A temperature rise co-monitoring and cooling control system

Abstract

The application discloses a temperature rise cooperative monitoring and cooling control system, which comprises a multi-dimensional cooperative perception module, a predictive cooperative decision module, a multi-modal adaptive cooling execution module and an edge computing unit, and realizes data interaction with the edge computing unit through an industrial Ethernet bus; the edge computing unit is also bidirectionally connected with a controller of a water-cooled high-voltage reactive power compensation device to realize device operation state data acquisition and load limiting signal transmission. The application belongs to the technical field of temperature control, and through multi-dimensional global perception, long-time sequence temperature rise accurate prediction and fault linkage emergency response, the application can avoid core component overheating risk in advance, quickly dispose of precursors of faults, effectively reduce the failure rate of faults such as thermal breakdown and short circuit, guarantee the stable operation of a power grid, and improve the operation safety and monitoring accuracy.

Description

Technical Field

[0001] This invention relates to the field of temperature control technology, specifically a temperature rise coordinated monitoring and cooling control system. Background Technology

[0002] Water-cooled high-voltage reactive power compensation devices are core equipment for power factor optimization and voltage stability control in power grids. Their internal core components, such as reactors and capacitors, are prone to generating a large amount of heat under high voltage and high current conditions. The temperature rise control effect directly determines the service life of the device and the safety of power grid operation.

[0003] Existing temperature rise control technologies face three major technical challenges: 1. Limited monitoring dimensions and insufficient accuracy: It only collects the temperature of a single point on the surface of the core component, without linking it to operating parameters such as working current, operating voltage, and vibration. It cannot distinguish between normal temperature rise and fault temperature rise, which is prone to misjudgment. Moreover, single-point temperature measurement cannot cover the temperature rise distribution across the entire area, making it difficult to detect potential local overheating hazards.

[0004] 2. Passive and delayed cooling control: It adopts a fixed threshold mode that starts when the temperature reaches the target, and has no ability to predict the temperature rise trend; when faced with sudden load fluctuations that cause a sudden increase in temperature, the cooling system responds slowly, which can easily lead to overheating and damage to components.

[0005] 3. Poor coordination of multimodal cooling: Existing multimodal solutions for water cooling and air cooling are mostly fixed combination operation or independent control, without dynamic adaptation according to the temperature rise risk level; high power water cooling is turned on in low-risk conditions, resulting in excessive energy consumption, while relying on a single cooling method in high-risk conditions leads to insufficient efficiency.

[0006] Therefore, there is an urgent need to develop a temperature rise coordinated monitoring and cooling control system to solve the problems in the existing technology. Summary of the Invention

[0007] The purpose of this invention is to provide a temperature rise coordinated monitoring and cooling control system that can improve the safety of device operation, reduce cooling energy consumption, enhance adaptability to operating conditions and environment, and has a simple structure and is easy to use, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A temperature rise collaborative monitoring and cooling control system includes a multi-dimensional collaborative sensing module, a predictive collaborative decision-making module, a multi-modal adaptive cooling execution module, and an edge computing unit. The system interacts with the edge computing unit via an industrial Ethernet bus. The edge computing unit is also bidirectionally connected to the controller of a water-cooled high-voltage reactive power compensation device to acquire device operating status data and send load limiting signals. The multi-dimensional collaborative sensing module is used to synchronously collect multi-dimensional physical quantity data. After processing by the fusion model, it outputs the global temperature rise distribution and core temperature rise indicators, providing input for the predictive collaborative decision-making module. Based on the temperature rise index, the predictive collaborative decision-making module uses a predictive model to predict the temperature rise trend, classify risk levels, generate dynamic cooling collaborative strategies, and output risk levels, cooling commands, and control parameters. The multimodal adaptive cooling execution module executes cooling commands and collects cooling effect data, which is then fed back to the preceding module to form a closed-loop control. The fusion model is a fusion model combining attention mechanism and CNN-LSTM, used to enhance key features and filter interference; the prediction model is an improved LSTM model that introduces multi-head attention mechanism and residual connection to solve the problem of gradient decay in long-term prediction.

[0009] By adopting the above technical solutions, and through the collaborative interaction of the four core modules and the edge computing unit, the system can accurately process temperature rise-related data of the water-cooled high-voltage reactive power compensation device, reliably predict temperature rise trends, and dynamically generate cooling strategies. At the same time, it relies on dedicated fusion and prediction models to solve the problems of feature interference filtering and long-term prediction gradient decay, providing basic architectural support for the stable and efficient operation of the entire system.

[0010] As a further aspect of the present invention: the multi-dimensional collaborative sensing module includes multiple types of distributed sensing units, including a temperature sensing unit, a load condition sensing unit, and an environmental sensing unit. The temperature sensing unit adopts distributed temperature sensors, which are laid along the reactor winding surface, capacitor shell, power module heat dissipation surface and busbar area of ​​the water-cooled high-voltage reactive power compensation device at a preset interval. At the same time, high-precision temperature sensors are arranged at the inlet and outlet of the water-cooling system to collect the inlet and outlet temperatures of the water-cooling medium. The load condition sensing unit includes a current sensor, a voltage sensor, and a vibration sensor. The current sensor is connected in series with the main circuit of the device, the voltage sensor is connected in parallel with the input terminal of the device, and the vibration sensor is attached to preset positions on the device housing, reactor base, and capacitor bracket, respectively, for collecting operating current, operating voltage, and vibration data. The environmental sensing unit is located in the industrial plant where the device is located. It includes temperature and humidity sensors and wind speed sensors, which are used to collect ambient temperature and airflow speed data to correct the environmental interference factors of the prediction model.

[0011] By adopting the above technical solutions, comprehensive collection of physical quantities such as device temperature, load conditions and environment can be achieved, ensuring that the collected data covers the core heat-generating area of ​​the device and key operating scenarios. At the same time, it provides basic data for the prediction model that can correct for environmental interference, ensuring the accuracy of subsequent temperature rise index calculations.

