Load Matching Optimization Method for Multi-Channel Gate Driver and Related Devices

Abstract

The present invention relates to a load matching optimization method for a multi-channel gate driver and related devices. The method includes: collecting gate current, gate voltage, and switching frequency of a multi-channel parallel gate driver to obtain driver load operation data; performing regional division processing on the operation space to obtain a safe operation region and a transition region; inputting into a first adaptive prediction model for feature extraction to obtain a load balance prediction result; performing parallel optimization calculation on the control strategy to obtain control parameters for each drive channel, where the control parameters for each drive channel include a gate drive voltage adjustment amount and a drive timing adjustment amount; performing real-time adjustment on the control parameters for each drive channel to obtain a load balance adjustment instruction; performing performance index evaluation calculation to obtain performance optimization parameters, and updating the first adaptive prediction model to obtain a second adaptive prediction model. The present invention realizes dynamic balance of loads on each channel through coordinated control of voltage regulation and timing compensation.

Description

Technical Field

[0001] The present invention relates to the technical field of gate drivers, and particularly to a method for optimizing load matching of a multi-channel gate driver and related devices. Background Art

[0002] With the rapid development of high-power frequency converter technology, multi-channel parallel gate drive technology has become an important means to improve power density and switching performance. However, in practical applications, due to factors such as device parameter discreteness, wiring asymmetry, and uneven temperature distribution, there are often load imbalance problems between drive channels, resulting in a decrease in drive ability and an increase in switching losses, seriously affecting the reliability of the system.

[0003] Traditional load balancing methods mainly rely on static compensation and empirical parameter adjustment, and it is difficult to cope with load fluctuations under dynamic working conditions. At the same time, there are complex electromagnetic coupling and thermal coupling effects between drive channels, and these coupling effects will change dynamically with the change of switching frequency and load conditions, bringing great challenges to load balancing control. Summary of the Invention

[0004] The main object of the present invention is to provide a method for optimizing load matching of a multi-channel gate driver and related devices. The present invention realizes the dynamic balance of loads of each channel through the coordinated control of voltage regulation and timing compensation.

[0005] To achieve the above object, the present invention provides a method for optimizing load matching of a multi-channel gate driver, including the following steps:

[0006] Collect the gate current, gate voltage, and switching frequency of the multi-channel parallel gate driver to obtain the drive load operation data, and the drive load operation data includes impedance parameters, coupling parameters, and temperature parameters;

[0007] Based on the drive load operation data, perform regional division processing on the operation space to obtain a safe operation area and a transition area;

[0008] Input the drive load operation data into a first adaptive prediction model for feature extraction to obtain a load balance prediction result, and the load balance prediction result includes the predicted value of the load distribution between channels and the predicted value of load fluctuation;

[0009] According to the load balance prediction result, perform parallel optimization calculation on the control strategy to obtain the control parameters of each drive channel, and the control parameters of each drive channel include the gate drive voltage adjustment amount and the drive timing adjustment amount;

[0010] Perform real-time adjustment on the control parameters of each drive channel to obtain a load balance adjustment instruction;

[0011] Perform performance index evaluation and calculation based on the load balancing adjustment instruction to obtain performance optimization parameters, and update the first adaptive prediction model according to the performance optimization parameters to obtain a second adaptive prediction model.

[0012] The present invention also provides a load matching optimization system for a multi-channel gate driver, including:

[0013] An acquisition module, configured to acquire the gate current, gate voltage, and switching frequency of a multi-channel parallel gate driver to obtain driver load operation data, where the driver load operation data includes impedance parameters, coupling parameters, and temperature parameters;

[0014] A region division module, configured to perform region division processing on the operation space based on the driver load operation data to obtain a safe operation region and a transition region;

[0015] A feature extraction module, configured to input the driver load operation data into a first adaptive prediction model for feature extraction to obtain a load balancing prediction result, where the load balancing prediction result includes an inter-channel load distribution prediction value and a load fluctuation prediction value;

[0016] A calculation module, configured to perform parallel optimization calculation on the control strategy according to the load balancing prediction result to obtain control parameters for each drive channel, where the control parameters for each drive channel include a gate drive voltage adjustment amount and a drive timing adjustment amount;

[0017] An adjustment module, configured to perform real-time adjustment on the control parameters for each drive channel to obtain a load balancing adjustment instruction;

[0018] An update module, configured to perform performance index evaluation and calculation based on the load balancing adjustment instruction to obtain performance optimization parameters, and update the first adaptive prediction model according to the performance optimization parameters to obtain a second adaptive prediction model.

[0019] The present invention also provides a computer device, including a memory and a processor, where a computer program is stored in the memory, and when the processor executes the computer program, the steps of the method described in any one of the above are implemented.

[0020] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in any one of the above are implemented.

[0021] In summary, the technical solution provided by the present invention accurately obtains the load operation characteristics of a multi-channel parallel gate driver by adopting high-precision data acquisition and parameter extraction methods, combined with the comprehensive analysis of impedance parameters, coupling parameters, and temperature parameters. Based on the regional division and dynamic boundary optimization of the operation space, a determination mechanism for the safe operation area and the transition area is established, effectively preventing the system from entering an unsafe working state; an adaptive prediction model is used for load feature extraction, and through multi-level feature decoupling and fusion, the load distribution and fluctuation trend between channels are accurately predicted, providing a decision basis for the optimization of control strategies; a parallel optimization algorithm is introduced to perform real-time calculations on control strategies, and through the coordinated control of voltage regulation and timing compensation, the dynamic balance of the loads of each channel is achieved; a hierarchical control mechanism and a priority scheduling strategy are designed to ensure the smooth execution of control instructions and avoid system instability caused by mutations during the adjustment process; a complete performance evaluation and model update mechanism is established, and through online learning and parameter adaptive adjustment, the performance of the prediction model is continuously optimized to ensure the long-term stable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a schematic diagram of the steps of a load matching optimization method for a multi-channel gate driver according to an embodiment of the present invention;

[0023] Figure 2 is a block diagram of the structure of a load matching optimization system for a multi-channel gate driver according to an embodiment of the present invention;

[0024] Figure 3 is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0025] The implementation, functional features, and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] In order to make the object, technical solution, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0027] Referring to Figure 1 , this embodiment provides a load matching optimization method for a multi-channel gate driver, including the following steps:

[0028] S1. Collect the gate current, gate voltage, and switching frequency of the multi-channel parallel gate driver to obtain the drive load operation data, where the drive load operation data includes impedance parameters, coupling parameters, and temperature parameters;

[0029] Among them, the operating parameters of the multi-channel parallel gate driver are collected and processed, mainly including the measurement of gate current, gate voltage and switching frequency, so as to obtain the load operation data of the driver. For the sampling of the gate current, a high-precision current sensor and a high-speed data acquisition module are used to capture the current changes of the multi-channel gate driver at different time points, forming a gate current sampling sequence. This sequence represents the dynamic characteristics of the current over time in a digital way. At the same time, the gate voltage is sampled by configuring a high-bandwidth voltage probe and a precise sampling circuit, and the voltage signal of the multi-channel gate driver is converted into a digital gate voltage sampling sequence. This sequence records the voltage waveform changes of each drive channel and is used for subsequent calculation of resistance, capacitance and inductance parameters. Based on the collected gate current sampling sequence and gate voltage sampling sequence, through the formula (where is impedance, and are voltage and current respectively) the impedance parameters of the driver are calculated. To refine the results, the gate resistance value, Miller capacitance value and parasitic inductance value are extracted by combining the circuit model. The gate resistance reflects the power loss characteristics of the driver, the Miller capacitance describes the phase relationship between voltage and current, and the parasitic inductance reveals the dynamic characteristics of the circuit under high-frequency conditions. After obtaining the impedance parameters of a single drive channel, the coupling calculation is performed on adjacent drive channels. The coupling calculation includes the analysis of electromagnetic coupling and thermal coupling. The calculation of the electromagnetic coupling coefficient is realized by measuring the mutual inductance and capacitance values between adjacent channels, and using the formula (where is the electromagnetic coupling coefficient, is the mutual inductance, and are the self-inductances of adjacent channels respectively) to quantify the degree of electromagnetic interference between channels. The calculation of the thermal coupling coefficient is achieved by establishing a heat conduction model, combining the heat power consumption and temperature rise of each channel, and analyzing the heat transfer rate and influence degree of heat energy between adjacent channels. These coupling parameters help to identify the interference and mismatch problems that occur during the operation of the multi-channel driver. The temperature of each drive channel of the multi-channel parallel gate driver is monitored. By arranging high-precision temperature sensors at key nodes of each channel, the temperature changes are continuously recorded to form a temperature monitoring data sequence. Based on this sequence, the temperature distribution of each drive channel is calculated by using the spatial interpolation method and the heat distribution model to obtain specific temperature parameters. The temperature parameters include the temperature values of each channel and also include the temperature gradient value, that is, the change rate of temperature between channels. This information can reflect the thermal stability and heat dissipation balance of the driver. The impedance parameters, coupling parameters and temperature parameters are subjected to data standardization processing. Through the data normalization method, the data with different dimensions are converted into dimensionless values to generate the load operation data of the driver.

