Fuel cell air compressor heat dissipation method, device and equipment

Abstract

The invention provides a heat dissipation method, device and equipment for a fuel cell air compressor. The method comprises the following steps: collecting a multi-dimensional heat data set of the fuel cell air compressor in a layered manner; performing feature extraction on the multi-dimensional heat data set, and inputting the extracted feature data set into a neural network model based on deep transfer learning to obtain a corresponding heat dissipation efficiency attenuation prediction model; establishing a multi-objective optimization equation of the air compressor based on the multi-dimensional heat data set and the heat dissipation efficiency attenuation prediction model, and solving the multi-objective optimization equation to obtain a heat dissipation control strategy set; and dynamically partitioning a heat dissipation system of the air compressor, and cooperatively adjusting the flow and the flow direction of cooling liquid in each dynamically partitioned region according to the heat dissipation control strategy set to realize heat dissipation of the air compressor. According to the invention, the problems of cooling efficiency attenuation and non-uniform resource distribution caused by cooling liquid quality change and integral control in the prior art are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment heat dissipation, and in particular to a method, device and equipment for heat dissipation of a fuel cell air compressor. Background Art

[0002] Fuel cells are a highly efficient and clean energy conversion device that has been widely used in new energy vehicles and distributed power generation systems. As the core auxiliary equipment in the fuel cell system, the air compressor is mainly responsible for providing the right amount of compressed air to the fuel cell stack. Its working stability directly affects the overall performance of the fuel cell.

[0003] During the operation of the fuel cell air compressor, a large amount of heat is generated due to high-speed rotation and compression, which requires heat management through a heat dissipation system. The existing heat dissipation method for fuel cell air compressors mainly adopts a single coolant circulation system, and performs heat dissipation management through a fixed cooling strategy and unified cooling control parameters. This heat dissipation method has the following technical problems: due to the lack of real-time monitoring and consideration of changes in coolant quality, when the coolant quality changes, the heat dissipation efficiency will drop significantly; at the same time, the traditional heat dissipation system adopts an integrated control strategy, which is unable to perform fine management of the heat load differences in different areas inside the air compressor, resulting in uneven distribution of heat dissipation resources, which not only affects the heat dissipation effect, but also causes energy waste. Summary of the invention

[0004] The main purpose of the present invention is to solve the technical problems of heat dissipation efficiency attenuation and uneven resource allocation caused by the use of an integrated control strategy in the existing fuel cell air compressor heat dissipation system, which cannot perform refined management of the heat load differences in different areas inside the air compressor; A first aspect of the present invention provides a fuel cell air compressor heat dissipation method, the fuel cell air compressor heat dissipation method comprising: The direct thermal characteristic parameters, indirect thermal characteristic parameters and coolant quality characteristic parameters of the fuel cell air compressor are collected in layers to obtain a multi-dimensional thermal data set of the air compressor; Extracting features from the multidimensional thermal data set, and inputting the extracted feature data set into a neural network model based on deep transfer learning to obtain a corresponding heat dissipation efficiency attenuation prediction model; Establishing a multi-objective optimization equation for the air compressor based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and solving the multi-objective optimization equation to obtain a heat dissipation control strategy set; The heat dissipation system of the air compressor is dynamically partitioned, and the flow rate and flow direction of the coolant in each area of ​​the dynamic partition are coordinated and adjusted according to the heat dissipation control strategy set to achieve the heat dissipation of the air compressor.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the direct thermal characteristic parameters, indirect thermal characteristic parameters and coolant quality characteristic parameters of the air compressor of the fuel cell are collected in layers to obtain a multidimensional thermal data set of the air compressor, including: A first layer of data collection points is set at a heat exchange position where the coolant of the air compressor of the fuel cell flows through, to collect the direct thermal characteristic parameters, wherein the direct thermal characteristic parameters include temperature data, flow rate data and pressure data; A second layer of data collection points is set at the preset core components of the air compressor to collect the indirect thermal characteristic parameters, wherein the indirect thermal characteristic parameters include speed data, current data, and vibration data; A third layer of data collection points is set in the coolant circulation system of the air compressor heat dissipation system to collect coolant quality characteristic parameters, wherein the coolant quality characteristic parameters include conductivity, pH value and turbidity data; The direct thermal characteristic parameters, the indirect thermal characteristic parameters and the coolant quality characteristic parameters are subjected to time synchronization and data preprocessing to obtain a multi-dimensional thermal data set of the air compressor.

[0006] Optionally, in a second implementation of the first aspect of the present invention, extracting features from the multidimensional thermal data set, inputting the extracted feature data set into a neural network model based on deep transfer learning, and obtaining a corresponding heat dissipation efficiency attenuation prediction model includes: Performing time domain feature extraction, frequency domain feature extraction and time-frequency feature extraction on each parameter in the multidimensional thermal data set to obtain target domain features; The heat dissipation efficiency attenuation mode features of the existing air compressors in the preset source domain knowledge base are used as source domain features, and the feature mapping algorithm of the attention mechanism is used to perform adaptive conversion between the source domain features and the target domain features; The target domain features are fused with the adapted source domain features, and the fused features are dynamically weighted by a hierarchical attention network to obtain a feature data set; The feature data set is input into a neural network model based on deep transfer learning for training, and the heat dissipation efficiency attenuation prediction model is obtained by minimizing the distribution difference between the source domain features and the target domain features and the heat dissipation efficiency prediction error.

[0007] Optionally, in a third implementation of the first aspect of the present invention, establishing a multi-objective optimization equation for the air compressor based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and solving the multi-objective optimization equation to obtain a heat dissipation control strategy set includes: According to the multidimensional thermal data set and the heat dissipation efficiency attenuation prediction model, a thermodynamic model of the air compressor heat dissipation system is established, and the multi-objective optimization equation is constructed based on the thermodynamic model; Establishing a set of constraints for the multi-objective optimization equation, wherein the set of constraints includes upper temperature limits of each component, coolant flow range constraints, pump power limit constraints, and dynamic constraints based on coolant quality characteristic parameters; Inputting the multi-objective optimization equation and the constraint condition set thereof into a preset multi-objective dung beetle optimizer to solve and obtain a Pareto optimal solution set; The current operating condition is determined according to the multidimensional thermal data set, and the Pareto optimal solution set is dynamically screened according to the current operating condition, and a corresponding heat dissipation control strategy set is generated according to the screened Pareto optimal solution set.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, the step of inputting the multi-objective optimization equation and the set of constraint conditions thereof into a preset multi-objective dung beetle optimizer to solve and obtain a Pareto optimal solution set comprises: Inputting the multi-objective optimization equation and the constraint condition set thereof into a preset multi-objective dung beetle optimizer, mapping the objective function space of the multi-objective optimization equation into the decision variable space through the multi-objective dung beetle optimizer, and obtaining an initial solution population of the optimization problem; Calculating the objective function value for each solution in the initial solution population to obtain an evaluation value matrix of the solution, and calculating the dominance relationship between any two solutions based on the evaluation value matrix, hierarchically sorting the solutions based on the dominance relationship, and obtaining a position update vector of each solution; Applying the position update vector to each solution in the initial solution population to generate an updated solution set, and performing constraint checking on the solutions in the updated solution set, and when a solution in the updated solution set violates the constraint condition set, correcting the position of the corresponding solution by a preset penalty function to obtain a candidate solution set that satisfies the constraint; Calculating the crowding distance of the candidate solution set, sorting the solutions with the same non-dominated level based on the crowding distance, and selecting high-quality solutions through an elite retention strategy to obtain a non-dominated solution set in an iterative process; According to the preset iterative termination condition, the optimization process of the multi-objective dung beetle optimizer is iterated cyclically. When the termination condition is met, the Pareto frontier is extracted from the final non-dominated solution set to obtain the Pareto optimal solution set.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the heat dissipation system of the air compressor is dynamically partitioned, and the flow rate and flow direction of the coolant in each area of ​​the dynamic partition are collaboratively adjusted according to the heat dissipation control strategy set to achieve the heat dissipation of the air compressor, which includes: Extracting temperature field distribution information according to direct thermal characteristic parameters in the multidimensional thermal data set, and calculating heat flow characteristics according to the direct thermal characteristic parameters and the indirect thermal characteristic parameters; Adopting an adaptive K-means clustering algorithm to cluster the temperature field distribution information and heat flow characteristics, and dynamically partitioning the heat dissipation system of the air compressor based on the clustering result to obtain N control sub-areas, where N is an integer from 2 to 8; Calculating a heat load index for each of the control sub-regions based on the direct heat characteristic parameters, and constructing an inter-region heat transfer matrix to characterize the heat transfer relationship between the control sub-regions; Determine the cooling resource allocation strategy for each of the control sub-areas using a hierarchical progressive control method according to the heat dissipation control strategy set, the heat load index and the heat transfer matrix; The coolant flow regulating valve and the flow direction control valve in the heat dissipation system are used to adjust the coolant parameters of each control sub-area in real time according to the cooling resource allocation strategy, and the cooling intensity of adjacent areas is collaboratively controlled based on the heat transfer matrix to achieve heat dissipation of the air compressor.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, the coolant flow regulating valve and the flow direction control valve in the heat dissipation system adjust the coolant parameters of each of the control sub-areas in real time according to the cooling resource allocation strategy, and the cooling intensity of adjacent areas is collaboratively controlled based on the heat transfer matrix, including: Prioritizing each of the control sub-areas according to the heat load index, allocating a control cycle and a response order according to the ranking result, and obtaining a control priority list; Performing proportional-integral-differential control calculation on the flow parameters and flow direction parameters in the cooling resource allocation strategy, converting the control signal into a valve opening instruction, and obtaining a control sequence of the actuator; Calculating the thermal coupling influence coefficient between each of the control sub-areas based on the heat transfer matrix, adjusting the cooling resource allocation of adjacent areas according to the thermal coupling influence coefficient, and obtaining a coordinated control compensation amount; The control sequence and the coordinated control compensation are integrated to generate valve control instructions, and real-time closed-loop correction is performed according to the dynamic response of the system to obtain control signals for the coolant flow regulating valve and the flow direction control valve; The control signal is applied to the coolant flow regulating valve and the flow direction control valve to adjust the coolant parameters of each of the control sub-areas in real time.

