and the control cycle is over 2 minutes, unable to respond promptly to rapid load changes. Second, insufficient load forecasting accuracy: Existing forecasting models rely solely on macroscopic load curves and meteorological data, neglecting to consider equipment start-up and shutdown events, resulting in forecast errors as high as 15%–22%. Furthermore, they generally ignore the correlation between real-time solar intensity and photovoltaic output and load, leading to photovoltaic absorption prediction deviations >25%. Third, traditional LSTM models, with a time step of 15 minutes, cannot capture the 30ms-level instantaneous power changes during equipment start-up and shutdown, resulting in delayed prediction of load abrupt changes. Conflict between anti-backflow and capacity utilization: Traditional anti-backflow solutions mainly rely on direct curtailment of solar power or restriction of photovoltaic output, which cannot fully utilize photovoltaic capacity and result in significant revenue loss. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic control method for industrial and commercial energy storage based on user electricity consumption behavior prediction, which solves the problems existing in the background technology.

[0005] To solve the above technical problems, the present invention adopts the following technical solution: The present invention provides a dynamic control method for industrial and commercial energy storage based on user electricity consumption behavior prediction, including: S1. Edge-cloud collaborative control system deployment: constructing a hierarchical control architecture and clarifying the functions of each level.

[0006] S2. Multi-source data and feature acquisition: Acquire real-time data.

[0007] S3. Load Forecasting Model Construction: Using... Hybrid neural networks are used for load prediction.

[0008] S4. Dynamic Optimization Control: Model Predictive Control Dynamically optimize the charging and discharging power of energy storage.

[0009] S5. Backflow Prevention and Demand Control: Implement a comprehensive backflow prevention and demand control strategy, based on energy storage status, And the grid connection point power can be dynamically adjusted to allow for energy storage charging and discharging or photovoltaic output.

[0010] Preferably, the method for constructing a hierarchical control architecture and defining the functions of each level is as follows: the edge-cloud collaborative control system includes an edge layer, a cloud layer, and a communication architecture.

[0011] The edge layer adopts The server is used as an edge computing device, utilizing Alibaba Cloud. As a cloud platform, it is used for data preprocessing, rapid response to backflow prevention, and gradual power control.

[0012] The cloud layer is deployed on Alibaba Cloud. Platform, for Model prediction, Rolling optimization and life health assessment.

[0013] Preferably, the communication architecture is a data transmission channel connecting the device layer, edge layer, and cloud layer.

[0014] The device layer adopts Protocol, edge-cloud adoption Protocol, communication security Encryption and two-way certificate authentication ensure security.

[0015] Preferably, the real-time data includes load data, equipment status, environmental parameters, photovoltaic data, and electricity price information.

[0016] The load data is collected through smart meters, and the load power calculation formula is as follows: ,in This indicates the current point in time, i.e., the instantaneous moment. Indicates the sampling time window. express The instantaneous voltage value at time t. express The instantaneous value of the current at a given moment.

[0017] The device status is determined by identifying device start-up and shutdown events using a smart meter.

[0018] The environmental parameters are collected via temperature from a weather station or a sensor built into the photovoltaic inverter. And light intensity H.

[0019] The photovoltaic data is uploaded via photovoltaic inverter communication to obtain photovoltaic power data.

[0020] The electricity price information is transmitted through the power grid. The interface retrieves the time-of-use electricity price matrix in real time.

[0021] Preferably, the use The specific method for load forecasting using hybrid neural networks is as follows: Hybrid neural networks consist of an input layer, layer, The input layer and output layer are connected sequentially. The input layer receives a feature matrix containing multiple time steps and multiple feature dimensions. The layer contains a first number of neuronal units used to extract long-term trend features of electrical load. The first layer contains a second number of neurons, focusing on capturing short-term power fluctuations such as device start-up and shutdown. The output layer contains multiple prediction nodes, outputting load prediction curves for a predetermined future time period.

[0022] Model training uses a loss function, expressed as follows: ,in for Loss, used for stable training, To amplify the loss at the demand point, To enhance the loss due to the instantaneous characteristics of equipment start-up and shutdown, the coefficient and Used to adjust the importance of different loss terms.

[0023] The training process is deployed on Alibaba Cloud. The platform uses preset learning rate and batch size parameters to accelerate model convergence. The inference phase is deployed at the edge layer. Servers and edge devices upload real-time data to the cloud at predetermined intervals. The cloud updates its model based on historical data for a predetermined time period and sends load prediction results for the next predetermined time period to the edge layer to support dynamic control decisions.

