An intelligent fast-charging equalization regulation device and method for a self-adaptive load forklift charger

Abstract

The present application relates to the technical field of battery charging control, in particular to a kind of self-adapting load's intelligent fast charging equalization regulation and control device and method for forklift charger, including the acquisition battery pack each single voltage, temperature and factory power grid load prediction information, dynamic price signal and forklift chargeable window duration;Based on the reduced-order digital twin model of battery electro-thermal-aging coupling, the state of charge, health state and internal concentration distribution of each single body are deduced in real time;Using deep reinforcement learning online solution, the total power instruction of the current moment and the differential single equalization current distribution vector are generated.The deep integration of battery electro-thermal-aging digital twin model and deep reinforcement learning realizes the multi-objective dynamic collaborative optimization of charging speed, battery life and electricity cost;Through the collaborative control of multi-channel active equalization module and main charging current, the accurate allocation of each single differential charging current is realized.

Description

Technical Field

[0001] This invention relates to the field of battery charging control technology, specifically to an intelligent fast charging equalization control device and method for an adaptive load forklift charger. Background Technology

[0002] As core equipment in modern logistics systems, the charging management of the power lithium batteries in industrial electric forklifts directly impacts operational efficiency, operating costs, and battery lifespan. The following pain points exist in charging industrial electric forklifts: traditional fast charging uses a fixed charging curve, which cannot adapt to peak and off-peak electricity prices; inconsistencies among individual cells within the battery pack worsen with use, often leading to low charging efficiency and rapid battery lifespan degradation due to the "weakest link" effect; and the varying lengths of charging windows provided by forklift operation schedules necessitate a trade-off between time constraints and grid peak avoidance.

[0003] Traditional forklift charging solutions often employ fixed charging curves. For example, patent application CN109050327A discloses a charging control system and method for large-tonnage electric forklifts, using a distributed contactor layout and a self-designed charging controller to reduce current surges; patent application CN110224189B discloses an intelligent fast pulse charging method for electric forklift power supplies, reducing charging time from 12-16 hours to 6-7 hours. However, these existing solutions all use preset charging strategies, which cannot be flexibly adjusted according to grid load fluctuations and dynamic electricity price signals, resulting in high charging costs and difficulty in achieving a positive interaction with the grid.

[0004] The inconsistency between individual cells within a battery pack due to manufacturing tolerances and variations in operating conditions intensifies with each charge-discharge cycle. This "weakest link" effect limits the overall usable capacity of the pack to the weakest cell, resulting in low charging efficiency and rapid battery life degradation. Patent application CN120677605A discloses a battery balancing circuit, control method, and electronic device. However, most of these balancing schemes employ voltage comparison strategies, performing energy transfer only at the end of charging or discharging. They lack fine-grained control based on the battery's internal physical state, making it difficult to achieve dynamic, differentiated balancing control coordinated with the main charging current during fast charging. The contradiction between balancing efficiency and charging speed remains unresolved.

[0005] In recent years, digital twin technology and deep reinforcement learning have gradually gained attention in the field of battery management. Patent application CN111611750A discloses a lithium-ion battery charging and thermal management method using digital twin technology, which adapts to changes in battery state and operating environment by establishing a digital twin of the battery. Patent application CN121224500A discloses a charge-discharge control strategy based on power battery performance degradation, using an electrochemical-thermal-aging coupled model to train a deep reinforcement learning algorithm to obtain the optimal charging strategy for the battery. However, the above research mainly focuses on electric vehicles or energy storage scenarios and does not specifically address the scheduling constraints of forklifts in this particular application scenario. Furthermore, in existing solutions, the digital twin model and deep reinforcement learning decision-making are often loosely coupled or even operate independently, lacking a tightly coupled multi-objective adaptive optimization architecture that deeply integrates grid load demand, peak-valley electricity price signals, and the forklift's charging window.

[0006] In summary, existing technologies generally lack adaptive charging solutions that deeply integrate grid load, real-time battery health status, and forklift logistics scheduling information, making it difficult to achieve dynamic and optimal coordination among charging speed, battery life, and electricity costs. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, on the one hand, the purpose of this invention is to provide an intelligent fast charging equalization control device and method for forklift chargers with adaptive load, which achieves multi-objective dynamic optimization of charging speed, lifespan and cost through digital twin and deep reinforcement learning, and actively balances and adaptively reduces power with the power grid, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides an intelligent fast charging equalization control method for an adaptive load forklift charger, comprising the following steps: Collect data on the voltage and temperature of each cell in the battery pack, as well as factory grid load forecasts, dynamic electricity price signals, and forklift charging window duration; Based on a reduced-order digital twin model of battery electro-thermal-aging coupling, the state of charge of each cell is simulated in real time. Core temperature, health status and internal concentration distribution; With the goal of optimizing the charging speed, battery life degradation and electricity cost in a coordinated manner, a Markov decision process composed of the inferred state and external information is constructed. Deep reinforcement learning is used to solve the problem online to generate the total charging power command and differentiated single-cell equalization current allocation vector at the current moment. The bidirectional reconfigurable power conversion unit is controlled to perform pulse fast charging, and the multi-channel active balancing module is driven to perform energy transfer between any cells in a high-frequency soft-switching manner according to the equalization current distribution vector. When a grid load spike is detected, the total charging power is adaptively reduced, thereby achieving intelligent fast charging control that is adaptive to both the grid-side load and the battery-side health status.

