Battery charging control method and system based on artificial intelligence

Abstract

The embodiment of the application provides a battery charging control method and system based on artificial intelligence, which obtains battery state data in the running process of a battery, wherein the battery state data comprises voltage state data, current state data, temperature state data and impedance state data of the battery. The state of health (SOH) of the battery is calculated. A target feature sequence is constructed according to the battery state data and the SOH, and the target feature sequence is input into a pre-trained remaining useful life prediction model to obtain a remaining useful life prediction result of the battery. According to the SOH of the battery and the remaining useful life prediction result, the charging strategy of the battery is adjusted to obtain a target charging strategy, and the battery is charged based on the target charging strategy, thereby improving the stability and life of the battery system.

Description

Technical Field

[0001] This application relates to the field of battery technology, and more specifically, to a battery charging control method and system based on artificial intelligence. Background Technology

[0002] Critical infrastructure such as communication base stations and data centers place extremely high demands on the stability and lifespan of power systems. Lithium-ion batteries are widely used in backup power systems due to their high energy density and excellent cycle performance. However, current charging management still commonly employs constant voltage / current limiting (CV / CC) methods, neglecting the aging characteristics of batteries in varying environments and during long-term operation. Traditional battery management systems (BMS) rely primarily on simple temperature and voltage threshold judgments, failing to provide refined prediction and dynamic control. This not only increases the potential failure risk of battery systems but also limits further improvements in their cycle life.

[0003] Therefore, improving the stability and lifespan of battery systems is an urgent problem to be solved. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a battery charging control method and system based on artificial intelligence to improve the stability and lifespan of the battery system.

[0005] In a first aspect, this application provides a battery charging control method based on artificial intelligence, comprising:

[0006] Acquire battery status data during battery operation, the battery status data including battery voltage status data, current status data, temperature status data and impedance status data;

[0007] Calculate the state of health (SOH) of the battery;

[0008] Based on the battery state data and the SOH, a target feature sequence is constructed, and the target feature sequence is input into a pre-trained remaining life prediction model to obtain the remaining life prediction result of the battery.

[0009] Based on the battery's state of health (SOH) and the predicted remaining lifespan, the charging strategy for the battery is adjusted to obtain the target charging strategy.

[0010] The battery is charged based on the target charging strategy.

[0011] Optionally, calculating the state of health (SOH) of the battery includes:

[0012] Obtain the nominal capacity of the battery;

[0013] The SOH of the battery is calculated based on the battery's nominal capacity, temperature state data, current state data, and preset parameters, including preset weighting coefficients, activation energy constant, and gas constant.

[0014] Optionally, the step of constructing a target feature sequence based on the battery state data and the SOH, and inputting the target feature sequence into a pre-trained remaining life prediction model to obtain the remaining life prediction result of the battery, includes:

[0015] Local features of the feature sequence are extracted using a one-dimensional convolutional neural network in the remaining useful life prediction model.

[0016] Temporal dependency features are extracted from the feature sequence using a bidirectional gated recurrent unit network in the remaining useful life prediction model.

[0017] Feature weights are determined using the attention mechanism in the remaining useful life prediction model.

[0018] The remaining lifespan prediction result of the battery is generated based on the local features, the temporal dependency features, and the feature weights.

[0019] Optionally, adjusting the charging strategy of the battery based on the battery's SOH and the predicted remaining lifespan to obtain a target charging strategy includes:

[0020] If the predicted remaining service life is less than or equal to the preset remaining service life, or the SOH is less than or equal to the preset SOH, then the target charging current is obtained based on the SOH, the SOH health threshold, and the SOH change rate.

[0021] Based on the target charging current, the target charging output duty cycle is obtained;

[0022] The target charging strategy is determined based on the target charging current and the target charging output duty cycle.

[0023] Optionally, obtaining the target charging current based on the SOH, the SOH health threshold, and the SOH change rate includes:

[0024] Obtain the first empirical parameter and the second empirical parameter;

[0025] Calculate the quotient of the SOH and the SOH health threshold;

[0026] The target charging current is calculated based on the first empirical parameter, the second empirical parameter, the quotient of the SOH and the SOH health threshold, and the SOH change rate.

[0027] Optionally, if the predicted remaining useful life is less than or equal to a preset remaining useful life, or the SOH is less than or equal to a preset SOH, the method further includes:

[0028] Output charging strategy prompt information, which is used to prompt the user that the battery will switch to the extended life slow charging mode.

