A multi-scenario adaptive dynamic optimization method for energy storage-photovoltaic-electrolytic aluminum system

Abstract

The application discloses a kind of energy storage-photovoltaic-electrolytic aluminium system dynamic optimization methods suitable for multiple scenarios, it is related to intelligent power grid technical field, including acquisition and preprocessing energy storage-photovoltaic-electrolytic aluminium system data, obtain photovoltaic, load and price characteristics, construct equipment mathematical model, predict state of charge, battery cycle aging cost, photovoltaic output power, load baseline power and adjustable flexible interval, construct system state prediction model, economic dispatch optimization model and adaptive control model, output power prediction report, day-ahead optimal economic dispatch plan and energy storage power control instruction, energy storage power control instruction is transmitted to energy storage converter after multiple scene logic determination is executed and feedback is obtained, and by integrated digital lookboard, visualized.The application realizes the collaborative optimization of economy, reliability and green electricity consumption efficiency by intelligent prediction and adaptive control, improves the comprehensive performance of energy storage-photovoltaic-electrolytic aluminium system under variable working conditions.

Description

Technical Field

[0001] This invention relates to the field of smart grid technology, and specifically to a dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios. Background Technology

[0002] The energy storage-photovoltaic-electrolytic aluminum system is a new type of power system that integrates energy storage technology, photovoltaic power generation, and electrolytic aluminum production. This system utilizes user-side energy storage projects to charge the system with photovoltaic power absorbed during the day and discharges it during peak grid load periods, achieving stable power supply and peak shaving / valley filling. Simultaneously, the system actively promotes the development of the energy storage industry, increases the proportion of green electricity use, and reduces electricity costs for enterprises. In the electrolytic aluminum production process, the system profits by storing photovoltaic power and utilizing peak-valley electricity price differences, helping enterprises achieve green, low-carbon, and high-quality development, and serving as a model for the development of the energy storage industry.

[0003] To address the challenges of high electricity costs, stringent power supply stability requirements, and difficulties in integrating fluctuating photovoltaic (PV) energy during the energy-intensive electrolytic aluminum production process, existing technologies employ peak-valley arbitrage strategies based on fixed time periods and simple threshold judgment logic. However, these approaches suffer from rigid control strategies, an inability to adapt to real-time fluctuations in PV output and actual load changes, leading to insufficient green energy integration, suboptimal overall system economic efficiency, and difficulty in ensuring reliable power supply under varying operating conditions. To resolve these issues, a dynamic optimization method for energy storage-PV-electrolytic aluminum systems adapted to multiple scenarios is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic optimization method for energy storage-photovoltaic-electrolytic aluminum systems that can be adapted to multiple scenarios, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios, comprising the following steps: Step 1: Collect and preprocess data from the energy storage-photovoltaic-electrolytic aluminum system to obtain preprocessed data. Extract photovoltaic characteristics, load characteristics, and electricity price characteristics from the preprocessed data. The energy storage-photovoltaic-electrolytic aluminum system data includes energy storage system data, photovoltaic system data, electrolytic aluminum load power data, power grid data, and meteorological data. Step 2: Input the preprocessed energy storage system data and energy storage power control commands into the energy storage system model, and output the state of charge and battery cycle aging cost at the next moment; Based on the preprocessed meteorological data and long short-term memory network, train the photovoltaic power generation model, and output the photovoltaic output power for the next 24 hours through the trained photovoltaic power generation model; Input the load characteristics and factory production plan into the electrolytic aluminum load model, and output the load baseline power and adjustable flexible range using the baseline and flexible dichotomy method; Step 3: Using the system state prediction model, the outputs of the photovoltaic power generation model and the electrolytic aluminum load model are used to obtain a power prediction report; using the economic dispatch optimization model, based on electricity price characteristics, the power prediction report, and the output of the energy storage system model, mixed integer linear programming is used to output the day-ahead optimal economic dispatch plan; using the adaptive control model, based on the preprocessed photovoltaic system data, energy storage system data, electrolytic aluminum load power data, and the day-ahead optimal economic dispatch plan, deep reinforcement learning technology is used to output energy storage power control commands. Step 4: Combining grid data and meteorological data, the energy storage power control command is transmitted to the energy storage converter for execution after multi-scenario logic judgment, and the execution effect of the energy storage power control command is obtained. Step 5: Visualize the energy storage state of charge, photovoltaic power, and photovoltaic and load performance curve data through an integrated digital dashboard, and feed back the execution effect of energy storage power control commands to the adaptive control model.