[0009] As a further improvement to this technical solution, the reduced-order digital twin model of the battery electro-thermal-aging coupling adopts the following reduced-order expression: The electrical characteristics of a single unit are modeled using a second-order RC equivalent circuit, and its state differential equation is as follows: ; ; Single-unit terminal voltage ; in, , These are the polarization voltages of the two RC circuits, representing the transient voltage response caused by the electrochemical reactions and diffusion processes inside the battery; , These are electrochemically polarized resistance and capacitance, corresponding to fast dynamic processes with small time constants, namely charge transfer reactions; , For concentration polarization resistance and capacitance, corresponding to a slow dynamic process with a large time constant, namely solid-phase lithium-ion diffusion; This refers to the charging and discharging current flowing through the cell; a positive value indicates charging, and a negative value indicates discharging. This differential equation describes the dynamic evolution of the polarization voltage under current excitation, and is used for real-time estimation. The foundation; This refers to the directly measurable terminal voltage of a single unit; It is the open-circuit voltage. The nonlinear function, calibrated experimentally. Obtained by looking up the curve in a table; The internal resistance is ohmic, which characterizes the instantaneous voltage drop of components such as electrolyte, diaphragm, and current collector; The terminal voltage equals the open-circuit voltage minus the ohmic voltage drop and the two polarization voltages. This equation is used for comparison and correction between the model output and the measured voltage. The thermal model uses a lumped-parameter heat balance equation: ; in, Indicates the rate of change of monomer temperature. For monomer mass, This is the equivalent specific heat capacity; This is ohmic heat, an irreversible Joule heating caused by internal resistance; Polarization heat is the heat generated by electrochemical polarization and concentration polarization. For heat dissipation, The equivalent heat transfer coefficient, For effective heat dissipation area, Ambient temperature; This equation achieves a two-way coupling of electrical and thermal properties: temperature affects resistance and aging parameters, and conversely, current and polarization generate heat and raise the temperature. The aging model uses a cumulative capacity loss model: ; in, The cumulative irreversible capacity loss is a measure of battery health. Core metrics; This is a pre-exponential factor, a constant related to battery materials and design processes; The activation energy characterizes the sensitivity of the aging reaction to temperature. This is the universal gas constant; This refers to the absolute temperature of the battery cell. The cumulative ampere-hour throughput is the total amount of charge flowing through the battery during charging and discharging. The power-law exponent, typically between 0.5 and 0.6, reflects the kinetic characteristics of side reactions such as SEI film growth; The model shows that the higher the temperature and the greater the throughput, the faster the capacity decays, and the decay exhibits a non-linear accelerating trend. The above partial differential-ordinary differential coupled system is achieved through The above partial differential-ordinary differential coupled system is approximately simplified into a reduced-order state-space model that can be solved in real time by the edge controller.

[0010] As a further improvement to this technical solution, the deep reinforcement learning adopts a deep deterministic policy gradient algorithm with a safety layer, whose state space... Including each individual Vector, core temperature of each individual cell, individual cells Power grid load index Dynamic electricity price and remaining charging window duration Action space Including: Total charging current amplitude Pulse duty cycle ,frequency and the equalization current distribution of each individual cell The vector formed; the reward function is designed as follows: ; in, For the charging speed target item, The average state of charge at the end of charging. For the target state of charge, this penalty is applied for insufficient charging completion; the square form means the penalty increases the further the deviation from the target; weighting coefficients. The importance of adjusting charging speed, the more urgent the window. The larger; For battery life degradation target items, The capacity loss increment for this charging cycle is obtained from the aging model; this term directly penalizes charging behaviors that cause irreversible damage to battery health; weighting coefficients. Reflecting the level of importance placed on battery life, this setting can be configured by the user or based on... Dynamic adjustment; For electricity cost target item, This is the dynamic electricity price for the current period. To control the step size, This is the total voltage of the battery pack. This refers to the charging energy consumption within the current control step; this penalty for electricity costs drives the system to charge or reduce power to avoid peak hours during off-peak electricity periods; weighting coefficient. It can be adjusted according to the user's sensitivity to cost-effectiveness; For temperature safety targets, For the first The core temperature of each individual unit This represents the highest core temperature among all cells; this penalty penalizes excessively high cell temperatures to prevent thermal runaway and accelerated aging; weighting coefficient. It automatically increases in the high-temperature range to form a soft constraint protection; To balance the consistency objective term, each individual entity represents... The sum of the absolute values ​​of the deviations of C from the mean measures the degree of inconsistency. It is a single-cell charged state. The average state of charge; this penalty When the C-dispersion is too large, the active balancing module prioritizes energy transfer to reduce the disparity; weighting coefficients Determine the priority of consistency protection to avoid the "weakest link" effect limiting available capacity; For grid overload penalties, For grid interaction power, For power grid peak limits; when Exceed A penalty is incurred when the value is below the limit; this item is zero when the value is below the limit. This item ensures that the charging system actively responds to the grid load demand and automatically reduces power to avoid peak periods.