[0029] Optionally, after charging the battery based on the target charging strategy, the method further includes:

[0030] During the charging process, the core temperature and heat dissipation point temperature of the battery are obtained.

[0031] If the difference between the core temperature and the heat dissipation point temperature is greater than or equal to a preset temperature fluctuation threshold, the heat to be controlled is calculated based on the core temperature and the heat dissipation point temperature.

[0032] Temperature control is performed on the charging process of the battery based on the heat to be controlled.

[0033] Optionally, the method further includes:

[0034] Obtain the actual remaining lifespan of the battery;

[0035] Based on the actual remaining useful life, the predicted remaining useful life, the self-learning weight parameter set, and the regularization term weight factor, the model optimization objective function of the remaining useful life prediction model is obtained.

[0036] The remaining useful life prediction model is optimized based on the objective function of the model.

[0037] Optionally, the step of obtaining the model optimization objective function of the remaining useful life prediction model based on the actual remaining useful life, the remaining useful life prediction result, the self-learning weight parameter set, and the regularization term weight factor includes:

[0038] The error between the actual remaining useful life and the predicted remaining useful life is calculated.

[0039] The self-learning weight parameter set is adjusted by the regularization term weight factor to obtain the adjusted self-learning weight parameter set;

[0040] Based on the error results and the adjusted set of self-learning weight parameters, the objective function of the remaining useful life prediction model is obtained.

[0041] Secondly, this application provides an artificial intelligence-based battery charging control system, which includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions. When the processor executes the machine-executable instructions, the artificial intelligence-based battery charging control system implements the aforementioned artificial intelligence-based battery charging control method.

[0042] The AI-based battery charging control method and system provided in this application acquires battery state data during battery operation, including voltage, current, temperature, and impedance data. The State of Health (SOH) of the battery is calculated. A target feature sequence is constructed based on the battery state data and SOH, and input into a pre-trained remaining lifespan prediction model to obtain a predicted remaining lifespan. The charging strategy is adjusted based on the SOH and the predicted remaining lifespan to obtain a target charging strategy. The battery is then charged according to this target strategy. By dynamically adapting the upper and lower limits of charging voltage, current intensity, and charging duration, the charging process always aligns with the battery's current health and expected lifespan. This avoids the overcharging risks caused by fixed parameters in traditional constant voltage and current limiting strategies, or the exacerbated side reactions and thermal runaway risks from using high-intensity charging during battery aging. This ensures more stable chemical reactions and a more stable structure throughout the battery's lifespan, reducing unnecessary energy loss and degradation damage, thereby improving the stability and lifespan of the battery system. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0044] Figure 1 A flowchart illustrating an artificial intelligence-based battery charging control method provided in this application embodiment;

[0045] Figure 2 This is a schematic diagram of a battery charging control system based on artificial intelligence, provided as an embodiment of this application.

[0046] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0047] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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.

[0048] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0049] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0050] Figure 1 This is a flowchart illustrating an artificial intelligence-based battery charging control method provided in an embodiment of this application. It should be understood that in other embodiments, the order of some steps in the artificial intelligence-based battery charging control method of this embodiment can be shared according to actual needs, or some steps can be omitted or maintained. Figure 1 As shown, the method may include the following steps:

[0051] S110. Obtain battery status data during battery operation, wherein the battery status data includes the battery's voltage status data, current status data, temperature status data, and impedance status data.

[0052] In this step, the battery can be, for example, a battery cluster composed of lithium iron phosphate cells or ternary lithium cells. The acquisition of battery state data is accomplished by a data acquisition unit, which is located at each cell string circuit, output circuit, and individual cell interface of the battery module. This unit includes a temperature sensor, a voltage detection chip, a current detection resistor, a Hall element, and an impedance spectrum module. Specifically, the temperature sensor acquires temperature state data T(t), the voltage detection chip acquires voltage state data V(t) (including individual cell voltage and total voltage), the current detection resistor, in conjunction with the Hall element, acquires current state data I(t) (including charging and discharging current), and the impedance spectrum module acquires impedance state data Z(t) (containing impedance values ​​at multiple frequency points) in the 10Hz-10kHz range using AC excitation. This data is periodically transmitted to the embedded AI controller.