[0024] Preferably, the model-based predictive control The specific method for dynamically optimizing the charging and discharging power of energy storage is as follows: the dynamic optimization control is achieved through model predictive control. Implementation, executed by edge devices The optimization strategy dynamically adjusts the energy storage charging and discharging power, and the edge layer monitors the anti-reverse current state with a preset short cycle. The anti-reverse current state includes the photovoltaic power generation. Load power Energy storage system state.

[0025] The Optimization focuses on minimizing the objective function, which is expressed as follows: ,in For time step index, To predict the total number of steps in the time domain, This is the demand-based electricity price weighting factor. This is the weighting factor for time-of-use electricity charges. This is the photovoltaic revenue weighting coefficient. For demand-based electricity pricing, For time-of-use electricity pricing, For the revenue from photovoltaic power generation, Power at the grid connection point (assessment point), This refers to the photovoltaic power generation capacity.

[0026] The optimization process is subject to the following constraints: ,in For charging power, For discharge power, For energy storage converter Rated power For charging efficiency, For discharge efficiency, , This is the boundary of the safe operating range for energy storage in its state of charge. Preset proportional coefficient for charging power, The preset proportional coefficient of discharge power.

[0027] Preferably, the specific method for implementing the integrated anti-reverse flow and demand control strategy is as follows: real-time acquisition of photovoltaic power generation. Real-time charging power of energy storage Real-time discharge power of energy storage Grid connection point power Load power and energy storage systems state.

[0028] Based on real-time charging power of energy storage Real-time discharge power of energy storage To determine the charging and discharging status of the energy storage system, combined with the power at the grid connection point. and energy storage systems In each state, demand control strategy and anti-backflow control strategy are executed respectively.

[0029] Preferably, the method for determining the charging and discharging state of the energy storage system is as follows: obtaining a power threshold from a database. Real-time charging power of energy storage is detected through the energy storage converter. and real-time discharge power of energy storage Combined with power threshold Determine the current running status: when At that time, it is determined that the energy storage system is in a charging state.

[0030] when At that time, it is determined that the energy storage system is in a discharging state.

[0031] when and At that time, the energy storage system is determined to be in standby mode.

[0032] Preferably, the specific method for implementing the demand control strategy is as follows: when the energy storage system is in a charging state, if the demand is satisfied... Then set the charging power. Otherwise, it will operate according to the charging power issued by the cloud. To preset the demand warning coefficient, The upper limit of power is controlled according to demand.

[0033] Preferably, the method for implementing the anti-reverse current control strategy is as follows: when the energy storage system is in a discharging state or a standby state, if and Then set .

[0034] like and Then set the photovoltaic power generation capacity. .

[0035] like and Then set .

[0036] like and The energy storage system then operates according to cloud-based strategies, in which... This represents the maximum safe operating boundary for the energy storage's state of charge. To preset the anti-backflow adjustment coefficient, To control the upper limit of power according to demand, This is the rated power of the energy storage converter.

[0037] The beneficial effects of the present invention are: (1) significantly improved prediction accuracy: through The hybrid model combines equipment start-up and shutdown events with the characteristics of solar-load coupling, reducing the load prediction error from the traditional 15% to 22% to 6.8%, and the photovoltaic consumption prediction error to <6%. (2) Photovoltaic consumption and anti-reverse flow coordination: Through the hierarchical control strategy, the photovoltaic consumption rate is increased from the traditional 70% to 80% to 98%, while avoiding reverse flow and reducing curtailment losses. (3) Optimized economic benefits: After implementation, the peak-valley arbitrage benefit of the energy storage system is increased by about 8%, and the overall economic benefits are increased by about 11%. (4) Enhanced real-time response capability: The 500ms response cycle of the edge layer can capture minute-level or even 30ms-level load mutations, solving the problem of adjustment lag caused by the traditional control cycle being too long (≥2min). Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the implementation steps of the method of the present invention.

[0040] Figure 2 This is a diagram of the edge-cloud collaborative control system architecture.

[0041] Figure 3 This is a flowchart of the dynamic control method.

[0042] Figure 4 for Schematic diagram of a hybrid neural network structure.

[0043] Figure 5 Logic diagram for preventing backflow and controlling demand.

[0044] Figure 6 This is a comparison chart of the prediction effects of embodiments of the present invention.