[0011] As a further improvement to this technical solution, the reward function is negative overall, and maximizing the cumulative reward is equivalent to minimizing the weighted sum of the penalties of each item; the items are optimized collaboratively through weight coefficients, and the weights can be dynamically adjusted according to the operating conditions; the original actions output by the strategy network must be mapped to the feasible domain allowed by hard constraints through a safety projection layer to ensure that the individual voltage does not exceed the preset upper limit, the temperature does not exceed the safety threshold, and the equalization current does not exceed the maximum capacity of the circuit, forming a dual protection mechanism of soft constraints guiding the optimization direction and hard constraints providing a safety net.

[0012] As a further improvement to this technical solution, the current trajectory of the pulse fast charging includes a positive pulse phase, a negative pulse depolarization phase, and an intermittent phase. Its parameters are adjusted online in real time by the deep reinforcement learning. Furthermore, the bidirectional reconfigurable power conversion unit automatically switches to an interleaved parallel or full-bridge phase-shifting topology based on the current battery terminal voltage to maintain high efficiency over a wide voltage range.

[0013] As a further improvement to this technical solution, the energy transfer of the multi-channel active balancing module is achieved through an addressable switching matrix and a bidirectional high-frequency soft-switching DC / DC converter. The balancing energy path and transmission power are directly determined by the differentiated individual unit balancing current allocation vector, prioritizing the transfer of high-voltage power. Energy transfer from high internal resistance monomers to low internal resistance monomers Low-temperature monomers to minimize temperature difference and Difference.

[0014] On the other hand, the present invention provides an intelligent fast charging equalization control device for an adaptive load forklift charger, which is applied to the above-mentioned intelligent fast charging equalization control method for an adaptive load forklift charger. It includes a bidirectional reconfigurable power conversion unit, with its input end connected to the AC power grid and its output end connected to the battery pack, for performing pulse fast charging and receiving power commands. A multi-channel active balancing module is connected across each cell in the battery pack to perform bidirectional energy transfer between any cells using a high-frequency soft-switching method. The edge computing controller communicates with the power conversion unit and the active equalization module, runs a reduced-order digital twin model and deep reinforcement learning inference, and outputs control commands. The 5G communication interface is connected to the edge computing controller to receive factory power grid load forecast information, dynamic electricity prices and forklift charging window data, and to synchronize and update the high-fidelity digital twin model in the cloud.

[0015] As a further improvement to this technical solution, the bidirectional reconfigurable power conversion unit consists of a front-stage AC / DC rectifier stage and a rear-stage reconfigurable DC / DC stage. The rear stage includes a transformer or inductor matrix switched by a switching network to achieve zero-voltage switching and minimum circulating current within the range of battery voltage variations.

[0016] As a further improvement to this technical solution, the multi-channel active equalization module includes a multi-winding high-frequency transformer, an H-bridge circuit for each channel, and a corresponding isolating switch. It achieves arbitrary path scheduling of energy between individual units through phase shift control, and operates at a frequency greater than 100kHz.

[0017] As a further improvement to this technical solution, the edge computing controller runs program instructions through a computer-readable storage medium to realize the above-mentioned intelligent fast charging equalization control method for adaptive load forklift chargers.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The intelligent fast charging equalization control device and method for the adaptive load forklift charger achieves multi-objective dynamic collaborative optimization of charging speed, battery life and electricity cost through the deep integration of battery electrical-thermal-aging digital twin model and deep reinforcement learning. It can adaptively respond to grid load peak and valley and dynamic electricity price signals, effectively reduce charging costs and alleviate grid burden.

[0019] 2. The adaptive load forklift charger uses an intelligent fast charging equalization control device and method. Through the coordinated control of the multi-channel active equalization module and the main charging current, it realizes the precise allocation of differentiated charging current for each cell. It prioritizes the transfer of energy from high SOC and high internal resistance cells to low SOC and low temperature cells, which significantly improves the consistency of the battery pack and delays the overall lifespan degradation caused by the weakest link effect.

[0020] 3. The adaptive load forklift charger uses an intelligent fast charging equalization control device and method. It performs hard constraint mapping on deep reinforcement learning actions through a safety projection layer, combined with soft constraint guidance of each safety item in the reward function, to form a dual-layer protection mechanism to ensure that the voltage, temperature and current of each cell do not exceed the safety threshold during the charging process. Attached Figure Description

[0021] The accompanying drawings described herein are for illustrative purposes only.

[0022] Figure 1 This is a schematic diagram of the device structure of the present invention; Figure 2 This is a flowchart of the overall method of the present invention; Figure 3 This is a flowchart illustrating the derivation of the reduced-order digital twin model of the present invention; Figure 4This is a schematic diagram of the deep reinforcement learning decision-making and security constraints of the present invention. Detailed Implementation

[0023] Under the guidance of this invention, any possible variations of this invention by those skilled in the art should be considered to fall within the scope of this invention.

[0024] Please see Figures 1-4 As shown, the present invention provides an intelligent fast charging equalization and control device for an adaptive load forklift charger, including a bidirectional reconfigurable power conversion unit, with its input end connected to the AC power grid and its output end connected to the battery pack, for performing pulse fast charging and receiving power commands; A multi-channel active balancing module is connected across each cell in the battery pack to perform bidirectional energy transfer between any cells using a high-frequency soft-switching method. The edge computing controller communicates with the power conversion unit and the active equalization module, runs a reduced-order digital twin model and deep reinforcement learning inference, and outputs control commands. The 5G communication interface connects to the edge computing controller to receive factory power grid load forecast information, dynamic electricity prices, and forklift charging window data, and synchronizes and updates the high-fidelity digital twin model in the cloud.