[0006] A further improvement to the technical solution of this invention lies in the following: the energy storage-photovoltaic-electrolytic aluminum system data includes energy storage system data, photovoltaic system data, electrolytic aluminum load power data, power grid data, and meteorological data. The process of collecting and preprocessing the energy storage-photovoltaic-electrolytic aluminum system data to obtain preprocessed energy storage-photovoltaic-electrolytic aluminum system data, and extracting photovoltaic characteristics, load characteristics, and electricity price characteristics from the preprocessed energy storage-photovoltaic-electrolytic aluminum system data includes: Inside the integrated energy storage converter and boost converter in the project plant area, the integrated energy storage converter and battery management system are used to collect the real-time charging and discharging power and state of charge of the energy storage system as energy storage system data. The energy management system periodically polls the energy storage converter and battery management system through the internal communication network in accordance with standard protocols to obtain the energy storage system data. By installing multi-functional energy meters at the grid connection points of the photovoltaic areas within the project plant, the photovoltaic power time series of the photovoltaic system is measured in real time as photovoltaic system data. The energy management system reads the data from the multi-functional energy meters through the communication interface to grasp the real-time photovoltaic power generation situation. A high-precision power quality monitoring device is installed on the feeder circuit of the power distribution room in the project plant area that supplies power to the electrolytic aluminum load. The power quality monitoring device monitors the real-time active power and current flowing to the electrolytic aluminum load as the power data of the electrolytic aluminum load. The energy management system obtains the power consumption of the core load by reading the data from the power quality monitoring device. The power, frequency and voltage of the power grid are collected by the measurement and control protection device at the grid connection point of the project plant area. The time-of-use electricity price meter is statically configured in the energy management system as the basic information, and the power, frequency and voltage of the power grid and the time-of-use electricity price meter are used as the power grid data. By deploying small automated weather stations in the photovoltaic area, real-time total solar irradiance, ambient temperature, relative humidity and wind speed are collected at the project site as meteorological data. The collected data from energy storage systems, photovoltaic systems, electrolytic aluminum load power data, power grid data, and meteorological data are appended with a unified timestamp and then transmitted to the historical database of the energy management system for centralized storage. The energy storage-photovoltaic-electrolytic aluminum system data is then cleaned. The cleaning process includes identifying outlier data points caused by sensor jumps and communication interruptions using the interquartile range method, filling out the outlier data points with linear interpolation using valid data points from adjacent time points, aligning the cleaned energy storage-photovoltaic-electrolytic aluminum system data on time scales, and performing maximum-minimum normalization processing to obtain preprocessed energy storage system data, photovoltaic system data, electrolytic aluminum load power data, power grid data, and meteorological data. Photovoltaic characteristics include photovoltaic power change rate and short-term volatility characteristics. The photovoltaic power change rate is obtained by calculating the difference between the power value at the current moment and the power value at the previous moment based on the preprocessed photovoltaic system data. The short-term volatility characteristics are obtained by calculating the standard deviation of the photovoltaic system data within a preset time window. Load characteristics include load baseline level and high-frequency disturbance component characteristics. The real-time active power is smoothed by a moving average filter. The output of the moving average filter is regarded as the load baseline level. The difference sequence obtained by subtracting the load baseline level from the real-time active power is used to characterize the high-frequency disturbance component characteristics. Electricity price characteristics include peak-valley time distribution and peak-valley price difference characteristics within the next 24 hours. The program parses the time-of-use electricity price table, identifies and outputs the start and end times of peak, valley and flat periods within the next 24 hours, forming the peak-valley time distribution within the next 24 hours. Based on the peak-valley time distribution within the next 24 hours, the difference between the peak period electricity price and the valley period electricity price is extracted and calculated from the configured time-of-use electricity price table to obtain the peak-valley price difference characteristics.

[0007] A further improvement to the technical solution of this invention lies in the process of inputting preprocessed energy storage system data and energy storage power control commands into the energy storage system model, and outputting the state of charge and battery cycle aging cost at the next moment, which includes: The energy storage system model receives energy storage power control commands from the adaptive control model. If the energy storage power control command is negative, it is considered as charging power. If the energy storage power control command is positive, it is considered as the discharge power. The rated capacity of the energy storage system is preset. Charging efficiency and discharge efficiency Calculate the state of charge at the next time step. The calculation process is as follows: ; in, This represents the state of charge at the previous moment. This refers to the time step for calculation; Based on the pre-set total investment cost of the project plant area and the number of cycles throughout the battery's lifespan. Calculate the equivalent cycle cost of completing one full charge and discharge cycle. Based on the charging and discharging energy at the current moment Calculate the equivalent number of cycles for charge and discharge energy. Multiply the equivalent number of cycles by the equivalent cycle cost to obtain the battery cycle aging cost within the current time step. The calculation process is as follows: ; ; in, The total energy throughput of a single full charge.

[0008] A further improvement to the technical solution of this invention lies in the following: the process of training a photovoltaic power generation model based on preprocessed meteorological data and a long short-term memory network, and outputting the photovoltaic output power for the next 24 hours, includes: A photovoltaic power generation model based on a long short-term memory network was trained and fixed in an energy management system, and executed at a fixed frequency. Meteorological data was used as the input features of the photovoltaic power generation model and divided into training and validation sets. The total output power of the actual photovoltaic power generation system at the same time was used as the output label. The long short-term memory network includes an input layer, a hidden layer and an output layer. The number of nodes in the input layer corresponds to the dimension of the meteorological data; The hidden layer consists of a long short-term memory layer containing a gating structure. The gating structure selectively remembers and forgets historical information received from the input layer, thereby capturing the nonlinear mapping relationship between meteorological data and photovoltaic output power. The number of nodes in the output layer corresponds to the future time step that needs to be predicted, and outputs the photovoltaic output power for the next 24 hours. The training set is input into the trained Long Short-Term Memory (LSTM) network. The mean square error between the output photovoltaic power output for the next 24 hours and the actual output label is calculated. Adaptive estimation is used, and the weight parameters inside the LSM network are adjusted according to the backpropagation of the mean square error. The calculation process of the training set is iterated until the mean square error converges. The generalization ability of the photovoltaic power generation model is tested using the validation set.

[0009] A further improvement to the technical solution of this invention lies in the following: the process of inputting load characteristics and factory production plans into the electrolytic aluminum load model, and outputting the load baseline power and adjustable flexible range using the baseline and flexible dichotomy method includes: The load baseline level in the load characteristics is taken as the load baseline power; The distribution range of high-frequency disturbance components during normal production is statistically analyzed to obtain an initial adjustment range driven by historical data. The initial adjustment range is compared with the rigid safety boundary set according to the factory production plan and the electrolytic cell process requirements. The intersection of the initial adjustment range and the rigid safety boundary is taken as the output adjustable flexible range.

[0010] A further improvement to the technical solution of this invention lies in the process of obtaining a power prediction report using the output of the photovoltaic power generation model and the output of the electrolytic aluminum load model, which includes: The system state prediction model calls the photovoltaic output power time series for the next 24 hours output by the photovoltaic power generation model and the load baseline power output by the electrolytic aluminum load model to extrapolate the most recent load baseline power to the next 24 hours, forming the load baseline power time series for the next 24 hours. The load baseline power time series and the photovoltaic output power time series are aligned and calculated, using the load baseline power time series as the basis. Subtract photovoltaic output power time series The net load forecast sequence of the energy storage-photovoltaic-electrolytic aluminum system for the next 24 hours is obtained, and the photovoltaic output power time series, the load baseline power time series and the net load forecast sequence are integrated into a power forecast report.