[0011] A second aspect of the present invention provides a heat dissipation device for a fuel cell air compressor, the heat dissipation device for a fuel cell air compressor comprising: A data acquisition module is used to perform layered acquisition of direct thermal characteristic parameters, indirect thermal characteristic parameters and coolant quality characteristic parameters of the air compressor of the fuel cell to obtain a multi-dimensional thermal data set of the air compressor; A pattern recognition module, used for extracting features from the multidimensional thermal data set, inputting the extracted feature data set into a neural network model based on deep transfer learning, and obtaining a corresponding heat dissipation efficiency attenuation prediction model; A strategy generation module, used to establish a multi-objective optimization equation for the air compressor based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and solve the multi-objective optimization equation to obtain a heat dissipation control strategy set; The partition control module is used to dynamically partition the heat dissipation system of the air compressor, and coordinately adjust the flow rate and flow direction of the coolant in each area of ​​the dynamic partition according to the heat dissipation control strategy set to achieve the heat dissipation of the air compressor.

[0012] The third aspect of the present invention provides a fuel cell air compressor cooling device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the fuel cell air compressor cooling device executes the steps of the above-mentioned fuel cell air compressor cooling method.

[0013] The above-mentioned fuel cell air compressor heat dissipation method, device, and equipment collect the multidimensional thermal data set of the fuel cell air compressor in layers; extract features from the multidimensional thermal data set, input the extracted feature data set into a neural network model based on deep transfer learning, and obtain a corresponding heat dissipation efficiency attenuation prediction model; establish a multi-objective optimization equation for the air compressor based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and solve the multi-objective optimization equation to obtain a heat dissipation control strategy set; dynamically partition the heat dissipation system of the air compressor, and coordinately adjust the flow rate and flow direction of the coolant in each area of ​​the dynamic partition according to the heat dissipation control strategy set to achieve heat dissipation of the air compressor. The present invention solves the problems of heat dissipation efficiency attenuation and uneven resource allocation caused by changes in coolant quality and integrated control in the prior art.

[0014] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0015] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of a first embodiment of a method for cooling a fuel cell air compressor in an embodiment of the present invention; Figure 2 A schematic diagram of an embodiment of a heat dissipation device for a fuel cell air compressor in an embodiment of the present invention; Figure 3 It is a schematic diagram of an embodiment of a heat dissipation device for a fuel cell air compressor in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0019] To facilitate understanding of this embodiment, a fuel cell air compressor heat dissipation method disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps: 101. Perform layered collection of direct thermal characteristic parameters, indirect thermal characteristic parameters, and coolant quality characteristic parameters of the air compressor of the fuel cell to obtain a multidimensional thermal data set of the air compressor; In one embodiment of the present invention, the direct thermal characteristic parameters, indirect thermal characteristic parameters and coolant quality characteristic parameters of the air compressor of the fuel cell are collected in layers to obtain a multi-dimensional thermal data set of the air compressor, including: setting a first layer of data collection points at a heat exchange position through which the coolant of the air compressor of the fuel cell flows, to collect the direct thermal characteristic parameters, wherein the direct thermal characteristic parameters include temperature data, flow rate data and pressure data; setting a second layer of data collection points at a preset core component of the air compressor, to collect the indirect thermal characteristic parameters, wherein the indirect thermal characteristic parameters include speed data, current data and vibration data; A third layer of data collection points is set in the coolant circulation system of the heat dissipation system of the air compressor to collect coolant quality characteristic parameters, which include conductivity, pH value and turbidity data; the direct thermal characteristic parameters, the indirect thermal characteristic parameters and the coolant quality characteristic parameters are time synchronized and data preprocessed to obtain a multidimensional thermal data set of the air compressor.

[0020] Specifically, the process of collecting multi-dimensional thermal characteristic data of the air compressor first sets the first layer of data collection points at the key heat exchange positions where the coolant of the fuel cell air compressor flows. These key heat exchange positions include the coolant inlet and outlet, the radiator interface, and the area where the heat load is concentrated. High-precision temperature sensors, flow meters, and pressure sensors are installed at these positions to collect direct thermal characteristic parameters. The temperature data is collected by a PT100 platinum resistance temperature sensor with a measurement range of -50°C to 150°C, an accuracy of ±0.1°C, and a sampling frequency of 100Hz, which can capture subtle temperature fluctuations. The flow rate data is measured by an ultrasonic flowmeter with a measurement range of 0.5m / s to 10m / s and an accuracy of ±1%. It is also sampled at a frequency of 100Hz to ensure real-time monitoring of fluid dynamic changes. The pressure data is collected by a piezoresistive pressure sensor with a measurement range of 0MPa to 2MPa, an accuracy of ±0.5%, and a sampling frequency of 50Hz. These three types of direct thermal characteristic parameters constitute the basic physical quantities of the heat exchange process, which directly reflect the heat exchange efficiency and working status of the cooling system.

[0021] Specifically, the second layer of data collection points is set at the preset core components of the air compressor. These core components include the motor stator, rotor, bearings, and key structures such as the air inlet and compression chamber. Hall sensors are installed at the motor stator and rotor to collect the motor speed data. The measurement range is 0rpm to 20000rpm, the accuracy is ±0.5%, and the sampling frequency is set to 50Hz. A current transformer is installed on the motor power supply line to measure the current data. The range is 0A to 100A, the accuracy is ±0.5%, and the sampling frequency is also 50Hz. Piezoelectric accelerometers are installed on the bearings and housings of the air compressor to collect vibration data. The measurement range is 0Hz to 10kHz, the sensitivity is 100mV / g, and the sampling frequency is 100Hz to capture high-frequency vibration characteristics. Although these indirect thermal characteristic parameters do not directly measure heat, they are closely related to the load, efficiency and heat generation of the air compressor. Through these parameters, the causes and laws of heat generation can be reversely deduced.

[0022] Specifically, the third layer of data collection points is set in the coolant circulation system of the air compressor's heat dissipation system, mainly concentrated in the coolant storage tank, the circulation pump inlet and the main loop. Conductivity sensors, pH detectors and turbidity meters are installed at these points to collect coolant quality characteristic parameters. The conductivity sensor uses the four-electrode method to measure the conductivity of the coolant, with a measurement range of 0μS / cm to 2000μS / cm, an accuracy of ±1%, and a sampling frequency of 1Hz. The pH detector uses the glass electrode method to measure the pH of the coolant, with a measurement range of 0 to 14, an accuracy of ±0.1, and a sampling frequency of 1Hz. The turbidity meter uses the scattered light measurement principle, with a measurement range of 0NTU to 100NTU, an accuracy of ±2%, and a sampling frequency of 1Hz. The collection frequency of coolant quality characteristic parameters is low because these parameters usually change slowly and do not require high-frequency monitoring. These parameters reflect the degree of contamination, corrosiveness and thermal conductivity of the coolant, which directly affect the efficiency and life of the heat dissipation system.

[0023] Specifically, time synchronization and data preprocessing of various types of collected data are key steps in constructing a multidimensional thermal data set. Time synchronization uses a unified timestamp marking method to align all sensor data according to the system's unified clock to eliminate the deviation of different sensor sampling times. Data preprocessing includes three steps: outlier detection, missing value filling, and data normalization. Outlier detection uses the triple standard deviation rule to mark data that exceeds the mean ±3σ range as outliers, and corrects them in combination with the sliding window median replacement method. Missing value filling uses a combination of linear interpolation and forward filling to supplement missing points in the time series. Data normalization uses the minimum-maximum standardization method to uniformly map feature data of different dimensions and orders of magnitude to the [0,1] interval, which is convenient for subsequent feature extraction and model training.

[0024] 102. Extract features from the multidimensional thermal data set, input the extracted feature data set into a neural network model based on deep transfer learning, and obtain a corresponding heat dissipation efficiency attenuation prediction model; In one embodiment of the present invention, the feature extraction of the multidimensional thermal data set, inputting the extracted feature data set into a neural network model based on deep transfer learning, and obtaining a corresponding heat dissipation efficiency attenuation prediction model includes: performing time domain feature extraction, frequency domain feature extraction, and time-frequency feature extraction on each parameter in the multidimensional thermal data set to obtain target domain features; using the heat dissipation efficiency attenuation pattern features of the existing air compressors in the preset source domain knowledge base as source domain features, and using the feature mapping algorithm of the attention mechanism to perform adaptive conversion on the source domain features and the target domain features; fusing the target domain features with the adaptively converted source domain features, and dynamically assigning weights to the fused features through a hierarchical attention network to obtain a feature data set; inputting the feature data set into a neural network model based on deep transfer learning for training, and obtaining the heat dissipation efficiency attenuation prediction model by minimizing the distribution differences and heat dissipation efficiency prediction errors between the source domain features and the target domain features.