[0012] As a further aspect of the present invention: the attention-based CNN-LSTM fusion model includes an input layer, an attention mechanism layer, a CNN feature extraction layer, an LSTM temporal processing layer, and an output layer, with each layer connected sequentially; The input features of the input layer include distributed temperature data, operating current, operating voltage, vibration data, ambient temperature, and airflow speed. The input is time-series data of a preset duration. The attention mechanism layer adopts a scaled dot product attention mechanism, which strengthens the weight of key features and filters out invalid interference by calculating the correlation weight between each input feature and the temperature feature, and presets the weight threshold of key features. The CNN feature extraction layer includes several convolutional layers and a max pooling layer, used to extract spatial correlation features between multiple features; The LSTM timing processing layer includes several LSTM units, with Dropout layers between the units to handle dynamic changes in timing data. The output layer outputs pixel matrix data of the global temperature rise distribution heat map of the device, the maximum temperature rise value of the core component, and the temperature rise rate through the fully connected layer.

[0013] By adopting the above technical solutions, the overscaling dot product attention mechanism strengthens key features and filters out invalid interference. By combining CNN and LSTM to extract spatial correlation features and temporal dynamic features respectively, the final output is an accurate global temperature rise distribution and core temperature rise index, providing high-quality input data for predictive collaborative decision-making.

[0014] As a further aspect of the present invention: the training process of the attention-based CNN-LSTM fusion model includes the following steps: S1: Data preparation: Collect sample data of water-cooled high-voltage reactive power compensation device under different operating conditions, different environments, and different fault states. Each sample includes the input features corresponding to the input layer of the fusion model and manually labeled global temperature rise distribution, maximum temperature rise, and temperature rise rate. S2: Data preprocessing, normalizing the sample data, mapping the input feature data to a preset interval; using preset criteria to remove abnormal data caused by sensor failures; and using preset algorithms to expand the sample data of fault states and other low-sample data to ensure a balanced proportion of samples for each working condition. S3: Model training, the preprocessed samples are divided into training set, validation set and test set according to the preset ratio; the preset optimizer and multi-task loss function are selected, the preset training rounds and batch size are set, and the early stopping strategy is adopted to avoid model overfitting; S4: Model Validation and Optimization. Validate the model performance on the test set. If the preset validation metrics are not met, adjust the model parameters and retrain until the metrics are met. After successful validation, solidify the model and deploy it to the edge computing unit.

[0015] By adopting the above technical solutions, the fusion model is ensured to have stable performance under different working conditions, environments and fault states. The trained and optimized model can accurately adapt to the target device, and after solidification and deployment, it can realize real-time and efficient fusion processing of multi-dimensional data.

[0016] As a further aspect of the present invention: the improved LSTM temperature rise prediction model includes an input layer, a multi-head attention mechanism layer, an improved LSTM layer, and an output layer, with each layer connected sequentially; The input features of the input layer include the current maximum temperature rise of the core component, the current temperature rise rate, the operating current, and the voltage level. The input is time-series data of a preset duration. The multi-head attention mechanism layer sets up several parallel attention heads, each of which independently calculates the feature weights for different time series stages. The outputs of each head are fused through the splicing layer to strengthen the weight of recent temperature rise data and improve the accuracy of long-term time series prediction. The improved LSTM layer includes several LSTM units. A residual connection is added to the output of each LSTM unit, and the unit input is directly superimposed on the output to solve the gradient decay problem in long-term training. A Dropout layer is set after the last LSTM unit. The output layer outputs the predicted maximum temperature rise and average temperature rise rate of the core components within a preset time period through a fully connected layer.

[0017] By adopting the above technical solution, the weight of recent temperature rise data is strengthened through a multi-head attention mechanism, and the problem of gradient decay in long-term prediction is solved by combining residual connection. This enables accurate prediction of the maximum temperature rise and average temperature rise rate of core components within a preset time period, providing a reliable basis for risk level classification.

[0018] As a further aspect of the present invention, the training process of the improved LSTM temperature rise prediction model includes the following steps: SQ1: Data preparation, collecting sample data. The sample source is the same as the training data of the CNN-LSTM fusion model based on the attention mechanism. Each set of samples contains the input features corresponding to the input layer of the improved LSTM temperature rise prediction model and manually labeled the maximum temperature rise and average temperature rise rate within a preset time period. SQ2: Data preprocessing, normalizing the input data to make it conform to a preset distribution; using the sliding window method to expand the sample to ensure the continuity of time series data; and removing samples with abnormal temperature rise rate fluctuations. SQ3: Model training, dividing the samples into training set, validation set and test set according to a preset ratio; selecting a preset optimizer and loss function, setting a preset training round and batch size, and using an early stopping strategy to control the training process; SQ4: Model Validation and Optimization. Validate model performance on the test set. If the preset validation metrics are not met, adjust the number of LSTM units and attention heads and retrain. After successful validation, deploy to the edge computing unit to run in collaboration with the fusion model.

[0019] By adopting the above technical solutions and ensuring the stability of the prediction model through a standardized training process, the model can accurately output temperature rise prediction results after parameter adjustment and verification optimization. When running in conjunction with the fusion model, it can achieve seamless connection between perception and prediction.

[0020] As a further aspect of the present invention: the temperature rise risk level determination logic of the predictive collaborative decision-making module is as follows: based on the output results of the improved LSTM temperature rise prediction model, combined with the preset comfortable operating temperature threshold and preset safe temperature threshold of the core components in the water-cooled high-voltage reactive power compensation device, three risk levels are determined, wherein: Low risk: The predicted maximum temperature rise is no greater than the preset comfortable operating temperature threshold, and the temperature rise rate is no greater than the first preset rate; Medium risk: The predicted maximum temperature rise is between the preset comfortable working temperature threshold and the preset safe temperature threshold, and the temperature rise rate is between the first preset rate and the second preset rate, satisfying any one of them; High risk: The predicted maximum temperature rises above the preset safe temperature threshold, and the temperature rise rate is higher than the second preset rate, satisfying either one of these conditions.

[0021] By adopting the above technical solution, the output results of the prediction model are transformed into intuitive low, medium and high risk levels, providing a clear basis for matching subsequent dynamic cooling coordination strategies and ensuring the pertinence and rationality of cooling control.