[0011] A further improvement to the technical solution of this invention lies in the following: The process of outputting the day-ahead optimal economic dispatch plan using a mixed-integer linear programming approach, based on electricity price characteristics, power forecast reports, and the output of the energy storage system model through an economic dispatch optimization model, includes: Economic dispatch optimization model based on power grid switching power The real-time time-of-use electricity price provided by the time-of-use electricity price meter and battery cycle aging costs The problem employs mixed-integer linear programming, with the objective function being to minimize the daily operating cost of the energy storage-photovoltaic-electrolytic aluminum system. Optimization is performed to find the solution, where... The positive time indicates the purchase of electricity. A negative value indicates electricity sales and backfeeding to the grid, and T represents the total number of time steps in the optimization cycle; With charging power Discharge power Using binary variables that cause charging and discharging to occur at different times as decision variables, power balance constraints and energy storage system operation constraints are established. Power balance constraints This is used to balance the power generation and consumption of the energy storage system; Energy storage system operating constraints include state of charge update constraints, state of charge boundary constraints, and charge / discharge power constraints. State of charge update constraints... This is used to update the state of charge (SPC) of an energy storage system at any time t based on the charging and discharging power; SPC boundary constraints. This is used to preset the state of charge of the energy storage system between its minimum and maximum values, and to constrain the charging and discharging power. , This is used to limit the charging and discharging power of the energy storage system within a preset maximum range, and charging and discharging cannot occur simultaneously. and These represent the minimum and maximum states of charge. and These are the maximum values ​​of charging power and discharging power, respectively. By integrating the objective function, decision variables, and energy storage system operating constraints, the sequence of decision variables that minimizes the objective function value is calculated using mixed-integer linear programming. This sequence is then used as the optimal charging and discharging power of the energy storage system at each time step within the next 24 hours. Finally, the optimal charging and discharging power at each time step within the next 24 hours is formatted as the day-ahead optimal economic dispatch plan.

[0012] A further improvement to the technical solution of this invention lies in the following: The process of outputting energy storage power control commands using an adaptive control model, based on preprocessed photovoltaic system data, energy storage system data, electrolytic aluminum load power data, and the day-ahead optimal economic dispatch plan, and employing deep reinforcement learning technology, includes: Define the states, actions, and rewards required for the adaptive control model to make decisions on energy storage power control commands. The states include the real-time photovoltaic power of the photovoltaic system, the real-time electrolytic aluminum load power, and the power deviation from the day-ahead optimal economic dispatch plan. The actions are the energy storage power control commands for the next time step, with values ​​constrained by the rated power of the energy storage converter. The rewards are... The negative value associated with real-time operating costs, where, This represents the actual cost of purchasing electricity from the grid at time t; A digital twin system is composed of an energy storage system model, a photovoltaic power generation model, and an electrolytic aluminum load model. An agent is trained in the digital twin system. The agent outputs energy storage power control commands, and the digital twin system calculates the state and reward for the next moment and feeds them back to the agent. The agent is trained using a deep deterministic policy gradient algorithm. The agent performs simulated interactions in a digital twin system. In each simulated interaction, the agent selects an action based on the current state. The digital twin system provides feedback on the reward and the new state. The agent then adjusts the weight parameters of its internal neural network based on the reward signal. The training process of the agent is iterated repeatedly until the agent's policy converges and the agent continuously obtains the maximum cumulative reward. The mature offline-trained agent policy network is deployed in the energy management system. At each time step of actual operation, the energy management system inputs the real-time collected state data into the agent policy network. The agent policy network outputs the optimal energy storage power control command for the current state through a forward propagation calculation.

[0013] A further improvement to the technical solution of this invention lies in the following: The multi-scenario approach includes conventional scenarios, extreme weather scenarios, and grid fault scenarios. Combining grid data and meteorological data, the process of transmitting the energy storage power control command to the energy storage converter for execution after multi-scenario logical judgment includes: When the grid data is within the normal operating range and the energy management system does not receive any external extreme weather warning signals, the current scenario is determined to be a normal scenario, and the energy storage power control command output by the adaptive control model is directly transmitted to the energy storage converter for execution. When the energy management system receives an external warning based on meteorological data, it determines that the current scenario is an extreme weather scenario, temporarily suspends the energy storage power control command output by the adaptive control model, executes the emergency plan, generates a safety-oriented energy storage power control command, and transmits the safety-oriented energy storage power control command to the energy storage converter for execution. When the monitoring and protection device at the grid connection point detects an anomaly in the grid data, it determines that the current scenario is a grid fault scenario, triggers the highest priority protection and control logic, issues a command to disconnect the circuit breaker connected to the public grid, causes the plant microgrid to enter islanded operation mode, stops the energy storage power control command output by the adaptive control model, switches to the islanded operation power supply control logic to maintain the voltage and frequency stability of the grid within the island, and prioritizes ensuring continuous power supply to the core load of electrolytic aluminum, generates a balanced energy storage power control command generated by the islanded operation controller according to the real-time power balance requirements, and transmits the balanced energy storage power control command to the energy storage converter.

[0014] A further improvement to the technical solution of this invention lies in the following: the process of visualizing the energy storage state of charge, photovoltaic power, and photovoltaic-load performance curve data through an integrated digital dashboard, and feeding back the execution effect of energy storage power control commands to the adaptive control model, includes: The front-end application of the energy management system requests and aggregates data on energy storage state of charge, photovoltaic power, and photovoltaic and load performance curves from historical and real-time databases. After structuring, the data is visualized on the human-machine interface through dynamic charts, dashboards, and status indicators. After each energy storage power control command is executed, the energy management system records an experience data unit containing the state before execution, the actions performed, the actual reward, and the new state after execution. The experience data unit is stored in the experience playback database. Data is periodically extracted from the experience playback database to retrain and optimize the policy network of the adaptive control model offline. The updated policy network is then redeployed back to the energy management system.