[0045] Figure 7 This is a comparison chart showing the effects of optimized control strategies in embodiments of the present invention. Detailed Implementation

[0046] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Reference Figure 1 As shown, the present invention provides a dynamic control method for industrial and commercial energy storage based on user electricity consumption behavior prediction, including: S1. Deployment of edge-cloud collaborative control system: constructing a hierarchical control architecture and clarifying the functions of each level; S2. Multi-source data and feature acquisition: Acquiring real-time data; S3. Load Forecasting Model Construction: Using... Hybrid neural networks for load forecasting; S4. Dynamic Optimization Control: Model Predictive Control Dynamically optimize energy storage charging and discharging power; S5. Backflow Prevention and Demand Control: Implement a comprehensive backflow prevention and demand control strategy, based on energy storage status, And the grid connection point power can be dynamically adjusted to allow for energy storage charging and discharging or photovoltaic output.

[0048] It should be noted that, This indicates the remaining state of charge of an energy storage system (such as a battery), usually measured as a percentage (0%-100%). In reverse current prevention control... Photovoltaic power generation needs to be limited (to avoid overcharging) in demand control. Constrained charge and discharge power (e.g., requirements during discharge) ); It should be noted that the grid connection point power refers to the real-time power at the connection point (assessment point) between the energy storage system and the power grid. A positive grid connection point power indicates that the system is drawing power from the grid (user electricity consumption), while a negative grid connection point power indicates that the system is feeding power back to the grid (reverse flow, which should be avoided).

[0049] In one specific embodiment, the method for constructing a hierarchical control architecture and clarifying the functions of each layer is as follows: the edge-cloud collaborative control system includes an edge layer, a cloud layer, and a communication architecture. The edge layer adopts The server is used as an edge computing device, utilizing Alibaba Cloud. As a cloud platform, it is used for data preprocessing, fast response to backflow prevention, and power gradient control; It should be noted that the anti-backflow rapid response refers to the edge layer. The server is The ability to detect and prevent photovoltaic power generation from flowing back to the grid in real time within 500ms; power gradient control refers to the edge layer smoothly adjusting the charging and discharging power of energy storage to avoid power surges that could impact the grid and equipment. The cloud layer is deployed on Alibaba Cloud. Platform, for Model prediction, Rolling optimization and life health assessment; It should be noted that the edge layer response time is ≤500ms, and the cloud layer optimization time is 15 minutes. The cloud layer updates the prediction model and distributes optimization parameters every 15 minutes. Secure communication between the edge layer and the cloud is achieved through a 5G industrial router, and the communication protocol adopts... .

[0050] In one specific embodiment, the communication architecture is a data transmission channel connecting the device layer, edge layer, and cloud layer; the device layer adopts... Protocol, edge-cloud adoption Protocol, communication security Encryption and two-way certificate authentication ensure security.

[0051] It should be noted that the equipment layer includes high-precision smart meters (1ms sampling), photovoltaic inverters (outputting photovoltaic power), and energy storage converters. (Controlling charging and discharging power), meteorological sensors (collecting temperature and light intensity), energy storage (Provide energy storage) state); It should be noted that the following is adopted: Protocol, transmission delay ;use Protocol, transmission delay .

[0052] In one specific embodiment, the real-time data includes load data, equipment status, environmental parameters, photovoltaic data, and electricity price information; The load data is collected through smart meters, and the load power calculation formula is as follows: ,in This indicates the current point in time, i.e., the instantaneous moment. Indicates the sampling time window. express The instantaneous voltage value at time t. express The instantaneous value of the current at a given moment; It should be noted that, This indicates the sampling time window, with a sampling time of 1ms. Historical load data is obtained through communication with the project monitoring system. Load data is collected using high-frequency smart meters and transmitted via... The communication protocol transmits data to ; use The communication protocol communicates with weather stations and inverters to acquire data; use Communication protocols and and communication, acquisition and Power and other data; The device status is determined by identifying device start / stop events through a smart meter. It should be noted that the determination condition for equipment start-up and shutdown events is the change in power. ≥15% × rated power, and duration ≥30ms; a 15% threshold is used to filter measurement noise. <15% × rated power is considered normal fluctuation and not equipment start-up / shutdown; a 30ms threshold ensures that real equipment actions are captured; The environmental parameters are collected via temperature from a weather station or a sensor built into the photovoltaic inverter. and light intensity H; It should be noted that temperature An air conditioning load correction model can be established using light intensity H. ,in Indicates the air conditioning cooling load power. This represents the real-time light intensity, 0.082 represents the light-to-power conversion coefficient (obtained by fitting historical data), and 500 represents the baseline threshold for light intensity. Indicates the basic air conditioning load; The photovoltaic data is uploaded via photovoltaic inverter communication to display photovoltaic power data. The electricity price information is transmitted through the power grid. The interface retrieves the time-of-use electricity price matrix in real time.