[0022] As a further aspect of the present invention: the multimodal cooling dynamic coordination strategy of the predictive collaborative decision-making module is as follows: based on the divided temperature rise risk level, dynamically match a coordination mode with water cooling as the main method and air cooling combined with phase change cooling as the auxiliary method, and optimize the control parameters with the goal of achieving the best cooling efficiency and the lowest energy consumption. Low-risk operating conditions: A combined mode of natural heat dissipation and low-power air cooling is adopted. The water cooling system is turned off, the phase change cooling components are in a natural heat storage state, and the fan is controlled to run at a preset low power speed. Medium-risk operating conditions: A synergistic mode of high-power air cooling combined with phase change cooling is adopted, and the water cooling system operates at a preset basic flow rate; the phase change cooling components actively contact the core heat-generating components through the drive mechanism, and the fan operates at a preset high-power speed; High-risk operating conditions: Adopting a three-mode collaborative mode of water cooling combined with high-power air cooling and phase change cooling; the water cooling system operates at the preset optimal flow rate, and the water cooling medium temperature is controlled at the preset target temperature through the heat exchange medium adjustment component; the fan operates at the preset limit power speed; the phase change cooling component fully covers the core heat-generating area, enhancing the latent heat cooling effect.

[0023] By adopting the above technical solutions, energy consumption is reduced in low-risk operating conditions, cooling efficiency is guaranteed in medium- and high-risk operating conditions, and cooling effect and energy economy are balanced, thus solving the problem of insufficient adaptability of traditional fixed cooling modes.

[0024] As a further aspect of the present invention: the predictive collaborative decision-making module also includes fault linkage logic: when the multi-dimensional collaborative sensing module detects a combination of abnormal temperature rise, sudden voltage change, excessive current fluctuation, and excessive vibration, it determines that there is an internal fault in the device. Among them, abnormal temperature rise is when the temperature rise rate is higher than the preset abnormal rate and there is no corresponding load change; voltage sudden change is when the single-unit voltage change rate is higher than the preset voltage change rate threshold; excessive current fluctuation is when the current change rate is higher than the preset current change rate threshold; and excessive vibration is when the vibration acceleration is higher than the preset vibration acceleration threshold. At this point, the enhanced cooling strategy is triggered, switching to a high-risk tri-modal collaborative mode; simultaneously, a load limiting signal is sent to the device controller to limit the operating current to a preset ratio of the rated value, and a fault alarm is triggered through the audible and visual alarm device.

[0025] By adopting the above technical solutions, a multi-parameter linkage fault determination and emergency response logic can be constructed, which can quickly identify the early signs of internal faults in the device. By immediately switching to a high-level cooling strategy, limiting the load, and combining audible and visual alarms, the risk of fault expansion can be minimized and the operational safety of the device can be improved.

[0026] As a further aspect of the present invention: the multimodal adaptive cooling execution module includes an air-cooled execution component, a water-cooled execution component, a phase-change cooling execution component, and an execution control unit; The air-cooled execution component includes a speed-regulating fan and an adjustable airflow guiding structure. The adjustable airflow guiding structure adjusts the angle through a drive component and guides the airflow to the local overheated area based on the global temperature rise distribution thermal map output by the multi-dimensional collaborative sensing module. The water-cooled actuator includes a microchannel heat exchange structure, a speed-regulating water pump, a heat exchange medium regulating component, and a heat exchanger. The microchannel heat exchange structure is attached to the bottom of the core component. The speed-regulating water pump adjusts the flow rate according to control parameters. The heat exchange medium regulating component stabilizes the water-cooled medium at a preset target temperature through a temperature control unit. The phase change cooling actuator includes a composite phase change material assembly and an electromagnetic drive mechanism. The composite phase change material assembly is made into a thin sheet and fills the gap between the components. The electromagnetic drive mechanism controls the adhesion and separation of the material and the components according to the command. The execution control unit adopts an industrial-grade microcontroller, which controls the fan speed and water pump flow rate through pulse width modulation signals, and controls the start and stop of the electromagnetic drive mechanism and water cooling system through switching signals; the integrated sampling module collects the cooling medium temperature and the operating current of each component in real time, and feeds back the data to the edge computing unit through the industrial Ethernet bus.

[0027] By adopting the above technical solution, the core components of the cooling execution module and the function of each component are clearly defined. Through the coordinated action of components such as adjustable speed fan and microchannel heat exchange structure, cooling commands are executed accurately. At the same time, cooling effect data is collected and fed back in real time to ensure closed-loop optimization of cooling control and improve the accuracy and reliability of cooling execution.

[0028] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention, through multi-dimensional full-domain perception, long-term temperature rise accurate prediction and fault linkage emergency response, can avoid the risk of overheating of core components in advance, quickly deal with fault precursors, effectively reduce the occurrence rate of faults such as thermal breakdown and short circuit, ensure the stable operation of the power grid, and improve operational safety and monitoring accuracy.

[0029] 2. This invention is based on a multi-modal cooling coordination mode that dynamically matches risk levels, avoids ineffective energy consumption in low-risk conditions, ensures cooling efficiency in medium- and high-risk conditions, significantly reduces the average annual cooling energy consumption of the device, improves the economic efficiency of energy utilization, and achieves a balance between cooling efficiency and energy consumption.

[0030] Other features and advantages of the present invention will be disclosed in detail in the following detailed description and accompanying drawings. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of an overall structure in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] In this embodiment of the invention, a temperature rise coordinated monitoring and cooling control system is described, see [link to relevant documentation]. Figure 1 As shown, it includes the following: 1. Multi-dimensional collaborative perception module Module function: Synchronously collect multi-dimensional physical quantity data from the device, process them through a fusion model, and output accurate global temperature rise distribution and core temperature rise indicators, providing reliable input for subsequent prediction and decision-making.

[0034] The input consists of raw data from each sensor and operating status data transmitted by the device controller; the output consists of a thermal map of the temperature rise distribution across the entire device, the maximum temperature rise value of the core components, and the temperature rise rate. The data update frequency is 100ms.

[0035] The detailed design of the non-public common knowledge section of this module is as follows: (1) Precise layout design of multi-dimensional sensing units Unlike existing single-temperature sensor designs, this module adopts a multi-type, distributed sensor layout, suitable for high-voltage and strong electromagnetic industrial environments. The specific layout logic and parameters are as follows: Temperature sensing unit: A distributed fiber optic temperature sensor is selected because high-voltage scenarios are subject to strong electromagnetic interference, while fiber optic sensors have the characteristics of strong anti-electromagnetic interference and distributed temperature measurement.