[0025] The bidirectional reconfigurable power conversion unit consists of a front-end AC / DC rectifier stage and a rear-end reconfigurable DC / DC stage. The rear-end stage includes a transformer or inductor matrix switched by a switching network to achieve zero-voltage switching and minimum circulating current within the range of battery voltage variations.

[0026] The multi-channel active equalization module includes a multi-winding high-frequency transformer, an H-bridge circuit for each channel, and a corresponding isolating switch. It achieves arbitrary path scheduling of energy between individual units through phase shift control, and operates at a frequency greater than 100kHz.

[0027] Furthermore, the edge computing controller executes program instructions through a computer-readable storage medium to implement the aforementioned intelligent fast charging equalization control method for forklift chargers with adaptive load. This control device is installed in the forklift intelligent charging system, which is also equipped with a cloud-based digital twin platform and a warehouse management system. The cloud-based digital twin platform interacts with the edge computing controller via a 5G network, continuously calibrating the reduced-order model parameters and distributing updated deep reinforcement learning strategy network weights. These are existing technologies and will not be elaborated further here.

[0028] The intelligent fast charging equalization control method for adaptive load forklift chargers of the present invention includes the following steps: Collect data on the voltage and temperature of each cell in the battery pack, as well as factory grid load forecasts, dynamic electricity price signals, and forklift charging window duration; Based on a reduced-order digital twin model of battery electro-thermal-aging coupling, the state of charge of each cell is simulated in real time. Core temperature, health status and internal concentration distribution; With the goal of optimizing the charging speed, battery life degradation and electricity cost in a coordinated manner, a Markov decision process composed of the inferred state and external information is constructed. Deep reinforcement learning is used to solve the problem online to generate the total charging power command and differentiated single-cell equalization current allocation vector at the current moment. The bidirectional reconfigurable power conversion unit is controlled to perform pulse fast charging, and the multi-channel active balancing module is driven to perform energy transfer between any cells in a high-frequency soft-switching manner according to the equal current distribution vector. When a grid load spike is detected, the total charging power is adaptively reduced, thereby achieving intelligent fast charging control that is adaptive to both the grid-side load and the battery-side health status.

[0029] In summary, after charging begins, the system first collects individual cell voltage / temperature and grid / window information; then, it uses a reduced-order digital twin model to deduce SOC / temperature / SOH; next, it solves multi-objective optimization through deep reinforcement learning, outputting total power commands and differentiated balancing commands; then, it executes pulse fast charging (power conversion unit) and active balancing (high-frequency soft-switching energy transfer) in parallel; simultaneously, it continuously monitors grid load peaks and adaptively reduces power when necessary; the process ends after determining whether charging is complete or interrupted; if not completed, it returns to re-decision, forming a closed-loop control.

[0030] Specifically, the reduced-order digital twin model of the battery's electro-thermal-aging coupling adopts the following reduced-order expression: The electrical characteristics of a single unit are modeled using a second-order RC equivalent circuit, and its state differential equation is as follows: ; ; Single-unit terminal voltage ; in, , These are the polarization voltages of the two RC circuits, representing the transient voltage response caused by the electrochemical reactions and diffusion processes inside the battery; , These are electrochemically polarized resistance and capacitance, corresponding to fast dynamic processes with small time constants, namely charge transfer reactions; , For concentration polarization resistance and capacitance, there is a slow dynamic process with a large time constant, namely solid-phase lithium-ion diffusion; the solid-phase lithium-ion diffusion process in the battery is described by Fick's second law, which is an infinite-dimensional partial differential equation. This refers to the charging and discharging current flowing through the cell; a positive value indicates charging, and a negative value indicates discharging. This differential equation describes the dynamic evolution of the polarization voltage under current excitation, and is used for real-time estimation. The foundation; This refers to the directly measurable terminal voltage of a single unit; It is the open-circuit voltage. The nonlinear function, calibrated experimentally. Obtained by looking up the curve in a table; The internal resistance is ohmic, which characterizes the instantaneous voltage drop of components such as electrolyte, diaphragm, and current collector; The terminal voltage equals the open-circuit voltage minus the ohmic voltage drop and the two polarization voltages. This equation is used for comparison and correction between the model output and the measured voltage. The thermal model uses a lumped-parameter heat balance equation: ; in, Indicates the rate of change of monomer temperature. For monomer mass, This is the equivalent specific heat capacity; This is ohmic heat, an irreversible Joule heating caused by internal resistance; Polarization heat is the heat generated by electrochemical polarization and concentration polarization. For heat dissipation, The equivalent heat transfer coefficient, For effective heat dissipation area, Ambient temperature; This equation achieves a two-way coupling of electrical and thermal properties: temperature affects resistance and aging parameters, and conversely, current and polarization generate heat and raise the temperature. The aging model uses a cumulative capacity loss model: ; in, The cumulative irreversible capacity loss is a measure of battery health. Core metrics; This is a pre-exponential factor, a constant related to battery materials and design processes; The activation energy characterizes the sensitivity of the aging reaction to temperature. This is the universal gas constant; This refers to the absolute temperature of the battery cell. The cumulative ampere-hour throughput is the total amount of charge flowing through the battery during charging and discharging. The power-law exponent, typically between 0.5 and 0.6, reflects the kinetic characteristics of side reactions such as SEI film growth; The model shows that the higher the temperature and the greater the throughput, the faster the capacity decays, and the decay exhibits a non-linear accelerating trend. The above partial differential-ordinary differential coupled system is achieved through The above partial differential-ordinary differential coupled system is approximated as a reduced-order state-space model that can be solved in real time by an edge controller. After approximate order reduction processing, the state variables such as SOC vector, core temperature, SOH capacity loss, polarization voltage, lithium ion concentration distribution and internal resistance change are output in real time, and the parameters are periodically corrected through 5G and cloud high-fidelity model.

