Individual differentiation intelligent charging and discharging optimization method and system for communication power storage battery

Abstract

The application provides an individual differentiation intelligent charging and discharging optimization method and system for communication power storage batteries. The method comprises collecting all storage battery monomer data, determining the state of charge, category and abnormal flag of the storage battery monomer based on all the storage battery monomer data; constructing a monomer state feature through the state of charge, category and abnormal flag of the storage battery monomer, determining a comprehensive reward function of the storage battery monomer based on the monomer state feature and a co-evolution reinforcement learning algorithm, and performing online policy optimization based on the reward function to construct a storage battery monomer differentiation control strategy model; inputting real-time data of each storage battery monomer into the storage battery monomer differentiation control strategy model, outputting the optimal charging and discharging strategy of each storage battery monomer, and implementing individual differentiation intelligent charging and discharging optimization for the communication power storage batteries. The application improves the monomer state perception accuracy of the communication power storage batteries, and improves the service life and charging and discharging reliability of the communication power storage batteries.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, and more specifically, it relates to the monitoring and analysis of the performance parameters of individual communication power supply batteries, as well as the construction and execution of charging and discharging strategies based on individual differences. The invention can be applied to the operation and maintenance of communication power systems, such as the operation and maintenance of batteries in communication operator's computer room, backup power management of data centers, and power supply guarantee for communication base stations, to help accurately control the charging and discharging of batteries, extend their service life, and ensure stable power supply. Background Technology

[0002] With the rapid development of communication technology, the large-scale construction of 5G base stations, and the continuous expansion of data centers, the stable operation of communication power supply battery packs, as the core backup force to ensure uninterrupted power supply to communication systems, is becoming increasingly critical. However, in communication scenarios, battery packs are subjected to complex operating conditions for extended periods. Affected by charging and discharging frequencies, ambient temperatures, and power grid fluctuations, the performance differences between individual batteries are constantly widening, posing challenges to the reliable operation and maintenance of communication power supplies.

[0003] Current battery charging and discharging management for communication power supplies mostly adopts a unified control strategy based on the entire battery pack, ignoring the individual differences in capacity, internal resistance, and self-discharge characteristics of individual cells. Under a unified charging and discharging mode, high-performance batteries cannot fully realize their potential, while low-performance batteries are prone to accelerated aging due to overcharging and over-discharging. This not only shortens the lifespan of the entire battery pack but may also lead to communication interruptions due to individual cell failure during power outages. Furthermore, existing management methods struggle to accurately identify individual cell differences and dynamically adjust strategies in a timely manner.

[0004] Existing technical document 1 (CN103401031B) discloses a method and device for monitoring and maintaining battery faults. Its shortcomings are that it only statically adjusts the charging and discharging current and voltage parameters for different scenarios, without taking into account the dynamic changes in the state of health (SOH) and state of charge (SOC) of the individual battery cells for differentiated adaptation. The adjustment logic is simple and lacks a closed-loop optimization mechanism, which cannot solve the operation and maintenance problems caused by the widening performance differences of individual cells.

[0005] Prior art document 2 (CN120315294B) discloses a method and related device for integrated remote capacity control of communication power supply and storage battery based on an improved fish swarm algorithm. Its drawback is that the fish swarm algorithm only focuses on charging current adjustment and discharge threshold setting, resulting in limited adaptability and intelligence. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides an intelligent charging and discharging optimization method suitable for the individualized management of communication power supply batteries. The method collects data from all individual battery cells, determines their state of charge (SOC), category, and anomaly indicators, and constructs individual cell state features based on these features. Then, using these features and a co-evolutionary reinforcement learning algorithm, a comprehensive reward function for each battery cell is determined. Online strategy optimization is performed based on this reward function to build a differentiated control strategy model for each individual battery cell. Real-time data from each individual battery cell is input into this model, which outputs the optimal charging and discharging strategy for each cell, enabling intelligent, individualized charging and discharging optimization of the communication power supply batteries. This method achieves accurate identification of individual battery cell performance differences and dynamic adaptation of charging and discharging strategies, achieving intelligent control of the entire communication power supply battery pack through "on-demand charging and discharging, and differentiated management." This invention can be applied to scenarios such as communication equipment room battery maintenance, base station backup power management, and data center uninterruptible power supply battery optimization, providing technical support for the stable operation of communication power systems. By accurately identifying the differences in performance parameters of individual batteries, and dynamically constructing charging and discharging actions adapted to each individual, differentiated charging and discharging control can be achieved. This solves the problems of short battery life and power supply reliability being constrained by the performance of individual batteries under a unified charging and discharging mode, and provides technical support for the reliable operation and maintenance of communication power supply batteries.

[0007] The present invention adopts the following technical solution.

[0008] The first aspect of the present invention provides an individualized intelligent charging and discharging optimization method for communication power supply batteries, comprising: The state of charge, category, and abnormal indicators of individual battery cells are obtained to construct individual cell state features; based on the individual cell state features, a co-evolutionary reinforcement learning algorithm is used to determine the comprehensive reward function of the individual battery cells; based on the comprehensive reward function, online policy optimization is performed to construct a differentiated control policy model for individual battery cells; Real-time data from each battery cell is input into the battery cell-specific control strategy model, which outputs the optimal charging and discharging strategy for each battery cell to implement individualized intelligent charging and discharging optimization for communication power batteries.

