Electric forklift battery pack management method based on genetic algorithm

Abstract

The invention relates to the technical field of electric forklift battery pack management, and discloses an electric forklift battery pack management method based on a genetic algorithm, which comprises the following steps of: firstly, acquiring single voltage, current, temperature, SOC (State of Charge) and working load data of an electric forklift battery pack through a sensor network; judging whether the voltage or the temperature of the single body is out of limit or not, if the voltage or the temperature of the single body is out of limit, immediately cutting off a charging and discharging loop through a hardware circuit, otherwise, judging whether a trigger event is generated or not, if the trigger event occurs, optimizing parameters by utilizing a genetic algorithm, and if the trigger event does not occur, stopping the charging and discharging loop; according to the method, when a specific trigger event occurs, the management strategy is switched from fuzzy logic dynamic adjustment to genetic algorithm optimization, so that the high efficiency of daily operation is ensured, the re-optimization of global parameters is realized at key opportunities, and the management strategy is more intelligent and efficient.

Description

A Genetic Algorithm-Based Battery Pack Management Method for Electric Forklifts Technical Field

[0001] This invention relates to the field of electric forklift battery pack management technology, specifically to an electric forklift battery pack management method based on a genetic algorithm. Background Technology

[0002] As a key piece of equipment in modern logistics and manufacturing, the performance, lifespan, and safety of the lithium-ion battery pack—the power source of electric forklifts—directly determine the equipment's operating costs and reliability. A battery pack consists of dozens to hundreds of individual cells connected in series. Due to differences in manufacturing processes, operating environments, and charge / discharge cycles, inconsistencies in voltage, capacity, and state of charge (SOC) are inevitable among individual cells—a phenomenon known as "inconsistency." If this inconsistency is not effectively managed, it can lead to overcharging or over-discharging of some cells, accelerating the overall capacity decay of the battery pack and potentially causing serious safety incidents such as thermal runaway.

[0003] Traditional battery pack management methods primarily rely on monitoring and protection based on fixed thresholds. For example, when any cell voltage or temperature exceeds a preset absolute safety limit, the system cuts off the circuit. In terms of balancing management, passive balancing or active balancing strategies based on fixed voltage thresholds are commonly used. The limitations of these methods are twofold: First, their management parameters (such as balancing start / stop thresholds and maximum charge / discharge current) are static and preset, unable to adaptively adjust according to the battery pack's real-time health status, ambient temperature, and dynamically changing workloads (such as heavy load increases and no-load driving). For instance, in low-temperature environments, battery internal resistance increases, and using the normal-temperature charging current may induce lithium plating; under heavy-load conditions, a fixed discharge current limit may not meet peak power requirements or lead to excessive voltage drop. Second, existing optimization methods are relatively simplistic. Some studies attempt to apply fuzzy logic or genetic algorithms for parameter optimization, but these often only target a single objective (such as minimizing imbalance), and the optimization strategies are offline or periodic, failing to deeply couple with the real-time state changes of the battery system. While fuzzy logic can achieve a certain degree of self-adaptation, its rule base relies on expert experience and is difficult to self-evolve under complex multi-objective constraints.

[0004] Furthermore, advanced battery management strategies require the synergistic optimization of several conflicting objectives: balancing speed, mitigating battery aging, and system energy efficiency. Currently, there is a lack of a mechanism that can dynamically weigh the weights of these objectives. In addition, the management strategy itself should have the ability to "learn," extracting better control rules from long-term operational data, rather than always relying on the initial design.

[0005] Therefore, existing technologies have the following core problems: 1) rigid management strategies that cannot adapt to complex dynamic working conditions; 2) safety protection and performance optimization are separated and lack coordination; 3) insufficient multi-objective optimization capabilities and fixed objective weights; 4) lack of an integrated intelligent management framework that can integrate real-time adaptive control and event-triggered deep optimization and can evolve itself through machine learning.