[0036] The sensors are laid out in a serpentine pattern along the surface of the reactor winding, the capacitor housing, the heat dissipation surface of the power module, and the busbar area, with a spacing of 2cm. This spacing was determined through simulation verification, which can achieve full coverage of the temperature rise distribution without monitoring blind spots, and at the same time, it will not cause excessive cost due to too small a spacing.

[0037] One high-precision platinum resistance temperature sensor with an accuracy of ±0.1℃ is installed at the inlet and outlet of the water cooling system to collect the inlet and outlet temperatures of the water cooling medium and realize closed-loop feedback of the cooling effect.

[0038] Load condition sensing unit: Deploys 2 sets of high-voltage adaptive Hall current sensors, with an adaptive voltage level ≥10kV, a measurement range of 0~1000A, and an accuracy of ±1%, connected in series with the main circuit of the device to collect the operating current; deploys 2 sets of high-voltage capacitive voltage sensors, with an adaptive voltage level ≥10kV, a measurement range of 0~35kV, and an accuracy of ±0.5%, connected in parallel with the input terminal of the device to collect the operating voltage; deploys 3 piezoelectric vibration sensors, with a measurement range of 0~10kHz and a sensitivity of 100mV / g, respectively attached to the middle of the device housing, the reactor base, and the capacitor bracket.

[0039] The bonding positions are determined through modal analysis. These three positions are the areas where vibration transmission is most uniform during device operation, and can accurately capture vibration signals from abnormal operating conditions such as component loosening or malfunction.

[0040] Environmental sensing unit: One temperature and humidity sensor and one wind speed sensor are installed in the industrial plant where the device is located. The temperature and humidity sensor has a measurement range of -20℃ to 85℃ and 0 to 100%RH, and the wind speed sensor has a measurement range of 0 to 20m / s with an accuracy of ±0.1m / s. They are used to collect ambient temperature and airflow speed data to correct the environmental interference factors of the temperature rise prediction model.

[0041] (2) Design of CNN-LSTM fusion model based on attention mechanism Existing technologies often employ simple weighted fusion or single temporal models to process data, failing to accurately distinguish between valid and interfering features. This invention innovatively introduces an attention mechanism, constructing a fusion model combining the attention mechanism with CNN-LSTM to achieve accurate fusion of multi-dimensional data.

[0042] Among them, attention mechanism refers to the algorithm structure that strengthens key features and weakens interference features by calculating the association weights between each input feature and the target feature; CNN, or Convolutional Neural Network, is good at extracting spatial association features of data; LSTM, or Long Short-Term Memory Network, is good at processing dynamic changes in time series data.

[0043] The model structure and core parameters are as follows: Input layer: The input feature dimension is 6, which are distributed temperature data, operating current, operating voltage, vibration data, ambient temperature, and airflow velocity.

[0044] The distributed temperature data extracts the temperature values ​​of each sensing node, totaling 50 feature points; the vibration data is the combined effective value of three sensors.

[0045] The input time series length is 30s and the sampling interval is 100ms. These parameters were determined experimentally. The 30s time series data can cover a complete cycle of device load fluctuation, ensuring that the model captures the dynamic temperature rise pattern.

[0046] Attention Mechanism Layer: Employs a scaled dot product attention mechanism. The core logic is to strengthen key features and filter out invalid interference by calculating the association weights between each input feature and the temperature feature.

[0047] This includes: first, mapping each feature to a query vector Q, a key vector K, and a value vector V, where Q is the vector corresponding to the temperature feature, and K and V are the vectors corresponding to other features; then, obtaining the association weights through the weight coefficient calculation formula; and finally, multiplying each feature value by its corresponding weight coefficient and inputting it into the subsequent network.

[0048] The preset weight thresholds are: operating current and vibration data weights ≥ 0.6, as these two types of data are most strongly correlated with temperature rise; ambient temperature and airflow velocity weights ≤ 0.2, to avoid interference from instantaneous environmental fluctuations.

[0049] CNN Feature Extraction Layer: Set 2 convolutional layers and 1 max pooling layer. The convolutional kernel size is 3×3, the stride is 1, and the padding is 1. This parameter setting can ensure that the feature map size remains unchanged. The first convolutional layer has 32 output channels, and the second convolutional layer has 64 output channels. The pooling kernel size is 2×2, and the stride is 2.

[0050] Its core function is to extract spatial correlation features among multiple features, such as the spatial correspondence between sudden increases in operating current and local temperature rises in reactors.

[0051] LSTM temporal processing layer: Set up 2 LSTM units, each unit has 64 hidden neurons; add a Dropout layer between the two LSTM units with a dropout rate of 0.2.

[0052] The dropout rate was determined experimentally and can effectively prevent model overfitting without affecting the model's fitting ability.

[0053] Its core function is to process dynamic changes in time-series data, such as the continuous trend of temperature rise rate.

[0054] Output layer: The fully connected layer outputs three types of results: pixel matrix data of the thermal map of the temperature rise distribution of the entire device, the maximum temperature rise value of the core component, and the temperature rise rate. The resolution of the thermal map is 128×128.

[0055] (3) The complete training process of the fusion model Data preparation phase: Collect sample data of water-cooled high-voltage reactive power compensation device under different operating conditions, different environments, and different fault states, with a total sample size of ≥100,000 groups.

[0056] Operating conditions cover operating current from 0 to rated current, voltage level from 10kV to 35kV, and load fluctuation frequency from 0.1Hz to 1Hz; ambient temperature covers from -20℃ to 85℃, and airflow speed from 0 to 20m / s; fault conditions include blockage of heat dissipation pipes, loose reactors, and poor capacitor contact, with the degree of blockage of heat dissipation pipes covering 20% ​​to 80%.

[0057] Each sample group contains 6-dimensional input features and 3-dimensional labels. The input features correspond to the model input layer, and the labels are manually labeled global temperature rise distribution, maximum temperature rise, and temperature rise rate.

[0058] Data preprocessing stage: The first step is normalization, which uses min-max normalization to map all input feature data to the [0,1] interval, using the formula... Implementation, in which The original data, , These are the minimum and maximum values ​​of the feature, respectively.

[0059] The second step is to remove outlier samples. The 3σ criterion is used to remove abnormal data caused by sensor malfunctions, such as temperatures exceeding the reasonable range of -40℃ to 150℃ or current suddenly dropping to 0. The 3σ criterion is a statistical method for removing data that deviates from the mean by more than three times the standard deviation.