[0009] Preferably, obtaining the state of charge, type, and abnormal indicators of individual battery cells includes: Data from all individual battery cells was used; Based on the filtered current and open-circuit voltage in the individual battery cell data, the state of charge of the individual battery cell is determined using the ampere-hour integration method. Based on the actual available capacity and filtered current in the individual battery cell data, the health status of the individual battery cell is determined; based on the health status of the individual battery cell, the category of the individual battery cell is determined and the abnormal indicators of the individual battery cell are identified.

[0010] Preferably, the health status of a single battery cell is determined by the following formula:

[0011] In the formula, This indicates the health status of battery cell j at time t. This indicates the weight of the capacity component in SOH. This represents the actual capacity of monomer j at time t. This indicates the rated capacity of battery cell j. This indicates the weight of the internal resistance component in SOH. This represents the internal resistance penalty coefficient. This represents the rate of increase in internal resistance of a single battery cell j over time t.

[0012] Preferably, determining the category of a single battery cell includes: when , and If both conditions are met, the battery cell j at time t is classified as a Class 1 battery; when , and If both conditions are met, the battery cell j at time t is classified as a Class 2 battery; Except for Class 1 and Class 2 batteries, the remaining batteries are classified as Class 3 batteries; in, This represents the SOH threshold for Class 1 batteries. This represents the threshold value for the internal resistance ratio of Class 1 batteries. This represents the temperature coefficient of battery cell j. This represents the temperature coefficient threshold for Class 1 batteries. This indicates the minimum SOH value for the second type of battery. This indicates the maximum SOH value for the second type of battery. This indicates the minimum internal resistance ratio of the second type of battery. This indicates the maximum internal resistance ratio of the second type of battery. This represents the minimum temperature coefficient of the second type of battery. This indicates the maximum temperature coefficient of the second type of battery.

[0013] Preferably, the abnormal indicators for determining a single battery cell include: when , or If a battery cell j is found to be abnormal at time t, the abnormality flag for battery cell j at time t is set to 1; otherwise, if a battery cell j is found to be normal at time t, the abnormality flag for battery cell j at time t is set to 0.

[0014] Preferably, the construction of a differentiated control strategy model for individual battery cells includes: Based on the co-evolutionary reinforcement learning algorithm, multiple parallel battery cell co-evolutionary policy networks are constructed. The individual cell state features are input into the multiple parallel battery cell co-evolutionary policy networks, and the original actions are output. The penalty term for each battery cell is determined based on its original action, and a comprehensive reward function for each battery cell is constructed based on the penalty term. The fitness of the strategy network is determined based on the comprehensive reward function of the individual battery cells. The optimal strategy network of each generation is retained based on the fitness and merged with the strategy network after crossover and mutation to obtain the strategy network population set. PPO updates are performed on each policy network in the policy network population set to optimize the policy online until a set number of training iterations are reached, resulting in a trained policy network, which serves as the differentiated control policy model for individual battery cells. Construct a policy knowledge base for all trained policy networks. The policy knowledge base includes the fitness, class distribution, number of batteries in each class, average SOC, average SOH, and average internal resistance ratio of each policy network.

[0015] Preferably, the penalty term for a single battery cell is expressed by the following formula:

[0016] In the formula, The policy network i represents the penalty term obtained by battery cell j at time t. This indicates an indicator function; it returns 1 if the condition is true and 0 otherwise. This indicates the abnormal condition of battery cell j at time t. This represents the penalty coefficient for abnormal conditions. This indicates the threshold for low SOC conditions. When it is determined to be low SOC, This indicates a low SOC penalty coefficient. This indicates the threshold for high SOC conditions. It is judged as high SOC at that time. This indicates a high SOC penalty coefficient. This represents the original charging current coefficient. This represents the original discharge weight.

[0017] Preferably, the optimal charging and discharging strategy for each battery cell includes: Extract real-time data from each battery cell, perform similarity matching between the real-time data and the strategy knowledge base, and obtain the similarity between the real-time data of each battery cell and each strategy in the knowledge base. If the maximum similarity between the real-time data of each battery cell and all policy networks in the policy knowledge base is greater than the policy selection similarity threshold, a policy network is selected from the policy knowledge base based on the similarity to obtain the best matching policy in the policy knowledge base, which is then set as the optimal charging and discharging policy for each battery cell. Otherwise, the optimal strategy obtained from the training of the battery cell differentiated control strategy model is input into the real-time data and set as the optimal charging and discharging strategy for each battery cell.

[0018] Preferably, similarity matching is expressed by the following formula:

[0019] In the formula, This represents the similarity between the real-time data of each battery cell and the k-th policy network in the policy knowledge base. This indicates the real-time health status distribution of each individual battery cell. This represents the health status distribution of the k-th policy network in the policy knowledge base. This represents the real-time average SOC of each individual battery cell. The average SOC of the k-th policy network in the policy knowledge base is represented. This represents the real-time average SOH of each battery cell. The average state of interest (SOH) of the k-th policy network in the policy knowledge base is represented. This represents the real-time average internal resistance ratio of each individual battery cell. This represents the average internal resistance ratio of the k-th policy network in the policy knowledge base. Indicates the bandwidth of category distribution similarity. Indicates the bandwidth of SOC similarity. Indicates the SOH similarity bandwidth. This indicates the bandwidth of the internal resistance ratio similarity. This represents an exponential function.