[0030] S2. Perform regional division processing on the operating space based on the drive load operating data to obtain a safe operating area and a transition area;

[0031] Specifically, analyze the impedance parameters in the drive load operating data, extract the key characteristics of each drive channel, and calculate its maximum allowable power value. By combining the impedance characteristics such as resistance, capacitance, and inductance of each channel, as well as the dynamic current and voltage data during operation, based on the formula (where is the maximum allowable power, and are the gate voltage and impedance respectively) to obtain the power limit of each channel. On this basis, construct a load deviation threshold matrix using the power deviation relationship between channels. This matrix is used to describe the load deviation degree of each channel and provides a basic basis for delimiting the operating area. Through threshold optimization of the load deviation threshold matrix, the boundary threshold of the operating area is obtained. Based on optimization algorithms, such as through linear programming or non-linear optimization techniques, map the data in the deviation matrix to a reasonable boundary range to ensure that the divided safe area can meet the requirements of operating stability. The operating area boundary threshold is used to mark the safe range of the load at this stage and simultaneously identify the boundary conditions for entering the transition area. Based on the coupling parameters in the drive load operating data, quantitatively calculate the interference degree between each drive channel to generate a channel coupling strength matrix. Analyze key parameters such as the electromagnetic coupling coefficient and the thermal coupling coefficient. Through the formula (where is the coupling strength between channels and , and Quantify the interference intensity (which are the electric field and magnetic field intensities between two channels respectively), and represent the mutual coupling relationship of all channels in matrix form. The channel coupling intensity matrix can reflect the interaction characteristics of the driver under different operating states. Use this matrix to correct the boundary threshold of the operating region to ensure that the influence of interference between channels is taken into account. After completing the coupling correction, introduce the temperature parameter to optimize the boundary threshold of the operating region. Based on the temperature information in the driver load operation data, perform thermal stress analysis on the temperature distribution of each drive channel to calculate the thermal stress distribution matrix. The calculation of the thermal stress distribution is based on the mechanics of materials and heat conduction models, combined with actual temperature monitoring data, to analyze the influence of the temperature gradient on the channel reliability. By converting the temperature gradient value into a stress distribution matrix, implement temperature compensation for the boundary threshold of the operating region, so as to more accurately reflect the safe range under actual operating conditions. Based on the boundary threshold of the operating region and the load deviation threshold matrix, initially divide the initial boundary of the operating space. Extract multi-dimensional features from the initial boundary of the operating space to comprehensively describe the characteristics of the operating space. By calculating indicators such as load balance degree, temperature uniformity, and coupling strength, extract the feature vector of the operating space. Among them, the load balance degree represents the uniformity of the load distribution among channels, the temperature uniformity reflects the thermal distribution state of the driver, and the coupling strength describes the degree of interference between channels. Optimize the region division result according to the feature vector of the operating space. By adjusting boundary conditions, re-evaluating feature weights, etc., obtain the optimized operating space division result. Based on the optimized operating space division result, by real-time monitoring the driver's load operation data, dynamically match it with the optimized region division result to determine whether the current operating state is in the safe region or the transition region. If the system operating state is in the safe region, the driver can maintain efficient and stable operation; if it enters the transition region, trigger the warning mechanism and start further control strategy adjustment to prevent system instability.

[0032] S3. Input the driver load operation data into the first adaptive prediction model for feature extraction to obtain the load balance prediction result. The load balance prediction result includes the predicted values of the load distribution between channels and the load fluctuation prediction value.

[0033] It should be noted that the driver load operating data is input into the load decoupling layer of the first adaptive prediction model. This layer uses an autoencoder structure to perform nonlinear separation of the impedance, coupling, and temperature parameters in the driver operating data. This separation process, through the autoencoder's hidden layer embedding operation, decomposes the original complex multidimensional input data into a set of load feature components with independent physical meaning, effectively eliminating redundant information in the data while retaining key features. The load feature components are then input into the multi-channel complementary feature extraction layer of the first adaptive prediction model. In this layer, the model analyzes the independent characteristics of each driver channel using parallel channel feature subnetworks. Each subnetwork uses a residual connection structure to avoid the vanishing gradient problem and improve feature extraction efficiency. Furthermore, to capture inter-channel correlation information, these subnetworks communicate through lateral information transfer paths, allowing the characteristics of a channel to dynamically adjust to the status of other channels, thereby extracting more comprehensive channel correlation features. Channel correlation features can reflect the mutual coupling between driver channels during operation. The channel correlation features are then input into the dynamic coupling modeling layer of the first adaptive prediction model, which is designed using a graph convolutional network structure. Graph convolutional networks (GCNs) can model complex multi-channel relationships based on node and edge representations. They dynamically construct a correlation graph between channels to capture the load coupling effects of the drive. Through graph convolution operations, the model extracts global and local load characteristics within different graph structures, generating coupling dynamic features that reflect the coupling relationships of multi-channel systems under complex operating conditions. These coupling dynamic features are input into a multi-scale temporal feature fusion layer, which comprises multiple parallel groups of long short-term memory units. Each group processes features at a specific time scale, enabling the extraction of load temporal characteristics from both short-term fluctuations and long-term trends. A cross-scale attention mechanism weightedly fuses features at different time scales, ensuring a comprehensive and accurate description of temporal dynamics. The fused multi-scale temporal features reflect the temporal variations of the drive load. In the thermal-electric coupling analysis layer, the model employs a two-stream network structure to process thermal and electrical features separately. The thermal feature stream focuses on the distribution and variation of temperature parameters, analyzing the drive's operating performance under thermal conditions; the electrical feature stream analyzes the dynamic characteristics of the electrical load based on impedance and coupling parameters. Through an adaptive weight fusion, these two feature streams are merged into a unified thermoelectric coupling feature, achieving a comprehensive description of the operating state of the multi-channel drive under thermal and electrical conditions. The thermoelectric coupling feature is input into the distribution prediction layer and the fluctuation analysis layer, respectively. In the distribution prediction layer, the model uses a conditional variational autoencoder structure to learn the implicit distribution of the current load state to generate a probabilistic model of the inter-channel load distribution and obtain a predicted value for the inter-channel load distribution. This value describes the load distribution across the channels in a probabilistic manner. In the fluctuation analysis layer, the model uses a wavelet neural network structure to capture load fluctuations of different frequencies through multi-resolution analysis to generate a load fluctuation prediction value.The advantage of the wavelet neural network lies in its ability to simultaneously process high-frequency short-term fluctuations and low-frequency long-term trends. The predicted values of the load distribution between channels and the predicted values of the load fluctuations are integrated into the load balance prediction result, which reflects the current load distribution state and provides the ability to warn of future fluctuations.