[0053] In one specific embodiment, the adoption The specific method for load forecasting using hybrid neural networks is as follows: Hybrid neural networks consist of an input layer, layer, The layers and output layer are connected in sequence; the input layer receives a feature matrix containing multiple time steps and multiple feature dimensions. The layer contains a first number of neuron units used to extract long-term trend features of electrical load; The layer contains a second number of neurons, focusing on capturing short-term power fluctuations such as device start-up and shutdown; the output layer contains multiple prediction nodes, outputting load prediction curves for a preset future time period. It should be noted that the feature matrix includes daily data at multiple time steps (e.g., 1440 time steps per day) obtained at a preset high-frequency sampling granularity (e.g., per minute), as well as data including load data, environmental parameters, equipment status, photovoltaic data, and energy storage system data. The system includes six types of feature parameters: electricity price information, etc.; the first quantity can be set to 96, the second quantity can be set to 48, and the preset time length can be set to 24 hours. Model training uses a loss function, expressed as follows: ,in for Loss, used for stable training, To amplify the loss at the demand point, To enhance the loss due to the instantaneous characteristics of equipment start-up and shutdown, the coefficient and Used to adjust the importance of different loss terms; It should be noted that, It can be set to greater than The value is used to place greater emphasis on the accuracy of the required measurement points; It can be set to 3.0. It can be set to 2.0, which places more emphasis on the accuracy of the measurement points; The training process is deployed on Alibaba Cloud. The platform uses preset learning rate and batch size parameters to accelerate model convergence; the inference phase is deployed at the edge layer. Servers and edge devices upload real-time data to the cloud at predetermined intervals. The cloud updates its model based on historical data for a predetermined time period and sends load prediction results for the next predetermined time period to the edge layer to support dynamic control decisions.

[0054] It should be noted that the preset learning rate and batch size parameters can be configured with a learning rate of 0.001 and a batch size of 64. The preset period can be set to every 15 minutes, the preset time length of historical data can be set to the historical data of the past year (1 minute granularity per day), and the preset time length of the future can be set to the next 24 hours.

[0055] In one specific embodiment, the model-based predictive control The specific method for dynamically optimizing the charging and discharging power of energy storage is as follows: the dynamic optimization control is achieved through model predictive control. Implementation, executed by edge devices The optimization strategy dynamically adjusts the energy storage charging and discharging power, and the edge layer monitors the anti-reverse current state with a preset short cycle. The anti-reverse current state includes the photovoltaic power generation. Load power Energy storage system state; It should be noted that the preset short period can be set to 500ms. The Optimization focuses on minimizing the objective function, which is expressed as follows: ,in For time step index, To predict the total number of steps in the time domain, This is the demand-based electricity price weighting factor. This is the weighting factor for time-of-use electricity charges. This is the photovoltaic revenue weighting coefficient. For demand-based electricity pricing, For time-of-use electricity pricing, For the revenue from photovoltaic power generation, Power at the grid connection point (assessment point), Photovoltaic power generation; The optimization process is subject to the following constraints: ,in For charging power, For discharge power, For energy storage converter Rated power For charging efficiency, For discharge efficiency, , This is the boundary of the safe operating range for energy storage in its state of charge. Preset proportional coefficient for charging power, The preset proportional coefficient of discharge power.

[0056] It should be noted that the key parameter values ​​are: , , , Yuan / kw·month =0.1, =1.0, =0.5, =0.5, other parameters are derived from load forecast data and edge layer Obtained from the server.

[0057] In one specific embodiment, the method for implementing the integrated anti-reverse flow and demand control strategy is as follows: real-time acquisition of photovoltaic power generation. Real-time charging power of energy storage Real-time discharge power of energy storage Grid connection point power Load power and energy storage systems state; Based on real-time charging power of energy storage Real-time discharge power of energy storage To determine the charging and discharging status of the energy storage system, combined with the power at the grid connection point. and energy storage systems In each state, demand control strategy and anti-backflow control strategy are executed respectively.

[0058] It should be noted that the demand control strategy is used to optimize demand electricity costs, while the anti-reverse flow control strategy is used to avoid power feeding into the grid and improve the photovoltaic absorption rate. The two work together to respond to the real-time monitoring signal of the edge layer with a period of 500ms.