[0053] For example, the temperature sensors are arranged in a distributed manner, with eight temperature sensors set at different locations in the battery module to collect temperature status data at different cell or module locations. The sampling frequency of each sensor is 10Hz, so the temperature status data T(t) is an 8-dimensional time series. For example, at time t, T(t) can be represented as [T1(t), T2(t), T3(t), T4(t), T5(t), T6(t), T7(t), T8(t)], where each element corresponds to the temperature value at a different location, in degrees Celsius.

[0054] The voltage detection chip is used to collect the voltage of each cell in a string and the total output voltage of the battery module. The sampling frequency is 20Hz. The voltage state data V(t) is a multi-dimensional sequence, which includes the voltage of each individual cell and the total voltage. For example, for a battery module composed of 16 cells in a string, V(t) can be represented as [V1(t), V2(t), ..., V16(t), Vtotal(t)], with the unit being volts. V1(t) to V16(t) are the voltages of each individual cell at time t, and Vtotal(t) is the total voltage of the module.

[0055] The current sensing resistor and Hall element work together to collect current state data I(t) at a sampling frequency of 20Hz. The current state data includes charging current and discharging current. When the battery is charging, I(t) is positive and when discharging, it is negative. The unit is amperes. The data is in the form of a one-dimensional time series. However, since the sampling time is continuous, it forms a set of sequences with multiple time points, such as [I(t-2Δt), I(t-Δt), I(t)], where Δt is the sampling interval. Here, Δt=0.05s (corresponding to a sampling frequency of 20Hz).

[0056] The impedance spectrum module acquires impedance state data Z(t) through AC excitation. It can obtain Z(f) curves in the range of 10Hz to 10kHz. The sampling frequency is 1Hz, and an impedance sequence containing multiple frequency points is acquired every second. Therefore, Z(t) is a multi-dimensional sequence. For example, at time t, Z(t) can be represented as [Z(10Hz,t), Z(100Hz,t), Z(1kHz,t), Z(5kHz,t), Z(10kHz,t)], with the unit being ohms, containing impedance values ​​at different frequencies.

[0057] S120. Calculate the state of health (SOH) of the battery.

[0058] S121. Obtain the nominal capacity of the battery.

[0059] In this step, the nominal capacity Cr of the battery is the capacity parameter marked when the battery leaves the factory. For example, the nominal capacity of a lithium-ion battery commonly used in communication backup power can be set to 100Ah. This parameter is stored in the embedded AI controller as the basic parameter for calculating SOH.

[0060] S122. Calculate the SOH of the battery based on the nominal capacity of the battery, the temperature state data, the current state data, and preset parameters, whereby the preset parameters include a preset weighting coefficient, an activation energy constant, and a gas constant.

[0061] In this step, SOH can be calculated using the following formula (1):

[0062] (1);

[0063] The integration interval is from the initial time 0 when the battery starts to be used to the current time t. The preset parameters include the preset weighting coefficient α (dimensionless, such as 0.001), the activation energy constant Ea (such as 30000 J / mol), and the gas constant. |I(τ)| represents the absolute value of the current in the current state data, T(τ) represents the temperature value in the temperature state data, and the value of SOH is between 0 and 1.

[0064] S130. Construct a target feature sequence based on the battery state data and the SOH, and input the target feature sequence into a pre-trained remaining life prediction model to obtain the remaining life prediction result of the battery.

[0065] The remaining lifetime prediction model can be computed using an NPU acceleration chip, which supports FP16 tensor operations. The model has fewer than or equal to 3 M parameters and can complete a remaining lifetime prediction inference within 100ms.

[0066] In this embodiment, the target feature sequence is first constructed based on the battery state data and SOH. The battery state data includes voltage state data V(t), current state data I(t), temperature state data T(t), and impedance state data Z(t). Assume that V(t) is a 17-dimensional sequence containing the voltages of 16 individual cells and a total voltage, taking the data from the three most recent moments: V(t-2Δt), V(t-Δt), and V(t). Each moment's V data is 17-dimensional, therefore the voltage-related feature is 3 × 17 = 51 dimensions. I(t) is the current sequence, taking the data from the three most recent moments: I(t-2Δt), I(t-Δt), and I(t), which is 3-dimensional. T(t) is the temperature data from eight temperature sensors, taking the data from the current moment t, which is 8-dimensional. Z(t) is the impedance data from five frequency points, taking the data from the current moment t, which is 5-dimensional. SOH is the currently calculated 0.5, which is 1-dimensional. These data are concatenated in sequence to construct the target feature sequence x, which has a dimension of 51+3+8+5+1=68, i.e. x=[V(t-2Δt), V(t-Δt), V(t), I(t-2Δt), I(t-Δt), I(t), T(t), Z(t), SOH], where each element is a corresponding data sequence or value.