[0025] Specifically, the feature extraction of multidimensional thermal data sets begins with time domain feature extraction, frequency domain feature extraction, and time-frequency feature extraction. In the process of time domain feature extraction, the system calculates statistical features for direct thermal feature parameters, indirect thermal feature parameters, and coolant quality feature parameters, including mean, standard deviation, skewness, kurtosis, interquartile range, and entropy. Taking temperature data as an example, the system takes every 10 seconds of data as a time window, calculates the mean of the temperature in the window to reflect the overall temperature level, the standard deviation reflects the degree of temperature fluctuation, and the skewness and kurtosis reflect the asymmetry and peak degree of temperature distribution, respectively. For signals with large fluctuations such as current and vibration, the peak factor, waveform factor, and pulse factor are additionally calculated to capture the instantaneous change characteristics of the signal. For parameters such as flow rate and pressure, the rate of change and trend characteristics are extracted, and characterized by calculating the first-order difference and linear regression coefficient. Although the sampling frequency of coolant quality parameters is low, the sliding window method is still used to extract its change trend characteristics within 24 hours. These time domain characteristics together constitute a comprehensive description of the thermal characteristics of the air compressor in the time dimension, which can effectively reflect the basic statistical characteristics and changing laws of each parameter.

[0026] Specifically, the frequency domain feature extraction focuses on analyzing the periodic variation of each parameter. The system performs fast Fourier transform (FFT) on temperature, flow rate, pressure, speed, current and vibration data respectively, and converts the time domain signal to the frequency domain. The frequency domain features extracted for each parameter include the main frequency and its amplitude, power spectrum density, frequency band energy distribution and spectrum moment. Specifically, the system decomposes the vibration signal into three frequency bands of 0-100Hz, 100-1000Hz and 1000-10000Hz, and extracts low-frequency, medium-frequency and high-frequency vibration energy respectively. These features correspond to the vibration characteristics of different parts of the air compressor. The temperature and flow rate data focus on the spectral features in the range of 0-1Hz, extracting the frequency values ​​and amplitudes of the first five main frequency components, as well as the energy proportion in the range of 0.1-0.5Hz. The frequency domain analysis of current data focuses on the power supply frequency (such as 50Hz) and its harmonic components, and reflects the working status of the motor by calculating the total harmonic distortion rate. These frequency domain features can identify periodic patterns hidden in time domain data, and are important indicators of cooling efficiency degradation caused by bearing failures, motor anomalies, etc.

[0027] Specifically, the time-frequency feature extraction adopts a method combining wavelet transform and Hilbert-Huang transform, aiming to capture the non-stationary characteristics of parameters at different time scales. The system performs wavelet decomposition on signals such as temperature, flow rate, and vibration, selects Daubechies (db4) wavelet as the basis function, decomposes the signal into 5 scales, and extracts the wavelet energy entropy, wavelet packet entropy, and singular value decomposition features of each scale. For vibration signals, empirical mode decomposition (EMD) is additionally applied to decompose the signal into several intrinsic mode functions (IMFs), and the instantaneous frequency and instantaneous amplitude features of each IMF are extracted. For temperature and flow rate data, continuous wavelet transform is used to calculate the time-frequency spectrum, and the time-frequency energy distribution features are extracted from it. These time-frequency features can capture the energy distribution changes of the signal at different time scales, and have significant advantages in detecting the sudden changes and gradual changes in the working state of the air compressor. Through these three types of feature extraction methods, the system extracts hundreds of dimensional target domain features from the multidimensional thermal data set to comprehensively characterize the thermal characteristics of the current air compressor.

[0028] Specifically, the heat dissipation efficiency attenuation pattern features of existing air compressors in the preset source domain knowledge base are used as source domain features, and the adaptation conversion through the feature mapping algorithm of the attention mechanism is the core link of transfer learning. The source domain knowledge base stores the historical operation data of multiple air compressors with different working conditions and different usage time and their corresponding heat dissipation efficiency attenuation features, which also include time domain, frequency domain and time-frequency features. The feature mapping algorithm of the attention mechanism first aligns the source domain features and the target domain features in feature space, and quantifies the distribution difference between the two domains through the maximum mean difference (MMD) criterion. Then, the algorithm constructs a multi-head attention network, and each attention head is responsible for learning the importance of a specific type of features, such as one head focusing on temperature-related features and the other head focusing on vibration-related features. Each attention head calculates the similarity matrix between the query vector (from the target domain) and the key vector (from the source domain), generates attention weights, and performs weighted summation on the value vector (source domain features) based on this to obtain the adapted feature representation. The output of the multi-head attention is spliced ​​and linearly transformed to obtain the final source domain feature adaptation result. This attention-based feature mapping method can adaptively learn the feature correspondence between the source domain and the target domain, deal with the distribution shift problem between the two domains, and achieve effective knowledge transfer.

[0029] Specifically, the fusion of target domain features with adapted source domain features requires a well-designed hierarchical attention network for dynamic weight allocation. This fusion process first groups features according to physical meaning, such as temperature group, fluid group, mechanical group and electrical group, each group contains relevant time domain, frequency domain and time-frequency features. The hierarchical attention network adopts a three-layer structure: the bottom layer focuses on the fusion of features within the group, the middle layer focuses on the fusion of features between groups, and the top layer focuses on the fusion of source domain and target domain features. At the bottom layer, a self-attention mechanism is used within each feature group to learn the relationship between different representations of the same physical quantity (such as time domain, frequency domain, time-frequency). At the middle layer, the mutual influence between different feature groups is learned through the cross-attention mechanism, such as the influence of temperature change on current change. At the top layer, the weights of source domain features and target domain features are dynamically adjusted through the gated fusion unit. In specific implementation, the attention calculation of each layer includes three steps: attention score calculation, softmax normalization and weighted summation. The final fused features are further enhanced through residual connection and layer normalization to enhance the feature expression ability. This hierarchical attention fusion method fully considers the physical correlation and hierarchical structure between features, so that the fused feature dataset not only retains the specificity of the target domain, but also utilizes the rich knowledge of the source domain, providing high-quality input for subsequent model training.

[0030] Specifically, inputting the feature dataset into the neural network model based on deep transfer learning for training requires a carefully designed network structure and training strategy. The neural network model adopts an encoder-decoder architecture. The encoder part contains three main modules: feature extraction module, domain adaptation module and task specific module. The feature extraction module consists of 4 layers of stacked bidirectional GRU (gated recurrent unit), with 128 neurons in each layer, capturing the long-term and short-term dependencies of temporal features. The domain adaptation module uses a combination of gradient reversal layer and domain classifier to make the extracted features invariant to the source domain and target domain through adversarial training. The task specific module consists of 2 layers of fully connected layers, with 256 and 128 neurons in each layer, using GELU activation function. The decoder part adopts a multi-task learning framework to simultaneously predict the heat dissipation efficiency value and decay trend category. During the training process, the loss function contains four parts: mean square error loss for heat dissipation efficiency prediction, cross entropy loss for decay trend classification, maximum mean difference loss for feature distribution and adversarial loss for domain classification. The training adopts the Adam optimizer with a batch size of 64, an initial learning rate of 0.001, and a cosine annealing strategy to dynamically adjust the learning rate. To avoid overfitting, weight decay and early stopping mechanisms are introduced. During the training process, the weight of the source domain data gradually decreases with the increase of the number of iterations, while the weight of the target domain data gradually increases, achieving a smooth transition from source domain knowledge to target domain adaptation. Through this deep transfer learning method, even if the target domain data is limited, an accurate heat dissipation efficiency attenuation prediction model can be trained. The model can not only predict the specific value of the heat dissipation efficiency, but also determine the trend and severity of the attenuation.

[0031] 103. Based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, a multi-objective optimization equation for the air compressor is established, and the heat dissipation control strategy set is obtained by solving the multi-objective optimization equation; In one embodiment of the present invention, the multi-objective optimization equation of the air compressor is established based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and the heat dissipation control strategy set is obtained by solving the multi-objective optimization equation, including: establishing a thermodynamic model of the air compressor heat dissipation system according to the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and constructing the multi-objective optimization equation based on the thermodynamic model; establishing a constraint condition set of the multi-objective optimization equation, wherein the constraint condition set includes upper temperature limit constraints of each component, coolant flow range constraints, pump power limit constraints and dynamic constraints based on coolant quality characteristic parameters; inputting the multi-objective optimization equation and its constraint condition set into a preset multi-objective dung beetle optimizer for solving to obtain a Pareto optimal solution set; determining the current operating condition according to the multi-dimensional thermal data set, dynamically screening the Pareto optimal solution set according to the current operating condition, and generating a corresponding heat dissipation control strategy set according to the screened Pareto optimal solution set.

[0032] Specifically, the establishment of a thermodynamic model of the air compressor heat dissipation system is a strategy generation process that requires the comprehensive use of multidimensional thermal data sets and the output of the heat dissipation efficiency attenuation prediction model. The thermodynamic model contains three submodules: heat conduction module, heat convection module and heat radiation module. The heat conduction module describes the heat transfer between solid components inside the air compressor. It is constructed based on Fourier heat conduction theory and simplifies the air compressor into multiple thermal particles. Each particle represents a key component such as a motor stator, rotor, bearing, etc. The thermal resistance between each particle is calculated based on the thermal conductivity, contact area and distance of the component material, while considering the nonlinear effect of component temperature on thermal resistance. The thermal convection module describes the heat exchange between solid components and coolant. It is modeled using Newton's cooling law. The convective heat transfer coefficient when the coolant flows through each heat exchange surface is determined by dimensionless parameters such as the Reynolds number and the Prandtl number. For areas with large flow rate changes, the system uses computational fluid dynamics methods to pre-simulate the convective heat transfer coefficient at different flow rates and makes a lookup table for real-time calculation. The thermal radiation module describes the radiation heat exchange between components. Although it accounts for a small proportion in the water cooling system, it still needs to be considered in the high temperature area and is calculated using the Stefan-Boltzmann law. The parameters of these three submodules are fitted and optimized using the historical data in the multidimensional thermal data set. At the same time, the parameters are dynamically adjusted based on the efficiency change trend predicted by the heat dissipation efficiency attenuation prediction model to reflect the characteristics of the system changing over time. The resulting thermodynamic model can accurately describe the temperature distribution and heat flow of each component of the air compressor under different working conditions.