[0020] A second aspect of the present invention provides an individualized intelligent charge-discharge optimization system for communication power batteries, which operates the individualized intelligent charge-discharge optimization method for communication power batteries described in the first aspect, comprising: The data acquisition module is used to acquire the state of charge, category, and abnormal indicators of individual battery cells to construct individual cell state characteristics; The model building module is used to determine the comprehensive reward function of a battery cell based on its individual state characteristics and using a co-evolutionary reinforcement learning algorithm; and to perform online policy optimization based on the comprehensive reward function to build a differentiated control strategy model for the battery cell. The strategy generation module is used to input the real-time data of each battery cell into the battery cell differentiated control strategy model, and output the optimal charging and discharging strategy for each battery cell to implement individual differentiated intelligent charging and discharging optimization of communication power batteries.

[0021] The third aspect of the present invention discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the computer program, when loaded onto the processor, implements the individualized intelligent charging and discharging optimization method for the communication power supply battery.

[0022] The fourth aspect of the present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the individualized intelligent charging and discharging optimization method for the communication power supply battery.

[0023] Compared with the prior art, the beneficial effects of the present invention include at least the following: This invention comprehensively acquires the real-time status of each individual cell through multi-dimensional data acquisition. The internal resistance directly reflects the aging differences of the cells, and the temperature identifies the differences in the thermal characteristics of the cells. It extracts the differentiated performance parameters of different individual batteries, solving the problem of "difficulty in accurately identifying individual cell differences" in the existing technology and improving the accuracy of individual cell status perception of communication power supply batteries. This invention uses an improved ampere-hour integration method to refine the estimation of the State of Charge (SOC) by considering temperature and current compensation for coulombic efficiency, and determines the State of OH (SOH) by considering capacitance and internal resistance. The Health (health status) module addresses the technical problem that "existing technologies do not incorporate the dynamic changes in the health status and state of charge of individual battery cells." It calculates the SOC and SOH of each cell in real time, providing a quantitative basis for differentiated control, dynamically tracks cell state changes, and supports adaptive strategy adjustments. Based on a three-level classification system of SOH, internal resistance growth rate, and temperature coefficient, it detects anomalies through SOH mutations, internal resistance mutations, and temperature mutations. This solves the technical problems of "weak cells being prone to accelerated aging due to overcharging and over-discharging" and "cell failure causing communication interruption risks." It identifies weak-performing cells through SOH assessment, detects potential faulty cells early through anomaly detection, and isolates high-risk cells through a three-level classification mechanism, providing a basis for safety protection decisions. This improves the accuracy of cell state estimation, classification accuracy, and anomaly detection sensitivity of communication power supply batteries. Quantifying the aging degree of each cell enhances the cell health status assessment capability. Precise SOC estimation and SOH assessment reduce the risk of overcharging and over-discharging. Accurate state assessment supports optimized charging and discharging strategies, extending the lifespan of communication power supply batteries. Early anomaly detection and classification isolation reduce the risk of communication interruption caused by cell failure leading to overall battery failure. This invention employs a co-evolutionary reinforcement learning (RL) framework, utilizing parallel evolution of the policy network to maintain policy diversity. A closed-loop optimization mechanism within reinforcement learning continuously improves the policy based on reward feedback, enabling online Proximal Policy Application (PPO). Optimization (near-end policy optimization) optimizes real-time adaptation to environmental changes, solving the technical problem of "single adjustment logic and lack of closed-loop optimization mechanism". Deep reinforcement learning provides highly intelligent decision-making capabilities, co-evolution mechanism improves algorithm adaptability and robustness, and policy knowledge base enables experience accumulation and rapid transfer, solving the technical problem of "limited adaptability and intelligence level of existing technology". It improves the intelligence level of charging and discharging strategies and enhances the reliability of communication power supply batteries. It provides customized decisions for different performance units through differentiated state vectors, and the reward function encourages optimization of overall performance rather than simple averaging. The policy network learns how to allocate resources according to the characteristics of different units, solving the technical problem of "good performance batteries cannot fully realize their potential" in existing technologies. Intelligent management reduces the operation and maintenance costs of communication power supply batteries by reducing the need for manual inspection and maintenance. The policy knowledge base supports rapid switching to historical excellent strategies, improving the emergency response capability of communication power supply batteries. It achieves a balance among multiple objectives such as lifespan, efficiency, and reliability, and improves the overall performance optimization capability of communication power supply batteries. This invention addresses the technical problem that "in existing technologies, under a uniform charge / discharge mode, high-performance batteries cannot fully realize their potential, while low-performance batteries are prone to accelerated aging due to overcharging and over-discharging" by assigning differentiated charging current and discharge weights to each individual cell and considering the impact of temperature on charging current. This reduces energy consumption. It ensures power quality through reliability rewards during mains power outages; isolates abnormal cells and implements penalties and safety limits to ensure voltage remains within acceptable ranges during mains power outages, addressing the risk of communication interruptions due to individual cell failures during mains power interruptions. This improves power supply success rate and reliability during mains power outages, reduces communication interruption risks, and significantly enhances the safety of communication power supply batteries. It also addresses the technical problem of "existing management methods struggling to accurately identify individual cell differences and dynamically adjust strategies" by rapidly matching or generating optimal strategies based on the current state and dynamically adjusting strategies according to performance evaluation and environmental changes, with feedback guiding subsequent optimization. Automated intelligent control reduces manual intervention and lowers maintenance costs. Finally, it addresses the technical problem of "existing technologies shortening the overall battery pack lifespan" by guiding voltage and temperature control optimization through lifespan rewards to slow aging and encouraging periodic deep discharge to delay sulfation through maintenance rewards, thus improving battery lifespan. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the individualized intelligent charging and discharging optimization method for communication power batteries provided in accordance with an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0026] like Figure 1 As shown, Embodiment 1 of the present invention provides an individualized intelligent charging and discharging optimization method for communication power supply batteries, comprising: Step 1: Collect data from all individual battery cells.