[0006] The present invention aims to overcome the above-mentioned defects and proposes a hierarchical, adaptive, multi-objective collaborative optimization intelligent battery pack management method. Summary of the Invention

[0007] The purpose of this invention is to address the problems in existing technologies, such as rigid management strategies that cannot adapt to complex dynamic working conditions; the separation and lack of coordination between safety protection and performance optimization; insufficient multi-objective optimization capabilities and fixed objective weights; and the lack of an integrated intelligent management framework that can integrate real-time adaptive control and event-triggered deep optimization and can evolve through machine learning. Therefore, this invention proposes a method for managing electric forklift battery packs based on genetic algorithms.

[0008] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A method for managing electric forklift battery packs based on genetic algorithms, comprising the following steps: S1, firstly, collecting data on the individual cell voltage, current, temperature, SOC, and workload of the electric forklift battery pack through a sensor network; S2, determining whether the individual cell voltage or temperature exceeds the limit. If the individual cell voltage or temperature exceeds the limit, the charging and discharging circuit is immediately cut off through hardware circuitry. If neither the individual cell voltage nor the temperature exceeds the limit, then determining whether a trigger event has occurred. If a trigger event occurs, the battery balancing threshold is optimized using a genetic algorithm. Equalization time interval Maximum discharge current and maximum charging current If no triggering event occurs, the battery balancing threshold is dynamically adjusted based on the fuzzy logic controller. Equalization time interval Maximum discharge current and maximum charging current .

[0009] Based on the above technical solution, the present invention can be further improved as follows.

[0010] Preferably, the triggering event includes the maximum SOC difference between individual battery cells. Exceeding the dynamic threshold : ;in, The base imbalance threshold is a parameter that is dynamically or statically configured based on the battery pack's state of equilibrium (SOH) and / or historical workload patterns. , For ambient temperature, This is the temperature sensitivity coefficient.

[0011] Preferably, the genetic algorithm performs the following operations: (a) initializing the population, wherein the population adopts a two-layer chromosome structure, the upper chromosome encodes long-term strategy parameters, and the lower chromosome encodes short-term strategy rules; (b) evaluating individuals based on a multi-objective fitness function, wherein the fitness function is: ;in, This represents the individual fitness value. , , , These are dynamic weights for imbalance optimization, aging suppression, and energy efficiency, respectively. This represents the maximum SOC difference between individual cells. For aging rate, For the maximum permissible aging rate, For energy efficiency, For maximum theoretical energy efficiency, (c) Perform adaptive evolutionary operations, using crowding distance to preserve Pareto front solutions, arithmetic crossover for upper-level chromosomes, regular exchange for lower-level chromosomes, and adaptive Gaussian mutation for mutation operations, where the mutation rate is dynamically adjusted using the following formula: ;in, The initial mutation rate, (d) Update the optimal policy obtained through evolution to the fuzzy rule base.

[0012] Preferably, the imbalance optimization dynamic weights, aging suppression dynamic weights, and energy efficiency dynamic weights of the genetic algorithm are dynamically adjusted through deep reinforcement learning, as shown in the following formula: ; ;in, For weight adjustment offset, For multi-target phase angle.

[0013] Preferably, the The formula for calculating the weight adjustment offset is as follows: ;in, The attenuation coefficient is and satisfies , This is the temperature-load coupling factor. This is the historical performance degradation factor.

[0014] Preferably, the deep reinforcement learning employs the PPO algorithm, whose reward function includes a Pareto dominance penalty term: ;in, The total reward value, For lifetime gain weighting, For lifespan increment, For efficiency gain weights, For efficiency gain, For safety violation penalty coefficient, For the number of safety violations, Let KL divergence be the distribution of the old and new strategies. The policy smoothing coefficient is and satisfies .

[0015] Preferably, the short-term strategy rule of the lower chromosome includes: when the load type is "heavy load", the rule instruction dynamically adjusts the maximum discharge current using the following formula: ;in, The rated capacity of the battery pack. This is the temperature compensation coefficient. Real-time temperature of the battery pack. The reference temperature is used; when the SOC is between 20% and 40%, the rule command dynamically adjusts the maximum charging current using the following formula: ;in, for Compensation slope.