[0059] In one specific embodiment, the method for determining the charging and discharging state of the energy storage system is as follows: obtaining a power threshold from a database. Real-time charging power of energy storage is detected through the energy storage converter. and real-time discharge power of energy storage Combined with power threshold Determine the current running status: when At that time, it is determined that the energy storage system is in a charging state; when At that time, it is determined that the energy storage system is in a discharging state; when and At that time, the energy storage system is determined to be in standby mode.

[0060] It should be noted that, and Data is collected via an energy storage converter at a sampling frequency of at least 1ms, with a power threshold set to 5W to eliminate signal noise interference. Charging and discharging are physically mutually exclusive. and Not both are valid integrity, if Lasting 100ms If the duration is 50ms, it is determined to be in charging state.

[0061] In one specific embodiment, the method for implementing the demand control strategy is as follows: when the energy storage system is in a charging state, if the demand control strategy is satisfied... Then set the charging power. Otherwise, it will operate according to the charging power issued by the cloud. To preset the demand warning coefficient, The upper limit of power is controlled according to demand.

[0062] It should be noted that the preset demand warning coefficient The upper limit of power can be set to 95% based on demand control. Choose 1400kw.

[0063] In one specific embodiment, the method for implementing the anti-reverse current control strategy is as follows: when the energy storage system is in a discharging state or a standby state, if and Then set ; like and Then set the photovoltaic power generation capacity. ; like and Then set ; like and The energy storage system then operates according to cloud-based strategies, in which... This represents the maximum safe operating boundary for the energy storage's state of charge. To preset the anti-backflow adjustment coefficient, To control the upper limit of power according to demand, This is the rated power of the energy storage converter.

[0064] It should be noted that, Preset anti-backflow adjustment coefficient The rated power of the energy storage converter can be set to 95%. Choose 400kW.

[0065] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A dynamic control method for industrial and commercial energy storage based on user electricity consumption behavior prediction, characterized in that, include: S1. Edge-Cloud Collaborative Control System Deployment: Construct a hierarchical control architecture and clarify the functions of each level; S2. Multi-source data and feature acquisition: Acquiring real-time data; S3. Load Forecasting Model Construction: Using... Hybrid neural networks for load forecasting; S4. Dynamic Optimization Control: Model Predictive Control Dynamically optimize energy storage charging and discharging power; S5. Backflow Prevention and Demand Control: Implement a comprehensive backflow prevention and demand control strategy, based on energy storage status, And the grid connection point power can be dynamically adjusted to allow for energy storage charging and discharging or photovoltaic output.

2. The method for dynamic control of industrial and commercial energy storage based on user electricity consumption behavior prediction as described in claim 1, characterized in that, The specific method for constructing a hierarchical control architecture and clarifying the functions of each level is as follows: The edge-cloud collaborative control system includes an edge layer, a cloud layer, and a communication architecture; The edge layer adopts The server is used as an edge computing device, utilizing Alibaba Cloud. As a cloud platform, it is used for data preprocessing, fast response to backflow prevention, and power gradient control; The cloud layer is deployed on Alibaba Cloud. Platform, for Model prediction, Rolling optimization and life health assessment.

3. The method for dynamic control of industrial and commercial energy storage based on user electricity consumption behavior prediction according to claim 2, characterized in that, The communication architecture is a data transmission channel connecting the device layer, edge layer, and cloud layer; The device layer adopts Protocol, edge-cloud adoption Protocol, communication security Encryption and two-way certificate authentication ensure security.

4. The method for dynamic control of industrial and commercial energy storage based on user electricity consumption behavior prediction according to claim 1, characterized in that, The real-time data includes load data, equipment status, environmental parameters, photovoltaic data, and electricity price information; The load data is collected through smart meters, and the load power calculation formula is as follows: ,in This indicates the current point in time, i.e., the instantaneous moment. Indicates the sampling time window. express The instantaneous voltage value at time t. express The instantaneous value of the current at a given moment; The device status is determined by identifying device start / stop events through a smart meter. The environmental parameters are collected via temperature from a weather station or a sensor built into the photovoltaic inverter. and light intensity H; The photovoltaic data is uploaded via photovoltaic inverter communication to display photovoltaic power data. The electricity price information is transmitted through the power grid. The interface retrieves the time-of-use electricity price matrix in real time.