[0031] Approximation is a mathematical method that approximates a given function using a rational function (the division of two polynomials). It approximates the transfer function of the diffusion into a low-order rational fraction, thereby transforming the PDE into a finite-dimensional system of ordinary differential equations. The reduced-order electro-thermal-aging coupled model is usually compressed to the 6th to 10th order, and a complete state derivation can be completed every 100 milliseconds on an ARM Cortex-A72-level processor. At the same time, it retains sufficient approximation accuracy for charging dynamics (0.01 to 10 Hz frequency band), providing physically consistent state input for real-time optimal control.

[0032] Specifically, deep reinforcement learning employs the Deep Deterministic Policy Gradient (DDPG) algorithm with a safety layer, where the policy network uses an Actor-Critic architecture; the Actor network stores the state space... Mapped to deterministic actions, the Critic network evaluates the expected cumulative reward of state-action pairs; state space Including each individual Vector, core temperature of each individual cell, individual cells Power grid load index Dynamic electricity price and remaining charging window duration ; Action space Including: Total charging current amplitude Pulse duty cycle ,frequency and the equalization current distribution of each individual cell The vector formed; The deep deterministic policy gradient algorithm introduces an experience replay mechanism: it uses the transition tuples generated by online interaction ( , , , Stored in the experience pool During training, random small-batch sampling is performed to break data correlation; The deep deterministic policy gradient algorithm introduces a target network and a soft update mechanism: an Actor target network and a Critic target network are constructed respectively, and the main network parameters are tracked by a soft update method.

[0033] Specifically, the reward function is designed as follows: ; in, For the charging speed target item, The average state of charge at the end of charging. For the target state of charge, this penalty is applied for insufficient charging completion; the square form means the penalty increases the further the deviation from the target; weighting coefficients. The importance of adjusting charging speed, the more urgent the window. The larger; For battery life degradation target items, The capacity loss increment for this charging cycle is obtained from the aging model; this term directly penalizes charging behaviors that cause irreversible damage to battery health; weighting coefficients. Reflecting the level of importance placed on battery life, this setting can be configured by the user or based on... Dynamic adjustment; For electricity cost target item, This is the dynamic electricity price for the current period. To control the step size, This is the total voltage of the battery pack. This refers to the charging energy consumption within the current control step; this penalty for electricity costs drives the system to charge or reduce power to avoid peak hours during off-peak electricity periods; weighting coefficient. It can be adjusted according to the user's sensitivity to cost-effectiveness; For temperature safety targets, For the first The core temperature of each individual unit This represents the highest core temperature among all cells; this penalty penalizes excessively high cell temperatures to prevent thermal runaway and accelerated aging; weighting coefficient. It automatically increases in the high-temperature range to form a soft constraint protection; To balance the consistency objective term, each individual entity represents... The sum of the absolute values ​​of the deviations of C from the mean measures the degree of inconsistency. It is a single-cell charged state. The average state of charge; this penalty When the C-dispersion is too large, the active balancing module prioritizes energy transfer to reduce the disparity; weighting coefficients Determine the priority of consistency protection to avoid the "weakest link" effect limiting available capacity; For grid overload penalties, For grid interaction power, For power grid peak limits; when Exceed A penalty is incurred when the value is below the limit; this item is zero when the value is below the limit. This item ensures that the charging system actively responds to the grid load demand and automatically reduces power to avoid peak periods.

[0034] In summary, the decision-making process of deep reinforcement learning is based on a state space. (SOC / T / SOH vector, grid load) Electricity price ,window The input is a policy network (DDPG / SAC architecture, including Actor-Critic, experience replay, and target network) that generates the original action, which is then passed through a safety projection layer (hard constraints: , , After mapping (equalization current limiting), the output executes the action. ( , , , Meanwhile, environmental feedback is used for experience replay in policy network updates, forming a complete training loop of decision-making, execution, feedback, and optimization.

[0035] Furthermore, the reward function is negative overall, and maximizing the cumulative reward is equivalent to minimizing the weighted sum of the penalties for each item; the items are optimized collaboratively through weight coefficients, and the weights can be dynamically adjusted according to the working conditions, such as increasing when the window is tight. Increased when battery aging is severe The original actions output by the strategy network must be mapped to the feasible domain allowed by hard constraints through the safety projection layer to ensure that the individual voltage does not exceed the preset upper limit, the temperature does not exceed the safety threshold, and the balancing current does not exceed the maximum capacity of the circuit, forming a dual protection mechanism of soft constraints guiding the optimization direction and hard constraints providing a safety net.