[0015] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows: 1. This invention provides a dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system that is adaptable to multiple scenarios. By constructing a prediction model and an adaptive control model, it achieves accurate prediction of photovoltaic output and real-time dynamic optimization of energy storage charging and discharging strategies. It can maximize the use of photovoltaic green electricity for peak-valley arbitrage based on actual lighting conditions, and comprehensively consider the battery cycle aging cost, thus significantly improving the overall economic benefits of the entire system.

[0016] 2. This invention provides a dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system that adapts to multiple scenarios. By introducing multi-scenario judgment logic, the system can adaptively switch control objectives between different modes such as normal economic operation, extreme weather, and grid failure. This achieves a dynamic balance from a single economic objective to multiple objectives that take reliability into account, prioritizing the power supply stability and safety of the core load of electrolytic aluminum, and effectively enhancing the system's operational reliability under complex operating conditions.

[0017] 3. This invention provides a dynamic optimization method for energy storage-photovoltaic-electrolytic aluminum systems that is adaptable to multiple scenarios. By establishing a complete intelligent workflow from data acquisition and feature extraction to closed-loop feedback, it realizes the coordinated and refined management of photovoltaic, energy storage and load units, solves the problem of insufficient green energy consumption caused by the inability of fixed control strategies to match real-time operating conditions, and comprehensively improves energy utilization efficiency and system automation level. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 The flowchart shows a dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adapted to multiple scenarios, provided by the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0021] Example 1, such as Figure 1 As shown, this invention provides a dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios, including the following steps: Step 1: Collect and preprocess data from the energy storage-photovoltaic-electrolytic aluminum system to obtain preprocessed data. Extract photovoltaic characteristics, load characteristics, and electricity price characteristics from the preprocessed data. The data includes energy storage system data, photovoltaic system data, electrolytic aluminum load power data, power grid data, and meteorological data.

[0022] In some embodiments, inside the integrated energy storage converter and boost converter in the project plant area, the integrated energy storage converter and battery management system are used to collect the real-time charging and discharging power and state of charge of the energy storage system as energy storage system data. The energy management system periodically polls the energy storage converter and battery management system through the internal communication network in accordance with standard protocols to obtain the energy storage system data.

[0023] In some embodiments, by installing multi-functional energy meters at the grid-connected junction points of the photovoltaic areas within the project plant area, the photovoltaic power time series of the photovoltaic system is measured in real time as photovoltaic system data. The energy management system reads the data from the multi-functional energy meters through the communication interface to grasp the real-time photovoltaic power generation situation.

[0024] In some embodiments, a high-precision power quality monitoring device is installed on the feeder circuit of the power distribution room in the project plant area that supplies power to the electrolytic aluminum load. The power quality monitoring device monitors the real-time active power and current flowing to the electrolytic aluminum load as the power data of the electrolytic aluminum load. The energy management system obtains the power consumption of the core load by reading the data from the power quality monitoring device.

[0025] In some embodiments, the power exchange, frequency and voltage of the power grid are collected by the measurement and control protection device at the grid connection point of the project plant area, and the time-of-use electricity price meter is statically configured in the energy management system as basic information, with the power exchange, frequency and voltage of the power grid and the time-of-use electricity price meter as the power grid data.

[0026] In some embodiments, small automated weather stations deployed in photovoltaic areas collect real-time total solar irradiance, ambient temperature, relative humidity, and wind speed as meteorological data at the project site.

[0027] In some embodiments, the collected energy storage system data, photovoltaic system data, electrolytic aluminum load power data, power grid data, and meteorological data are appended with a unified timestamp and then transmitted to the historical database of the energy management system for centralized storage. The energy storage-photovoltaic-electrolytic aluminum system data is then cleaned. The cleaning process includes identifying outlier data points caused by sensor jumps and communication interruptions using the interquartile range method, filling out the outlier data points with linear interpolation using valid data points at adjacent times, aligning the cleaned energy storage-photovoltaic-electrolytic aluminum system data on time scales, and performing max-min normalization processing to obtain preprocessed energy storage system data, photovoltaic system data, electrolytic aluminum load power data, power grid data, and meteorological data.

[0028] In some embodiments, photovoltaic characteristics include photovoltaic power change rate and short-term volatility characteristics. The photovoltaic power change rate is obtained by calculating the difference between the power value at the current moment and the power value at the previous moment based on preprocessed photovoltaic system data. The short-term volatility characteristics are obtained by calculating the standard deviation of photovoltaic system data within a preset time window.

[0029] In some embodiments, the load characteristics include a load baseline level and high-frequency disturbance component characteristics. The real-time active power is smoothed by a moving average filter, and the output of the moving average filter is regarded as the load baseline level. The difference sequence obtained by subtracting the load baseline level from the real-time active power is used to characterize the high-frequency disturbance component characteristics.

[0030] In some embodiments, the electricity price characteristics include the peak-valley time distribution and peak-valley price difference characteristics within the next 24 hours. By parsing the time-of-use electricity price table through a program, the start and end times of peak, valley, and flat periods within the next 24 hours are identified and output to form the peak-valley time distribution within the next 24 hours. Based on the peak-valley time distribution within the next 24 hours, the difference between the peak-hour electricity price and the valley-hour electricity price is extracted and calculated from the configured time-of-use electricity price table to obtain the peak-valley price difference characteristics.

[0031] Step 2: Input the preprocessed energy storage system data and energy storage power control commands into the energy storage system model, and output the state of charge and battery cycle aging cost at the next moment; Based on the preprocessed meteorological data and long short-term memory network, train the photovoltaic power generation model, and output the photovoltaic output power for the next 24 hours through the trained photovoltaic power generation model; Input the load characteristics and factory production plan into the electrolytic aluminum load model, and output the load baseline power and adjustable flexible range using the baseline and flexible dichotomy method.

[0032] In some embodiments, the energy storage system model receives energy storage power control commands from the adaptive control model. If the energy storage power control command is negative, it is considered as charging power. If the energy storage power control command is positive, it is considered as the discharge power. The rated capacity of the energy storage system is preset. Charging efficiency and discharge efficiency Calculate the state of charge at the next time step. The specific calculation formula is as follows:

[0033] in, This represents the state of charge at the previous moment. This represents the time step for the calculation.