[0067] Next, the target feature sequence is input into the pre-trained remaining useful life prediction model. This model employs a combined structure of 1D-CNN, BiGRU, and Attention, and its specific structure and operation are as follows:

[0068] S131. Extract local features of the feature sequence through a one-dimensional convolutional neural network in the remaining useful life prediction model.

[0069] In this step, the one-dimensional convolutional neural network (1D-CNN) contains multiple convolutional layers and pooling layers. The convolutional layers perform convolution operations on the target feature sequence using convolutional kernels of different sizes. For example, 32 convolutional kernels of size 3 are used to extract the first-level local features, and then 64 convolutional kernels of size 2 are used for further extraction. The pooling layers reduce the feature dimensionality through max pooling, and finally obtain local features containing local correlation information of voltage, current, temperature, and impedance.

[0070] For example, a 1D-CNN layer contains two convolutional layers and one pooling layer. The first convolutional layer has 32 kernels, a kernel size of 3 (i.e., processing 3 adjacent feature dimensions at a time), a stride of 1, and uses the ReLU activation function to convolve the input 68-dimensional target feature sequence, outputting 32 feature maps, each with a dimension of 68-3+1=66. The second convolutional layer has 64 kernels, a kernel size of 2, a stride of 1, and also uses the ReLU activation function to convolve the 32 66-dimensional feature maps output from the first convolutional layer, outputting 64 feature maps, each with a dimension of 66-2+1=65. Then, a max pooling layer with a pooling window size of 2 and a stride of 2 is used to perform pooling operations on 64 65-dimensional feature maps, outputting 64 32-dimensional (65 / / 2=32) local feature maps. These local feature maps extract local feature information from the target feature sequence, such as local voltage fluctuation features and local correlation features between current and temperature.

[0071] S132. Extract the temporal dependency features from the feature sequence through the bidirectional gated recurrent unit network in the remaining useful life prediction model.

[0072] In this step, the Bidirectional Gated Recurrent Unit Network (BiGRU) consists of two bidirectional layers, each containing 128 hidden units. The local feature map output from the 1D-CNN is flattened and then input into the BiGRU. The forward layer captures the temporal correlation from the past to the present, and the backward layer captures the temporal correlation from the present to the future. After processing by the two layers, the output is a concatenated temporal dependent feature vector, which contains long-term correlation information of state data at different times (such as the relationship between current change trends and aging rate).

[0073] For example, the 64 32-dimensional local feature maps output from the 1D-CNN layer are flattened to obtain a 64×32=2048-dimensional feature vector, which is used as the input to the BiGRU layer. The first BiGRU layer processes this feature vector in both forward and backward directions. Forward processing captures the temporal dependencies from the past to the present, and backward processing captures the temporal dependencies from the present to the future, outputting two 128-dimensional feature vectors, which are then concatenated to obtain a 256-dimensional feature vector. The second BiGRU layer takes the 256-dimensional feature vector output from the first layer as input, and similarly performs forward and backward processing, outputting two 128-dimensional feature vectors, which are then concatenated to obtain a 256-dimensional temporal dependency feature vector. This vector contains long-term temporal correlation information between data at different times in the target feature sequence, such as the temporal relationship between current change trends and battery aging rate, and the impact of temperature fluctuations on impedance over time.

[0074] S133. The feature weights are determined by the attention mechanism in the remaining useful life prediction model.

[0075] In this step, the attention mechanism maps the temporal-dependent feature vector output by BiGRU to a weight vector through a fully connected layer, and then normalizes it using softmax to obtain the weights of each feature dimension (the sum of the weights is 1). For example, feature dimensions related to SOH changes are given higher weights, while dimensions with less influence are given lower weights, thereby highlighting the impact of key features on remaining useful life prediction.

[0076] Specifically, for example, a fully connected layer can be used to map a 256-dimensional feature vector to a 256-dimensional weight vector. Then, a softmax function is used to normalize this weight vector, obtaining the weight value for each dimension, with the sum of the weight values ​​being 1. For instance, dimensions related to changes in SOH might be assigned higher weights (e.g., 0.1), while dimensions with less influence might be assigned lower weights (e.g., 0.001). Then, each dimension value in the time-dependent feature vector is multiplied by its corresponding weight value to obtain a weighted time-dependent feature vector.