[0060] The third step is sample balancing, which uses the SMOTE algorithm to expand the small sample data such as fault conditions. The SMOTE algorithm is an algorithm that balances the dataset by synthesizing new samples, which can ensure that the sample proportion of each working condition is balanced, with an error of ≤5% and avoid model bias.

[0061] Model training phase: The preprocessed samples are divided into training set, validation set and test set in a ratio of 7:2:1.

[0062] The Adam optimizer was chosen because it has adaptive learning rate characteristics, converges faster than the SGD optimizer, and is suitable for training the complex features of this model. The initial learning rate was 0.001, decreasing by 0.1 every 10 epochs. The loss function adopted was a multi-task loss function, which is a weighted sum of the MSE loss of the global temperature rise distribution, the MAE loss of the highest temperature rise, and the MAE loss of the temperature rise rate. The MSE loss is used to measure the pixel deviation of the heatmap, and the MAE loss is used to measure the numerical deviation, with weights of 0.5:0.3:0.2 respectively. The training epochs were 50, and the batch size was 64. An early stopping strategy was adopted during training, which means stopping training when the model's performance on the validation set no longer improves. Specifically, training was stopped when the validation set loss value did not decrease for 5 consecutive epochs to avoid overfitting.

[0063] Model validation and optimization phase: Validate the model performance on the test set. The preset validation indicators are: pixel matching degree of global temperature rise distribution ≥90%, maximum temperature rise prediction error ≤0.5℃, and temperature rise rate prediction error ≤0.05℃ / min.

[0064] If the target is not met, retrain the model by adjusting the number of LSTM units and the dropout rate until the target is met; after verification, solidify the model and deploy it to the edge computing unit.

[0065] 2. Predictive Collaborative Decision-Making Module Module Function: Based on the accurate temperature rise index output by the sensing module, the module predicts the temperature rise trend through a predictive model, classifies risk levels, and generates dynamic cooling coordination strategies.

[0066] The inputs are the global temperature rise distribution, the maximum temperature rise of the core components, the temperature rise rate, and the operating current and voltage level transmitted by the device controller, all output by the multi-dimensional collaborative sensing module; the outputs are the temperature rise risk level, multi-modal cooling collaborative commands, and control parameters of each cooling component, where the risk level is divided into three levels: low, medium, and high, and the control parameters include fan speed, water pump flow rate, etc.

[0067] The detailed design of the non-public common knowledge section of this module is as follows: (1) Design of an improved LSTM temperature rise prediction model To address the issue of gradient decay in existing time series prediction models during long-term predictions, this invention designs an improved LSTM model that combines a multi-head attention mechanism with residual connections to achieve accurate prediction of the temperature rise trend in the next 5 minutes.

[0068] Among them, the multi-head attention mechanism refers to calculating the weights of different time segments by using multiple parallel attention heads, and then fusing the results to improve prediction accuracy; residual connection refers to directly superimposing the input of the network layer onto the output to alleviate the gradient decay problem in long-term training.

[0069] The model structure and parameters are as follows: Input layer: The input feature dimension is 4, which are the current maximum temperature rise of the core component, the current temperature rise rate, the operating current, and the voltage level; the input time series length is 30s, and the sampling interval is 1s.

[0070] This parameter setting balances data volume and real-time forecasting, with 30 seconds of data covering key stages of temperature rise trends.

[0071] Multi-head attention mechanism layer: Four attention heads are set up. Each attention head independently calculates the feature weights of different time stages, and then the outputs of each head are merged through a splicing layer.

[0072] Its core function is to strengthen the weight of recent temperature rise data. For example, the weight of the phase of sudden change in temperature rise rate in the last 10 seconds is 2 to 3 times that of earlier data, thereby improving the accuracy of long-term time series prediction.

[0073] The specific logic is to divide the input time series data into 4 segments according to time, and each attention head is responsible for the weight calculation of one segment. The weight coefficient is determined by the correlation between the segment features and the prediction target.

[0074] Improved LSTM layer: Three LSTM units are used, each with 128 hidden neurons; residual connections are added to the output of each LSTM unit, meaning the input of that unit is directly superimposed on the output, using the formula... Implementation, in which For cell input, This design addresses the gradient decay problem in long-term training by providing unit outputs; a Dropout layer with a dropout rate of 0.25 is added after the last LSTM unit.

[0075] Output layer: Two types of results are output through the fully connected layer: the predicted maximum temperature rise of the core component and the predicted average temperature rise rate within the next 5 minutes.

[0076] (2) The complete training process of the improved LSTM model Data preparation phase: 80,000 sets of sample data were collected, and the sample sources were the same as the training data of the multi-dimensional collaborative perception module.

[0077] Each sample group contains 4-dimensional input features and 2-dimensional labels. The input features correspond to the model input layer, and the labels are manually labeled maximum and average temperature rise rates for the next 5 minutes.

[0078] Data preprocessing stage: Input data is normalized using Z-score, and processed using the formula... Implementation, in which The characteristic mean, The characteristic standard deviation is used, and this normalization method can make the data follow a standard normal distribution; the sliding window method is used to expand the sample with a window step of 1 second to ensure the continuity of time series data; samples with abnormal temperature rise rate fluctuations, such as samples with instantaneous fluctuations >2℃ / min, are removed.

[0079] Model training phase: The samples are divided into training set, validation set and test set in a ratio of 7:2:1.

[0080] The SGD optimizer was selected with an initial learning rate of 0.01 and momentum of 0.9. The SGD optimizer was chosen because it is more stable in long-term model training, and momentum can accelerate convergence. The loss function used was MSE, i.e., mean squared error. The number of training epochs was 80, and the batch size was 32. An early stopping strategy was adopted with patience=8, i.e., training was stopped if the loss on the validation set did not decrease for 8 consecutive epochs.

[0081] Model validation and optimization phase: The preset validation indicators are: the prediction error of the maximum temperature rise in the next 5 minutes ≤ 2℃ and the prediction error of the temperature rise rate ≤ 0.1℃ / min.

[0082] If the target is not met, adjust the number of LSTM units and attention heads and retrain; after successful verification, deploy to the edge computing unit to run in conjunction with the fusion model.