[0036] Furthermore, the current trajectory of pulse fast charging includes a positive pulse phase, a negative pulse depolarization phase, and an intermittent phase. Its parameters are adjusted online in real time by deep reinforcement learning, and the bidirectional reconfigurable power conversion unit automatically switches to interleaved parallel or full-bridge phase-shifting topology according to the current battery terminal voltage to maintain high efficiency over a wide voltage range.

[0037] Furthermore, the energy transfer of the multi-channel active balancing module is achieved through an addressable switching matrix and a bidirectional high-frequency soft-switching DC / DC converter. The balancing energy path and transmitted power are directly determined by the differentiated individual unit balancing current distribution vector, prioritizing the transfer of high-voltage power. Energy transfer from high internal resistance monomers to low internal resistance monomers Low-temperature monomers to minimize temperature difference and Difference.

[0038] Example 1 – Standard Operating Condition Verification Taking a 24-cell lithium iron phosphate battery pack as the model, the nominal capacity of each cell is 25Ah, the nominal voltage is 3.2V, the internal resistance is R0≈0.8mΩ, and the RC parameters are: R1=0.5mΩ, C1=15kF, R2=0.3mΩ, C2=100kF. Thermal parameters are: cell mass m=0.85kg, c_p=900J / (kg·K), heat dissipation area A=0.02m², and heat transfer coefficient h=10W / (m²·K). The aging model constants are B=12000, Ea=31500J / mol, z=0.55, and R_gas=8.314. The reduced-order model uses equilibrium truncation to compress the state variables to the 6th order, and is derived every 100ms on an ARM Cortex-A72 edge controller.

[0039] The deep reinforcement learning algorithm uses the DDPG algorithm, with a state dimension of 24×3+4=76 and an action dimension of 28 (including equalization). The weight coefficients are α=10, β=500, γ=0.05, δ=0.1, ε=20, and η=100. The grid load limit is P_limit=20kW. The policy converges after 5000 charging cycles.

[0040] The following scenario was used for verification: During a 30-minute charging window, the initial SOC was 20%, the target SOC was 90%, the real-time grid electricity price was 0.8 yuan / kWh, and a load peak occurred in the 10th minute, requiring power limitation to 15kW. Table 1 shows the comparison data.

[0041] Table 1

[0042] It is evident that this invention significantly reduces temperature rise, consistency, and capacity decay without extending charging time, while actively responding to grid peaks to reduce electricity costs.

[0043] Example 2 – Adaptability Verification for Different Charging Window Lengths To verify the adaptability of this invention under different forklift operation scheduling scenarios, three operating conditions were set with charging window durations of 15 minutes, 30 minutes, and 60 minutes, respectively. The initial SOC was 20% for all conditions, the target SOC was 90% for all conditions, the grid electricity price was constant at 0.8 yuan / kWh, and there were no grid peak restrictions. Comparative data are shown in Table 2.

[0044] Table 2 Adaptability verification for different charging window durations

[0045] As shown in Table 2, the present invention can automatically adjust the charging strategy according to the charging window duration: when the window is tight (15 min), the strategy network increases the weight of charging speed to charge as much power as possible within the window; when the window is ample (60 min), the strategy network reduces the charging rate to reduce temperature rise and capacity loss. This demonstrates the dynamic adaptive capability of the present invention for forklift operation scheduling.

[0046] Example 3 – Adaptability Verification to Different Ambient Temperatures To verify the robustness of this invention under different ambient temperatures, three operating conditions were set: ambient temperature of -10℃, 25℃, and 45℃, with a charging window of 30 minutes, an initial SOC of 20%, a target SOC of 90%, and a constant grid electricity price of 0.8 yuan / kWh. Comparative data are shown in Table 3.

[0047] Table 3. Adaptability verification under different environmental temperatures

[0048] Table 3 shows that: in low-temperature environments (-10℃), the strategy network automatically reduces the charging current to avoid the risk of lithium plating, and the charging speed decreases slightly but remains within a safe range; in high-temperature environments (45℃), the temperature safety item... The weight of the core is automatically increased, and the strategy network actively reduces the charging power to keep the core temperature within a safe threshold. This demonstrates the adaptive protection capability of this invention against extreme environmental temperatures.

[0049] Example 4 – Battery Aging Full Life Cycle Adaptability Verification To verify the adaptability of this invention throughout the entire battery lifecycle, charging verification was conducted under three conditions: new battery (SOH=100%), moderately aged (SOH=85%), and deeply aged (SOH=70%). The charging window was 30 minutes, the initial SOC was 20%, the target SOC was 90%, and the grid electricity price was constant at 0.8 yuan / kWh. Comparative data are shown in Table 4.

[0050] Table 4 Battery Aging Life Cycle Adaptability Verification

[0051] As shown in Table 4, as the battery ages, the strategy network automatically reduces the maximum charging rate to control the rate of battery life degradation. (The weighting is increased), and the balancing current is increased to compensate for the increased inconsistency of individual cells caused by aging. This demonstrates the adaptive health management capability of this invention throughout the entire battery lifecycle.