[0033] Specifically, based on the established thermodynamic model, the system constructs a multi-objective optimization equation. The optimization equation contains three conflicting objective functions: heat dissipation efficiency maximization function, energy consumption minimization function, and equipment life maximization function. The heat dissipation efficiency maximization function is defined as the degree to which the average temperature of key components deviates from the target temperature. The closer the temperature is to the target value, the higher the heat dissipation efficiency. The target temperature is set to different values ​​according to the type of component, such as the target temperature of the motor stator is 75℃, and the target temperature of the bearing is 65℃. The energy consumption minimization function calculates the total energy consumption of the heat dissipation system, including the electric power consumption of the coolant pump and the energy consumption of the control system. The power consumption of the pump is proportional to the flow rate and pressure, and is calculated through the characteristic curve of the pump. The equipment life maximization function is constructed based on the Arrhenius formula, taking the temperature fluctuation amplitude, temperature gradient and temperature cycle number of key components as influencing factors, and establishing a relationship model with the material aging rate. It is usually difficult to achieve the optimal of these three objective functions at the same time. For example, improving heat dissipation efficiency usually requires increasing the coolant flow rate, but this will increase energy consumption; reducing temperature fluctuations is conducive to extending the life of the equipment, but may require more frequent adjustments and increase the burden on the control system. The decision variables of the optimization equation include the coolant flow rate in each area, the opening of the flow control valve, and the action frequency of the heat dissipation control. These variables together determine the operation strategy of the heat dissipation system.

[0034] Specifically, the solution of the multi-objective optimization equation requires a series of constraints to ensure the practical feasibility of the solution. The set of constraints established by the system first includes the upper temperature limit constraints of each component, such as the motor stator temperature not exceeding 120°C and the bearing temperature not exceeding 80°C. These constraints are hard constraints set based on the heat resistance of the material and safety standards, and cannot be violated under any circumstances. The coolant flow range constraint specifies the upper and lower limits of the flow of each cooling circuit, such as the main circuit flow range of 10-50L / min and the branch circuit flow range of 5-20L / min, taking into account the pipe diameter, pump capacity and fluid dynamics. The pump power limit constraint ensures that the required pump power does not exceed the equipment rating to prevent overload operation. The dynamic constraint based on the coolant quality characteristic parameters is an innovation that adjusts the maximum allowable flow rate in real time according to the conductivity, pH value and turbidity of the coolant. For example, when the turbidity exceeds 30NTU, the maximum flow rate of the main circuit is reduced from 50L / min to 30L / min to reduce the deposition of pollutants and pipeline corrosion under high-speed flow. When the pH value is lower than 6.5 or higher than 8.5, the allowable range of temperature fluctuation is reduced to reduce the risk of corrosion. These constraint sets ensure the practical operability and system safety of the optimization results, while taking into account the potential impact of changes in coolant quality on the cooling system.

[0035] Specifically, inputting the multi-objective optimization equation and its constraint set into the preset multi-objective dung beetle optimizer for solving is the key step to obtain the Pareto optimal solution set. The multi-objective dung beetle optimizer is a swarm intelligence algorithm inspired by the behavior of dung beetles in nature in finding, rolling and burying dung balls. It is particularly suitable for solving complex optimization problems with multiple conflicting objectives. The workflow of the optimizer first encodes the decision variables of the optimization problem into the position vectors of dung beetle individuals. The initial population contains 200 randomly distributed individuals. In each iteration, the algorithm simulates three dung beetle behaviors: foraging, rolling and burying. Foraging behavior corresponds to global search. Each dung beetle updates its search direction according to its own position and the global optimal position. The search radius decreases with the number of iterations, achieving a smooth transition from exploration to development. Rolling behavior corresponds to local refinement. The dung beetle rolls the found solution along the gradient direction and decides whether to accept the new position by comparing the non-dominated relationship between the new and old positions. The burial behavior corresponds to the preservation and diversity maintenance of solutions. The algorithm maintains an external archive to store non-dominated solutions, and maintains the diversity of the solution set through grid division and crowding distance calculation. When processing constraints, an improved constraint processing technique is used to convert the degree of constraint violation into a penalty term and add it to the objective function. The maximum number of iterations in the optimization process is set to 1000, and the convergence condition is that the change in the Pareto front for 50 consecutive iterations is less than the preset threshold. Finally, the optimizer outputs a set of evenly distributed Pareto optimal solutions, each of which represents a parameter configuration of a heat dissipation control strategy. These solutions perform differently on the three objective functions, and it is impossible to simply say which one is absolutely optimal.

[0036] Specifically, after obtaining the Pareto optimal solution set, it is necessary to select the most suitable solution as the actual control strategy according to the current working condition. The current working condition is determined by extracting the data of the most recent time window (usually 10 minutes) from the multidimensional thermal data set, calculating the mean, standard deviation and change trend of each key parameter, and forming a working condition feature vector. The working condition feature vector contains information such as motor load status, ambient temperature conditions, coolant quality status, etc., which fully reflects the current system operating environment. A pre-trained random forest classifier is used to identify and classify the working condition, and the current working condition is classified into predefined working condition categories, such as low load steady state, high load steady state, load fluctuation state, temperature rise state and temperature drop state. Different working condition categories correspond to different heat dissipation demand priorities. For example, under high load steady state, heat dissipation efficiency has the highest priority; under frequent start-stop fluctuation state, equipment life has the highest priority; under low ambient temperature, energy consumption priority is increased. According to the identified working condition category, the system calculates the weight coefficients of the three objective functions to form a comprehensive evaluation function. Then, each solution in the Pareto optimal solution set is scored using the evaluation function, and the top N solutions with the highest scores (usually N=3-5) are selected to form a candidate solution set. Finally, the system considers the smooth switching requirements of the control strategy to avoid drastic changes in control parameters. By comparing with the control strategy at the previous moment, the solution with the smallest change and meeting the performance requirements is selected as the final heat dissipation control strategy. The strategy includes parameters such as the coolant flow set value of each area, the flow control valve opening, and the control cycle, forming a complete heat dissipation control strategy set.

[0037] Furthermore, the step of inputting the multi-objective optimization equation and the set of constraints thereof into a preset multi-objective dung beetle optimizer for solving to obtain a Pareto optimal solution set includes: inputting the multi-objective optimization equation and the set of constraints thereof into a preset multi-objective dung beetle optimizer, mapping the objective function space of the multi-objective optimization equation to the decision variable space through the multi-objective dung beetle optimizer, and obtaining an initial solution population of the optimization problem; calculating the objective function value for each solution in the initial solution population to obtain an evaluation value matrix of the solution, and calculating the dominance relationship between any two solutions based on the evaluation value matrix, and hierarchically sorting the solutions based on the dominance relationship to obtain a position update vector for each solution; and The vector is applied to each solution in the initial solution population to generate an updated solution set, and the solutions in the updated solution set are subjected to constraint checking. When a solution in the updated solution set violates the constraint condition set, the position of the corresponding solution is corrected by a preset penalty function to obtain a candidate solution set that satisfies the constraint; the crowding distance is calculated for the candidate solution set, solutions of the same non-dominated level are sorted based on the crowding distance, and high-quality solutions are selected by an elite retention strategy to obtain a non-dominated solution set in an iterative process; according to a preset iterative termination condition, the optimization process of the multi-objective dung beetle optimizer is iterated cyclically, and when the termination condition is met, the Pareto frontier is extracted from the final non-dominated solution set to obtain the Pareto optimal solution set.

[0038] Specifically, in the embodiment, inputting the multi-objective optimization equation and its constraint set into the preset multi-objective dung beetle optimizer is the starting point for solving complex heat dissipation control problems. The multi-objective dung beetle optimizer first receives three objective functions: heat dissipation efficiency maximization function, energy consumption minimization function and equipment life maximization function, as well as a complete set of constraints, including temperature upper limit constraints, flow range constraints, power limit constraints and coolant quality dynamic constraints. The optimizer converts the optimization problems in these high-dimensional spaces into a form suitable for dung beetle algorithm processing through a coding mechanism. The mapping from the objective function space to the decision variable space adopts a direct coding method, that is, each decision variable (such as the coolant flow rate in each area, the flow control valve opening, etc.) is represented as a dimension in the individual position vector of the dung beetle. For continuous variables, such as flow values, real number coding is directly used; for discrete variables, such as valve opening gears, integer coding is used. This coding method retains the physical meaning of the decision variables and facilitates the interpretation and application of subsequent solutions. The generation of the initial solution population adopts the Latin hypercube sampling method to generate 200 candidate solutions uniformly distributed in the decision space. Compared with pure random sampling, this sampling method can more effectively cover the decision space and increase the diversity of the initial population. At the same time, a small amount of Gaussian noise is added to each initial solution to prevent the solutions from being exactly the same in certain dimensions, further enhancing diversity. In terms of boundary processing, a reflection strategy is adopted. When the sampling point exceeds the allowable range of the decision variable, it is reflected back to the feasible domain along the boundary to ensure that all initial solutions are within the legal range of the decision variable.