[0034] In some embodiments, based on the preset total investment cost of the project plant area and the number of cycles throughout the battery's lifespan. Calculate the equivalent cycle cost of completing one full charge and discharge cycle. Based on the charging and discharging energy at the current moment Calculate the equivalent number of cycles for charge and discharge energy. Multiply the equivalent number of cycles by the equivalent cycle cost to obtain the battery cycle aging cost within the current time step. The specific calculation formula is as follows:

[0035]

[0036] in, The total energy throughput of a single full charge.

[0037] In some embodiments, a photovoltaic power generation model based on a long short-term memory network is trained and fixed in an energy management system, and executed on a rolling basis at a fixed frequency. Meteorological data is used as the input feature of the photovoltaic power generation model and divided into a training set and a validation set. The total output power of the actual photovoltaic power generation system at the same time is used as the output label. The long short-term memory network includes an input layer, a hidden layer and an output layer.

[0038] In some embodiments, the number of nodes in the input layer corresponds to the dimension of the meteorological data.

[0039] In some embodiments, the hidden layer consists of a long short-term memory layer containing gating structures that selectively memorize and forget historical information received from the input layer, thereby capturing the nonlinear mapping relationship between meteorological data and photovoltaic output power.

[0040] In some embodiments, the number of nodes in the output layer corresponds to the future time step that needs to be predicted, and the photovoltaic output power for the next 24 hours is output.

[0041] In some embodiments, the training set is input into the trained Long Short-Term Memory (LSTM) network, the mean square error between the output photovoltaic power output for the next 24 hours and the actual output label is calculated, adaptive estimation is used, the weight parameters inside the LSM network are adjusted according to the backpropagation of the mean square error, the training set calculation process is iterated until the mean square error converges, and the generalization ability of the photovoltaic power generation model is tested using the validation set.

[0042] In some embodiments, the load baseline level in the load characteristics is used as the load baseline power.

[0043] In some embodiments, the distribution range of high-frequency disturbance components during normal production is statistically analyzed to obtain an initial adjustment range driven by historical data. The initial adjustment range is compared with a rigid safety boundary set according to the factory production plan and the electrolytic cell process requirements. The intersection of the initial adjustment range and the rigid safety boundary is taken as the output adjustable flexible range.

[0044] Step 3: Using the system state prediction model, the outputs of the photovoltaic power generation model and the electrolytic aluminum load model are used to obtain a power prediction report; using the economic dispatch optimization model, based on the electricity price characteristics, the power prediction report, and the output of the energy storage system model, mixed integer linear programming is used to output the day-ahead optimal economic dispatch plan; using the adaptive control model, based on the preprocessed photovoltaic system data, energy storage system data, electrolytic aluminum load power data, and the day-ahead optimal economic dispatch plan, deep reinforcement learning technology is used to output energy storage power control commands.

[0045] In some embodiments, the system state prediction model calls the photovoltaic output power time series for the next 24 hours output by the photovoltaic power generation model and the load baseline power output by the electrolytic aluminum load model to extrapolate the most recent load baseline power to the next 24 hours, forming a load baseline power time series for the next 24 hours.

[0046] In some embodiments, the load baseline power time series and the photovoltaic output power time series are aligned for calculation, using the load baseline power time series as the reference. Subtract photovoltaic output power time series The net load forecast sequence of the energy storage-photovoltaic-electrolytic aluminum system for the next 24 hours is obtained, and the photovoltaic output power time series, the load baseline power time series and the net load forecast sequence are integrated into a power forecast report.

[0047] In some embodiments, the economic dispatch optimization model is based on grid switching power. The real-time time-of-use electricity price provided by the time-of-use electricity price meter and battery cycle aging costs The problem employs mixed-integer linear programming, with the objective function being to minimize the daily operating cost of the energy storage-photovoltaic-electrolytic aluminum system. Optimization is performed to find the solution, where... The positive time indicates the purchase of electricity. A negative value indicates electricity sales and backfeeding to the grid, and T represents the total number of time steps in the optimization cycle.

[0048] In some embodiments, with charging power Discharge power Using binary variables that prevent charging and discharging from occurring simultaneously as decision variables, power balance constraints and energy storage system operation constraints are established.

[0049] In some embodiments, power balance constraints This is used to balance the power generation and consumption of the energy storage system.

[0050] In some embodiments, the operating constraints of the energy storage system include state of charge update constraints, state of charge boundary constraints, and charge / discharge power constraints. The state of charge update constraints... This is used to update the state of charge (SPC) of an energy storage system at any time t based on the charging and discharging power; SPC boundary constraints. This is used to preset the state of charge of the energy storage system between its minimum and maximum values, and to constrain the charging and discharging power. , This is used to limit the charging and discharging power of the energy storage system within a preset maximum range, and charging and discharging cannot occur simultaneously. and These represent the minimum and maximum states of charge. and These are the maximum values ​​of charging power and discharging power, respectively.

[0051] In some embodiments, the objective function, decision variables, and energy storage system operating constraints are integrated, and the sequence of decision variables that minimizes the objective function value is calculated by mixed integer linear programming. The sequence of decision variables that minimizes the objective function value is used as the optimal charging and discharging power of the energy storage system at each time step in the next 24 hours, and the optimal charging and discharging power at each time step in the next 24 hours is formatted as the day-ahead optimal economic dispatch plan.

[0052] In some embodiments, the adaptive control model is defined with respect to the states, actions, and rewards required for making energy storage power control command decisions. The states include the real-time photovoltaic power of the photovoltaic system, the real-time electrolytic aluminum load power, and the power deviation from the day-ahead optimal economic dispatch plan. The action is the energy storage power control command for the next time step, with values ​​constrained by the rated power of the energy storage converter. The reward is... The negative value associated with real-time operating costs, where, This represents the actual cost of purchasing electricity from the grid at time t.