[0034] S4. According to the load balance prediction result, perform parallel optimization calculations on the control strategy to obtain the control parameters of each drive channel. The control parameters of each drive channel include the gate drive voltage adjustment amount and the drive timing adjustment amount;

[0035] Specifically, the load distribution prediction values between channels are extracted from the load balancing prediction results, and the load imbalance of each driving channel is calculated through a data parsing method. The load imbalance reflects the degree to which the load of each driving channel deviates from the balanced state and is an important input variable for optimizing the driving voltage and timing control. Based on the load imbalance, the voltage compensation value required for each channel is calculated to adjust the voltage deviation of the channel load and improve its load balancing performance. At the same time, a dynamic compensation model is established based on the load fluctuation prediction value. The core goal of this model is to design a driving timing compensation strategy to cope with the dynamic load changes between different channels. The driving timing compensation strategy specifically includes the dead time adjustment amount and the switch delay compensation amount. The dead time adjustment amount is used to optimize the safety interval time of the power device during the switching process, while the switch delay compensation amount reduces the timing deviation caused by load fluctuations by fine-tuning the trigger time of the channel switch. These dynamic compensation strategies ensure the timing stability and load response ability of the driver during operation. The calculated load imbalance is input into the weight allocator to generate the voltage regulation coefficient matrix of each driving channel. This matrix determines the voltage regulation range of each channel through weighted calculation and provides a reference for the specific adjustment of the gate driving voltage. At the same time, the driving timing compensation strategy is decomposed in parallel to obtain the channel-level timing control sequence. The channel-level timing control sequence cooperates with the voltage regulation coefficient matrix to jointly optimize the voltage control amount and timing control amount of each channel, enabling it to achieve consistent dynamic performance under complex multi-channel load conditions. Orthogonal optimization is performed based on the voltage regulation coefficient matrix and the channel-level timing control sequence, and a multi-channel cooperative control matrix is generated through a multi-objective optimization algorithm. This matrix contains both the voltage control amount and the timing control amount and is used to comprehensively describe the optimization control strategy of the driver. In this process, the role of the orthogonal optimization algorithm is to ensure that the mutual influence between voltage and timing can be effectively processed, making the generated control strategy have optimal coordination globally. The multi-channel cooperative control matrix is input into the constraint processor to perform boundary constraint processing on the control parameters. The boundary constraints include the upper and lower limits of the voltage amplitude and the minimum and maximum allowable ranges of the timing interval, thus preventing the control parameters from exceeding the physical limits of the actual hardware. After the constraint processing, the initial values of the generated control parameters meet the basic hardware operation requirements. The initial values of the control parameters are optimized through multi-objective iteration. Through the multi-objective optimization algorithm, a balance is sought between the load balancing degree and the switching loss, ensuring that both the balanced distribution of the load between channels and the reduction of the power loss during the switching process can be achieved. The result of the iterative optimization is a set of optimal control parameter combinations, including the voltage adjustment amount and timing adjustment amount of each channel. Based on the optimal control parameter combination, the gate driving voltage and driving timing are allocated to specific driving units according to channels to obtain the control parameters of each driving channel.

[0036] S5. Real-time adjustment is performed on the control parameters of each driving channel to obtain a load balancing adjustment instruction;

[0037] Among them, the control parameters of each drive channel are monitored for their status. By collecting in real time the dynamic operation data such as the current, voltage, temperature, and timing of the driver, a real-time operation status matrix is constructed. This matrix includes the current load balance degree and temperature distribution data, which respectively reflect the balance of the multi-channel load distribution and the thermal status of each channel. Based on the real-time operation status matrix, the working area of the system is determined. By analyzing the matrix data, the current operation parameters are compared with the predefined safety operation area threshold to generate a region identification signal. The region identification signal is used to determine whether the system is in the safe operation area. If the system parameters exceed the boundary of the safe operation area, the identification signal will prompt to enter the transition area, triggering the adjustment of the corresponding control mode. The region identification signal is input into the control strategy selector to generate a control mode signal. The control mode signal is dynamically adjusted according to the current operation status of the system, including two types: the steady-state control mode and the fast response mode. When the system is in the safe operation area, the steady-state control mode will maintain the current control strategy and finely adjust the voltage and timing to maintain load balance; while when the system enters the transition area, the fast response mode will be immediately activated to quickly adjust the key parameters to avoid performance degradation or instability. According to the control mode signal, the control parameters of each drive channel are classified and processed, thereby generating a hierarchical control sequence. The hierarchical control sequence includes a voltage regulation sequence and a timing regulation sequence. Among them, the voltage regulation sequence calculates the adjustment amount of the gate voltage according to the load deviation of each channel, and the timing regulation sequence optimizes the switching timing of the drive signal according to the dynamic response requirements of the system. The classification and processing ensure that different types of control parameters can be optimized separately and solve different load imbalance problems targeted. The hierarchical control sequence is sorted by priority to generate a control execution order table. The control execution order table sorts the adjustment instructions according to the urgency of load balance, and places the adjustment instructions that have the greatest impact on the system performance in the position of priority execution. The judgment of the urgency is based on the magnitude of the load imbalance in the real-time operation status matrix and the change rate of the temperature gradient, so as to ensure that the most urgent problems can be solved in time. The control execution order table is input into the actuator scheduling module to generate a specific timing execution plan. The timing execution plan determines the specific execution time of the control instructions of each drive channel by comprehensively considering the hardware response time, the dependency relationship of the adjustment instructions, and the coordination among multiple channels. The formulation of this plan needs to balance the requirements of fast response and stable transition, avoiding conflicts caused by excessive synchronous execution while ensuring that the adjustment can take effect as soon as possible. Buffer management is performed on the timing execution plan to generate an instruction cache queue. The instruction cache queue realizes the smooth switching of instructions through the storage and scheduling of the adjustment instructions, thereby avoiding the instability of the system caused by frequent parameter adjustments. Based on the instruction cache queue, the adjustment instructions are synthesized, integrating the voltage and timing adjustment information of all channels to generate a load balance adjustment instruction.

[0038] S6, perform performance metric evaluation and calculation based on the load balancing adjustment instruction to obtain performance optimization parameters, and update the first adaptive prediction model according to the performance optimization parameters to obtain a second adaptive prediction model.

[0039] Specifically, the performance of the load balancing adjustment instruction is evaluated, and the performance index matrix is calculated by collecting the changes of key parameters of the system after adjustment. The performance index matrix includes three core dimensions: the load balancing degree index, the temperature consistency index, and the response time index. The load balancing degree index reflects the evenness of the load distribution among drive channels and is a direct manifestation of evaluating the load balancing optimization effect; the temperature consistency index describes the thermal distribution difference among channels and is a key measure of the system's thermal stability; the response time index is used to evaluate the time required for the system to achieve load adjustment from receiving the adjustment instruction to verify the real-time performance and efficiency of the adjustment. Based on the performance index matrix, model error analysis is carried out to quantify the prediction accuracy of the first adaptive prediction model under actual load conditions. By comparing the model prediction output with the actual operation results, a prediction error distribution is generated. The prediction error distribution includes two parts: the load distribution prediction error and the fluctuation prediction error. Among them, the load distribution prediction error reflects the prediction deviation of the model for the load balancing situation of each channel, while the fluctuation prediction error reflects the insufficient prediction ability of the model for the dynamic changes of the load. By analyzing these errors, the performance bottlenecks and deficiencies of the model in different scenarios are identified, providing a clear direction for subsequent optimization. The prediction error distribution is input into the adaptive corrector, and the model parameter correction amount is calculated. The adaptive corrector uses an error feedback mechanism to adjust the network parameters of each layer of the model specifically to improve the prediction ability of the model. The model parameter correction amount provides a specific adjustment basis for each layer and updates the parameters of each layer differently according to the characteristics of different error distributions. The corrected parameters are used to optimize the overall performance of the model to make it more adaptable to the actual operation requirements of the driver. Based on the model parameter correction amount, the load decoupling layer of the first adaptive prediction model is updated to optimize its load feature extraction structure. The load decoupling layer is the basic part of the whole model and realizes the non-linear separation of the drive operation data through an autoencoder structure. During the update process, the corrected parameters will improve the decomposition accuracy of the decoupling layer for impedance parameters, coupling parameters, and temperature parameters, generating more accurate load feature components. The dynamic weight adjustment of the multi-channel complementary feature extraction layer of the first adaptive prediction model is carried out to optimize the channel feature network. The multi-channel complementary feature extraction layer extracts the independent characteristics of each drive channel through parallel channel feature sub-networks and captures the mutual influence between channels through horizontal information transfer. Through dynamic weight adjustment, the information transfer weights between sub-networks are optimized according to the feedback of the performance index matrix, thereby improving the complementary ability and robustness of the channel feature network under complex operation conditions. Based on the performance index matrix, the graph structure of the dynamic coupling modeling layer of the first adaptive prediction model is optimized. The dynamic coupling modeling layer models the coupling effect between channels through a graph convolutional network structure, and its performance directly determines the model's prediction ability for load dynamic changes.By adjusting the node representation and edge weights in the graph structure, the optimized coupled modeling network can more accurately capture the correlation characteristics between channels, thereby improving the modeling accuracy of the system's dynamic behavior. Integrate the updated load feature extraction structure, the optimized channel feature network, and the updated coupled modeling network to construct the second adaptive prediction model. With the support of parameter optimization and structure adjustment at each layer, the updated model can more accurately reflect the complex operating state of the multi-channel gate driver.