[0039] Specifically, calculating the objective function value for each solution in the initial solution population is a basic step in evaluating the quality of the solution. The system inputs the decision variable configuration represented by each solution into the previously constructed thermodynamic model, and simulates and calculates the corresponding heat dissipation efficiency, energy consumption, and equipment life indicators. The heat dissipation efficiency is calculated by the mean absolute error of the temperature of key components deviating from the target temperature. The smaller the value, the higher the heat dissipation efficiency; the energy consumption index is calculated by comprehensively calculating factors such as pump power and control system power consumption; the equipment life index estimates the material aging rate based on factors such as temperature cycle and temperature gradient. These three index values ​​constitute the objective function evaluation vector of each solution, and the evaluation vectors of all solutions together constitute the evaluation value matrix, with the matrix dimension of [number of solutions × number of objective functions]. Based on the evaluation value matrix, the algorithm calculates the dominance relationship between any two solutions. In multi-objective optimization, if solution A is not inferior to solution B in all objective functions (that is, the objective function values ​​are not worse than B), and is strictly better than solution B in at least one objective function, then solution A is said to dominate solution B. For example, if the heat dissipation efficiency and energy consumption of solution A are not inferior to solution B, and the equipment life index is strictly better than solution B, then A dominates B. By traversing and comparing all pairs of solutions, the number of times each solution is dominated (domination count) and the set of solutions dominated by the solution (dominated set) are calculated. According to the domination count, all solutions are sorted into non-dominated order, and solutions with a domination count of 0 (i.e., solutions that are not dominated by any other solution) are classified into the first frontier (level 1), and then solutions with a domination count of 0 are found again from the remaining solutions and classified into the second frontier (level 2), and so on, until all solutions are assigned levels. Based on this classification result, combined with the position update mechanism of the dung beetle optimization algorithm, the position update vector of each solution is calculated. The calculation of the position update vector is divided into three parts: the global optimal guidance term, the random exploration term, and the local refinement term. The global optimal guidance term comes from a randomly selected solution in the current non-dominated solution set, the random exploration term adds appropriate random perturbations, and the local refinement term considers understanding its own historical optimal position.

[0040] Specifically, applying the position update vector to each solution in the initial solution population is the core iterative step of the dung beetle optimization algorithm. For each solution, a new candidate position is generated based on its current position and the calculated position update vector. The position update adopts an update strategy with inertia weight, that is, the new position is not only affected by the current update vector, but also partially retains the information of the original position. This strategy helps to balance the exploration and development capabilities of the algorithm. The inertia weight is dynamically adjusted with the number of iterations. The initial value is set to 0.9 and linearly decreases to 0.4 as the iteration proceeds, prompting the algorithm to gradually shift from global exploration to local refinement. The updated solution set needs to undergo strict constraint checks to ensure that all constraints are met. When a solution violates a constraint, the solution is corrected by a static penalty method. Specifically, for solutions that violate the upper temperature limit constraint, the amount exceeding the upper temperature limit is calculated, and the coolant flow rate is increased proportionally until the temperature constraint is met; for solutions that violate the flow range constraint, the flow value exceeding the range is directly truncated to the boundary value; for solutions that violate the pump power limit, the flow rate in each area is proportionally reduced until the total power meets the limit; for solutions that violate the dynamic constraint of coolant quality, the flow rate upper limit is dynamically adjusted according to the current coolant quality parameters, and the excess part is truncated. This constraint processing method not only ensures the feasibility of the solution, but also improves the convergence efficiency of the algorithm through a correction strategy with clear physical meaning. After strict constraint checking, each solution in the candidate solution set meets all constraints, ensuring the practical operability of the optimization results.

[0041] Specifically, calculating the crowding distance of the candidate solution set is an important means to maintain the diversity of the solution set. The crowding distance reflects the distribution density of the solution in the target space. The larger the distance, the sparser the solutions around the solution. The calculation process first sorts the solutions according to each objective function value, and then calculates the normalized distance of the adjacent solutions on the objective function. The final crowding distance of the solution is the sum of the distances on all objective functions. The boundary solutions (i.e., the solutions that achieve extreme values ​​on any objective function) are assigned infinite crowding distances to ensure that they can be retained. Based on the crowding distance and the previous non-dominated level, the system implements an elite retention strategy. This strategy first merges the parent solution set and the child solution set, and then selects solutions in order from low to high non-dominated levels (the lower the level, the better) until the upper limit of the population size is reached. When selecting a solution of a certain level will cause the population size to exceed, the solutions within the level are selected from large to small crowding distances, and the solutions distributed in the sparse area are preferentially retained. This elite retention strategy ensures the retention of high-quality solutions and the uniformity of the distribution of the solution set, preventing the algorithm from converging to the local Pareto front too early. In the actual implementation, the fast non-dominated sorting algorithm is used to improve the sorting efficiency, which is particularly important for large-scale problems. The solutions selected by the elite retention strategy constitute the next generation of non-dominated solution sets, which achieve different balances on the three objective functions and represent different heat dissipation control strategy choices.

[0042] Specifically, the iteration process of the multi-objective dung beetle optimizer is cycled according to the preset termination conditions. The termination conditions set three criteria: the maximum number of iterations (1000 times), the stability of the Pareto front, and the calculation time limit. The stability of the Pareto front is measured by the change rate of the hypervolume index of the Pareto front less than 0.1% in 50 consecutive iterations. This index comprehensively reflects the convergence and diversity of the non-dominated solution set. In each iteration, the system records the current non-dominated solution set and updates the global Pareto front. The convergence behavior in the iterative process is manifested in that the Pareto front moves quickly to the ideal point of the objective function in the early stage, the front boundary continues to expand in the middle stage, and the front distribution gradually becomes uniform in the late stage. In order to improve the stability of the algorithm, an adaptive control parameter strategy is adopted to dynamically adjust the key parameters of the dung beetle optimizer such as the rolling ball radius and the burial depth according to the progress of the search. When any termination condition is met, the algorithm stops iterating and extracts the Pareto front from the final non-dominated solution set. This front consists of solutions that are not dominated by any other solution and represents the optimal set of trade-off solutions that can be achieved under the current problem setting. In practical applications, in order to facilitate the selection and implementation of control strategies, the extracted Pareto frontier is post-processed, including solution sparsification (selecting representative solutions through clustering methods) and solution mapping (mapping the solutions in the decision space back to specific control parameter configurations). The final Pareto optimal solution set usually contains 30-50 different solutions, each of which represents a specific heat dissipation control strategy that achieves different degrees of balance between the three goals of heat dissipation efficiency, energy consumption, and equipment life.

[0043] 104. Dynamically partition the heat dissipation system of the air compressor, and coordinately adjust the flow rate and flow direction of the coolant in each area of ​​the dynamic partition according to the heat dissipation control strategy set to achieve heat dissipation of the air compressor.

[0044] In one embodiment of the present invention, the heat dissipation system of the air compressor is dynamically partitioned, and the flow rate and flow direction of the coolant in each area of ​​the dynamic partition are coordinated according to the heat dissipation control strategy set to achieve the heat dissipation of the air compressor, including: extracting temperature field distribution information according to the direct thermal characteristic parameters in the multidimensional thermal data set, and calculating the heat flow characteristics according to the direct thermal characteristic parameters and the indirect thermal characteristic parameters; clustering the temperature field distribution information and heat flow characteristics using an adaptive K-means clustering algorithm, and dynamically partitioning the heat dissipation system of the air compressor based on the clustering results to obtain N control sub-areas, where N is an integer from 2 to 8. ; Based on the direct thermal characteristic parameters, the heat load index is calculated for each of the control sub-areas, and an inter-area heat transfer matrix is ​​constructed to characterize the heat transfer relationship between the control sub-areas; according to the heat dissipation control strategy set, the heat load index and the heat transfer matrix, a hierarchical progressive control method is used to determine the cooling resource allocation strategy for each of the control sub-areas; the coolant parameters of each of the control sub-areas are adjusted in real time according to the cooling resource allocation strategy through the coolant flow regulating valve and the flow direction control valve in the heat dissipation system, and the cooling intensity of adjacent areas is collaboratively controlled based on the heat transfer matrix to achieve heat dissipation of the air compressor.

[0045] Specifically, the dynamic partitioning of the cooling system of the air compressor first requires the extraction of temperature field distribution information from the multidimensional thermal data set. This step is achieved by processing the temperature data in the direct thermal characteristic parameters, and the specific process includes data denoising, interpolation and reconstruction. Data denoising uses the wavelet transform method, and the db4 wavelet basis function is selected to decompose the temperature signal into 5 layers. After removing the high-frequency noise, the signal is reconstructed to obtain smooth temperature data. The interpolation processing uses the radial basis function (RBF) interpolation method to expand the discrete temperature sampling points into a continuous temperature field. The interpolation grid density is set to 5mm×5mm, which is sufficient to capture the temperature gradient changes inside the air compressor. Through this processing, the system obtains the complete temperature field distribution information inside the air compressor, including the surface temperature, internal temperature and temperature gradient of each component. At the same time, the system also needs to calculate the heat flow characteristics, which is a process that comprehensively utilizes direct thermal characteristic parameters and indirect thermal characteristic parameters. The heat flow characteristic calculation adopts the heat flux estimation method, combining direct thermal characteristic parameters such as flow rate and pressure and indirect thermal characteristic parameters such as speed and current, and inverts the heat flux distribution of each part of the air compressor through the thermodynamic model. Specifically, the heat flux of the motor part is calculated based on the current, resistance and motor efficiency, the heat flux of the compression chamber is calculated based on the compression ratio, intake temperature and air flow, and the heat flux of the bearing part is estimated based on the speed, load and friction coefficient. These heat flow characteristics describe the generation and flow of heat in the air compressor, and together with the temperature field distribution information, they form the data basis for dynamic partitioning.