[0027] In a preferred but non-limiting embodiment of the present invention, step 1 includes: Step 1.1.1: Record the initial capacity, internal resistance, rated voltage and other baseline values ​​of all battery cells, establish individual files, and configure the sampling frequency (default 10 seconds / time), communication interface parameters (baud rate 9600bps) and safety thresholds (such as overcharge voltage 2.45V, over-discharge voltage 1.70V) of the monitoring module.

[0028] Step 1.1.2: Collect all individual battery cell parameters at the set frequency, and upload them to the system database after preprocessing; perform integrity verification on all individual battery cell parameters, and trigger a hardware fault alarm if a certain battery cell fails to collect parameters three times in a row.

[0029] The parameters of each battery cell include, but are not limited to, the terminal voltage (range 0~16V, accuracy ±0.02V), charging and discharging current (range -100A~+100A, accuracy ±0.1A), internal resistance (range 0~100mΩ, resolution 0.01mΩ), and surface temperature (range -30℃~85℃, accuracy ±0.5℃). The timestamps of the parameter acquisition are also recorded.

[0030] The system adopts a distributed acquisition method, with each individual battery having its own independent acquisition interface and equipped with a high-precision sensing chip (such as the ADS1115 for voltage acquisition and a dedicated AC injection chip for internal resistance testing). Data is transmitted through an isolated RS485 bus to avoid electromagnetic interference.

[0031] Step 2: Determine the state of charge, category, and abnormal indicators of each battery cell based on all individual battery cell data.

[0032] In a preferred but non-limiting embodiment of the present invention, step 2 includes: Step 2.1: Based on the filtered current and open-circuit voltage in the battery cell data, determine the state of charge of the battery cell using the ampere-hour integration method; based on the actual usable capacity and filtered current in the battery cell data, determine the health status of the battery cell.

[0033] More preferably, step 2.1 includes: Step 2.1.1: Based on the filtered current and open-circuit voltage in the individual battery cell data, determine the state of charge of the individual battery cell using the ampere-hour integration method, expressed by the following formula:

[0034] In the formula, This indicates the state of charge of battery cell j at time t. Indicates the control cycle. This represents the value of the abnormal current removed from battery cell j at time t. This indicates the rated capacity of battery cell j. The coulombic efficiency of a single battery cell j at time t is expressed by the following formula:

[0035] In the formula, This represents the baseline coulombic efficiency, used to describe the efficiency of a typical lithium-ion battery under ideal conditions. The coefficient of the quadratic term representing the effect of current on efficiency. This represents the magnification parameter. The coefficient of the first-order term representing the effect of temperature on efficiency. This represents the temperature value of battery cell j at time t. Indicates the standard test temperature.

[0036] Step 2.1.2: Based on the actual usable capacity and filtered current in the individual battery cell data, determine the health status of the individual battery cell, expressed by the following formula:

[0037] In the formula, This indicates the health status of battery cell j at time t. This indicates the weight of the capacity component in SOH. This indicates the weight of the internal resistance component in SOH. This represents the internal resistance penalty coefficient. The actual capacity of monomer j at time t is expressed by the following formula:

[0038] In the formula, This represents the Kalman gain of battery cell j at time t. The rate of increase in internal resistance of a single battery cell j over time t is expressed by the following formula:

[0039] In the formula, This represents the measured internal resistance of battery cell j at time t. This represents the initial internal resistance of battery cell j.

[0040] Step 2.2: Determine the category of the battery cell and the abnormal indicators of the battery cell based on the health status of the battery cell.

[0041] More preferably, step 2.2 includes: The classification of battery cells is determined based on their health status, including: when , and Simultaneously satisfying the condition that battery cell j is of category at time t. It was classified as a Class 1 battery; when , and If both conditions are met, the battery cell j at time t is classified as a Class 2 battery; Apart from Class 1 and Class 2 batteries, the remaining batteries were classified as Class 3 batteries.

[0042] in, This represents the SOH threshold for Class 1 batteries, with a value of [value missing]. , This represents the threshold value for the internal resistance ratio of Class 1 batteries, with a value of 1.1. This represents the temperature coefficient threshold for Class 1 batteries. The temperature coefficient of battery cell j is expressed by the following formula:

[0043] In the formula, This indicates the number of sampling points within a set time period, where the set time period is 1 hour. Indicates that the battery cell j is in time The filtered voltage, This represents the average voltage of battery cell j over a set period of time. This indicates the average temperature of battery cell j over a set period of time.

[0044] This indicates the minimum SOH value for the second type of battery. This indicates the maximum SOH value for the second type of battery. This indicates the minimum internal resistance ratio of the second type of battery. This indicates the maximum internal resistance ratio of the second type of battery. This represents the minimum temperature coefficient of the second type of battery. This indicates the maximum temperature coefficient of the second type of battery.

[0045] The abnormal indicators of a battery cell are determined based on its health status, including: when , or Determine if battery cell j is abnormal at time t, and identify the abnormality flag of battery cell j at time t. If the value is 1, then the battery cell j is considered normal at time t; otherwise, the abnormal flag for battery cell j at time t is determined. It is 0.