5. The method for dynamic control of industrial and commercial energy storage based on user electricity consumption behavior prediction according to claim 1, characterized in that, The use The specific method for load forecasting using hybrid neural networks is as follows: The Hybrid neural networks consist of an input layer, layer, The layers and output layer are connected in sequence; the input layer receives a feature matrix containing multiple time steps and multiple feature dimensions. The layer contains a first number of neuron units used to extract long-term trend features of electrical load; The layer contains a second number of neurons, focusing on capturing short-term power fluctuations such as device start-up and shutdown; the output layer contains multiple prediction nodes, outputting load prediction curves for a preset future time period. Model training uses a loss function, expressed as follows: ,in for Loss, used for stable training, To amplify the loss at the demand point, To enhance the loss due to the instantaneous characteristics of equipment start-up and shutdown, the coefficient and Used to adjust the importance of different loss terms; The training process is deployed on Alibaba Cloud. The platform uses preset learning rate and batch size parameters to accelerate model convergence; the inference phase is deployed at the edge layer. Servers and edge devices upload real-time data to the cloud at predetermined intervals. The cloud updates its model based on historical data for a predetermined time period and sends load prediction results for the next predetermined time period to the edge layer to support dynamic control decisions.

6. The method for dynamic control of industrial and commercial energy storage based on user electricity consumption behavior prediction as described in claim 1, characterized in that, Model-based predictive control The specific method for dynamically optimizing the charging and discharging power of energy storage is as follows: The dynamic optimization control is achieved through model predictive control. Implementation, executed by edge devices The optimization strategy dynamically adjusts the energy storage charging and discharging power, and the edge layer monitors the anti-reverse current state with a preset short cycle. The anti-reverse current state includes the photovoltaic power generation. Load power Energy storage system state; The Optimization focuses on minimizing the objective function, which is expressed as follows: ,in For time step index, To predict the total number of steps in the time domain, This is the demand-based electricity price weighting factor. This is the weighting factor for time-of-use electricity charges. This is the photovoltaic revenue weighting coefficient. For demand-based electricity pricing, For time-of-use electricity pricing, For the revenue from photovoltaic power generation, Power at the grid connection point (assessment point), Photovoltaic power generation; The optimization process is subject to the following constraints: ,in For charging power, For discharge power, For energy storage converter Rated power For charging efficiency, For discharge efficiency, , This is the boundary of the safe operating range for energy storage in its state of charge. Preset proportional coefficient for charging power, The preset proportional coefficient of discharge power.

7. The method for dynamic control of industrial and commercial energy storage based on user electricity consumption behavior prediction according to claim 1, characterized in that, The specific method for implementing the integrated anti-backflow and demand control strategy is as follows: Real-time acquisition of photovoltaic power generation Real-time charging power of energy storage Real-time discharge power of energy storage Grid connection point power Load power and energy storage systems state; Based on real-time charging power of energy storage Real-time discharge power of energy storage To determine the charging and discharging status of the energy storage system, combined with the power at the grid connection point. and energy storage systems In each state, demand control strategy and anti-backflow control strategy are executed respectively.

8. The method for dynamic control of industrial and commercial energy storage based on user electricity consumption behavior prediction according to claim 7, characterized in that, The specific method for determining the charging and discharging state of the energy storage system is as follows: Retrieve power threshold from database Real-time charging power of energy storage is detected through the energy storage converter. and real-time discharge power of energy storage Combined with power threshold Determine the current running status: when At that time, it is determined that the energy storage system is in a charging state; when At that time, it is determined that the energy storage system is in a discharging state; when and At that time, the energy storage system is determined to be in standby mode.

9. The method for dynamic control of industrial and commercial energy storage based on user electricity consumption behavior prediction according to claim 7, characterized in that, The specific method for implementing the demand control strategy is as follows: When the energy storage system is in a charging state, if the following conditions are met Then set the charging power. Otherwise, it will operate according to the charging power issued by the cloud. To preset the demand warning coefficient, The upper limit of power is controlled according to demand.

10. A dynamic control method for industrial and commercial energy storage based on user electricity consumption behavior prediction according to claim 7, characterized in that, The specific method for implementing the anti-backflow control strategy is as follows: When the energy storage system is in a discharging or standby state, if and Then set ; like and Then set the photovoltaic power generation capacity. ; like and Then set ; like and The energy storage system then operates according to cloud-based strategies, in which... This represents the maximum safe operating boundary for the energy storage's state of charge. To preset the anti-backflow adjustment coefficient, To control the upper limit of power according to demand, This is the rated power of the energy storage converter.

Images