[0052] Example 5 – Dynamic Electricity Price Response Verification To verify the responsiveness of this invention to dynamic electricity pricing, time-of-use pricing scenarios were set: 0:00-8:00 yuan / kWh (off-peak), 8:00-12:00 yuan / kWh (peak), and 12:00-14:00 yuan / kWh (flat). The forklift charging windows were set for two scenarios: 8:00-9:00 (only 1 hour, peak) and 13:00-15:00 (2 hours, flat). The initial State of Charge (SOC) was 30% for both scenarios, and the target SOC was 90% for both. Comparative data are shown in Table 5.

[0053] Table 5 Verification of Dynamic Electricity Price Response

[0054] Table 5 shows that when the charging window is during peak electricity price periods and the window is tight (Scenario 1), the policy network in the electricity price penalty term... By balancing the charging speed term α, the system completes charging within the window at a higher rate. When the charging window is in a flat and relatively ample period (Scenario 2), the strategy network proactively reduces the charging rate and extends the charging time to take advantage of the flat-price period, resulting in a reduction of electricity costs of approximately 33% compared to traditional solutions. This demonstrates the intelligent response capability of this invention to dynamic electricity prices.

[0055] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A smart fast charging equalization control method for an adaptive load forklift charger, characterized in that, Includes the following steps: Collect data on the voltage and temperature of each cell in the battery pack, as well as factory grid load forecasts, dynamic electricity price signals, and forklift charging window duration; Based on a reduced-order digital twin model of battery electro-thermal-aging coupling, the state of charge of each cell is simulated in real time. Core temperature, health status and internal concentration distribution; With the goal of optimizing the charging speed, battery life degradation and electricity cost in a coordinated manner, a Markov decision process composed of the inferred state and external information is constructed. Deep reinforcement learning is used to solve the problem online to generate the total charging power command and differentiated single-cell equalization current allocation vector at the current moment. The bidirectional reconfigurable power conversion unit is controlled to perform pulse fast charging, and the multi-channel active balancing module is driven to perform energy transfer between any cells in a high-frequency soft-switching manner according to the equalization current distribution vector. When a grid load spike is detected, the total charging power is adaptively reduced, thereby achieving intelligent fast charging control that is adaptive to both the grid-side load and the battery-side health status.

2. The intelligent fast charging equalization control method for an adaptive load forklift charger according to claim 1, characterized in that: The reduced-order digital twin model of the battery's electro-thermal-aging coupling adopts the following reduced-order expression: The electrical characteristics of a single unit are modeled using a second-order RC equivalent circuit, and its state differential equation is as follows: ； ； Single-unit terminal voltage ; in, , These are the polarization voltages of the two RC circuits, representing the transient voltage response caused by the electrochemical reactions and diffusion processes inside the battery; , These are electrochemically polarized resistance and capacitance, corresponding to fast dynamic processes with small time constants, namely charge transfer reactions; , For concentration polarization resistance and capacitance, corresponding to a slow dynamic process with a large time constant, namely solid-phase lithium-ion diffusion; This represents the charging and discharging current flowing through the cell; a positive value indicates charging, and a negative value indicates discharging. This differential equation describes the dynamic evolution of the polarization voltage under current excitation, and is used for real-time estimation. The foundation; This refers to the directly measurable terminal voltage of a single unit; It is the open-circuit voltage. The nonlinear function, calibrated experimentally. Obtained by looking up the curve in a table; The internal resistance is ohmic, which characterizes the instantaneous voltage drop of components such as electrolyte, diaphragm, and current collector; The terminal voltage equals the open-circuit voltage minus the ohmic voltage drop and the two polarization voltages. This equation is used for comparison and correction between the model output and the measured voltage. The thermal model uses a lumped-parameter heat balance equation: ； in, Indicates the rate of temperature change of the monomer. For monomer mass, This is the equivalent specific heat capacity; This is ohmic heat, an irreversible Joule heating caused by internal resistance; Polarization heat is the heat generated by electrochemical polarization and concentration polarization. For heat dissipation, The equivalent heat transfer coefficient, For effective heat dissipation area, The ambient temperature; This equation achieves a two-way coupling of electrical and thermal properties: temperature affects resistance and aging parameters, and conversely, current and polarization generate heat and raise the temperature. The aging model uses a cumulative capacity loss model: ; in, The cumulative irreversible capacity loss is a measure of battery health. Core metrics; This is a pre-exponential factor, a constant related to battery materials and design processes; The activation energy characterizes the sensitivity of the aging reaction to temperature. This is the universal gas constant; This refers to the absolute temperature of the battery cell. The cumulative ampere-hour throughput is the total amount of charge flowing through the battery during charging and discharging. The power-law exponent, typically between 0.5 and 0.6, reflects the kinetic characteristics of side reactions such as SEI film growth; The model shows that the higher the temperature and the greater the throughput, the faster the capacity decays, and the trend is non-linear and accelerating. The above partial differential-ordinary differential coupled system is achieved through The above partial differential-ordinary differential coupled system is approximately simplified into a reduced-order state-space model that can be solved in real time by the edge controller.