[0046] Step 3: Construct individual cell state features by analyzing the state of charge, category, and anomaly indicators of individual battery cells. Based on these individual cell state features and a co-evolutionary reinforcement learning algorithm, determine the comprehensive reward function for each individual battery cell. Then, perform online policy optimization based on the reward function to construct a differentiated control strategy model for each individual battery cell.

[0047] Step 3.1: Determine the individual cell status characteristics based on the state of charge, category, and anomaly indicators of each battery cell. Individual cell status characteristics include: category, state of charge, health status, internal resistance growth rate, temperature coefficient, anomaly indicators, filtered voltage, initial temperature, DC bus voltage, mains power status, communication equipment power requirements, and load current for all battery cells.

[0048] Step 3.2: Construct multiple parallel battery cell co-evolutionary policy networks based on the co-evolutionary reinforcement learning algorithm. Input the individual cell state features into the multiple parallel battery cell co-evolutionary policy networks and output the original actions. The original actions include the original charging current coefficient and the original discharging weight.

[0049] More preferably, step 3.2 includes: The individual cell state characteristics are input into a network of multiple parallel battery cell co-evolutionary strategies, and the original actions are output, expressed by the following formula:

[0050] In the formula, This indicates that policy network i represents the original action generated by individual entity j. , This represents the original charging current coefficient. Indicates the original discharge weight. The parameter is The policy network, The parameters represent the policy network i. This represents the state vector of battery cell j at time t. Indicates the number of policy networks.

[0051] Step 3.3: Determine the penalty term for each battery cell based on its initial action, and construct a comprehensive reward function for each battery cell based on the penalty term, expressed by the following formula:

[0052] In the formula, Let policy network i represent the total reward obtained by battery cell j at time t. , , , and These represent the weights for lifespan bonus, efficiency bonus, reliability bonus, maintenance bonus, and penalty, respectively. The policy network i represents the lifetime reward obtained by battery cell j at time t, expressed by the following formula:

[0053] In the formula, This represents the voltage deviation penalty coefficient. Represents ideal voltage. This indicates the penalty coefficient for temperature deviation. Indicates the ideal temperature. This indicates that the SOH deviates from the penalty coefficient. This represents the rate of change of SOH in battery cell j. This indicates taking the maximum value.

[0054] The policy network i represents the efficiency reward obtained by battery cell j at time t, expressed by the following formula:

[0055] In the formula, Indicates reference efficiency. This represents the coulombic efficiency of battery cell j at time t.

[0056] The policy network i represents the reliability reward obtained by battery cell j at time t, expressed by the following formula:

[0057] In the formula, Indicates the status of mains power. Indicates the conditions for a mains power outage. Indicates the DC bus voltage. and These represent the minimum allowable bus voltage and the maximum allowable bus voltage, respectively. This indicates a reward for reliable power supply during a power outage. This indicates a penalty for unreliable power supply during a power outage. Indicates the task completion status, used to describe the completion of the charging and discharging task. This indicates that the task is complete. This indicates the reward for completing the task. This indicates that the task has failed. This indicates a penalty for failing the task.

[0058] The policy network i represents the maintenance reward that battery cell j receives at time t, expressed by the following formula:

[0059] In the formula, This indicates the best maintenance SOC reward. Indicates the maintenance cycle indicator. A value of 1 is used to describe a period of more than 30 days since the last deep discharge. and These represent the minimum and maximum values ​​for maintaining the optimal SOC, respectively. and These represent the minimum and maximum values ​​for maintaining an acceptable State of Charge (SOC), respectively. This indicates that the SOC maintenance reward is acceptable.

[0060] The policy network i represents the penalty term obtained by battery cell j at time t, expressed by the following formula:

[0061] In the formula, This indicates an indicator function; it returns 1 if the condition is true and 0 otherwise. This indicates the abnormal condition of battery cell j at time t. This represents the penalty coefficient for abnormal conditions. This indicates the threshold for low SOC conditions. When it is determined to be low SOC, This indicates a low SOC penalty coefficient. This indicates the threshold for high SOC conditions. It is judged as high SOC at that time. This indicates a high SOC penalty coefficient.

[0062] Step 3.4: Determine the fitness of the strategy network based on the comprehensive reward function of the individual battery cells; retain the optimal strategy network of each generation based on the fitness, and merge it with the strategy network after crossover and mutation to obtain the strategy network population set.

[0063] More preferably, step 3.4 includes: Step 3.4.1: Determine the fitness of the policy network based on the comprehensive reward function of each battery cell, expressed by the following formula:

[0064] In the formula, Let represent the fitness of policy network i in the g-th generation. Indicates the discount factor. Indicates the number of evaluation steps. Indicates the time step in the evaluation. This indicates the total number of individual battery cells.

[0065] The optimal policy network for each generation is retained based on fitness, as expressed by the following formula:

[0066] In the formula, This represents the optimal policy network in the g-th generation. This indicates that the top K best strategies are selected, where K represents the number of best strategies.

[0067] Step 3.4.2: Perform policy cross-validation on the policy networks of the parent and child generations to obtain the cross-generated policy network; Step 3.4.3: Mutate each generation of the policy network to obtain the mutated policy network; Step 3.4.4: Merge the optimal policy network of each generation, the policy network generated by crossover, and the policy network generated by mutation to obtain a policy population set.