[0016] Preferably, the genetic algorithm further includes a cloud-edge co-evolution step: establishing a communication connection between the cloud-based battery management knowledge base and the edge forklift controller. The cloud-based battery management knowledge base aggregates the non-dominated solution sets of multiple electric forklifts within the same fleet. During population initialization in step (a), in addition to random initialization, historical global Pareto front solutions are downloaded from the cloud-based battery management knowledge base and injected into the local population as elite populations. The injection ratio is dynamically adjusted according to the local battery pack's state of health (SOH). in, Based on injection rate, To adjust the injection rate, This is the steepness coefficient. For reference to health status; after the local genetic algorithm completes the evolution, the new non-dominated solutions generated locally are uploaded to the cloud battery management knowledge base after being privately encrypted.

[0017] Preferably, after optimizing the battery balancing threshold, balancing time interval, maximum discharge current, and maximum charging current using a multi-objective genetic algorithm, and before executing the optimization strategy, a digital twin verification step is further included: constructing a battery digital twin model that maps to the physical battery pack in real time; inputting the optimal strategy parameters output by the multi-objective genetic algorithm into the battery digital twin model for pre-simulation; and calculating the state prediction error during the pre-simulation process. The calculation formula is as follows: ;in, and These represent the voltage of the i-th cell in the digital twin model and the physical battery pack, respectively. and These are the temperature predicted by the digital twin model and the measured temperature, respectively. This is the temperature error weighting coefficient. The number of monomers; if If the value is less than the set safety confidence threshold, the optimal strategy will be sent to the physical battery pack for execution; if... If the risk exceeds the set safety confidence threshold, the policy rollback mechanism is triggered, the control policy of the previous cycle is used and the sample is marked as a high-risk sample, and then fed back to the fitness function of the multi-objective genetic algorithm for punishment.

[0018] Compared with the prior art, the technical solution of this application has the following beneficial technical effects: 1. The present invention adds a layer of judgment based on trigger events on the basis of fuzzy logic control. When a specific trigger event occurs, the management strategy switches from dynamic adjustment of fuzzy logic to optimization of genetic algorithm, forming a two-layer intelligent decision-making structure that combines conventional adaptive adjustment and event-triggered global optimization. This not only ensures the efficiency of daily operation, but also realizes the re-optimization of global parameters at critical moments, making the management strategy more intelligent and efficient.

[0019] 2. This invention defines the threshold of "maximum SOC difference between individual cells" for triggering genetic algorithm optimization as a dynamic value related to ambient temperature. This dynamic threshold can adaptively adjust the sensitivity of trigger optimization according to ambient temperature, making the triggering mechanism more scientific and reasonable. It avoids unnecessary frequent optimization due to an overly strict fixed threshold under harsh conditions such as low temperature, or missing the best equilibrium opportunity due to an overly lenient threshold under high temperature.

[0020] 3. This invention sets up a multi-objective fitness function that simultaneously considers imbalance, aging suppression, energy efficiency, and safety penalty, and introduces dynamic weights. The dynamic weights can flexibly adjust the priority between different objectives, which helps to find Pareto optimal solutions that satisfy multiple constraints, resulting in better overall performance.

[0021] 4. This invention uses deep reinforcement learning to dynamically adjust the three dynamic weights of a multi-objective fitness function. Through continuous interaction with the battery management environment, deep reinforcement learning learns how to balance equilibrium, lifetime, and efficiency goals under different system states to obtain the maximum long-term cumulative reward. This is superior to manual setting or adjustments based on simple rules, and can automatically discover complex and non-intuitive weight adjustment strategies.

[0022] 5. This invention concretizes the "short-term strategy rules" encoded by the lower chromosome into formula-based dynamic current adjustment rules for two typical operating conditions: "heavy load" and "low SOC range". Under heavy load, the maximum discharge current is dynamically calculated through formula, taking temperature compensation into account. It can appropriately limit the current at high temperatures to protect the battery, which is more scientific than a simple fixed current limit value. When charging in the low SOC range, the maximum charging current is dynamically adjusted through formula, and a compensation slope is adopted to achieve a gentler charging strategy, which is beneficial to protecting the battery in an energy-deficient state and extending its cycle life. These formulaic rules make the "strategy" optimized by the genetic algorithm no longer an abstract parameter, but an immediately executable and precise control command, which greatly enhances the practicality of the method. Attached Figure Description