[0040] In one example, the gate current, gate voltage, and switching frequency of a multi-channel parallel gate driver are collected to obtain the drive load operation data, which includes impedance parameters, coupling parameters, and temperature parameters, including:

[0041] Sample the gate current of the multi-channel parallel gate driver to obtain a gate current sampling sequence, and sample the gate voltage of the multi-channel parallel gate driver to obtain a gate voltage sampling sequence;

[0042] Based on the gate current sampling sequence and the gate voltage sampling sequence, perform impedance calculation to obtain impedance parameters, which include gate resistance value, Miller capacitance value, and parasitic inductance value;

[0043] Perform coupling calculation on the impedance parameters of adjacent drive channels to obtain coupling parameters, which include electromagnetic coupling coefficient and thermal coupling coefficient;

[0044] Monitor the temperature of each drive channel of the multi-channel parallel gate driver to obtain a temperature monitoring data sequence, and perform temperature distribution calculation based on the temperature monitoring data sequence to obtain temperature parameters, which include the temperature value of each drive channel and the temperature gradient value;

[0045] Perform data standardization processing on the impedance parameters, coupling parameters, and temperature parameters to obtain the drive load operation data.

[0046] In this example, the gate current of the multi-channel parallel gate driver is sampled by an accurate sensor system. Assume that the sampled gate current signal is , representing the gate current value at time . After analog-to-digital conversion, this current signal forms a digital gate current sampling sequence , where each represents the current value at the discrete time point . At the same time, use a high-precision voltage probe to sample the gate voltage to obtain a sampling signal , which also undergoes analog-to-digital conversion to generate a gate voltage sampling sequence , where represents the discrete time point The voltage value. Using the sampled gate current and voltage data, impedance calculations are performed to obtain the impedance parameters of the driver. According to the impedance definition , where is the impedance value, is the voltage value, is the current value. By calculating point by point a sequence of impedance values at corresponding time points is obtained . To extract specific impedance parameters, the impedance is decomposed into the gate resistance , the Miller capacitance and the parasitic inductance using a circuit model. The gate resistance is obtained by steady-state measurement as , where and represent the average values of voltage and current respectively. The Miller capacitance is calculated by frequency response analysis as , where represents the change in gate charge, represents the change in gate voltage. The parasitic inductance is derived using the phase relationship between high-frequency voltage and current as , where is the change in voltage across the inductor, is the change in the rate of change of current. After obtaining the impedance parameters of a single channel, the coupling relationship between adjacent drive channels is analyzed, and the electromagnetic coupling coefficient and the thermal coupling coefficient are calculated. The electromagnetic coupling coefficient is defined by analyzing the ratio of the mutual inductance between channel and channel to the self-inductances , as , where is the mutual inductance between channels, representing the magnetic field effect generated by the current in channel on channel . The thermal coupling coefficient is calculated based on the heat conduction equation, through the temperature gradient between adjacent channels and the heat transfer power as , where is the heat flow power from channel to channel . By means of temperature sensors installed on each drive channel, the temperature data of the driver is monitored in real time to generate a temperature monitoring data sequence , where represents the temperature value of channel . The temperature gradient is calculated by analyzing the spatial distribution of the temperature data, thereby generating temperature parameters, including the temperature value and temperature gradient value of each channel. These temperature parameters can intuitively reflect the thermal distribution state and thermal imbalance degree of the multi-channel system. The obtained impedance parameters, coupling parameters, and temperature parameters are subjected to data normalization processing to generate drive load operation data. Suppose a certain parameter set is , and the normalization processing is carried out through the linear normalization formula , where represents the value after normalization, and represent the minimum and maximum values of the original data respectively. Through processing, parameters with different dimensions and ranges are unified into the range of [0,1].

[0047] In an example, based on the drive load operation data, the operation space is divided into regions to obtain a safe operation region and a transition region, including:

[0048] Analyze the impedance parameters in the drive load operation data to obtain the maximum allowable power value of each drive channel, and calculate the load deviation threshold matrix based on the maximum allowable power value;

[0049] Based on the load deviation threshold matrix, perform threshold optimization to obtain the operation region boundary threshold, and the operation region boundary threshold is used to divide the safe operation region and the transition region;

[0050] Quantitatively calculate the interference degree between each drive channel based on the coupling parameters in the drive load operation data to obtain the channel coupling strength matrix, and the channel coupling strength matrix is used to correct the operation region boundary threshold;

[0051] According to the temperature parameters in the drive load operation data, perform thermal stress analysis on the temperature distribution of each drive channel to obtain the thermal stress distribution matrix, and perform temperature compensation on the operation region boundary threshold based on the thermal stress distribution matrix;

[0052] Based on the operation region boundary threshold and the load deviation threshold matrix, perform region division to obtain the initial operation space boundary, and perform multi-dimensional feature extraction on the initial operation space boundary to obtain the operation space feature vector. The operation space feature vector includes load balance degree, temperature uniformity, and coupling strength;

[0053] According to the operation space feature vector, perform region optimization to obtain the optimized operation space division result, and based on the optimized operation space division result, perform real-time determination on the operation state to obtain the safe operation region and the transition region.

[0054] In this example, by analyzing the impedance parameters in the drive load operation data, the maximum allowable power value of each drive channel is obtained. The maximum allowable power value It is calculated based on the impedance parameters of the channels and reflects the maximum power load that the channels can withstand. The formula is

[0055] ;

[0056] Among them, is the maximum allowable voltage of the channel , is the gate resistance of the channel , is the switching frequency, and is the Miller capacitance. By calculating the of each channel, the maximum power carrying capacity of the system is obtained, providing boundary conditions for load balancing and safe operation. Based on the maximum allowable power value, the load deviation threshold matrix is calculated. Assume that there are channels in the system, and the actual load power value is . Each element of the deviation threshold matrix is defined as:

[0057] ;

[0058] Among them, represents the relative power deviation between channel and channel . By constructing the matrix , the balance degree of load distribution between different channels is quantified. If the matrix element exceeds the preset threshold, it means that the system enters the transition region. The threshold of the load deviation threshold matrix is optimized to obtain the boundary threshold of the operating region. The optimization process of the boundary threshold is based on the optimization algorithm. For example, the threshold combination that can maximize the safe operating space is found through linear programming. Assume that the optimization goal is to minimize the proportion of the transition region. The optimization problem is described as:

[0059] ;

[0060] Among them, is the indicator function, which takes the value of 1 when the condition is satisfied and 0 otherwise; is the boundary threshold variable, and its optimal value is determined through optimization calculation. At the same time, based on the coupling parameters in the drive load operation data, the interference degree between each drive channel is quantified and calculated to generate the channel coupling strength matrix . The electromagnetic coupling strength between channels is calculated through the formula:

[0061] ;

[0062] is calculated, where is the mutual inductance between channel and , and are the self-inductances of the channels, respectively. Meanwhile, the thermal coupling strength is calculated by the formula:

[0063] ;

[0064] where is the heat flow between channels and and is the temperature difference between the two channels. The electromagnetic coupling and thermal coupling strengths together constitute the coupling strength matrix , which is used to correct the boundary threshold of the operating region. To enhance the accuracy of the operating region division, thermal stress analysis is performed on the temperature parameters in the drive load operating data to generate the thermal stress distribution matrix . The calculation of thermal stress is based on the thermoelastic theory and is calculated by the formula

[0065] ;

[0066] where is the coefficient of thermal expansion of the material, is Young's modulus, is the temperature gradient. The thermal stress distribution matrix can quantify the thermal load status of each channel and perform temperature compensation on the boundary threshold of the operating region. Through the above calculations, based on the boundary threshold of the operating region and the load deviation threshold matrix, the region division is completed to obtain the initial boundary of the operating space. On this basis, multi-dimensional feature extraction is performed on the initial boundary of the operating space to generate the operating space feature vector , where is the load balance degree, is the temperature uniformity, is the coupling strength. These feature vectors can comprehensively describe the operating state of the system. According to the operating space feature vectors, region optimization is performed to obtain the optimized operating space division result. By real-time determining the operating state, the system operating state is divided into a safe operating region or a transition region.