[0077] S134. Based on the local features, the temporal dependency features, and the feature weights, generate the predicted remaining lifespan of the battery.

[0078] In this step, the weighted temporal-dependent feature vector is input into the fully connected layer, which outputs the remaining lifetime prediction values ​​corresponding to multiple future time nodes, forming a multi-dimensional remaining lifetime prediction result (such as the remaining number of cycles, the remaining lifetime duration, etc.).

[0079] S140. Based on the battery's health status (SOH) and the predicted remaining lifespan, the charging strategy for the battery is adjusted to obtain the target charging strategy.

[0080] S141. If the predicted remaining service life is less than or equal to the preset remaining service life, or the SOH is less than or equal to the preset SOH, then the target charging current is obtained based on the SOH, the SOH health threshold, and the SOH change rate.

[0081] S1411, Obtain the first empirical parameter and the second empirical parameter.

[0082] In this step, the first empirical parameter k1 and the second empirical parameter k2 are preset adjustable parameters, such as k1=0.5, k2=0.3, etc., which are used to weight the influence of SOH deviation and health change rate on charging current.

[0083] S1412. Calculate the quotient of the SOH and the SOH health threshold.

[0084] In this step, the SOH health threshold SOH_thr is a set value (such as 0.8 or 0.9). The ratio of the current SOH to SOH_thr is calculated. For example, when SOH=0.5 and SOH_thr=0.8, the quotient is 0.5 / 0.8=0.625.

[0085] S1413. The target charging current is calculated based on the first empirical parameter, the second empirical parameter, the quotient of the SOH and the SOH health threshold, and the SOH change rate.

[0086] The SOH change rate can be obtained by linearly fitting the SOH values ​​over a preset time period. For example, the SOH change rate can be calculated by linearly fitting the SOH values ​​over the most recent 10 days. Assuming the SOH values ​​for the most recent 10 days are 0.55, 0.54, 0.53, 0.52, 0.51, 0.50, 0.49, 0.48, 0.47, and 0.46, the SOH change rate obtained by linear fitting is -0.01 / day (a decrease of 0.01 per day).

[0087] In this step, the target charging current can be calculated using the following formula (2):

[0088] (2);

[0089] Where dSOH / dt is the rate of change of SOH. For example, when k1=0.5, 1-SOH / SOH_thr=0.375, k2=0.3, and dSOH / dt=-0.01, the target charging current is 0.5×0.375+0.3×(-0.01)=0.1845A.

[0090] S142. Based on the target charging current, obtain the target charging output duty cycle.

[0091] In this step, the target charging output duty cycle D chg It can be calculated using the following formula (3):

[0092] (3);

[0093] Where I max This is the battery's maximum charging current (e.g., 10A). Assume I... chg * When the current is 0.1845A, the target charging output duty cycle D is... chg =0.1845 / 10=0.01845.

[0094] S143. Determine the target charging strategy based on the target charging current and the target charging output duty cycle.

[0095] In this step, the target charging strategy includes the target charging current (e.g., 0.1845A) and the target charging output duty cycle (e.g., 0.01845). The dynamic charge and discharge execution unit adjusts the PWM signal and current output according to the strategy to achieve charging control that balances charging efficiency and delays aging.

[0096] S150. Charge the battery based on the target charging strategy.

[0097] In this step, for example, the target charging strategy command can be received through a dynamic charge-discharge execution unit (such as a bidirectional DC-DC converter, MOSFET control unit, etc.). By adjusting the duty cycle of the PWM signal, the switching transistor is controlled to turn on and off, so that the actual charging current is stabilized near the target charging current, and the battery charging operation is completed.

[0098] In one possible implementation, if the predicted remaining lifespan is less than or equal to a preset remaining lifespan, or if the State of Balance (SOH) is less than or equal to a preset SOH, a charging strategy prompt can be output. This prompt alerts the user that the battery will switch to an extended lifespan slow charging mode. In this implementation, it is assumed that the preset remaining lifespan L... crit (e.g., 300 cycles) and preset SOH (SOH) crit (e.g., 0.6) is the judgment threshold. When the triggering condition is met, the embedded AI controller generates a prompt message (such as an electrical signal), transmits it to the monitoring terminal, and displays a text prompt to inform the user that the battery will switch to the extended life slow charging mode.