[0023] Figure 1 is a schematic diagram of the structure of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] A method for managing electric forklift battery packs based on genetic algorithms includes the following steps: S1. First, collect data on the individual cell voltage, current, temperature, SOC, and workload of the electric forklift battery pack through a sensor network; S2. Determine whether the individual cell voltage or temperature exceeds the limit. If either the individual cell voltage or temperature exceeds the limit, immediately cut off the charging and discharging circuit through hardware circuitry. If neither the individual cell voltage nor temperature exceeds the limit, then determine whether a trigger event has occurred. If a trigger event occurs, optimize the battery balancing threshold using a genetic algorithm. Equalization time interval Maximum discharge current and maximum charging current If no triggering event occurs, the battery balancing threshold is dynamically adjusted based on the fuzzy logic controller. Equalization time interval Maximum discharge current and maximum charging current .

[0026] The process of adjusting the dynamic control parameters of the battery pack based on fuzzy logic is executed by a fuzzy adaptive controller. First, the voltage difference, temperature difference, and load current between individual batteries are collected in real time as input variables. After being fuzzified by a triangular membership function, they are input into a preset fuzzy rule base for reasoning. This rule base is set according to the working conditions characteristics of the forklift. For example, when the voltage difference is large and the temperature difference is small, a smaller equalization threshold and a shorter equalization interval are output for aggressive equalization. When the temperature difference is large or the load current is high, a larger current limit value is output to ensure thermal safety and power demand. The reasoning result is defuzzified by the centroid method to obtain the initial fuzzy output values of the equalization threshold, equalization time interval, maximum discharge current, and maximum charge current. Subsequently, to ensure the physical safety boundary, the initial fuzzy output values of the maximum discharge current and maximum charge current are respectively compared with the physical formula limits calculated based on the battery temperature and SOC, and the minimum value is taken for safety clamping. Finally, the four calibrated control parameters are sent to the actuating mechanism to achieve real-time adaptive adjustment that takes into account equalization efficiency, thermal management, and power performance.

[0027] The triggering event includes the maximum SOC difference between individual batteries exceeding the dynamic threshold : ; Among them, is the basic unbalance threshold, which is a parameter dynamically or statically configured according to the battery pack SOH and / or historical workload patterns. , is the ambient temperature, is the temperature sensitivity coefficient.

[0028] Example 1 (standard working condition): For a new battery pack in a healthy state (SOH>90%) and working under standard cycle working conditions, the basic unbalance threshold is set, and the temperature sensitivity coefficient . At this time, the calculation formula for the dynamic threshold is: . This setting can avoid unnecessary frequent equalization actions while ensuring the equalization effect and optimize the system efficiency.

[0029] Example 2 (aging battery pack or mild working condition): For a battery pack with certain aging (80%<SOH≤90%) or working under light load and intermittent working conditions, the equalization trigger condition is moderately relaxed to reduce losses. Set , . The calculation formula is: . This helps to reduce the energy loss and component aging caused by excessive equalization and is more suitable for scenarios that are more sensitive to life attenuation.

[0030] Example 3 (High-load or extremely consistency-critical scenarios): For applications operating under heavy loads and continuous operation, or where battery pack consistency is highly sensitive, a more stringent balancing strategy is employed. (Settings...) , The calculation formula is: This setting triggers equalization earlier, keeping the differences between batteries within a smaller range, thereby improving high-current discharge capability and overall usable capacity, and ensuring performance and safety under high load.

[0031] The genetic algorithm performs the following operations: (a) initializes the population, which adopts a two-layer chromosome structure, where the upper chromosome encodes long-term strategy parameters and the lower chromosome encodes short-term strategy rules; (b) evaluates individuals based on a multi-objective fitness function, where the fitness function is: ;in, This represents the individual fitness value. , , , These are dynamic weights for imbalance optimization, aging suppression, and energy efficiency, respectively. This represents the maximum SOC difference between individual cells. For aging rate, For the maximum permissible aging rate, For energy efficiency, For maximum theoretical energy efficiency, (c) Perform adaptive evolutionary operations, using crowding distance to preserve Pareto front solutions, arithmetic crossover for upper-level chromosomes, regular exchange for lower-level chromosomes, and adaptive Gaussian mutation for mutation operations, where the mutation rate is dynamically adjusted using the following formula: ;in, The initial mutation rate, (d) Update the optimal policy obtained through evolution to the fuzzy rule base.