[0053] In some embodiments, a digital twin system is constituted by an energy storage system model, a photovoltaic power generation model, and an electrolytic aluminum load model. An agent is trained in the digital twin system, the agent outputs energy storage power control commands, and the digital twin system calculates the state and reward for the next moment and feeds them back to the agent.

[0054] In some embodiments, a deep deterministic policy gradient algorithm is used to train the agent. The agent performs simulated interactions in a digital twin system. In each simulated interaction, the agent selects an action based on the current state. The digital twin system provides feedback on the reward and the new state. The agent then adjusts the weight parameters of its internal neural network based on the reward signal. The agent training process is iterated repeatedly until the agent's policy converges and the agent continuously obtains the maximum cumulative reward.

[0055] In some embodiments, an offline-trained, mature agent policy network is deployed in an energy management system. At each time step of actual operation, the energy management system inputs the real-time collected state data into the agent policy network. The agent policy network outputs the optimal energy storage power control command for the current state through a forward propagation calculation.

[0056] Step 4: Combining grid data and meteorological data, and considering multiple scenarios including normal scenarios, extreme weather scenarios, and grid fault scenarios, the energy storage power control command is transmitted to the energy storage converter for execution after being judged by multi-scenario logic, and the execution effect of the energy storage power control command is obtained.

[0057] In some embodiments, when the grid data is within the normal operating range and the energy management system does not receive an external extreme weather warning signal, the current scenario is determined to be a normal scenario, and the energy storage power control command output by the adaptive control model is directly transmitted to the energy storage converter for execution.

[0058] In some embodiments, when the energy management system receives an external early warning based on meteorological data, it determines that the current scenario is an extreme weather scenario, temporarily suspends the energy storage power control command output by the adaptive control model, executes the emergency plan, generates a safety-oriented energy storage power control command, and transmits the safety-oriented energy storage power control command to the energy storage converter for execution.

[0059] In some embodiments, when the monitoring and protection device at the grid connection point detects an abnormality in the grid data, it determines that the current scenario is a grid fault scenario, triggers the highest priority protection and control logic, issues a command to disconnect the circuit breaker connected to the public grid, causes the plant microgrid to enter islanded operation mode, stops the energy storage power control command output by the adaptive control model, switches to the islanded operation power supply control logic to maintain the voltage and frequency stability of the grid within the island and prioritizes the continuous power supply to the core load of electrolytic aluminum, generates a balanced energy storage power control command generated by the islanded operation controller according to the real-time power balance requirements, and transmits the balanced energy storage power control command to the energy storage converter.

[0060] In some embodiments, when the monitoring and protection device at the grid connection point detects an abnormality in the grid data, it determines it to be a grid fault scenario, triggers the highest priority protection and control logic, issues a command to disconnect the circuit breaker connected to the public grid, causes the plant microgrid to enter islanded operation mode, stops the energy storage power control command output by the adaptive control model, switches to the islanded operation power supply control logic to maintain the voltage and frequency stability of the grid within the island and prioritizes the continuous power supply to the core load of electrolytic aluminum, generates the energy storage power control command generated by the islanded operation controller according to the real-time power balance requirements, and transmits the energy storage power control command to the energy storage converter.

[0061] Step 5: Visualize the energy storage state of charge, photovoltaic power, and photovoltaic and load performance curve data through an integrated digital dashboard, and feed back the execution effect of energy storage power control commands to the adaptive control model.

[0062] In some embodiments, the front-end application of the energy management system requests and aggregates energy storage state of charge, photovoltaic power, and photovoltaic and load performance curve data from historical and real-time databases, and then visualizes them on the human-machine interface through dynamic charts, dashboards, and status indicators after structuring.

[0063] In some embodiments, after each energy storage power control command is executed, the energy management system records an experience data unit containing the state before execution, the action performed, the actual reward, and the new state after execution, and stores the experience data unit in the experience playback database. The system periodically extracts data from the experience playback database to retrain and optimize the policy network of the adaptive control model offline, and redeploys the updated policy network back to the energy management system.