[0083] (3) Logic for determining the risk level of temperature rise Based on the output of the prediction model and combined with the thermodynamic characteristics of the core components of the water-cooled high-voltage reactive power compensation device, three risk level thresholds were preset. These thresholds were determined through experimental verification, as follows: Low risk: The predicted maximum temperature rise is ≤45℃, which is the upper limit of the comfortable operating temperature of the core components; and the temperature rise rate is ≤0.5℃ / min.

[0084] Medium risk: The predicted maximum temperature rise is >45℃ and ≤55℃, where 55℃ is the safe temperature threshold for core components; or the temperature rise rate is >0.5℃ / min and ≤1℃ / min.

[0085] High risk: Predicted maximum temperature rise > 55℃; or temperature rise rate > 1℃ / min.

[0086] (4) Multimodal cooling dynamic coordination strategy Breaking away from existing fixed-combination cooling modes, a collaborative strategy is adopted based on risk level, with water cooling as the primary method and air cooling combined with phase change cooling as a supplement. This approach balances cooling efficiency and energy consumption, and the specific strategy is as follows: Low-risk operating conditions: A combined mode of natural heat dissipation and low-power air cooling is adopted; the water cooling system is turned off to avoid ineffective energy consumption; the phase change cooling components are in a natural heat storage state. The control parameter is set to a fan speed of 1200 r / min. This speed is determined through energy consumption simulation, which can achieve the lowest energy consumption while meeting the heat dissipation requirements for low-risk temperature rise.

[0087] Medium-risk operating conditions: A synergistic mode of high-power air cooling combined with phase change cooling is adopted, and the water cooling system operates at the basic flow rate; the phase change cooling component actively contacts the core heat-generating component through an electromagnetic drive mechanism, with a contact gap of ≤0.5mm to ensure heat conduction efficiency; The control parameters are set as follows: fan speed is set to 2800 r / min, which is 3 times more efficient than the low-power mode; water pump flow rate is set to 4 L / min, which is the base flow rate to avoid excessive energy consumption.

[0088] High-risk operating conditions: A three-modal synergistic mode combining water cooling, high-power air cooling, and phase change cooling is adopted; the water cooling system operates at the optimal flow rate of 8 L / min, determined through CFD simulation, to achieve optimal heat exchange efficiency; the water cooling medium temperature is controlled at 25℃ through a heat exchange medium regulation component, which is the optimal heat dissipation target temperature for core components; the fan speed is set to 3500 r / min, representing the maximum power, with a maximum airflow of 200 m³ / min. 3 / h; The phase change cooling component fully covers the core heat-generating area to enhance the latent heat cooling effect.

[0089] (5) Fault linkage logic When the multi-dimensional collaborative sensing module detects abnormal temperature rise combined with sudden voltage changes, excessive current fluctuations, or excessive vibration, it determines that there is an internal fault in the device, such as decreased insulation of components or precursors to short circuits.

[0090] Among them, abnormal temperature rise is defined as a temperature rise rate > 1.5℃ / min without corresponding load change; sudden voltage change is defined as a single-unit voltage change rate > 0.1V / s; excessive current fluctuation is defined as a current change rate > 10A / s; and excessive vibration is defined as a vibration acceleration > 5g.

[0091] At this point, the enhanced cooling strategy is immediately triggered, directly switching to the high-risk tri-modal collaborative mode; at the same time, a load limiting signal is sent to the device controller to limit the operating current to 30% of the rated value; and a fault alarm is triggered through the audible and visual alarm.

[0092] 3. Multimodal adaptive cooling execution module Module Function: Accurately executes the cooling commands from the decision-making module, collects cooling effect data, and provides feedback.

[0093] The inputs are the cooling coordination commands, fan speed, water pump flow rate, and other control parameters output by the predictive collaborative decision-making module; the outputs are the inlet and outlet temperatures of the water cooling medium, fan operating current, water pump operating current, and the contact status of the phase change cooling components. The output data is fed back to the sensing module and the decision-making module.

[0094] The detailed design of this module is as follows: The air-cooled actuator includes two adjustable-speed industrial fans and an adjustable airflow structure.

[0095] The adjustable speed industrial fan has a power range of 0~200W and a speed adjustment range of 500~3500r / min; the airflow guide structure is made of aluminum alloy, which was selected because aluminum alloy has the characteristics of being lightweight and having good thermal conductivity; the angle is adjusted by a stepper motor, with an adjustment range of 0~90°.

[0096] Specifically, based on the global temperature rise distribution thermal map output by the sensing module, the airflow is directed to local overheated areas, such as the middle of the reactor, to improve local heat dissipation efficiency. Compared with a fixed flow guiding structure, the local heat dissipation efficiency can be improved by 40%.

[0097] The water-cooled actuator includes a dedicated microchannel heat exchange structure, an adjustable speed water pump, a heat exchange medium regulating component, and a plate heat exchanger.

[0098] The microchannel heat exchange structure is a self-designed non-standard component with a channel diameter of 2mm and a spacing of 5mm, which is attached to the bottom of the core component to increase the heat exchange area; the adjustable speed water pump has a flow rate adjustment range of 0~10L / min to adapt to the flow requirements of different risk levels; the heat exchange medium is a 50% ethylene glycol aqueous solution, selected based on the fact that the freezing point of the solution is -35℃, which is suitable for low-temperature industrial environments; the heat exchange medium adjustment component stabilizes the temperature of the water-cooled medium at a preset target temperature, such as 25℃ for high-risk conditions, through a PID temperature control unit.

[0099] The PID parameters are Kp=5.0, Ki=0.1, and Kd=0.5. These parameters were determined through experimental debugging and can ensure a temperature control accuracy of ±0.5℃.

[0100] The phase change cooling actuator includes a composite phase change material assembly and an electromagnetic drive mechanism.

[0101] The composite phase change material is a non-publicly known material with a self-developed formula, which is composed of PEG-6000 and expanded graphite, wherein PEG-6000 accounts for 70% by mass and expanded graphite accounts for 30% by mass.

[0102] The selection criteria are based on the phase transition temperature of PEG-6000 (42℃) and latent heat (200J / g), which can provide stable latent heat cooling; expanded graphite can increase the thermal conductivity of the material to 10W / (m·K), solving the problem of poor thermal conductivity of pure PEG. The material is made into a 3mm thin sheet and filled in the gap between the components; the electromagnetic drive mechanism has a response time of ≤0.3s and controls the adhesion or separation of the material and the components through GPIO signals.