[0032] The imbalance optimization, aging suppression, and energy efficiency dynamic weights of the genetic algorithm are dynamically adjusted using deep reinforcement learning. The specific formulas are as follows: ; ;in, For weight adjustment offset, For multi-target phase angle.

[0033] The The formula for calculating the weight adjustment offset is as follows: ;in, The attenuation coefficient is and satisfies , This is the temperature-load coupling factor. Its physical meaning is that when the temperature deviates from the ideal range or the load increases, the battery faces increased thermal stress, aging rate, and imbalance risks. Therefore, a greater degree of dynamic adjustment of the weights is needed. This is the historical performance degradation factor, which physically means that if recent historical performance continues to deviate from the target (e.g., aging too quickly, efficiency too low), then... The factor value will decrease (less than 1), thereby suppressing the weight adjustment offset. This causes the system to shift towards a more conservative, smaller-amplitude strategy adjustment, avoiding "oscillations" in ineffective directions. Conversely, if recent performance is good, When the value approaches 1, the system maintains its original regulatory initiative.

[0034] The temperature-load coupling factor The calculation process is as follows: ;in, For the optimal operating temperature of the battery pack, such as 25°C, Real-time temperature of the battery pack. The temperature deviation penalty coefficient represents the degree of impact on the weight adjustment requirement when the temperature deviates from the optimal value (e.g., ), This is the load sensitivity coefficient. It represents the degree to which the load size affects the demand for weight adjustment (e.g., ), load nonlinearity index (usually taken as ), used to amplify the effects under heavy load conditions.

[0035] The historical performance degradation factor The calculation process is as follows: ;in, The historical decay rate constant ( ), control the rate at which poor historical performance undermines confidence in current adjustments (e.g., ), For the performance dimension weight vector, For the target performance vector, This is the historical performance vector for the k-th period. Let be the norm of the target performance vector. This refers to the number of historical evaluation periods. It is a natural exponential function.

[0036] The deep reinforcement learning method employs the PPO algorithm, whose reward function includes a Pareto dominance penalty term. ;in, The total reward value, For lifetime gain weighting, For lifespan increment, For efficiency gain weights, For efficiency gain, For safety violation penalty coefficient, For the number of safety violations, Let KL divergence be the distribution of the old and new strategies. The policy smoothing coefficient is and satisfies .

[0037] The short-term strategy rules for the lower-level chromosomes include: when the load type is "heavy load", the rule instructions dynamically adjust the maximum discharge current using the following formula: ;in, The rated capacity of the battery pack. This is the temperature compensation coefficient. Real-time temperature of the battery pack. Reference temperature ( The rated maximum discharge rate represents the upper limit of safe continuous discharge, reflecting the battery's design capacity. When the SOC is between 20% and 40%, the regulations dynamically adjust the maximum charging current using the following formula: ;in, for Compensation slope ( For low The base charging rate is the rate at which the battery charges. The safe starting charging current at that time corresponds to the lower limit of lithium plating criticality.

[0038] The LSTM neural network predicts the SOC, SOH, and battery temperature for the next 5-10 minutes. The uncertainty range of the SOC prediction output by the LSTM is given. Calculated using the following formula: ;in, The number of dropout samples and satisfying Where z is the confidence coefficient; an electrochemical-thermal coupling model of the battery pack is constructed using digital twin technology, and the electrochemical parameters of the digital twin model are calibrated online using the following formula: Where η is the learning rate and satisfies 0.01≤η≤0.05; during the calibration of the digital twin model, electrochemical parameters include internal resistance. And the diffusion coefficient D, and the gradient term is calculated by the following formula: ;in, To output the gradient terms of electrochemical parameters in the simulation, The simulated electrical output of the digital twin model, The standard step size for internal resistance. is the standard step size for the diffusion coefficient.