[0068] Step 3.5: Perform PPO update for each policy network in the policy network population set to optimize the policy online until the set number of training iterations is reached, and obtain the trained policy network and the battery cell differentiated control policy model; construct a policy knowledge base for all trained policy networks, which includes the fitness, class distribution, number of batteries in each class, average SOC, average SOH and average internal resistance ratio of each policy network.

[0069] Step 4: Input the real-time data of each battery cell into the battery cell differentiated control strategy model, and output the optimal charging and discharging strategy for each battery cell to implement individual differentiated intelligent charging and discharging optimization of the communication power supply battery.

[0070] In a preferred but non-limiting embodiment of the present invention, step 4 includes: Step 4.1: Extract real-time data of each battery cell, and perform similarity matching between the real-time data and the strategy knowledge base to obtain the similarity between the real-time data of each battery cell and each strategy in the knowledge base.

[0071] More preferably, step 4.1 includes: Step 4.1.1: Extract real-time data for each battery cell, including: the real-time health status distribution of each battery cell (real-time number of Class 1 batteries, real-time number of Class 2 batteries, and real-time number of Class 3 batteries), the average SOC, real-time average SOH, and real-time average internal resistance ratio of each battery cell.

[0072] Step 4.1.2: Perform similarity matching between the real-time data and the strategy knowledge base in the battery cell differentiated control strategy model, expressed by the following formula:

[0073] In the formula, This represents the similarity between the real-time data of each battery cell and the k-th policy network in the policy knowledge base. This indicates the real-time health status distribution of each individual battery cell. This represents the health status distribution of the k-th policy network in the policy knowledge base. This represents the real-time average SOC of each individual battery cell. The average SOC of the k-th policy network in the policy knowledge base is represented. This represents the real-time average SOH of each battery cell. The average state of interest (SOH) of the k-th policy network in the policy knowledge base is represented. This represents the real-time average internal resistance ratio of each individual battery cell. This represents the average internal resistance ratio of the k-th policy network in the policy knowledge base. Indicates the bandwidth of category distribution similarity. Indicates the bandwidth of SOC similarity. Indicates the SOH similarity bandwidth. This indicates the bandwidth of the internal resistance ratio similarity. This represents an exponential function.

[0074] Step 4.2: If the maximum similarity between the real-time data of each battery cell and all policy networks in the policy knowledge base is greater than the policy selection similarity threshold, a policy network is selected from the policy knowledge base based on the similarity to obtain the best matching policy in the policy knowledge base, which is then set as the optimal charging and discharging policy for each battery cell. Otherwise, the optimal strategy obtained from the training of the real-time data input battery cell differentiated control strategy model is set as the optimal charging and discharging strategy for each battery cell, thereby realizing individual differentiated intelligent charging and discharging optimization of the communication power supply battery, as expressed by the following formula:

[0075] In the formula, This indicates the optimal charging and discharging strategy for each individual battery cell. This represents the maximum similarity between the real-time data of each battery cell and all policy networks in the policy knowledge base. This indicates the similarity threshold for strategy selection. This represents the best matching strategy in the strategy knowledge base. This represents the optimal policy obtained through training.

[0076] Embodiment 2 of the present invention provides an individualized intelligent charge-discharge optimization system for communication power batteries, which runs the individualized intelligent charge-discharge optimization method for communication power batteries described in Embodiment 1, including: The data acquisition module is used to acquire the state of charge, category, and abnormal indicators of individual battery cells to construct individual cell state characteristics; The model building module is used to determine the comprehensive reward function of a battery cell based on its individual state characteristics and using a co-evolutionary reinforcement learning algorithm; and to perform online policy optimization based on the comprehensive reward function to build a differentiated control strategy model for the battery cell. The strategy generation module is used to input the real-time data of each battery cell into the battery cell differentiated control strategy model, and output the optimal charging and discharging strategy for each battery cell to implement individual differentiated intelligent charging and discharging optimization of communication power batteries.

[0077] Embodiment 3 of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the individualized intelligent charging and discharging optimization method for the communication power supply battery according to Embodiment 1.

[0078] Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the individualized intelligent charging and discharging optimization method for communication power supply batteries according to Embodiment 1.