[0099] Optionally, after charging the battery based on the target charging strategy, the method may further include the following steps:

[0100] S210. During the charging process, the core temperature and heat dissipation point temperature of the battery are obtained.

[0101] In this step, the core temperature T_core is the temperature of the central cell of the battery module, and the heat dissipation point temperature T_sink is the temperature of the contact point between the heat dissipation structure and the outside environment. These temperatures are collected in real-time by temperature sensors located at the corresponding positions, and the unit is °C. This heat dissipation structure can be installed on the outer surface of the battery module or in the internal interlayer of the module, and may include, for example, heat pipes, thermal grease, or graphite heat sinks.

[0102] S220. If the difference between the core temperature and the heat dissipation point temperature is greater than or equal to a preset temperature fluctuation threshold, the heat to be controlled is calculated based on the core temperature and the heat dissipation point temperature.

[0103] In this step, the preset temperature fluctuation threshold is 2℃. When |T_core-T_sink|≥2℃, the heat to be controlled, q (in W), is calculated using the heat conduction formula q=λ×A×(T_core-T_sink) / R_th, where λ is the thermal conductivity of the heat dissipation material of the heat dissipation structure (e.g., ...). A is the heat transfer area of ​​the heat dissipation structure (e.g., 0.1m²), and R_th is the thermal resistance (e.g., 0.1K / W).

[0104] S230. Temperature control is performed on the charging process of the battery based on the heat to be controlled.

[0105] In this step, the battery's heat dissipation-temperature equalization structure (such as heat pipes, graphite heat equalization sheets, etc.) can start active (such as air cooling) and passive heat dissipation in accordance with the heat to be controlled. The heat pipes conduct core heat, and the air cooling accelerates heat dissipation, so that the temperature difference is reduced to ≤2℃, ensuring temperature balance during charging.

[0106] For example, assuming the heat to be controlled is q = 600W, a start command can be sent to the active cooling module (i.e., the air-cooling module) in the heat dissipation-temperature equalization structure. The fan of the air-cooling module starts running, and the fan speed is adjusted according to the amount of heat to be controlled. Here, since 600W of heat needs to be dissipated, the fan speed is set to 3000 rpm. At the same time, passive cooling components such as heat pipes, thermal grease, and graphite heat spreaders in the heat dissipation structure also work together. The heat pipes quickly conduct heat from the core area of ​​the battery module to the heat dissipation points, the thermal grease fills the gap between the heat pipes and the battery module to reduce contact thermal resistance, and the graphite heat spreader makes the temperature distribution on the surface of the battery module more uniform. By combining active and passive cooling, the temperature difference of the battery module gradually decreases. After a period of time, Tcore = 33℃, Tsink = 32.5℃, ΔT = 0.5℃ ≤ 2℃, and the temperature control process ends.

[0107] Optionally, the method may further include the following steps:

[0108] S310. Obtain the actual remaining lifespan of the battery.

[0109] In this step, the actual remaining service life L real For example, the number of cycles a battery can perform in actual use until its capacity decays to 20% of its nominal capacity can be obtained through long-term operation records or full life cycle testing, with the unit being the number of cycles.

[0110] S320. Based on the actual remaining useful life, the remaining useful life prediction result, the self-learning weight parameter set, and the regularization term weight factor, obtain the model optimization objective function of the remaining useful life prediction model.

[0111] S321. Calculate the error between the actual remaining useful life and the predicted remaining useful life.

[0112] In this step, the error result is the absolute value between the predicted value and the actual value, i.e., |L'-L real |, for example, L'=480 times, L real When the number of times is 500, the error result is 20.

[0113] S322. Adjust the self-learning weight parameter set by the regularization term weight factor to obtain the adjusted self-learning weight parameter set;

[0114] In this step, the self-learning weight parameter set θ consists of all trainable parameters in the model. The regularization weight factor γ (e.g., 0.0001) is used to adjust θ, and the adjusted set is... ,in Let θ be the L2 norm (the square root of the sum of the squares of the parameters).

[0115] S323. Based on the error results and the adjusted set of self-learning weight parameters, the model optimization objective function of the remaining useful life prediction model is obtained.

[0116] In this step, the model optimization objective function L is the sum of the error result and the adjusted set of self-learning weight parameters, i.e. This is used to balance prediction accuracy and model complexity, and to avoid overfitting.