[0039] When a trigger event occurs, the GA strategy is quickly pre-screened for effectiveness using the following formula: ;in, The simulated voltage output by the digital twin model. The voltage safety threshold is defined as follows: the population size of the genetic algorithm is 20-40, the number of iterations is 5-20 generations, the single run time is <2 seconds, and the activation frequency of the L2 layer is ≤10% of the total run time.

[0040] The genetic algorithm also includes a cloud-edge co-evolution step: establishing a communication connection between the cloud-based battery management knowledge base and the edge forklift controller. The cloud-based battery management knowledge base aggregates the non-dominated solution sets of multiple electric forklifts in the same fleet. In step (a) when initializing the population, in addition to random initialization, historical global Pareto front solutions are downloaded from the cloud-based battery management knowledge base and injected into the local population as elite populations. The injection ratio is dynamically adjusted according to the health status (SOH) of the local battery pack. ;in, Based on injection rate, To adjust the injection rate, This is the steepness coefficient. To reference the health status, after the local genetic algorithm completes its evolution, the new non-dominated solutions generated locally are privately encrypted and then uploaded to the cloud-based battery management knowledge base for updating the global policy library, thereby enabling cross-device knowledge transfer and species diversity enhancement.

[0041] After optimizing the battery balancing threshold, balancing time interval, maximum discharge current, and maximum charging current using a multi-objective genetic algorithm, and before executing the optimization strategy, a digital twin verification step is included: constructing a battery digital twin model that maps to the physical battery pack in real time; inputting the optimal strategy parameters output by the multi-objective genetic algorithm into the battery digital twin model for pre-simulation; and calculating the state prediction error during the pre-simulation process. The calculation formula is as follows: ;in, and These represent the voltage of the i-th cell in the digital twin model and the physical battery pack, respectively. and These are the temperature predicted by the digital twin model and the measured temperature, respectively. This is the temperature error weighting coefficient. The number of monomers; if If the value is less than the set safety confidence threshold, the optimal strategy will be sent to the physical battery pack for execution; if... If the risk exceeds the set safety confidence threshold, the policy rollback mechanism is triggered, the control policy of the previous cycle is used and the sample is marked as a high-risk sample, and then fed back to the fitness function of the multi-objective genetic algorithm for punishment.

[0042] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0043] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for managing battery packs in electric forklifts based on genetic algorithms, characterized in that, The process includes the following steps: S1. First, collect data on the individual cell voltage, current, temperature, SOC, and workload of the electric forklift battery pack through a sensor network; S2. Determine if the individual cell voltage or temperature exceeds the limit. If either limit is exceeded, immediately disconnect the charging / discharging circuit using hardware. If neither limit is exceeded, determine if a trigger event has occurred. If a trigger event occurs, optimize the battery balancing threshold using a genetic algorithm. Equalization time interval Maximum discharge current and maximum charging current If no triggering event occurs, the battery balancing threshold is dynamically adjusted based on the fuzzy logic controller. Equalization time interval Maximum discharge current and maximum charging current 。 2. The method for managing electric forklift battery packs based on genetic algorithms according to claim 1, characterized in that, The triggering event includes the maximum SOC difference between individual battery cells. Exceeding the dynamic threshold ： ;in, The base imbalance threshold is a parameter that is dynamically or statically configured based on the battery pack's state of equilibrium (SOH) and / or historical workload patterns. ， For ambient temperature, This is the temperature sensitivity coefficient.

3. The method for managing electric forklift battery packs based on genetic algorithms according to claim 1, characterized in that, The genetic algorithm performs the following operations: (a) initializes the population, which adopts a two-layer chromosome structure, where the upper chromosome encodes long-term strategy parameters and the lower chromosome encodes short-term strategy rules; (b) evaluates individuals based on a multi-objective fitness function, where the fitness function is: ;in, This represents the individual fitness value. ， 、 、 These are dynamic weights for imbalance optimization, aging suppression, and energy efficiency, respectively. This represents the maximum SOC difference between individual cells. For aging rate, For the maximum permissible aging rate, For energy efficiency, For maximum theoretical energy efficiency, (c) Perform adaptive evolutionary operations, using crowding distance to preserve Pareto front solutions, arithmetic crossover for upper-level chromosomes, regular exchange for lower-level chromosomes, and adaptive Gaussian mutation for mutation operations, where the mutation rate is dynamically adjusted using the following formula: ;in, The initial mutation rate, (d) Update the optimal policy obtained through evolution to the fuzzy rule base.