[0046] Specifically, clustering the temperature field distribution information and heat flow characteristics using an adaptive K-means clustering algorithm is the core technology for realizing dynamic partitioning. The traditional K-means algorithm has problems such as the difficulty in determining the number of clusters K, sensitivity to the initial cluster center, and unsuitability for non-spherical distribution. Therefore, this embodiment adopts an improved adaptive K-means clustering algorithm. The algorithm first normalizes the temperature field and heat flow characteristic data to eliminate the dimensional differences between different physical quantities. Then the optimal number of clusters K is automatically determined by the silhouette coefficient method. The specific process is to try K values ​​from 2 to 10, cluster and calculate the silhouette coefficient respectively, and select the K value with the largest silhouette coefficient as the final number of clusters. In order to solve the problem of initial cluster center sensitivity, the algorithm uses the density peak method to select the initial center point, that is, select a point with high local density and a long distance from the high-density point as the initial cluster center. During the iteration process, the system dynamically adjusts the feature weights, assigns higher weights to areas with large temperature gradients, and enables hot spots to form independent clusters. When the clustering result is stable or reaches the maximum number of iterations (usually set to 100 times), the clustering process ends. The final clustering result divides the air compressor physical space into N control sub-areas, where the value of N is usually between 2 and 8, depending on the complexity and working state of the air compressor. Each control sub-area has similar temperature characteristics and heat flow characteristics, and is suitable for a unified cooling control strategy. The partition results will be dynamically adjusted as the working state of the air compressor changes. For example, under high load conditions, the motor area may be divided into multiple sub-areas to achieve more refined control.

[0047] Specifically, calculating the heat load index for each control sub-region based on direct thermal characteristic parameters is an important step in evaluating the heat dissipation requirements of each region. The heat load index is a comprehensive indicator that reflects the urgency and importance of heat dissipation in the region. The calculation process first extracts the temperature characteristics of each control sub-region, including the average temperature, maximum temperature, temperature gradient, and temperature change rate. The average temperature is calculated by the arithmetic average of all temperature sampling points in the region, the maximum temperature is directly taken as the maximum temperature in the region, the temperature gradient is calculated by the finite difference method to calculate the modulus of the temperature spatial derivative, and the temperature change rate is obtained by dividing the temperature change in the time window by the time interval. Then the system sets the temperature weight coefficient according to the temperature resistance grade of the component, and key components such as motor stators and bearings have higher weights. The final heat load index is calculated by the weighted summation method, and the higher its value indicates that the heat dissipation demand of the region is more urgent. At the same time, the system constructs an inter-region heat transfer matrix to describe the heat exchange relationship between each control sub-region. The matrix is ​​an N×N square matrix, and each element represents the heat transfer coefficient from region i to region j. The heat transfer coefficient is calculated based on the contact area, material thermal conductivity, distance, and convection heat transfer conditions of the two regions. For example, the heat transfer coefficient between two adjacent metal parts is high, while the heat transfer coefficient between areas isolated by insulating materials is low. This heat transfer matrix reflects the flow path and intensity of heat inside the air compressor, providing an important basis for subsequent coordinated control.

[0048] Specifically, determining the cooling resource allocation strategy for each control sub-area based on the heat dissipation control strategy set, heat load index, and heat transfer matrix is ​​a key link in achieving precise heat dissipation. This step adopts a hierarchical progressive control method, dividing the control process into three levels: macro-overall strategy layer, meso-region coordination layer, and micro-execution control layer. At the macro-overall strategy layer, the system selects the optimal strategy under the current working conditions from the heat dissipation control strategy set to determine the allocation principle and control target of the total cooling resources (such as total flow, total power). At the meso-region coordination layer, the system performs preliminary resource allocation based on the heat load index of each control sub-area. The area with a higher heat load index obtains more cooling resources. The preliminary allocation result is corrected by the heat transfer matrix to consider the influence of heat exchange between regions. For example, if region A transfers a large amount of heat to region B, the cooling resources of region A will be appropriately increased or the cooling sequence of the two regions will be adjusted. The correction process adopts an iterative calculation method until the predicted temperature of each region meets the control target. At the micro-execution control layer, the system converts the resource allocation results into specific control parameters, including the flow set value, valve opening, flow direction switching timing, etc. of each region. The generation of control parameters takes into account the physical limitations and dynamic characteristics of the actuator, such as the response time of the valve and the measurement error of the flow meter, and improves the control accuracy through the combination of feedforward and feedback. The cooling resource allocation strategy finally formed not only meets the overall heat dissipation requirements, but also takes into account the personalized needs of each area, achieving the optimal configuration of cooling resources.

[0049] Specifically, the execution link of heat dissipation control is to adjust the coolant parameters of each control sub-area in real time according to the cooling resource allocation strategy through the coolant flow control valve and flow direction control valve in the heat dissipation system. The system first smoothes the cooling resource allocation strategy to avoid system oscillation caused by sudden changes in control parameters. The smoothing process uses a combination of sliding average and gradient limitation to ensure the continuity and stability of the control signal. The processed control signal is sent to each actuator through the communication interface, including the variable frequency pump, electric control valve and electromagnetic flow valve. The variable frequency pump adjusts the speed according to the total flow demand, the electric control valve adjusts the opening according to the allocation ratio, and the electromagnetic flow valve switches the flow direction of the cooling circuit as needed. The action of the actuator is monitored in real time and closed-loop controlled. The system adjusts the control signal and compensates for the execution error through the feedback information of the flow sensor, pressure sensor and temperature sensor. At the same time, the cooling intensity of adjacent areas is coordinated and controlled based on the heat transfer matrix to avoid local overcooling or overheating. For example, when it is detected that the temperature in a certain area rises abnormally, the system not only increases the cooling flow in the area, but also strengthens the cooling of the upstream area where the heat is mainly sourced. The collaborative control adopts a prediction-correction strategy to predict the temperature change trend of each area based on the heat transfer model, adjust the cooling parameters in advance, and prevent the temperature from exceeding the limit. Through this fine flow and flow direction control, the system realizes the precise management of the temperature of each component of the air compressor, ensuring that the key components work within the optimal temperature range, while avoiding the waste of cooling resources and unnecessary temperature fluctuations, extending the equipment life and reducing energy consumption.

[0050] Furthermore, the coolant flow regulating valve and the flow direction control valve in the heat dissipation system adjust the coolant parameters of each of the control sub-areas in real time according to the cooling resource allocation strategy, and at the same time, coordinately control the cooling intensity of adjacent areas based on the heat transfer matrix, including: prioritizing each of the control sub-areas according to the heat load index, allocating the control cycle and the response order according to the sorting result, and obtaining a control priority list; performing proportional-integral-differential control calculation on the flow parameters and flow direction parameters in the cooling resource allocation strategy, converting the control signal into a valve opening instruction, and obtaining a control sequence of the actuator; calculating the thermal coupling influence coefficient between each of the control sub-areas based on the heat transfer matrix, adjusting the cooling resource allocation of adjacent areas according to the thermal coupling influence coefficient, and obtaining a coordinated control compensation amount; fusing the control sequence and the coordinated control compensation amount to generate a valve control instruction, and performing real-time closed-loop correction according to the system dynamic response to obtain a control signal for the coolant flow regulating valve and the flow direction control valve; applying the control signal to the coolant flow regulating valve and the flow direction control valve to adjust the coolant parameters of each of the control sub-areas in real time.

[0051] Specifically, the heat load index, as a comprehensive indicator to measure the urgency of regional heat dissipation, directly determines the allocation priority of cooling resources. The sorting process first normalizes the heat load index to eliminate the differences in the magnitude of the index under different working conditions. The normalization uses the Min-Max method to map the heat load index of each region to the [0,1] interval. Then the system sets three priority thresholds: 0.8, 0.5, and 0.3, and divides the control sub-areas into four priority levels: emergency cooling area (index>0.8), priority cooling area (0.5<index≤0.8), conventional cooling area (0.3<index≤0.5), and monitoring area (index≤0.3). Within each priority level, detailed sorting is performed based on the precise value of the heat load index. Based on the priority sorting results, the system assigns different control cycles and response orders. The emergency cooling zone uses a fast control cycle of 100ms, and the control instructions are executed first; the priority cooling zone uses a control cycle of 200ms, and the second priority is execution; the conventional cooling zone uses a control cycle of 500ms; the monitoring zone only collects data without active regulation, and the control cycle is 1000ms. This hierarchical control method ensures that computing resources and control resources are concentrated on the areas that need the most attention, while ensuring the real-time performance of the overall system. The sorting results ultimately form a control priority list, which contains the ID, priority level, control cycle and execution order of each area. This list guides the generation and execution order of subsequent control signals.