[0103] The execution control unit uses an industrial-grade STM32H743 microcontroller with a main frequency of 480MHz, which can meet the requirements of real-time control of multiple parameters. The fan speed and water pump flow rate are controlled by PWM signals with a frequency of 10kHz. The duty cycle is linearly related to the speed and flow rate; specifically, for every 10% increase in the duty cycle, the fan speed increases by 300r / min and the water pump flow rate increases by 1L / min. The electromagnetic drive mechanism and water cooling system are started and stopped by switching signals. An integrated AD sampling module with a sampling frequency of 100Hz collects the cooling medium temperature and the operating current of each component in real time. The current can be used to determine whether a component is faulty; for example, if the water pump current is 0, it is considered a fault. Data is fed back to the edge computing unit via an industrial Ethernet bus with a baud rate of 500kbps.

[0104] 4. System closed-loop collaborative logic Each module achieves closed-loop coordination through the following logic to ensure control accuracy and stability, including: The perception, prediction, decision-making, and execution chain: The multi-dimensional collaborative perception module outputs temperature rise indicators every 100ms, the predictive collaborative decision-making module runs a prediction model every 1s and generates risk levels and cooling strategies, and the multi-modal adaptive cooling execution module executes instructions in real time and adjusts the parameters of cooling components.

[0105] Feedback optimization chain: The execution module provides feedback on cooling effect data every 500ms, the perception module updates the temperature rise index based on the feedback data, the prediction model corrects subsequent prediction results, and the decision module dynamically optimizes cooling parameters.

[0106] For example, when the rate of temperature decrease is greater than 1℃ / min, the fan speed should be reduced by 500r / min.

[0107] Shutdown coordination: After the device is shut down, the system continues to run for 3 minutes; if the temperature rises to below 30°C, all cooling components are turned off; otherwise, low-power air cooling is maintained at a fan speed of 1200 r / min until the temperature reaches the target.

[0108] This invention provides a temperature rise coordinated monitoring and cooling control system, which can improve the safety of device operation, reduce cooling energy consumption, enhance the adaptability to operating conditions and environment, and has high reliability.

[0109] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0110] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A temperature rise coordinated monitoring and cooling control system, characterized in that, It includes a multi-dimensional collaborative sensing module, a predictive collaborative decision-making module, a multi-modal adaptive cooling execution module, and an edge computing unit. It interacts with the edge computing unit through an industrial Ethernet bus. The edge computing unit is also bidirectionally connected to the controller of the water-cooled high-voltage reactive power compensation device to realize the acquisition of device operation status data and the transmission of load limit signals. The multi-dimensional collaborative sensing module is used to synchronously collect multi-dimensional physical quantity data. After processing by the fusion model, it outputs the global temperature rise distribution and core temperature rise indicators, providing input for the predictive collaborative decision-making module. Based on the temperature rise index, the predictive collaborative decision-making module uses a predictive model to predict the temperature rise trend, classify risk levels, generate dynamic cooling collaborative strategies, and output risk levels, cooling commands, and control parameters. The multimodal adaptive cooling execution module executes cooling commands and collects cooling effect data, which is then fed back to the preceding module to form a closed-loop control. The fusion model is a fusion model combining attention mechanism and CNN-LSTM, used to enhance key features and filter out interference; The prediction model is an improved LSTM model that incorporates a multi-head attention mechanism and residual connections to solve the problem of gradient decay in long-term predictions.

2. The temperature rise coordinated monitoring and cooling control system according to claim 1, characterized in that, The multi-dimensional collaborative sensing module includes multiple types of distributed sensing units, including temperature sensing units, load condition sensing units, and environmental sensing units. The temperature sensing unit adopts distributed temperature sensors, which are laid along the reactor winding surface, capacitor shell, power module heat dissipation surface and busbar area of ​​the water-cooled high-voltage reactive power compensation device at a preset interval. At the same time, high-precision temperature sensors are arranged at the inlet and outlet of the water-cooling system to collect the inlet and outlet temperatures of the water-cooling medium. The load condition sensing unit includes a current sensor, a voltage sensor, and a vibration sensor. The current sensor is connected in series with the main circuit of the device, the voltage sensor is connected in parallel with the input terminal of the device, and the vibration sensor is attached to preset positions on the device housing, reactor base, and capacitor bracket, respectively, for collecting operating current, operating voltage, and vibration data. The environmental sensing unit is located in the industrial plant where the device is located. It includes temperature and humidity sensors and wind speed sensors, which are used to collect ambient temperature and airflow speed data to correct the environmental interference factors of the prediction model.

3. The temperature rise coordinated monitoring and cooling control system according to claim 1, characterized in that, The attention-based CNN-LSTM fusion model includes an input layer, an attention mechanism layer, a CNN feature extraction layer, an LSTM temporal processing layer, and an output layer, with each layer connected sequentially. The input features of the input layer include distributed temperature data, operating current, operating voltage, vibration data, ambient temperature, and airflow speed. The input is time-series data of a preset duration. The attention mechanism layer adopts a scaled dot product attention mechanism, which strengthens the weight of key features and filters out invalid interference by calculating the correlation weight between each input feature and the temperature feature, and presets the weight threshold of key features. The CNN feature extraction layer includes several convolutional layers and a max pooling layer, used to extract spatial correlation features between multiple features; The LSTM timing processing layer includes several LSTM units, with Dropout layers between the units to handle dynamic changes in timing data. The output layer outputs pixel matrix data of the global temperature rise distribution heat map of the device, the maximum temperature rise value of the core component, and the temperature rise rate through the fully connected layer.