4. The method for managing electric forklift battery packs based on genetic algorithms according to claim 3, characterized in that, The imbalance optimization, aging suppression, and energy efficiency dynamic weights of the genetic algorithm are dynamically adjusted using deep reinforcement learning. The specific formulas are as follows: ； ;in, For weight adjustment offset, For multi-target phase angle.

5. The method for managing electric forklift battery packs based on genetic algorithms according to claim 4, characterized in that, The The formula for calculating the weight adjustment offset is as follows: ;in, The attenuation coefficient is and satisfies ， This is the temperature-load coupling factor. This is the historical performance degradation factor.

6. The method for managing electric forklift battery packs based on genetic algorithms according to claim 4, characterized in that, The deep reinforcement learning method employs the PPO algorithm, whose reward function includes a Pareto dominance penalty term. ;in, The total reward value, For lifetime gain weighting, For lifespan increment, For efficiency gain weights, For efficiency gain, For safety violation penalty coefficient, For the number of safety violations, Let KL divergence be the distribution of the old and new strategies. The policy smoothing coefficient is and satisfies 。 7. The method for managing electric forklift battery packs based on genetic algorithms according to claim 3, characterized in that, The short-term strategy rules for the lower-level chromosomes include: when the load type is "heavy load", the rule instruction dynamically adjusts the maximum discharge current using the following formula: ;in, The rated capacity of the battery pack. This is the temperature compensation coefficient. Real-time temperature of the battery pack. The reference temperature is used; when the SOC is between 20% and 40%, the rule command dynamically adjusts the maximum charging current using the following formula: ;in, for Compensation slope.

8. The method for managing electric forklift battery packs based on genetic algorithms according to claim 3, characterized in that, The genetic algorithm also includes a cloud-edge co-evolution step: establishing a communication connection between the cloud-based battery management knowledge base and the edge forklift controller. The cloud-based battery management knowledge base aggregates the non-dominated solution sets of multiple electric forklifts in the same fleet. In step (a) when initializing the population, in addition to random initialization, historical global Pareto front solutions are downloaded from the cloud-based battery management knowledge base and injected into the local population as elite populations. The injection ratio is dynamically adjusted according to the health status (SOH) of the local battery pack. ;in, Based on injection rate, To adjust the injection rate, This is the steepness coefficient. For reference to health status; after the local genetic algorithm completes the evolution, the new non-dominated solutions generated locally are uploaded to the cloud battery management knowledge base after being privately encrypted.

9. The method for managing electric forklift battery packs based on genetic algorithms according to claim 3, characterized in that, After optimizing the battery balancing threshold, balancing time interval, maximum discharge current, and maximum charging current using a multi-objective genetic algorithm, and before executing the optimization strategy, a digital twin verification step is included: constructing a battery digital twin model that maps to the physical battery pack in real time; inputting the optimal strategy parameters output by the multi-objective genetic algorithm into the battery digital twin model for pre-simulation; and calculating the state prediction error during the pre-simulation process. The calculation formula is as follows: ;in, and These represent the voltage of the i-th cell in the digital twin model and the physical battery pack, respectively. and These are the temperature predicted by the digital twin model and the measured temperature, respectively. This is the temperature error weighting coefficient. The number of monomers; if If the value is less than the set safety confidence threshold, the optimal strategy will be sent to the physical battery pack for execution; if... If the risk exceeds the set safety confidence threshold, the policy rollback mechanism is triggered, the control policy of the previous cycle is used and the sample is marked as a high-risk sample, and then fed back to the fitness function of the multi-objective genetic algorithm for punishment.

Images