3. The intelligent fast charging equalization control method for an adaptive load forklift charger according to claim 2, characterized in that: The deep reinforcement learning employs a deep deterministic policy gradient algorithm with a safety layer, whose state space... Including each individual Vector, core temperature of each individual cell, individual cells Power grid load index Dynamic electricity price and remaining charging window duration Action space Including: Total charging current amplitude Pulse duty cycle ,frequency and the equalization current distribution of each individual cell The vector formed; the reward function is designed as follows: ; in, For the charging speed target item, The average state of charge at the end of charging. For the target state of charge, this penalty is applied for insufficient charging completion; the square form means the penalty increases the further the deviation from the target; weighting coefficients. The importance of adjusting charging speed, the more urgent the window. The larger; For battery life degradation target items, The capacity loss increment for this charging cycle is obtained from the aging model; this term directly penalizes charging behaviors that cause irreversible damage to battery health; weighting coefficients. Reflecting the level of importance placed on battery life, this setting can be configured by the user or based on... Dynamic adjustment; For electricity cost target item, This is the dynamic electricity price for the current period. To control the step size, This is the total voltage of the battery pack. This refers to the charging energy consumption within the current control step; this penalty for electricity costs drives the system to charge or reduce power to avoid peak hours during off-peak electricity periods; weighting coefficient. It can be adjusted according to the user's sensitivity to cost-effectiveness; For temperature safety targets, For the first The core temperature of each individual unit This represents the highest core temperature among all cells; this penalty penalizes excessively high cell temperatures to prevent thermal runaway and accelerated aging; weighting coefficient. It automatically increases in the high-temperature range to form a soft constraint protection; To balance the consistency objective term, each individual entity represents... The sum of the absolute values ​​of the deviations of C from the mean measures the degree of inconsistency. It is a single-cell charged state. The average state of charge; this penalty When the C-dispersion is too large, the active balancing module prioritizes energy transfer to reduce the disparity; weighting coefficients Determine the priority of consistency protection to avoid the "weakest link" effect limiting available capacity; For grid overload penalties, For grid interaction power, For power grid peak limits; when Exceed A penalty is incurred when the value is below the limit; this item is zero when the value is below the limit. This item ensures that the charging system actively responds to the grid load demand and automatically reduces power to avoid peak periods.

4. The intelligent fast charging equalization control method for an adaptive load forklift charger according to claim 3, characterized in that: The reward function is negative overall, and maximizing the cumulative reward is equivalent to minimizing the weighted sum of the penalties of each item. The items are optimized in a multi-objective collaborative manner through weight coefficients, and the weights can be dynamically adjusted according to the operating conditions. The original actions output by the strategy network must be mapped to the feasible domain allowed by hard constraints through a safety projection layer to ensure that the individual voltage does not exceed the preset upper limit, the temperature does not exceed the safety threshold, and the equalization current does not exceed the maximum capacity of the circuit, forming a double-layer protection mechanism of soft constraints guiding the optimization direction and hard constraints providing a safety net.

5. The intelligent fast charging equalization control method for an adaptive load forklift charger according to claim 4, characterized in that: The current trajectory of the pulse fast charging includes a positive pulse phase, a negative pulse depolarization phase, and an intermittent phase. Its parameters are adjusted online in real time by the deep reinforcement learning. The bidirectional reconfigurable power conversion unit automatically switches to an interleaved parallel or full-bridge phase-shifting topology according to the current battery terminal voltage to maintain high efficiency over a wide voltage range.

6. The intelligent fast charging equalization control method for an adaptive load forklift charger according to claim 5, characterized in that: The energy transfer of the multi-channel active balancing module is achieved through an addressable switching matrix and a bidirectional high-frequency soft-switching DC / DC converter. The balancing energy path and transmission power are directly determined by the differentiated individual unit balancing current allocation vector, prioritizing high-frequency power transfer. Energy transfer from high internal resistance monomers to low internal resistance monomers Low-temperature monomers to minimize temperature difference and Difference.

7. An intelligent fast-charging equalization control device for an adaptive load forklift charger, applied to the intelligent fast-charging equalization control method for an adaptive load forklift charger as described in claim 6, characterized in that: It includes a bidirectional reconfigurable power conversion unit, with its input end connected to the AC power grid and its output end connected to the battery pack, used to perform pulse fast charging and receive power commands; A multi-channel active balancing module is connected across each cell in the battery pack to perform bidirectional energy transfer between any cells using a high-frequency soft-switching method. The edge computing controller communicates with the power conversion unit and the active equalization module, runs a reduced-order digital twin model and deep reinforcement learning inference, and outputs control commands. The 5G communication interface is connected to the edge computing controller to receive factory power grid load forecast information, dynamic electricity prices and forklift charging window data, and to synchronize and update the high-fidelity digital twin model in the cloud.

8. The intelligent fast charging equalization control device for an adaptive load forklift charger according to claim 7, characterized in that: The bidirectional reconfigurable power conversion unit consists of a front-end AC / DC rectifier stage and a rear-end reconfigurable DC / DC stage. The rear-end stage includes a transformer or inductor matrix switched by a switching network to achieve zero-voltage switching and minimum circulating current within the range of battery voltage variations.

9. The intelligent fast charging equalization control device for an adaptive load forklift charger according to claim 8, characterized in that: The multi-channel active equalization module includes a multi-winding high-frequency transformer, an H-bridge circuit for each channel, and a corresponding isolating switch. It achieves arbitrary path scheduling of energy between individual units through phase shift control, and operates at a frequency greater than 100kHz.

10. The intelligent fast charging equalization control device for an adaptive load forklift charger according to claim 9, characterized in that: The edge computing controller runs program instructions through a computer-readable storage medium to implement the above-mentioned intelligent fast charging equalization control method for adaptive load forklift chargers.

Images