[0064] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios, characterized in that, Includes the following steps: Data from an energy storage-photovoltaic-electrolytic aluminum system is collected and preprocessed to obtain preprocessed energy storage-photovoltaic-electrolytic aluminum system data. Photovoltaic characteristics, load characteristics, and electricity price characteristics are obtained from the preprocessed energy storage-photovoltaic-electrolytic aluminum system data. The energy storage-photovoltaic-electrolytic aluminum system data includes energy storage system data, photovoltaic system data, electrolytic aluminum load power data, power grid data, and meteorological data. The process of collecting and preprocessing data from the energy storage-photovoltaic-electrolytic aluminum system to obtain preprocessed data, and extracting photovoltaic characteristics, load characteristics, and electricity price characteristics from the preprocessed data, includes: Inside the integrated energy storage converter and boost converter in the project plant area, the real-time charging and discharging power and state of charge of the energy storage system are collected as data of the energy storage system by using the integrated energy storage converter and battery management system. By installing multi-functional energy meters at the grid-connected junction points of the photovoltaic areas within the project plant area, the photovoltaic power time series of the photovoltaic system is measured in real time and used as the photovoltaic system data. A power quality monitoring device is installed on the feeder circuit of the power distribution room in the project plant area that supplies power to the electrolytic aluminum load. The power quality monitoring device monitors the real-time active power and current flowing to the electrolytic aluminum load in real time as the power data of the electrolytic aluminum load. The power, frequency and voltage of the power grid are collected using the measurement and control protection device at the grid connection point of the project plant area. The time-of-use electricity price meter is statically configured in the preset energy management system as the basic information, and the power, frequency and voltage of the power grid and the time-of-use electricity price meter are used as the power grid data. By deploying small automated weather stations in the photovoltaic area, real-time total solar irradiance, ambient temperature, relative humidity, and wind speed at the project site are collected as meteorological data. The collected data from the energy storage system, photovoltaic system, electrolytic aluminum load power data, power grid data, and meteorological data are appended with a unified timestamp and then transmitted to the historical database of the energy management system for centralized storage. After undergoing maximum and minimum normalization processing, the pre-processed data from the energy storage system, photovoltaic system, electrolytic aluminum load power data, power grid data, and meteorological data are obtained. The photovoltaic characteristics include photovoltaic power change rate and short-term volatility characteristics. The photovoltaic power change rate is obtained by calculating the difference between the power value at the current moment and the power value at the previous moment based on the preprocessed photovoltaic system data. The short-term volatility characteristics are obtained by calculating the standard deviation of the photovoltaic system data within a preset time window. The load characteristics include the load baseline level and high-frequency disturbance component characteristics. The real-time active power is smoothed by a moving average filter, and the output of the moving average filter is regarded as the load baseline level. The difference sequence obtained by subtracting the load baseline level from the real-time active power is used to characterize the high-frequency disturbance component characteristics. The electricity price characteristics include the peak-valley time distribution and peak-valley price difference characteristics within the next 24 hours. The program parses the time-of-use electricity price table, identifies and outputs the start and end times of peak, valley and flat periods within the next 24 hours, forming the peak-valley time distribution within the next 24 hours. Based on the peak-valley time distribution within the next 24 hours, the difference between the peak period electricity price and the valley period electricity price is extracted and calculated from the configured time-of-use electricity price table to obtain the peak-valley price difference characteristics. The preprocessed energy storage system data and energy storage power control commands are input into the energy storage system model, and the state of charge and battery cycle aging cost at the next moment are output. Based on the preprocessed meteorological data and long short-term memory network, the photovoltaic power generation model is trained, and the photovoltaic output power for the next 24 hours is output through the trained photovoltaic power generation model. The load characteristics and factory production plan are input into the electrolytic aluminum load model, and the baseline and flexible dichotomy method is used to output the load baseline power and adjustable flexible range. A power forecast report is obtained using the output of the photovoltaic power generation model and the electrolytic aluminum load model through a system state prediction model. An economic dispatch optimization model is then used, based on the electricity price characteristics, the power forecast report, and the output of the energy storage system model, employing mixed-integer linear programming to output the day-ahead optimal economic dispatch plan. Finally, an adaptive control model is used, based on the preprocessed photovoltaic system data, energy storage system data, electrolytic aluminum load power data, and the day-ahead optimal economic dispatch plan, employing deep reinforcement learning techniques to output energy storage power control commands. The process of outputting energy storage power control commands using an adaptive control model, based on the preprocessed photovoltaic system data, energy storage system data, electrolytic aluminum load power data, and the day-ahead optimal economic dispatch plan, and employing deep reinforcement learning technology, includes the following: Define the state, action, and reward required for the adaptive control model to make the energy storage power control command decision. The state includes the real-time photovoltaic power of the photovoltaic system, the real-time electrolytic aluminum load power, and the power deviation from the day-ahead optimal economic dispatch plan. The action is the energy storage power control command for the next time step whose value range is constrained by the rated power of the energy storage converter. The reward is a negative value related to the real-time operating cost. A digital twin system is jointly constructed by the energy storage system model, the photovoltaic power generation model, and the electrolytic aluminum load model. An agent is trained in the digital twin system, the agent outputs the energy storage power control command, and the digital twin system calculates the state and reward at the next moment and feeds them back to the agent. The agent is trained using a deep deterministic policy gradient algorithm. The agent performs simulated interactions in the digital twin system. In each simulated interaction, the agent selects an action based on the current state. The digital twin system provides feedback on the reward and the new state. The agent then adjusts the weight parameters of its internal neural network based on the reward signal. The training process of the agent is iterated repeatedly until the agent's policy converges and the agent continuously obtains the maximum cumulative reward. The offline-trained and mature agent policy network is deployed in the energy management system. At each time step of actual operation, the energy management system inputs the real-time collected state data into the agent policy network. The agent policy network outputs the optimal energy storage power control command under the current state through one forward propagation calculation. By combining the power grid data and the meteorological data, the energy storage power control command is transmitted to the energy storage converter for execution after being determined by multi-scenario logic, and the execution effect of the energy storage power control command is obtained. The integrated digital dashboard visualizes the energy storage charge status, photovoltaic power, and photovoltaic-load performance curve data, and feeds back the execution effect of the energy storage power control command to the adaptive control model.

2. The dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios according to claim 1, characterized in that, The process of inputting the preprocessed energy storage system data and energy storage power control commands into the energy storage system model, and outputting the state of charge and battery cycle aging cost at the next moment, includes: The energy storage system model receives the energy storage power control command from the adaptive control model. If the energy storage power control command is negative, it is considered as charging power; if the energy storage power control command is positive, it is considered as discharging power. The state of charge at the next moment is calculated by applying the preset rated capacity, charging efficiency, and discharging efficiency of the energy storage system. Based on the preset total investment cost of the project plant and the number of cycles of the battery throughout its entire life cycle, the equivalent cycle cost of completing one full charge and discharge cycle is calculated. Based on the charge and discharge energy at the current moment, the equivalent cycle number of the charge and discharge energy is calculated. The equivalent cycle number is multiplied by the equivalent cycle cost to obtain the battery cycle aging cost within the current time step.

3. The dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios according to claim 1, characterized in that, The process of training a photovoltaic power generation model based on the preprocessed meteorological data and the long short-term memory network, and outputting the photovoltaic output power for the next 24 hours, includes: A photovoltaic power generation model based on a long short-term memory network is trained and fixed in the energy management system, and executed at a fixed frequency. The meteorological data is used as the input feature of the photovoltaic power generation model and divided into a training set and a validation set. The total output power of the actual photovoltaic power generation system at the same time is used as the output label. The long short-term memory network includes an input layer, a hidden layer and an output layer. The number of nodes in the input layer corresponds to the dimension of the meteorological data; The hidden layer consists of a long short-term memory layer containing a gating structure, which selectively memorizes and forgets historical information received from the input layer, thereby capturing the nonlinear mapping relationship between the meteorological data and the photovoltaic output power. The number of nodes in the output layer corresponds to the future time step that needs to be predicted, and outputs the photovoltaic output power for the next 24 hours. The training set is input into the trained Long Short-Term Memory (LSTM) network. The mean square error between the output photovoltaic power output for the next 24 hours and the actual output label is calculated. Adaptive estimation is used, and the weight parameters inside the LSM network are adjusted based on the mean square error through backpropagation. The calculation process of the training set is iterated until the mean square error converges. The generalization ability of the photovoltaic power generation model is then tested using the validation set.