[0052] Specifically, the proportional-integral-differential (PID) control calculation of the flow parameters and flow direction parameters in the cooling resource allocation strategy is a key step in converting high-level strategies into specific execution instructions. The process adopts a hierarchical PID control structure, including a main loop PID controller and multiple sub-loop PID controllers. The main loop controller takes the deviation between the average temperature of the area and the target temperature as input and outputs the total cooling required by the area; the sub-loop controller decomposes the total cooling amount into specific control quantities of the flow parameter and the flow direction parameter. For the flow parameter, the proportional coefficient Kp of the PID controller is set to 0.8-1.2 (dynamically adjusted according to the regional characteristics), the integral coefficient Ki is set to 0.1-0.3, and the differential coefficient Kd is set to 0.05-0.15. A larger proportional coefficient ensures that the system responds quickly to temperature changes, an appropriate integral coefficient eliminates steady-state errors, and a smaller differential coefficient provides the necessary advance correction. The control of the flow direction parameter adopts a discrete state switching method. The system determines the switching timing according to the temperature distribution. Dead zone and hysteresis mechanisms are introduced during the switching process to avoid system instability caused by frequent switching. PID calculation takes into account a variety of anti-saturation measures, including integral limiting, variable speed integral and integral separation, which effectively prevents the integral saturation problem. The output of the control calculation needs to be converted into an opening instruction that the actuator can understand. The conversion process takes into account the nonlinear characteristics of the valve and ensures the linear correspondence between the control quantity and the actual flow through linear compensation. The final generated actuator control sequence contains the opening instructions of all regulating valves and flow valves in each control sub-area, as well as the corresponding execution timestamps. These instructions will be sent directly to the actuator to complete the specific operation.

[0053] Specifically, calculating the thermal coupling influence coefficient between each control sub-region based on the heat transfer matrix is ​​the theoretical basis for realizing inter-regional coordinated control. The thermal coupling influence coefficient reflects the degree of influence of temperature change in one region on another region, and the calculation process is based on the heat transfer matrix and temperature response characteristics. The system first normalizes the heat transfer matrix to obtain the temperature response function under unit heat transfer. Then the thermal coupling influence coefficient is calculated by the temperature response function and the time constant. The coefficient is an N×N matrix (N is the number of control sub-regions), and each element Hij represents the intensity of the influence of temperature change in region i on region j. The calculation of the influence coefficient takes into account the time lag effect, that is, it takes a certain amount of time for heat transfer to be reflected in temperature changes. The time lag is described by the first-order lag model, and the time constant is determined according to the physical distance and material properties between regions, usually in the range of 10-300 seconds. According to the thermal coupling influence coefficient, the system adjusts the cooling resource allocation of adjacent regions. The adjustment principle is: if area A has a strong thermal coupling effect on area B, and the temperature of area B is close to the upper limit, then increase the cooling intensity of area A; if area A and area B have a thermal impact on each other, coordinate the cooling timing of the two areas to avoid strengthening or weakening the cooling at the same time. This collaborative control based on thermal coupling takes into account the overall thermal balance of the system, rather than just focusing on the local control of a single area. The calculation of the collaborative control compensation amount adopts a weighted summation method, which comprehensively considers the temperature state, thermal coupling intensity and time lag of each adjacent area to obtain a correction value for the original control amount. These compensation amounts ensure that the system will not cause new hot spots or waste energy while dealing with hot spot problems.

[0054] Specifically, the fusion of the control sequence and the cooperative control compensation is the process of generating the final control instruction. This process adopts a method combining weighted fusion and priority arbitration. For continuous control quantities such as flow parameters, the weighted fusion sums the original control sequence value and the cooperative control compensation in a ratio of 6:4 to obtain the fused control value. This ratio setting not only ensures the dominant position of the original control strategy, but also gives sufficient adjustment space for cooperative control. For discrete control quantities such as flow direction switching, the priority arbitration method is adopted, that is, when the priority of the cooperative control demand is higher than the original control sequence, the cooperative control instruction is adopted, otherwise the original control sequence remains unchanged. The fused control instruction is then smoothed to avoid sudden changes in the control quantity. The smoothing process adopts a method combining sliding window averaging and change rate limitation. The window size is dynamically adjusted according to the control cycle, and the change rate limitation is set according to the response capability of the actuator. The processed instruction is converted into a standard valve control command, including position instruction, speed instruction and acceleration instruction, to ensure the smoothness of valve movement. The control system performs closed-loop correction through a real-time feedback loop, collects feedback signals from flow sensors, pressure sensors and temperature sensors, calculates control deviations, and updates control parameters. The closed-loop correction adopts the model predictive control (MPC) method, which uses the system model to predict the system response in the future (usually 5-10 control cycles) and optimize the current control decision. This feedforward-feedback combined control architecture improves the system's ability to suppress disturbances and control accuracy. The final control signal contains complete timing information and control quantity information, which directly drives the movement of the coolant flow control valve and the flow direction control valve.

[0055] Specifically, applying control signals to the coolant flow control valve and the flow control valve is the final step in executing heat dissipation control. The control signal is sent to each execution unit through an industrial field bus such as Profibus-DP or Modbus. This process needs to consider communication delay and signal synchronization issues. The system uses time synchronization protocols such as PTP (Precision Time Protocol) to ensure the clock consistency of each execution unit. The control instructions are timestamped, and the execution unit determines the execution time based on the timestamp, ensuring the coordination of the actions of multiple actuators. The flow control valve uses an electric proportional valve, which is driven by a stepper motor or servo motor. The control accuracy reaches 1% of the full stroke of the valve, and the response time is less than 200ms. The flow control valve uses a solenoid valve or an electric ball valve to switch the flow direction of the cooling circuit. All valves are equipped with position feedback devices to report the current opening in real time to form a complete closed-loop control. To protect the actuator, the system implements multiple safety measures, including valve jam detection, overcurrent protection, and emergency shutdown logic. The valve stuck detection is realized by comparing the difference between the control signal and the position feedback signal. When the difference exceeds the threshold and the duration exceeds the set value, the system determines that the valve is stuck, sounds an alarm and switches to the backup circuit. The overcurrent protection monitors the motor drive current to prevent valve overload. The emergency shutdown logic switches the system to a safe state when a serious abnormality is detected, and all flow valves return to the preset safe position (designed according to the fail-safe principle). Through these sophisticated control and protection measures, the system can accurately and reliably adjust the coolant parameters of each control sub-area, achieve precise heat dissipation control, and ensure the safe and efficient operation of the air compressor under various working conditions.

[0056] In this embodiment, a multidimensional thermal data set of a fuel cell air compressor is collected in layers; features are extracted from the multidimensional thermal data set, and the extracted feature data set is input into a neural network model based on deep transfer learning to obtain a corresponding heat dissipation efficiency attenuation prediction model; a multi-objective optimization equation of the air compressor is established based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and the multi-objective optimization equation is solved to obtain a heat dissipation control strategy set; the heat dissipation system of the air compressor is dynamically partitioned, and the flow rate and flow direction of the coolant in each area of ​​the dynamic partition are coordinated and adjusted according to the heat dissipation control strategy set to achieve heat dissipation of the air compressor. The present invention solves the problems of heat dissipation efficiency attenuation and uneven resource allocation caused by changes in coolant quality and integrated control in the prior art.

[0057] The above describes the heat dissipation method of the fuel cell air compressor in the embodiment of the present invention. The following describes the heat dissipation device of the fuel cell air compressor in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a heat dissipation device for a fuel cell air compressor includes: The data acquisition module 201 is used to perform hierarchical acquisition of direct thermal characteristic parameters, indirect thermal characteristic parameters and coolant quality characteristic parameters of the air compressor of the fuel cell to obtain a multi-dimensional thermal data set of the air compressor; The pattern recognition module 202 is used to extract features from the multidimensional thermal data set, and input the extracted feature data set into a neural network model based on deep transfer learning to obtain a corresponding heat dissipation efficiency attenuation prediction model; A strategy generation module 203 is used to establish a multi-objective optimization equation for the air compressor based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and solve the multi-objective optimization equation to obtain a heat dissipation control strategy set; The partition control module 204 is used to dynamically partition the heat dissipation system of the air compressor, and coordinately adjust the flow rate and flow direction of the coolant in each area of ​​the dynamic partition according to the heat dissipation control strategy set to achieve heat dissipation of the air compressor.

[0058] In an embodiment of the present invention, the fuel cell air compressor heat dissipation device runs the above-mentioned fuel cell air compressor heat dissipation method, and the fuel cell air compressor heat dissipation device collects the multidimensional thermal data set of the fuel cell air compressor in layers; extracts features from the multidimensional thermal data set, and inputs the extracted feature data set into a neural network model based on deep transfer learning to obtain a corresponding heat dissipation efficiency attenuation prediction model; establishes a multi-objective optimization equation for the air compressor based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and solves the multi-objective optimization equation to obtain a heat dissipation control strategy set; dynamically partitions the heat dissipation system of the air compressor, and coordinately adjusts the flow rate and flow direction of the coolant in each area of ​​the dynamic partition according to the heat dissipation control strategy set to achieve heat dissipation of the air compressor. The present invention solves the problems of heat dissipation efficiency attenuation and uneven resource allocation caused by changes in coolant quality and integrated control in the prior art.

[0059] above Figure 2 The heat dissipation device of the fuel cell air compressor in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The heat dissipation device of the fuel cell air compressor in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0060] Figure 33 is a schematic diagram of the structure of a fuel cell air compressor heat dissipation device provided by an embodiment of the present invention. The fuel cell air compressor heat dissipation device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the fuel cell air compressor heat dissipation device 300. Furthermore, the processor 310 can be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the fuel cell air compressor heat dissipation device 300 to implement the steps of the above-mentioned fuel cell air compressor heat dissipation method.

[0061] The fuel cell air compressor cooling device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that Figure 3 The illustrated structure of the heat dissipation device for the fuel cell air compressor does not constitute a limitation on the heat dissipation device for the fuel cell air compressor provided by the present invention, and may include more or fewer components than illustrated, or a combination of certain components, or a different arrangement of components.