[0079] Compared with the prior art, the beneficial effects of the present invention include at least the following: This invention comprehensively acquires the real-time status of each individual cell through multi-dimensional data acquisition. The internal resistance directly reflects the aging differences of the cells, and the temperature identifies the differences in the thermal characteristics of the cells. It extracts the differentiated performance parameters of different individual batteries, solving the problem of "difficulty in accurately identifying individual cell differences" in the existing technology and improving the accuracy of individual cell status perception of communication power supply batteries. This invention addresses the technical problem of existing technologies failing to incorporate the dynamic changes in the state of health (SOH) and state of charge (SOC) of individual battery cells by employing an improved ampere-hour integral method, considering temperature and current compensation for coulombic efficiency, and determining SOH through capacity and internal resistance. It provides a quantitative basis for differentiated control by calculating the SOC and SOH of each cell in real time, dynamically tracking changes in cell state, and supporting adaptive strategy adjustments. Based on a three-level classification of SOH, internal resistance growth rate, and temperature coefficient, it detects anomalies through SOH mutations, internal resistance mutations, and temperature mutations, resolving the issues of weaker cells being prone to accelerated aging due to overcharging and over-discharging, and cell failure leading to… The technical problem of "communication interruption risk" is addressed by identifying weak-performing cells through SOH assessment, detecting potential faulty cells early through anomaly detection, and isolating high-risk cells through a three-level classification mechanism. This provides a basis for safety protection decisions, improves the accuracy of individual cell state estimation, classification accuracy, and anomaly detection sensitivity of communication power supply batteries, quantifies the aging degree of each cell to improve the cell health status assessment capability, reduces the risk of overcharging and over-discharging through accurate SOC estimation and SOH assessment, and supports optimized charging and discharging strategies to extend the life of communication power supply batteries through precise state assessment. Early anomaly detection and classification isolation reduce the risk of communication interruption caused by cell failure leading to overall battery failure. This invention employs a co-evolutionary reinforcement learning framework, utilizing parallel evolution of the policy network to maintain policy diversity. A closed-loop optimization mechanism in reinforcement learning continuously improves the policy based on reward feedback. Online PPO optimization adapts to environmental changes in real time, solving the technical problem of "single adjustment logic and lack of a closed-loop optimization mechanism." Deep reinforcement learning provides highly intelligent decision-making capabilities, the co-evolutionary mechanism enhances algorithm adaptability and robustness, and the policy knowledge base enables experience accumulation and rapid transfer, addressing the technical problem of "limited adaptability and intelligence level of existing technologies." This improves the intelligence level of charging and discharging strategies and enhances the reliability of communication power batteries. Differentiated state vectors provide customized decisions for different performance units, and the reward function encourages optimization of overall performance rather than simple averaging. The policy network learns how to allocate resources according to the characteristics of different units, solving the technical problem of "high-performance batteries not being able to fully realize their potential" in existing technologies. Intelligent management reduces the need for manual inspection and maintenance, lowering the operation and maintenance costs of communication power batteries. The policy knowledge base supports rapid switching to historically excellent strategies, improving the emergency response capability of communication power batteries. A balance is achieved among multiple objectives such as lifespan, efficiency, and reliability, enhancing the overall performance optimization capability of communication power batteries. This invention addresses the technical problem that "in existing technologies, under a uniform charge / discharge mode, high-performance batteries cannot fully realize their potential, while low-performance batteries are prone to accelerated aging due to overcharging and over-discharging" by assigning differentiated charging current and discharge weights to each individual cell and considering the impact of temperature on charging current. This reduces energy consumption. It ensures power quality through reliability rewards during mains power outages; isolates abnormal cells and implements penalties and safety limits to ensure voltage remains within acceptable ranges during mains power outages, addressing the risk of communication interruptions due to individual cell failures during mains power interruptions. This improves power supply success rate and reliability during mains power outages, reduces communication interruption risks, and significantly enhances the safety of communication power supply batteries. It also addresses the technical problem of "existing management methods struggling to accurately identify individual cell differences and dynamically adjust strategies" by rapidly matching or generating optimal strategies based on the current state and dynamically adjusting strategies according to performance evaluation and environmental changes, with feedback guiding subsequent optimization. Automated intelligent control reduces manual intervention and lowers maintenance costs. Finally, it addresses the technical problem of "existing technologies shortening the overall battery pack lifespan" by guiding voltage and temperature control optimization through lifespan rewards to slow aging and encouraging periodic deep discharge to delay sulfation through maintenance rewards, thus improving battery lifespan.

[0080] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for individualized intelligent charging and discharging optimization of communication power supply batteries, characterized in that: Obtain the state of charge, category, and abnormal indicators of individual battery cells to construct individual cell state characteristics; Based on the individual state characteristics, a co-evolutionary reinforcement learning algorithm is used to determine the comprehensive reward function of a battery cell. Online strategy optimization based on a comprehensive reward function is used to construct a differentiated control strategy model for individual battery cells, including: Based on the co-evolutionary reinforcement learning algorithm, a network of multiple parallel battery cell co-evolutionary strategies is constructed. The state features of the individual cells are input into the network of multiple parallel battery cell co-evolutionary strategies, and the original actions of the battery cells are output. The original actions include the original charging current coefficient and the original discharging weight. The penalty term for each battery cell is determined based on its original action, and a comprehensive reward function for each battery cell is constructed based on the penalty term. The fitness of the strategy network is determined based on the comprehensive reward function of the individual battery cells. The optimal strategy network of each generation is retained based on the fitness and then merged with the strategy networks generated by crossover and mutation to obtain a set of strategy network populations. PPO updates are performed on each policy network in the policy network population set to optimize the policy online until a set number of training iterations are reached, resulting in a trained policy network, which serves as the differentiated control policy model for individual battery cells. Construct a policy knowledge base for all trained policy networks. The policy knowledge base includes the fitness, class distribution, number of batteries in each class, average SOC, average SOH, and average internal resistance ratio of each policy network. Real-time data from each battery cell is input into the battery cell-specific control strategy model, which outputs the optimal charging and discharging strategy for each battery cell to implement individualized intelligent charging and discharging optimization for communication power batteries.

2. The individualized intelligent charging and discharging optimization method for communication power supply batteries according to claim 1, characterized in that: Obtaining the state of charge, category, and abnormal indicators of individual battery cells includes: Collect data from all individual battery cells; The state of charge of a battery cell is determined by the ampere-hour integration method based on the filtered current in the individual battery cell data. Based on the actual available capacity and filtered current in the individual battery cell data, the health status of the individual battery cell is determined; based on the health status of the individual battery cell, the category of the individual battery cell is determined and the abnormal indicators of the individual battery cell are identified.