4. The dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios according to claim 1, characterized in that, The process of inputting the load characteristics and factory production plan into the electrolytic aluminum load model, and outputting the load baseline power and adjustable flexible range using the baseline and flexible dichotomy method includes: The load baseline level in the load characteristics is taken as the load baseline power; The distribution range of the high-frequency disturbance component characteristics during normal production is statistically analyzed to obtain an initial adjustment range driven by historical data. The initial adjustment range is compared with the rigid safety boundary set according to the factory production plan and the electrolytic cell process requirements. The intersection of the initial adjustment range and the rigid safety boundary is taken as the output adjustable flexible range.

5. The dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios according to claim 1, characterized in that, The process of obtaining a power prediction report using the output of the photovoltaic power generation model and the output of the electrolytic aluminum load model includes: The system state prediction model calls the photovoltaic output power time series for the next 24 hours output by the photovoltaic power generation model and the load baseline power output by the electrolytic aluminum load model, and extrapolates the most recent load baseline power to the next 24 hours to form the load baseline power time series for the next 24 hours; The load baseline power time series and the photovoltaic output power time series are aligned and calculated. The photovoltaic output power time series is subtracted from the load baseline power time series to obtain the net load forecast sequence of the energy storage-photovoltaic-electrolytic aluminum system for the next 24 hours. The photovoltaic output power time series, the load baseline power time series and the net load forecast sequence are then integrated into a power forecast report.

6. The dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios according to claim 1, characterized in that, The process of outputting the day-ahead optimal economic dispatch plan using a mixed-integer linear programming approach based on the electricity price characteristics, the power forecast report, and the output of the energy storage system model through an economic dispatch optimization model includes: The economic dispatch optimization model, based on the power grid exchange power, the real-time time-of-use electricity price provided by the time-of-use electricity price table, and the battery cycle aging cost, adopts mixed integer linear programming and optimizes the solution with the objective function of minimizing the daily operating cost of the energy storage-photovoltaic-electrolytic aluminum system. Using charging power, discharging power, and binary variables that cause charging and discharging to occur at different times as decision variables, power balance constraints and energy storage system operation constraints are established. The power balance constraint is used to balance the power generation and consumption of the energy storage system; The energy storage system operation constraints include state of charge update constraints, state of charge boundary constraints, and charge / discharge power constraints. The state of charge update constraints are used to ensure that the state of charge of the energy storage system is updated at any time according to the charge / discharge power. The state of charge boundary constraints are used to preset the state of charge of the energy storage system between a minimum and a maximum value. The charge / discharge power constraints are used to limit the charge / discharge power of the energy storage system within a preset maximum value range, and charging and discharging cannot occur simultaneously. By integrating the objective function, the decision variables, and the energy storage system operating constraints, the sequence of decision variables that minimizes the objective function value is calculated using mixed integer linear programming. The sequence of decision variables that minimizes the objective function value is then used as the optimal charging and discharging power of the energy storage system at each time step in the next 24 hours. The optimal charging and discharging power at each time step in the next 24 hours is then formatted as the day-ahead optimal economic dispatch plan.

7. The dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios according to claim 1, characterized in that, The multiple scenarios include conventional scenarios, extreme weather scenarios, and grid failure scenarios. The process of combining the grid data and the weather data, and then transmitting the energy storage power control command to the energy storage converter for execution after multi-scenario logic determination, includes: When the power grid data is within the normal operating range and the energy management system does not receive an external extreme weather warning signal, the current scenario is determined to be the normal scenario, and the energy storage power control command output by the adaptive control model is directly transmitted to the energy storage converter for execution. When the energy management system receives an external warning based on the meteorological data, it determines that the current scenario is the extreme weather scenario, temporarily suspends the energy storage power control command output by the adaptive control model, executes the emergency plan, generates a safety-oriented energy storage power control command, and transmits the safety-oriented energy storage power control command to the energy storage converter for execution. When the monitoring and protection device at the grid connection point detects an anomaly in the grid data, it determines that the current scenario is a grid fault scenario, triggers the highest priority protection and control logic, issues a command to disconnect the circuit breaker connected to the public grid, causes the plant microgrid to enter islanded operation mode, stops the energy storage power control command output by the adaptive control model, switches to islanded operation power supply control logic to maintain the voltage and frequency stability of the grid within the island and prioritizes ensuring continuous power supply to the core load of electrolytic aluminum, generates a balanced energy storage power control command generated by the islanded operation controller according to the real-time power balance requirements, and transmits the balanced energy storage power control command to the energy storage converter.

8. The dynamic optimization method for an energy storage-photovoltaic-electrolytic aluminum system adaptable to multiple scenarios according to claim 1, characterized in that, The process of visualizing the energy storage state of charge, photovoltaic power, and photovoltaic-load performance curve data through an integrated digital dashboard, and feeding back the execution effect of the energy storage power control command to the adaptive control model, includes: The front-end application of the energy management system requests and aggregates the energy storage state of charge, photovoltaic power and photovoltaic-load performance curve data from historical and real-time databases, and then visualizes them on the human-machine interface through dynamic charts, dashboards and status indicator lights after structuring. After each energy storage power control command is executed, the energy management system records an experience data unit containing the state before execution, the action performed, the actual reward, and the new state after execution. The experience data unit is stored in the experience playback database. Data is periodically extracted from the experience playback database to retrain and optimize the policy network of the adaptive control model offline. The updated policy network is then redeployed back to the energy management system.

Images