[0067] In one example, the drive load operating data is input into the first adaptive prediction model for feature extraction to obtain the load balance prediction result, and the load balance prediction result includes the predicted values of the load distribution between channels and the predicted value of load fluctuation, including:

[0068] The drive load operating data is input into the load decoupling layer of the first adaptive prediction model, and the load decoupling layer adopts an autoencoder structure to non-linearly separate the impedance parameters, coupling parameters, and temperature parameters to obtain the load feature components;

[0069] The load feature components are input into the multi-channel complementary feature extraction layer of the first adaptive prediction model. The multi-channel complementary feature extraction layer includes parallel channel feature sub-networks. Each channel feature sub-network adopts a residual connection structure and establishes a horizontal information transmission path between channels to obtain channel correlation features.

[0070] The channel correlation features are input into the dynamic coupling modeling layer of the first adaptive prediction model. The dynamic coupling modeling layer adopts a graph convolutional network structure to capture the load coupling effect by dynamically constructing the correlation map between channels and obtain the coupling dynamic features.

[0071] The coupled dynamic features are input into the multi-scale temporal feature fusion layer of the first adaptive prediction model. The multi-scale temporal feature fusion layer contains parallel groups of long short-term memory units. Each group of units processes features at different time scales and fuses the features through a cross-scale attention mechanism to obtain multi-scale temporal features.

[0072] The multi-scale time series features are input into the thermoelectric coupling analysis layer of the first adaptive prediction model. The thermoelectric coupling analysis layer adopts a dual-stream network structure to process thermal features and electrical features respectively, and merges the two feature streams through an adaptive weight fusion to obtain thermoelectric coupling features.

[0073] The thermoelectric coupling feature is input into the distribution prediction layer of the first adaptive prediction model. The distribution prediction layer adopts a conditional variational autoencoder structure to generate a probability model of the inter-channel load distribution based on the current load state to obtain the inter-channel load distribution prediction value;

[0074] The thermoelectric coupling characteristics are input into the fluctuation analysis layer of the first adaptive prediction model. The fluctuation analysis layer adopts a wavelet neural network structure to capture the load fluctuation characteristics of different frequencies through multi-resolution analysis to obtain the load fluctuation prediction value.

[0075] Output the load balancing prediction result based on the load distribution prediction value and load fluctuation prediction value between channels.

[0076] In this example, the load operation data of the driver is input into the load decoupling layer of the first adaptive prediction model. This layer uses an autoencoder structure to reduce the dimension and perform nonlinear separation on complex multi-dimensional input data. The load operation data of the driver consists of impedance parameters, coupling parameters, and temperature parameters. Assume that the input data is a matrix ,in , respectively represent impedance related data (such as gate resistance , Miller capacitance , parasitic inductance ), coupling parameters (electromagnetic coupling coefficient , thermal coupling coefficient ) and temperature parameters (temperature distribution of each channel )。The encoder part of the autoencoder compresses the input data into the latent space through a non - linear mapping function, and the formula is , where are the load feature components, is the weight matrix of the encoder. The decoder part reconstructs the input data through the function to verify the accuracy of feature extraction. The extracted by the encoder can capture the non - linear features of the input data. The load feature components are input into the multi - channel complementary feature extraction layer, which is composed of multiple parallel channel feature sub - networks, and each sub - network processes the data of one driving channel respectively. To avoid the problem of gradient vanishing and improve the feature extraction efficiency, the sub - network adopts a residual connection structure, and its formula is , where is the feature representation of the layer, is the weight matrix is the activation function. To capture the dynamic correlation between channels, the channel feature sub - networks establish interactions through a horizontal information transfer path, and the weights of the interactions are represented by the shared parameter , generating the channel - associated feature matrix . Each element in the matrix represents the feature correlation between channel and channel , and can reflect the complementarity of the load distribution. The channel - associated feature is input into the dynamic coupling modeling layer, which dynamically constructs the association graph between channels through the graph convolutional network structure to capture the complex load coupling effect. The core formula of the graph convolutional network is , where is the node feature matrix of the layer, is the adjacency matrix of the graph, is the layer weight matrix, is the activation function. Through multi - layer graph convolutional operations, the model can effectively aggregate the neighborhood information of nodes, extract the dynamic association features between channels, and output the coupled dynamic features . The coupled dynamic features are input into the multi - scale temporal feature fusion layer, which processes features of different time scales through a parallel group of long short - term memory networks (LSTM). The state update formula of each LSTM unit is , where is the hidden state at time , and They are state weights and input weights respectively. By running multiple LSTMs in parallel, different units process short-term dynamic and long-term trend features. These features are weighted and fused through a cross-scale attention mechanism to generate multi-scale time series features. , which can comprehensively reflect the time dynamics of the load. Multi-scale time series characteristics It is transferred to the thermoelectric coupling analysis layer. This layer uses a dual-flow network structure to process and fuse thermal and electrical features separately. and current characteristics Extracted through independent sub-networks, the formula is ,in is the temperature data, The heat flow and current features are merged into a thermoelectric coupling feature through an adaptive weight fuser. , providing input for subsequent predictions. Input the distribution prediction layer and the fluctuation analysis layer respectively. In the distribution prediction layer, the conditional variational autoencoder generates a probability model of the load distribution by learning the potential distribution, and its formula is ,in is a hidden variable. The distribution prediction layer outputs the load distribution prediction value between channels to describe the distribution of load between channels. The fluctuation analysis layer captures load fluctuations of different frequencies based on wavelet neural network. Its formula is ,in is the wavelet transform function, is the wavelet kernel parameter, which outputs the load fluctuation prediction value and characterizes the dynamic behavior of the system. The inter-channel load distribution prediction value and the load fluctuation prediction value are combined as the output of the model to generate the load balancing prediction result.

[0077] In one example, a control strategy is optimized and calculated in parallel based on the load balancing prediction result to obtain control parameters of each drive channel. The control parameters of each drive channel include a gate drive voltage adjustment amount and a drive timing adjustment amount, including:

[0078] Perform data analysis on the inter-channel load distribution prediction value in the load balancing prediction result to obtain the load imbalance of each drive channel. The load imbalance is used to calculate the drive voltage compensation value;

[0079] A dynamic compensation model is established based on the load fluctuation prediction value to obtain a drive timing compensation strategy, which includes a dead time adjustment amount and a switch delay compensation amount.

[0080] The load imbalance is input into the weight distributor to obtain the voltage adjustment coefficient matrix of each driving channel. The voltage adjustment coefficient matrix is used to calculate the adjustment range of the gate driving voltage.

[0081] Parallel decomposition is performed on the driving timing compensation strategy to obtain a channel-level timing control sequence, and the channel-level timing control sequence cooperates with the voltage regulation coefficient matrix;

[0082] Orthogonal optimization is performed based on the voltage regulation coefficient matrix and the channel-level timing control sequence to obtain a multi-channel collaborative control matrix, and the multi-channel collaborative control matrix includes voltage control quantities and timing control quantities;

[0083] The multi-channel collaborative control matrix is input into a constraint processor to perform boundary constraints on the voltage amplitude and timing interval, and an initial value of the control parameter is obtained;

[0084] Multi-objective iterative optimization is performed on the initial value of the control parameter to obtain an optimal control parameter combination, and the optimal control parameter combination meets the balance requirements of the load balance degree and the switching loss;

[0085] Based on the optimal control parameter combination, the gate drive voltage and drive timing of each drive channel are allocated to obtain the control parameters of each drive channel.