[0117] S330. Optimize the remaining useful life prediction model based on the objective function of the model.

[0118] In this step, the stochastic gradient descent algorithm can be used to minimize the model optimization objective function L, and the self-learning weight parameter set θ is updated (fine-tuned online every 100 charge-discharge cycles). Through iterative optimization, the prediction accuracy and environmental adaptability of the model can be improved.

[0119] For example, the stochastic gradient descent algorithm can be used to optimize the model, with a learning rate of 0.001 and 100 training epochs. In each training epoch, the training set data is input into the model in batches, with each batch containing 32 data points. The model optimization objective function L is calculated for each batch of data, and then the self-learning weight parameter set θ of the model is updated according to L, with the update direction being the negative gradient direction of L. After each training epoch, the model is evaluated using validation set data, and the average error on the validation set is calculated. Training stops when the average error on the validation set no longer decreases for 10 consecutive training epochs. The resulting model is the trained remaining lifespan prediction model. Furthermore, during actual battery operation, an adaptive learning module is activated every 100 charge-discharge cycles to obtain the actual remaining lifespan and prediction results within those 100 cycles. The model parameters are then fine-tuned online according to the aforementioned model optimization objective function to adapt to changes in battery characteristics at different usage stages.

[0120] The structures described in the above embodiments achieve data transmission and control logic linkage through signal lines, control lines, and power lines, realizing a closed-loop structure of prediction-feedback-regulation. This method can operate stably in an ambient temperature range of -40℃ to +55℃, and under a charge / discharge rate of 0.5 C, the battery cycle life improvement factor reaches greater than or equal to 1.25 (i.e., the ratio of the cycle life of the battery controlled by this method to the cycle life of the battery under the traditional constant voltage and current limiting strategy is greater than or equal to 1.25). It can effectively improve battery life and reduce replacement frequency and maintenance costs while meeting backup power requirements.

[0121] The method provided in this application acquires battery state data during battery operation, including voltage state data, current state data, temperature state data, and impedance state data. The state of health (SOH) of the battery is calculated. A target feature sequence is constructed based on the battery state data and the SOH, and this target feature sequence is input into a pre-trained remaining lifespan prediction model to obtain a predicted remaining lifespan result for the battery. Based on the battery's SOH and the predicted remaining lifespan result, the battery's charging strategy is adjusted to obtain a target charging strategy. The battery is then charged based on this target charging strategy. By dynamically adapting the upper and lower limits of the charging voltage, current intensity, and charging duration, the charging process always aligns with the battery's current health state and expected lifespan. This avoids the overcharging risk caused by fixed parameters in traditional constant voltage and current limiting strategies, or the exacerbated side reactions and thermal runaway risks caused by using high-intensity charging during battery aging. This ensures more stable chemical reactions and a more stable structure throughout the battery's lifespan, reducing unnecessary energy loss and degradation damage, thereby improving the stability and lifespan of the battery system.

[0122] Figure 2 This is a schematic diagram of the structure of an artificial intelligence-based battery charging control system 100 provided in an embodiment of this application. Figure 2 As shown, the processor 120 can be used in the artificial intelligence-based battery charging control system 100 and to perform the functions in this invention.

[0123] The AI-based battery charging control system 100 can be a general-purpose server or a special-purpose server; both can be used to implement the AI-based battery charging control method of this invention. Although only one server is shown in this invention, for convenience, the functions described in this invention can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0124] For example, an AI-based battery charging control system 100 may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the AI-based battery charging control system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present invention can be implemented according to these program instructions. The AI-based battery charging control system 100 also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0125] For ease of explanation, only one processor is described in the AI-based battery charging control system 100. However, it should be noted that the AI-based battery charging control system 100 of the present invention may also include multiple processors, and therefore the steps performed by one processor described in the present invention may also be performed jointly or individually by multiple processors. For example, if the processor of the AI-based battery charging control system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0126] Optionally, the inner wall of the housing of the AI-based battery charging control system 100 may be coated with phase change thermal grease, which has a thermal capacity of... To suppress transient temperature rise.

[0127] Optionally, the AI-based battery charging control system 100 can be connected to the communication base station network management system via RS-485 or Ethernet interface, supporting the uploading of remaining service life prediction results and SOH curves using the Modbus-TCP protocol.