[0062] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0063] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0064] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for heat dissipation of a fuel cell air compressor, characterized in that: The fuel cell air compressor heat dissipation method comprises: The direct thermal characteristic parameters, indirect thermal characteristic parameters and coolant quality characteristic parameters of the fuel cell air compressor are collected in layers to obtain a multi-dimensional thermal data set of the air compressor; Extracting features from the multidimensional thermal data set, and inputting the extracted feature data set into a neural network model based on deep transfer learning to obtain a corresponding heat dissipation efficiency attenuation prediction model; Establishing a multi-objective optimization equation for the air compressor based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and solving the multi-objective optimization equation to obtain a heat dissipation control strategy set; The heat dissipation system of the air compressor is dynamically partitioned, and the flow rate and flow direction of the coolant in each area of ​​the dynamic partition are coordinated and adjusted according to the heat dissipation control strategy set to achieve the heat dissipation of the air compressor.

2. The method for heat dissipation of a fuel cell air compressor according to claim 1, characterized in that: The direct thermal characteristic parameters, indirect thermal characteristic parameters and coolant quality characteristic parameters of the air compressor of the fuel cell are collected in layers to obtain a multi-dimensional thermal data set of the air compressor, including: A first layer of data collection points is set at a heat exchange position where the coolant of the air compressor of the fuel cell flows through, to collect the direct thermal characteristic parameters, wherein the direct thermal characteristic parameters include temperature data, flow rate data and pressure data; A second layer of data collection points is set at the preset core components of the air compressor to collect the indirect thermal characteristic parameters, wherein the indirect thermal characteristic parameters include speed data, current data, and vibration data; A third layer of data collection points is set in the coolant circulation system of the air compressor heat dissipation system to collect coolant quality characteristic parameters, wherein the coolant quality characteristic parameters include conductivity, pH value and turbidity data; The direct thermal characteristic parameters, the indirect thermal characteristic parameters and the coolant quality characteristic parameters are subjected to time synchronization and data preprocessing to obtain a multi-dimensional thermal data set of the air compressor.

3. The method for heat dissipation of a fuel cell air compressor according to claim 1, characterized in that: The feature extraction of the multidimensional thermal data set and inputting the extracted feature data set into a neural network model based on deep transfer learning to obtain a corresponding heat dissipation efficiency attenuation prediction model includes: Performing time domain feature extraction, frequency domain feature extraction and time-frequency feature extraction on each parameter in the multidimensional thermal data set to obtain target domain features; The heat dissipation efficiency attenuation mode features of the existing air compressors in the preset source domain knowledge base are used as source domain features, and the feature mapping algorithm of the attention mechanism is used to perform adaptive conversion between the source domain features and the target domain features; The target domain features are fused with the adapted source domain features, and the fused features are dynamically weighted by a hierarchical attention network to obtain a feature data set; The feature data set is input into a neural network model based on deep transfer learning for training, and the heat dissipation efficiency attenuation prediction model is obtained by minimizing the distribution difference between the source domain features and the target domain features and the heat dissipation efficiency prediction error.

4. The method for heat dissipation of a fuel cell air compressor according to claim 1, characterized in that: The multi-objective optimization equation of the air compressor is established based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and the heat dissipation control strategy set is obtained by solving the multi-objective optimization equation, including: According to the multidimensional thermal data set and the heat dissipation efficiency attenuation prediction model, a thermodynamic model of the air compressor heat dissipation system is established, and the multi-objective optimization equation is constructed based on the thermodynamic model; Establishing a set of constraints for the multi-objective optimization equation, wherein the set of constraints includes upper temperature limits of each component, coolant flow range constraints, pump power limit constraints, and dynamic constraints based on coolant quality characteristic parameters; Inputting the multi-objective optimization equation and the constraint condition set thereof into a preset multi-objective dung beetle optimizer to solve and obtain a Pareto optimal solution set; The current operating condition is determined according to the multidimensional thermal data set, and the Pareto optimal solution set is dynamically screened according to the current operating condition, and a corresponding heat dissipation control strategy set is generated according to the screened Pareto optimal solution set.

5. The method for heat dissipation of a fuel cell air compressor according to claim 4, characterized in that: The step of inputting the multi-objective optimization equation and the constraint condition set thereof into a preset multi-objective dung beetle optimizer to solve and obtain a Pareto optimal solution set includes: Inputting the multi-objective optimization equation and the constraint condition set thereof into a preset multi-objective dung beetle optimizer, mapping the objective function space of the multi-objective optimization equation into the decision variable space through the multi-objective dung beetle optimizer, and obtaining an initial solution population of the optimization problem; Calculating the objective function value for each solution in the initial solution population to obtain an evaluation value matrix of the solution, and calculating the dominance relationship between any two solutions based on the evaluation value matrix, hierarchically sorting the solutions based on the dominance relationship, and obtaining a position update vector of each solution; Applying the position update vector to each solution in the initial solution population to generate an updated solution set, and performing constraint checking on the solutions in the updated solution set, and when a solution in the updated solution set violates the constraint condition set, correcting the position of the corresponding solution by a preset penalty function to obtain a candidate solution set that satisfies the constraint; Calculating the crowding distance of the candidate solution set, sorting the solutions with the same non-dominated level based on the crowding distance, and selecting high-quality solutions through an elite retention strategy to obtain a non-dominated solution set in an iterative process; According to the preset iterative termination condition, the optimization process of the multi-objective dung beetle optimizer is iterated cyclically. When the termination condition is met, the Pareto frontier is extracted from the final non-dominated solution set to obtain the Pareto optimal solution set.

6. The method for heat dissipation of a fuel cell air compressor according to claim 1, characterized in that: Dynamically partitioning the heat dissipation system of the air compressor, and collaboratively adjusting the flow rate and flow direction of the coolant in each area of ​​the dynamic partition according to the heat dissipation control strategy set to achieve the heat dissipation of the air compressor includes: Extracting temperature field distribution information according to direct thermal characteristic parameters in the multidimensional thermal data set, and calculating heat flow characteristics according to the direct thermal characteristic parameters and the indirect thermal characteristic parameters; Adopting an adaptive K-means clustering algorithm to cluster the temperature field distribution information and heat flow characteristics, and dynamically partitioning the heat dissipation system of the air compressor based on the clustering result to obtain N control sub-areas, where N is an integer from 2 to 8; Calculating a heat load index for each of the control sub-regions based on the direct heat characteristic parameters, and constructing an inter-region heat transfer matrix to characterize the heat transfer relationship between the control sub-regions; Determine the cooling resource allocation strategy for each of the control sub-areas using a hierarchical progressive control method according to the heat dissipation control strategy set, the heat load index and the heat transfer matrix; The coolant flow regulating valve and the flow direction control valve in the heat dissipation system are used to adjust the coolant parameters of each control sub-area in real time according to the cooling resource allocation strategy, and the cooling intensity of adjacent areas is collaboratively controlled based on the heat transfer matrix to achieve heat dissipation of the air compressor.

7. The method for heat dissipation of a fuel cell air compressor according to claim 6, characterized in that: The coolant flow regulating valve and the flow direction control valve in the heat dissipation system are used to adjust the coolant parameters of each control sub-area in real time according to the cooling resource allocation strategy, and the cooling intensity of adjacent areas is collaboratively controlled based on the heat transfer matrix, including: Prioritizing each of the control sub-areas according to the heat load index, allocating a control cycle and a response order according to the ranking result, and obtaining a control priority list; Performing proportional-integral-differential control calculation on the flow parameters and flow direction parameters in the cooling resource allocation strategy, converting the control signal into a valve opening instruction, and obtaining a control sequence of the actuator; Calculating the thermal coupling influence coefficient between each of the control sub-areas based on the heat transfer matrix, adjusting the cooling resource allocation of adjacent areas according to the thermal coupling influence coefficient, and obtaining a coordinated control compensation amount; The control sequence and the coordinated control compensation are integrated to generate valve control instructions, and real-time closed-loop correction is performed according to the dynamic response of the system to obtain control signals for the coolant flow regulating valve and the flow direction control valve; The control signal is applied to the coolant flow regulating valve and the flow direction control valve to adjust the coolant parameters of each control sub-area in real time.

8. A heat dissipation device for a fuel cell air compressor, characterized in that: The fuel cell air compressor heat dissipation device comprises: A data acquisition module is used to perform layered acquisition of direct thermal characteristic parameters, indirect thermal characteristic parameters and coolant quality characteristic parameters of the air compressor of the fuel cell to obtain a multi-dimensional thermal data set of the air compressor; A pattern recognition module, used for extracting features from the multidimensional thermal data set, inputting the extracted feature data set into a neural network model based on deep transfer learning, and obtaining a corresponding heat dissipation efficiency attenuation prediction model; A strategy generation module, used to establish a multi-objective optimization equation for the air compressor based on the multi-dimensional thermal data set and the heat dissipation efficiency attenuation prediction model, and solve the multi-objective optimization equation to obtain a heat dissipation control strategy set; The partition control module is used to dynamically partition the heat dissipation system of the air compressor, and coordinately adjust the flow rate and flow direction of the coolant in each area of ​​the dynamic partition according to the heat dissipation control strategy set to achieve the heat dissipation of the air compressor.

9. A fuel cell air compressor heat dissipation device, characterized in that: The fuel cell air compressor heat dissipation device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the fuel cell air compressor heat dissipation device executes the steps of the fuel cell air compressor heat dissipation method as described in any one of claims 1-7.

Images