[0086] In this example, data parsing is performed on the predicted values of the inter-channel load distribution in the load balance prediction result to extract the load information of each drive channel, so as to calculate the load imbalance amount. Assuming that the predicted load distribution is , where represents the load power of channel . By calculating the deviation of the load of each channel from the global average value, the load imbalance amount is expressed as

[0087] ;

[0088] where is the average value of the load powers of all channels. The load imbalance amount reflects the degree to which the load of each channel deviates from the balanced state and serves as the basis for calculating the drive voltage compensation value. Based on the predicted load fluctuation value , a dynamic compensation model is established for designing the driving timing compensation strategy. The predicted load fluctuation value represents the amplitude of the power fluctuation of channel . By analyzing the fluctuation characteristics, the dynamic compensation model includes the dead time adjustment amount and the switching delay compensation amount , and their calculation formulas are:

[0089] ;

[0090] where and are proportionality coefficients, is the load imbalance amount, is the maximum allowable power of the channel. These adjustment amounts are used to dynamically compensate the switching timings of each channel to reduce the load mismatch caused by fluctuations. The load imbalance is input to the weight allocator to generate the voltage regulation coefficient matrix of each drive channel , where represents the correlation between the voltage regulation coefficient of channel and the load imbalance of channel . The voltage regulation range is calculated by the following formula:

[0091] ;

[0092] where represents the gate drive voltage adjustment amount of channel . The voltage regulation coefficient matrix is optimized through weight allocation to ensure that the voltage adjustment of each channel is consistent with the overall load state of the system. The drive timing compensation strategy is decomposed in parallel to obtain the channel-level timing control sequence of each channel. The timing control sequence includes the switch dead time and the delay compensation time, and cooperates with the voltage regulation coefficient matrix to generate the complete control instruction. The voltage regulation coefficient matrix and the channel-level timing control sequence are input to the orthogonal optimization module, and the multi-channel cooperative control matrix is generated through the multi-objective optimization algorithm, where is the drive voltage control amount of channel , is the dead time adjustment amount, and is the delay compensation amount. The orthogonal optimization realizes the trade-off between load balancing and power consumption minimization through the following objective function:

[0093] ;

[0094] where and are the weight coefficients, which respectively adjust the priorities of the load balancing degree and the switching loss. The multi-channel cooperative control matrix is input to the constraint processor to perform boundary constraints on the voltage amplitude and the timing interval. The constraint conditions include the upper and lower limits of the voltage and the safe range of the timing interval . The initial values of the control parameters generated after processing conform to the operating limits of the physical hardware. Based on the initial values of the control parameters, through multi-objective iterative optimization, the optimal control parameter combination that satisfies the load balancing degree and the switching loss balance is calculated. The optimization algorithm adopts the genetic algorithm or the particle swarm optimization algorithm to iteratively update the parameters and gradually approach the global optimal solution. The final optimization result provides the voltage adjustment amount and the timing adjustment amount of each channel. According to the optimal control parameter combination, the gate drive voltage and the drive timing of each drive channel are allocated to obtain the control parameters of each drive channel.

[0095] In one example, real-time adjustment of the control parameters of each drive channel is performed to obtain a load balancing adjustment instruction, including:

[0096] Monitoring the status of the control parameters of each drive channel to obtain a real-time operating status matrix, where the real-time operating status matrix includes the current load balancing degree and temperature distribution data;

[0097] Determining the working area based on the real-time operating status matrix to obtain a region identification signal, where the region identification signal is used to determine whether the system is in a safe operating region;

[0098] Inputting the region identification signal into a control strategy selector to obtain a control mode signal, where the control mode signal includes a steady-state control mode and a fast response mode;

[0099] Classifying the control parameters of each drive channel based on the control mode signal to obtain a hierarchical control sequence, where the hierarchical control sequence includes a voltage adjustment sequence and a timing adjustment sequence;

[0100] Sorting the hierarchical control sequence by priority to obtain a control execution order table, where the control execution order table determines the execution order of each adjustment instruction according to the load balancing urgency;

[0101] Inputting the control execution order table into an actuator scheduling module to obtain a timing execution plan, where the timing execution plan includes the specific execution times of the control instructions for each drive channel;

[0102] Managing the buffer of the timing execution plan to obtain an instruction buffer queue, where the instruction buffer queue ensures the smooth switching of control instructions and performs instruction synthesis based on the instruction buffer queue to obtain a load balancing adjustment instruction.

[0103] In this example, the status of the control parameters of each drive channel is monitored to obtain a real-time operating status matrix. Assume there are drive channels in the system, and the operating status of each channel includes the real-time load power and temperature . The real-time operating status matrix is represented as:

[0104] ;

[0105] where is the real-time load power of channel , and is the real-time temperature of channel . By online collecting these data and dynamically updating the matrix , the operating status of the entire system is reflected in real time. Based on the real-time operating status matrix, calculate the load balancing degree and the temperature distribution uniformity , used to determine whether the system operation is within the safe range. The calculation formula for the load balance degree is:

[0106] ;

[0107] Where is the average load power of all channels, expressed as:

[0108] ;

[0109] The calculation of the temperature distribution uniformity is similar and is defined as:

[0110] ;

[0111] Where is the average temperature of all channels, expressed as:

[0112] ;

[0113] By judging the thresholds of and , a region identification signal is generated to indicate whether the current system is in the safe operation region. The rules of the region identification signal are:

[0114] ;

[0115] Where and are the safety thresholds of the load balance degree and the temperature distribution. The region identification signal is input into the control strategy selector to determine the control mode signal . The control strategies include the steady-state control mode and the fast response mode, and the signal rules are:

[0116] ;

[0117] In the steady-state mode, the system optimizes the control parameters with a smaller adjustment step; in the fast response mode, the system will preferentially adjust the emergency parameters to restore balance. According to the control mode signal, the control parameters of each drive channel are classified and processed to generate a hierarchical control sequence. Assuming that the control parameters include the voltage adjustment amount and the timing adjustment amount , the hierarchical control sequence is expressed as:

[0118] ;

[0119] The priority sorting of the hierarchical control sequence is based on the load balance urgency. The priority is determined by the load imbalance amount and is in accordance with Arrange in descending order to generate a control execution order list , indicating the execution order of the adjustment instructions. Input the control execution order list into the actuator scheduling module to generate a timing execution plan. Assume that each instruction has an execution time of , and the timing execution plan is as follows:

[0120] ;

[0121] where is the total number of instructions, ensuring that the instructions are executed in order. To ensure smooth switching of the control instructions, the timing execution plan generates an instruction cache queue through buffer management. The instruction cache queue is used to store unexecuted instructions and optimizes the execution efficiency of the instructions through queue management. The caching mechanism ensures that there are no conflicts among the adjustment instructions of multiple channels. Instruction synthesis is performed through the instruction cache queue to generate the final load balancing adjustment instruction . The adjustment instruction includes the voltage and timing adjustment amounts of all channels and is executed step by step according to the timing plan .

[0122] In one example, performance metric evaluation calculations are performed based on the load balancing adjustment instruction to obtain performance optimization parameters, and the first adaptive prediction model is updated according to the performance optimization parameters to obtain a second adaptive prediction model, including:

[0123] Perform performance evaluation on the execution effect of the load balancing adjustment instruction to obtain a performance metric matrix, and the performance metric matrix includes a load balancing degree metric, a temperature consistency metric, and a response time metric;

[0124] Based on the performance metric matrix, perform model error analysis to obtain a prediction error distribution, and the prediction error distribution includes a load distribution prediction error and a fluctuation prediction error;

[0125] [[ID=3�]]Input the prediction error distribution into the adaptive corrector to obtain a model parameter correction amount, and the model parameter correction amount is used to update the network parameters of each layer of the first adaptive prediction model;

[0126] Based on the model parameter correction amount, update the load decoupling layer of the first adaptive prediction model to obtain an updated load feature extraction structure;

[0127] Perform dynamic weight adjustment on the multi-channel complementary feature extraction layer of the first adaptive prediction model to obtain an optimized channel feature network;

[0128] Based on the performance metric matrix, perform graph structure optimization on the dynamic coupling modeling layer of the first adaptive prediction model to obtain an updated coupling modeling network;

[0129] Integrate the updated load feature extraction structure, the optimized channel feature network, and the updated coupling modeling network to obtain the second adaptive prediction model.