3. The individualized intelligent charging and discharging optimization method for communication power supply batteries according to claim 2, characterized in that: The health status of a single battery cell is determined by the following formula: In the formula, This indicates the health status of battery cell j at time t. This indicates the weight of the capacity component in SOH. This represents the actual capacity of monomer j at time t. This indicates the rated capacity of battery cell j. This indicates the weight of the internal resistance component in SOH. This represents the internal resistance penalty coefficient. This represents the rate of increase in internal resistance of a single battery cell j over time t.

4. The individualized intelligent charging and discharging optimization method for communication power supply batteries according to claim 3, characterized in that: Determining the category of a single battery cell includes: when and as well as If all conditions are met, the battery cell j at time t is classified as a Class 1 battery; when and as well as If all conditions are met, the battery cell j at time t is classified as a Class 2 battery. Except for Class 1 and Class 2 batteries, the remaining batteries are classified as Class 3 batteries; in, This represents the SOH threshold for Class 1 batteries. This represents the threshold value for the internal resistance ratio of Class 1 batteries. This represents the temperature coefficient of battery cell j. This represents the temperature coefficient threshold for Class 1 batteries. This indicates the minimum SOH value for the second type of battery. This indicates the maximum SOH value for the second type of battery. This indicates the minimum internal resistance ratio of the second type of battery. This indicates the maximum internal resistance ratio of the second type of battery. This represents the minimum temperature coefficient of the second type of battery. This indicates the maximum temperature coefficient of the second type of battery.

5. The individualized intelligent charging and discharging optimization method for communication power supply batteries according to claim 4, characterized in that: The abnormal indicators for identifying individual battery cells include: when or or Determine if battery cell j is abnormal at time t, and identify the abnormality flag of battery cell j at time t. If the value is 1, then the battery cell j is considered normal at time t; otherwise, the abnormal flag for battery cell j at time t is determined. The value is 0, where h is the hour and min is the minute.

6. The individualized intelligent charging and discharging optimization method for communication power supply batteries according to claim 1, characterized in that: The penalty for a single battery cell is expressed by the following formula: In the formula, The policy network i represents the penalty term obtained by battery cell j at time t. This indicates an indicator function; it returns 1 if the condition is true and 0 otherwise. This indicates an anomaly in battery cell j at time t. This represents the penalty coefficient for abnormal conditions. This indicates the threshold for low SOC conditions. It is judged as low SOC at that time. This indicates a low SOC penalty coefficient. This indicates the threshold for high SOC conditions. It is judged as high SOC at that time. This indicates a high SOC penalty coefficient. This represents the original charging current coefficient. This represents the original discharge weight.

7. The individualized intelligent charging and discharging optimization method for communication power supply batteries according to claim 1, characterized in that: The optimal charging and discharging strategies for each individual battery cell include: Extract real-time data from each battery cell, perform similarity matching between the real-time data and the strategy knowledge base, and obtain the similarity between the real-time data of each battery cell and each strategy in the knowledge base. If the maximum similarity between the real-time data of each battery cell and all policy networks in the policy knowledge base is greater than the policy selection similarity threshold, a policy network is selected from the policy knowledge base based on the similarity to obtain the best matching policy in the policy knowledge base, which is then set as the optimal charging and discharging policy for each battery cell. Otherwise, the optimal strategy obtained from the training of the battery cell differentiated control strategy model is input into the real-time data and set as the optimal charging and discharging strategy for each battery cell.

8. The individualized intelligent charging and discharging optimization method for communication power supply batteries according to claim 7, characterized in that: Similarity matching is expressed by the following formula: In the formula, This represents the similarity between the real-time data of each battery cell and the k-th policy network in the policy knowledge base. This indicates the real-time health status distribution of each individual battery cell. This represents the health status distribution of the k-th policy network in the policy knowledge base. This represents the real-time average SOC of each individual battery cell. The average SOC of the k-th policy network in the policy knowledge base is represented. This represents the real-time average SOH of each individual battery cell. The average state of interest (SOH) of the k-th policy network in the policy knowledge base is represented. This represents the real-time average internal resistance ratio of each individual battery cell. This represents the average internal resistance ratio of the k-th policy network in the policy knowledge base. Indicates the bandwidth of category distribution similarity. Indicates the bandwidth of SOC similarity. Indicates the SOH similarity bandwidth. This indicates the similarity bandwidth between the internal resistance ratio and the bandwidth. This represents an exponential function.

9. A personalized intelligent charge-discharge optimization system for communication power batteries, comprising the personalized intelligent charge-discharge optimization method for communication power batteries as described in any one of claims 1-8, characterized in that: The data acquisition module is used to acquire the state of charge, category, and abnormal indicators of individual battery cells to construct individual cell state characteristics; The model building module is used to determine the comprehensive reward function of a battery cell based on its individual state characteristics and using a co-evolutionary reinforcement learning algorithm; and to perform online policy optimization based on the comprehensive reward function to build a differentiated control strategy model for the battery cell. The strategy generation module is used to input the real-time data of each battery cell into the battery cell differentiated control strategy model, and output the optimal charging and discharging strategy for each battery cell to implement individual differentiated intelligent charging and discharging optimization of communication power batteries.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is loaded into the processor, it implements the individualized intelligent charging and discharging optimization method for the communication power supply battery as described in any one of claims 1 to 8.

11. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the individualized intelligent charging and discharging optimization method for the communication power supply battery as described in any one of claims 1 to 8.

Images