[0128] This invention discloses a computer read storage medium that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the steps in the artificial intelligence-based battery charging control method of the aforementioned embodiments.

[0129] This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the artificial intelligence-based battery charging control method of the foregoing embodiments.

[0130] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.

[0131] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electronically erasable rewritable read-only memory (EEPROM), compact optical disc (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to have or store data.

[0132] Finally, it should be noted that the above-disclosed embodiments are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence-based battery charging control method, characterized by, The method comprises: acquiring battery state data in the process of battery operation, the battery state data comprising voltage state data, current state data, temperature state data and impedance state data of the battery; calculating a state of health SOH of the battery; constructing a target feature sequence according to the battery state data and the SOH, and inputting the target feature sequence into a pre-trained remaining useful life prediction model to obtain a remaining useful life prediction result of the battery; adjusting a charging strategy of the battery according to the state of health SOH of the battery and the remaining useful life prediction result to obtain a target charging strategy, comprising: if the remaining useful life prediction result is less than or equal to a preset remaining useful life, or the SOH is less than or equal to a preset SOH, outputting a charging strategy prompt information, the charging strategy prompt information being used to prompt a user that the battery will switch to a life-extending slow charging mode, and obtaining a target charging current according to the SOH, an SOH health threshold and an SOH change rate, comprising: acquiring a first empirical parameter and a second empirical parameter; calculating a quotient of the SOH and the SOH health threshold; calculating the target charging current according to the first empirical parameter, the second empirical parameter, the quotient of the SOH and the SOH health threshold and the SOH change rate; obtaining a target charging output duty cycle based on the target charging current; determining the target charging strategy according to the target charging current and the target charging output duty cycle; charging the battery based on the target charging strategy. 2.The battery charging control method based on artificial intelligence according to claim 1, wherein, The calculation of the SOH of the battery comprises: acquiring a battery nominal capacity of the battery; calculating the SOH of the battery according to the battery nominal capacity, the temperature state data, the current state data and a preset parameter, the preset parameter comprising a preset weight coefficient, an activation energy constant and a gas constant. 3.The battery charging control method based on artificial intelligence according to claim 1, wherein, The construction of the target feature sequence according to the battery state data and the SOH, and the input of the target feature sequence into the pre-trained remaining useful life prediction model to obtain the remaining useful life prediction result of the battery comprise: extracting local features of the feature sequence through a one-dimensional convolutional neural network in the remaining useful life prediction model; extracting time sequence dependent features in the feature sequence through a bidirectional gated recurrent unit network in the remaining useful life prediction model; determining feature weights through an attention mechanism in the remaining useful life prediction model; generating the remaining useful life prediction result of the battery according to the local features, the time sequence dependent features and the feature weights. 4.The battery charging control method based on artificial intelligence according to claim 1, wherein, After the charging of the battery based on the target charging strategy, the method further comprises: acquiring a core temperature and a heat dissipation point temperature of the battery in the process of charging; if a difference between the core temperature and the heat dissipation point temperature is greater than or equal to a preset temperature fluctuation threshold, calculating a to-be-controlled heat quantity according to the core temperature and the heat dissipation point temperature; controlling the charging process of the battery according to the to-be-controlled heat quantity. 5.The battery charging control method based on artificial intelligence according to claim 1, wherein, The method further comprises: acquire an actual remaining use life of the battery; obtain a model optimization objective function of the remaining use life prediction model according to the actual remaining use life, the remaining use life prediction result, a self-learning weight parameter set and a regularization term weight factor; optimize the remaining use life prediction model based on the model optimization objective function. 6.The battery charging control method based on artificial intelligence according to claim 5, wherein, The obtaining of the model optimization objective function of the remaining use life prediction model according to the actual remaining use life, the remaining use life prediction result, the self-learning weight parameter set and the regularization term weight factor comprises: calculating an error result of the actual remaining use life and the remaining use life prediction result; adjusting the self-learning weight parameter set through the regularization term weight factor to obtain an adjusted self-learning weight parameter set; obtaining the model optimization objective function of the remaining use life prediction model based on the error result and the adjusted self-learning weight parameter set.

7. An artificial intelligence based battery charging control system characterized in that, The battery charging control method based on artificial intelligence comprises a processor and a computer readable storage medium, the computer readable storage medium stores machine executable instructions, and the machine executable instructions are executed by the computer to implement the battery charging control method based on artificial intelligence in any one of claims 1-6.

Images