[0130] In this example, the execution effect of the load balancing adjustment instruction is evaluated for performance, generating a performance metric matrix for evaluating the actual effect of the adjustment strategy. The performance metric matrix includes a load balancing degree metric, a temperature consistency metric, and a response time metric. Assume the system contains driver channels, and the actual load power of each channel is , and the ideal load power distribution is , then the load balancing degree metric is defined as:

[0131] ;

[0132] where represents the average load power. This formula calculates the deviation of the load distribution of each channel from the ideal distribution, used to quantify the load balancing effect. The temperature consistency metric is used to evaluate the temperature distribution uniformity of each driver channel, and its definition is:

[0133] ;

[0134] where is the actual temperature of channel , represents the average temperature. The response time metric represents the time required for the system to reach the target state from receiving the adjustment instruction. Assume the adjustment start time is , and the steady state time is , then:

[0135] ;

[0136] Through the above formula, the generated performance metric matrix is expressed as:

[0137] ;

[0138] Based on the performance metric matrix, perform model error analysis and calculate the prediction error distribution. The prediction error distribution includes the load distribution prediction error and the fluctuation prediction error. The load distribution prediction error is defined as the difference between the actual load and the model prediction value :

[0139] ;

[0140] while the fluctuation prediction error represents the actual load fluctuation Difference from predicted fluctuations :

[0141] ;

[0142] The predicted error distribution is represented as a vector

[0143] ;

[0144] The predicted error distribution is input into the adaptive corrector to generate the model parameter correction amount. The adaptive corrector is based on the gradient descent method and uses the predicted error to optimize the loss function of the model. Assume the loss function of the model is:

[0145] ;

[0146] where is the weight parameter of the fluctuation error. By taking the derivative of the loss function, the correction amount is obtained:

[0147] ;

[0148] where is the learning rate, is the set of model parameters. The correction amount is used to update the network parameters of each layer of the first adaptive prediction model. According to the correction amount, the load decoupling layer of the model is updated to optimize the load feature extraction structure. The load decoupling layer adopts an autoencoder structure, and its update formula is

[0149] ;

[0150] where is the weight matrix of the encoder, is the input data, is the updated feature representation. At the same time, the dynamic weight adjustment of the multi-channel complementary feature extraction layer is performed to optimize the channel feature network. The feature extraction network of each channel is based on the residual connection structure, and its dynamic adjustment formula is:

[0151] ;

[0152] where is the feature of the th layer, is the weight matrix, is the correction amount. The graph structure of the dynamic coupling modeling layer is optimized based on the performance metric matrix, and the adjacency matrix of the graph convolutional network is updated

[0153] ;

[0154] wherein represents the correction amount adjusted according to the performance index. Integrate the updated load feature extraction structure, the optimized channel feature network, and the updated coupling modeling network to generate the second adaptive prediction model.

[0155] Referring to Figure 2 , this embodiment provides a load matching optimization system for a multi-channel gate driver, including:

[0156] An acquisition module 1, configured to acquire the gate current, gate voltage, and switching frequency of the multi-channel parallel gate driver to obtain the driver load operation data, where the driver load operation data includes impedance parameters, coupling parameters, and temperature parameters;

[0157] A region division module 2, configured to perform region division processing on the operation space based on the driver load operation data to obtain a safe operation region and a transition region;

[0158] A feature extraction module 3, configured to input the driver load operation data into the first adaptive prediction model for feature extraction to obtain a load balance prediction result, where the load balance prediction result includes an inter-channel load distribution prediction value and a load fluctuation prediction value;

[0159] A calculation module 4, configured to perform parallel optimization calculation on the control strategy according to the load balance prediction result to obtain control parameters for each drive channel, where the control parameters for each drive channel include a gate drive voltage adjustment amount and a drive timing adjustment amount;

[0160] An adjustment module 5, configured to adjust the control parameters for each drive channel in real time to obtain a load balance adjustment instruction;

[0161] An update module 6, configured to perform performance index evaluation calculation based on the load balance adjustment instruction to obtain performance optimization parameters, and update the first adaptive prediction model according to the performance optimization parameters to obtain the second adaptive prediction model.

[0162] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to that described in the above method embodiment, and details are not described herein again.

[0163] Referring to Figure 3 , this embodiment of the present invention also provides a computer device, which may be a server, and its internal structure may be as Figure 3As shown in the figure. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected via a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0164] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0165] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0166] Those of ordinary skill in the art can understand that all or part of the processes in the above-described embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0167] It should be noted that in this text, the terms "including", "comprising", or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such a process, apparatus, article, or method. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of additional identical elements in the process, apparatus, article, or method comprising such an element.

[0168] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A load matching optimization method for a multi-channel gate driver, characterized in that It includes the following steps: Collect the gate current, gate voltage, and switching frequency of the multi-channel parallel gate driver to obtain the driver load operation data, where the driver load operation data includes impedance parameters, coupling parameters, and temperature parameters; Perform regional division processing on the operation space based on the driver load operation data to obtain a safe operation area and a transition area; Input the driver load operation data into the first adaptive prediction model for feature extraction to obtain a load balance prediction result, where the load balance prediction result includes the inter-channel load distribution prediction value and the load fluctuation prediction value; Perform parallel optimization calculation on the control strategy according to the load balance prediction result to obtain the control parameters of each drive channel, where the control parameters of each drive channel include the gate drive voltage adjustment amount and the drive timing adjustment amount; Perform real-time adjustment on the control parameters of each drive channel to obtain a load balance adjustment instruction; specifically including: monitoring the states of the control parameters of each drive channel to obtain a real-time operation state matrix, where the real-time operation state matrix includes the current load balance degree and temperature distribution data; determining the working area based on the real-time operation state matrix to obtain a region identification signal, where the region identification signal is used to determine whether the system is in the safe operation area; inputting the region identification signal into a control strategy selector to obtain a control mode signal, where the control mode signal includes a steady-state control mode and a fast response mode; classifying the control parameters of each drive channel based on the control mode signal to obtain a hierarchical control sequence, where the hierarchical control sequence includes a voltage adjustment sequence and a timing adjustment sequence; sorting the hierarchical control sequence by priority to obtain a control execution order table, where the control execution order table determines the execution order of each adjustment instruction according to the load balance urgency; inputting the control execution order table into an actuator scheduling module to obtain a timing execution plan, where the timing execution plan includes the specific execution times of the control instructions of each drive channel; performing buffer management on the timing execution plan to obtain an instruction cache queue, where the instruction cache queue ensures the smooth switching of control instructions and performs instruction synthesis based on the instruction cache queue to obtain a load balance adjustment instruction; Perform performance index evaluation calculation based on the load balance adjustment instruction to obtain performance optimization parameters, and update the first adaptive prediction model according to the performance optimization parameters to obtain a second adaptive prediction model.

2. The load matching optimization method of the multi-channel gate driver according to claim 1, characterized in that The step of collecting the gate current, gate voltage, and switching frequency of the multi-channel parallel gate driver to obtain the driver load operation data, where the driver load operation data includes impedance parameters, coupling parameters, and temperature parameters, includes: Sample the gate current of the multi-channel parallel gate driver to obtain a gate current sampling sequence, and sample the gate voltage of the multi-channel parallel gate driver to obtain a gate voltage sampling sequence; Perform impedance calculation based on the gate current sampling sequence and the gate voltage sampling sequence to obtain impedance parameters, where the impedance parameters include the gate resistance value, Miller capacitance value, and parasitic inductance value; Perform a coupling calculation on the impedance parameters of adjacent drive channels to obtain coupling parameters, where the coupling parameters include an electromagnetic coupling coefficient and a thermal coupling coefficient; Monitor the temperature of each drive channel of the multi-channel parallel gate driver to obtain a temperature monitoring data sequence, and perform a temperature distribution calculation based on the temperature monitoring data sequence to obtain temperature parameters, where the temperature parameters include the temperature values and temperature gradient values of each drive channel; Perform data standardization processing on the impedance parameters, the coupling parameters, and the temperature parameters to obtain drive load operation data.

3. The load matching optimization method for the multi-channel gate driver according to claim 2, characterized in that Perform a regional division process on the operation space based on the drive load operation data to obtain a safe operation area and a transition area, including: Analyze the impedance parameters in the drive load operation data to obtain the maximum allowable power values of each drive channel, and calculate a load deviation threshold matrix based on the maximum allowable power values; Perform threshold optimization based on the load deviation threshold matrix to obtain an operation area boundary threshold, where the operation area boundary threshold is used to divide the safe operation area and the transition area; Perform a quantitative calculation on the interference degree between each drive channel based on the coupling parameters in the drive load operation data to obtain a channel coupling strength matrix, where the channel coupling strength matrix is used to correct the operation area boundary threshold; Perform a thermal stress analysis on the temperature distribution of each drive channel according to the temperature parameters in the drive load operation data to obtain a thermal stress distribution matrix, and perform temperature compensation on the operation area boundary threshold based on the thermal stress distribution matrix; Perform a regional division based on the operation area boundary threshold and the load deviation threshold matrix to obtain an initial operation space boundary, and perform multi-dimensional feature extraction on the initial operation space boundary to obtain an operation space feature vector, where the operation space feature vector includes a load balance degree, a temperature uniformity degree, and a coupling strength; Perform regional optimization according to the operation space feature vector to obtain an optimized operation space division result, and perform real-time determination of the operation state based on the optimized operation space division result to obtain a safe operation area and a transition area.

4. The load matching optimization method of the multi-channel gate driver according to claim 3, characterized in that Input the drive load operation data into a first adaptive prediction model for feature extraction to obtain a load balance prediction result, where the load balance prediction result includes predicted values of the load distribution between channels and predicted values of load fluctuations, including: Input the drive load operation data into the load decoupling layer of the first adaptive prediction model, where the load decoupling layer adopts an autoencoder structure to non-linearly separate the impedance parameters, the coupling parameters, and the temperature parameters to obtain load feature components; Input the load feature components into the multi-channel complementary feature extraction layer of the first adaptive prediction model, where the multi-channel complementary feature extraction layer includes parallel channel feature sub-networks, each channel feature sub-network adopts a residual connection structure, and a horizontal information transfer path is established between channels to obtain channel correlation features; Inputting the channel correlation features into the dynamic coupling modeling layer of the first adaptive prediction model, wherein the dynamic coupling modeling layer adopts a graph convolutional network structure, captures the load coupling effect by dynamically constructing a correlation map between channels, and obtains coupling dynamic features; Inputting the coupled dynamic features into a multi-scale temporal feature fusion layer of the first adaptive prediction model, wherein the multi-scale temporal feature fusion layer comprises a parallel group of long short-term memory units, each group of units processes features at different time scales, and performs feature fusion through a cross-scale attention mechanism to obtain multi-scale temporal features; Inputting the multi-scale time series features into the thermoelectric coupling analysis layer of the first adaptive prediction model, the thermoelectric coupling analysis layer adopts a dual-stream network structure to process thermal features and electrical features respectively, and merges the two feature streams through an adaptive weight fuser to obtain thermoelectric coupling features; Inputting the thermoelectric coupling feature into the distribution prediction layer of the first adaptive prediction model, wherein the distribution prediction layer adopts a conditional variational autoencoder structure to generate a probability model of the inter-channel load distribution based on the current load state to obtain a predicted value of the inter-channel load distribution; Inputting the thermoelectric coupling characteristics into the fluctuation analysis layer of the first adaptive prediction model, wherein the fluctuation analysis layer adopts a wavelet neural network structure and captures load fluctuation characteristics of different frequencies through multi-resolution analysis to obtain a load fluctuation prediction value; A load balancing prediction result is output according to the inter-channel load distribution prediction value and the load fluctuation prediction value.

5. The load matching optimization method of the multi-channel gate driver according to claim 4, characterized in that The control strategy is optimized and calculated in parallel according to the load balancing prediction result to obtain control parameters of each drive channel, wherein the control parameters of each drive channel include a gate drive voltage adjustment amount and a drive timing adjustment amount, including: Performing data analysis on the inter-channel load distribution prediction value in the load balancing prediction result to obtain a load imbalance value of each driving channel, wherein the load imbalance value is used to calculate a driving voltage compensation value; Establishing a dynamic compensation model based on the load fluctuation prediction value to obtain a driving timing compensation strategy, wherein the driving timing compensation strategy includes a dead time adjustment amount and a switch delay compensation amount; Inputting the load imbalance amount into a weight distributor to obtain a voltage adjustment coefficient matrix for each driving channel, wherein the voltage adjustment coefficient matrix is used to calculate the adjustment range of the gate driving voltage; Parallel decomposition of the driving timing compensation strategy is performed to obtain a channel-level timing control sequence, wherein the channel-level timing control sequence cooperates with the voltage adjustment coefficient matrix; Performing orthogonal optimization based on the voltage regulation coefficient matrix and the channel-level timing control sequence to obtain a multi-channel collaborative control matrix, wherein the multi-channel collaborative control matrix includes a voltage control amount and a timing control amount; Inputting the multi-channel cooperative control matrix into a constraint processor, performing boundary constraints on the voltage amplitude and time interval, and obtaining initial values of control parameters; Performing multi-objective iterative optimization on the initial values of the control parameters to obtain an optimal control parameter combination, wherein the optimal control parameter combination satisfies the balance requirements of load balancing and switching loss; Based on the optimal control parameter combination, the gate drive voltages and drive timings of each drive channel are allocated to obtain the control parameters of each drive channel.

6. The load matching optimization method of the multi-channel gate driver according to claim 1, characterized in that The performance index evaluation calculation based on the load balancing adjustment instruction to obtain performance optimization parameters, and updating the first adaptive prediction model according to the performance optimization parameters to obtain a second adaptive prediction model, includes: Perform a performance evaluation on the execution effect of the load balancing adjustment instruction to obtain a performance index matrix, where the performance index matrix includes a load balancing degree index, a temperature consistency index, and a response time index; Based on the performance index matrix, perform model error analysis to obtain a prediction error distribution, where the prediction error distribution includes a load distribution prediction error and a fluctuation prediction error; Input the prediction error distribution into an adaptive corrector to obtain a model parameter correction amount, and the model parameter correction amount is used to update the network parameters of each layer of the first adaptive prediction model; Based on the model parameter correction amount, update the load decoupling layer of the first adaptive prediction model to obtain an updated load feature extraction structure; Perform dynamic weight adjustment on the multi-channel complementary feature extraction layer of the first adaptive prediction model to obtain an optimized channel feature network; Based on the performance index matrix, perform graph structure optimization on the dynamic coupling modeling layer of the first adaptive prediction model to obtain an updated coupling modeling network; Integrate the updated load feature extraction structure, the optimized channel feature network, and the updated coupling modeling network to obtain a second adaptive prediction model.

7. A load matching optimization system for a multi-channel gate driver, characterized in that, For implementing the steps of the method according to any one of claims 1 to 6, the system includes: An acquisition module, configured to acquire the gate current, gate voltage, and switching frequency of a multi-channel parallel gate driver to obtain drive load operation data, where the drive load operation data includes impedance parameters, coupling parameters, and temperature parameters; A region division module, configured to perform region division processing on the operation space based on the drive load operation data to obtain a safe operation region and a transition region; A feature extraction module, configured to input the drive load operation data into a first adaptive prediction model for feature extraction to obtain a load balancing prediction result, where the load balancing prediction result includes an inter-channel load distribution prediction value and a load fluctuation prediction value; A calculation module, configured to perform parallel optimization calculation on a control strategy according to the load balancing prediction result to obtain control parameters of each drive channel, where the control parameters of each drive channel include a gate drive voltage adjustment amount and a drive timing adjustment amount; An adjustment module, configured to perform real-time adjustment on the control parameters of each drive channel to obtain a load balancing adjustment instruction; An update module, configured to perform performance index evaluation calculation based on the load balancing adjustment instruction to obtain performance optimization parameters, and update the first adaptive prediction model according to the performance optimization parameters to obtain a second adaptive prediction model.

8. A computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images