4. The temperature rise coordinated monitoring and cooling control system according to claim 3, characterized in that, The training process of the attention-based CNN-LSTM fusion model includes the following steps: S1: Data preparation: Collect sample data of water-cooled high-voltage reactive power compensation device under different operating conditions, different environments, and different fault states. Each sample includes the input features corresponding to the input layer of the fusion model and manually labeled global temperature rise distribution, maximum temperature rise, and temperature rise rate. S2: Data preprocessing, normalizing the sample data, mapping the input feature data to a preset interval; using preset criteria to remove abnormal data caused by sensor failures; and using preset algorithms to expand the sample data of fault states and other low-sample data to ensure a balanced proportion of samples for each working condition. S3: Model training, the preprocessed samples are divided into training set, validation set and test set according to the preset ratio; the preset optimizer and multi-task loss function are selected, the preset training rounds and batch size are set, and the early stopping strategy is adopted to avoid model overfitting; S4: Model Validation and Optimization. Validate the model performance on the test set. If the preset validation metrics are not met, adjust the model parameters and retrain until the metrics are met. After successful validation, solidify the model and deploy it to the edge computing unit.

5. The temperature rise coordinated monitoring and cooling control system according to claim 1, characterized in that, The improved LSTM temperature rise prediction model includes an input layer, a multi-head attention mechanism layer, an improved LSTM layer, and an output layer, with each layer connected sequentially. The input features of the input layer include the current maximum temperature rise of the core component, the current temperature rise rate, the operating current, and the voltage level. The input is time-series data of a preset duration. The multi-head attention mechanism layer sets up several parallel attention heads, each of which independently calculates the feature weights for different time series stages. The outputs of each head are fused through the splicing layer to strengthen the weight of recent temperature rise data and improve the accuracy of long-term time series prediction. The improved LSTM layer includes several LSTM units. A residual connection is added to the output of each LSTM unit, and the unit input is directly superimposed on the output to solve the gradient decay problem in long-term training. A Dropout layer is set after the last LSTM unit. The output layer outputs the predicted maximum temperature rise and average temperature rise rate of the core components within a preset time period through a fully connected layer.

6. The temperature rise coordinated monitoring and cooling control system according to claim 5, characterized in that, The training process of the improved LSTM temperature rise prediction model includes the following steps: SQ1: Data preparation, collecting sample data. The sample source is the same as the training data of the CNN-LSTM fusion model based on the attention mechanism. Each set of samples contains the input features corresponding to the input layer of the improved LSTM temperature rise prediction model and manually labeled the maximum temperature rise and average temperature rise rate within a preset time period. SQ2: Data preprocessing, normalizing the input data to make it conform to a preset distribution; using the sliding window method to expand the sample to ensure the continuity of time series data; and removing samples with abnormal temperature rise rate fluctuations. SQ3: Model training, dividing the samples into training set, validation set and test set according to a preset ratio; selecting a preset optimizer and loss function, setting a preset training round and batch size, and using an early stopping strategy to control the training process; SQ4: Model Validation and Optimization. Validate model performance on the test set. If the preset validation metrics are not met, adjust the number of LSTM units and attention heads and retrain. After successful validation, deploy to the edge computing unit to run in collaboration with the fusion model.

7. The temperature rise coordinated monitoring and cooling control system according to claim 1, characterized in that, The temperature rise risk level determination logic of the predictive collaborative decision-making module is as follows: based on the output results of the improved LSTM temperature rise prediction model, combined with the preset comfortable operating temperature threshold and preset safe temperature threshold of the core components in the water-cooled high-voltage reactive power compensation device, three risk levels are determined, wherein: Low risk: The predicted maximum temperature rise is no greater than the preset comfortable operating temperature threshold, and the temperature rise rate is no greater than the first preset rate; Medium risk: The predicted maximum temperature rise is between the preset comfortable working temperature threshold and the preset safe temperature threshold, and the temperature rise rate is between the first preset rate and the second preset rate, satisfying any one of them; High risk: The predicted maximum temperature rises above the preset safe temperature threshold, and the temperature rise rate is higher than the second preset rate, satisfying either one of these conditions.

8. A temperature rise coordinated monitoring and cooling control system according to any one of claims 1 and 7, characterized in that, The multimodal cooling dynamic coordination strategy of the predictive collaborative decision-making module is as follows: based on the divided temperature rise risk level, dynamically match the coordination mode of water cooling as the main mode and air cooling combined with phase change cooling as the auxiliary mode, and optimize the control parameters with the goal of optimal cooling efficiency and minimum energy consumption. Low-risk operating conditions: A combined mode of natural heat dissipation and low-power air cooling is adopted. The water cooling system is turned off, the phase change cooling components are in a natural heat storage state, and the fan is controlled to run at a preset low power speed. Medium-risk operating conditions: A synergistic mode of high-power air cooling combined with phase change cooling is adopted, and the water cooling system operates at a preset basic flow rate; the phase change cooling components actively contact the core heat-generating components through the drive mechanism, and the fan operates at a preset high-power speed; High-risk operating conditions: Adopting a three-mode collaborative mode of water cooling combined with high-power air cooling and phase change cooling; the water cooling system operates at the preset optimal flow rate, and the water cooling medium temperature is controlled at the preset target temperature through the heat exchange medium adjustment component; the fan operates at the preset limit power speed; the phase change cooling component fully covers the core heat-generating area, enhancing the latent heat cooling effect.

9. The temperature rise coordinated monitoring and cooling control system according to claim 1, characterized in that, The predictive collaborative decision-making module also includes fault linkage logic: when the multi-dimensional collaborative sensing module detects a combination of abnormal temperature rise, sudden voltage change, excessive current fluctuation, and excessive vibration, it determines that there is an internal fault in the device. Among them, abnormal temperature rise is when the temperature rise rate is higher than the preset abnormal rate and there is no corresponding load change; voltage sudden change is when the single-unit voltage change rate is higher than the preset voltage change rate threshold; excessive current fluctuation is when the current change rate is higher than the preset current change rate threshold; and excessive vibration is when the vibration acceleration is higher than the preset vibration acceleration threshold. At this point, the enhanced cooling strategy is triggered, switching to a high-risk tri-modal collaborative mode; simultaneously, a load limiting signal is sent to the device controller to limit the operating current to a preset ratio of the rated value, and a fault alarm is triggered through the audible and visual alarm device.

10. The temperature rise coordinated monitoring and cooling control system according to claim 1, characterized in that, The multimodal adaptive cooling execution module includes an air-cooled execution component, a water-cooled execution component, a phase-change cooling execution component, and an execution control unit; The air-cooled execution component includes a speed-regulating fan and an adjustable airflow guiding structure. The adjustable airflow guiding structure adjusts the angle through a drive component and guides the airflow to the local overheated area based on the global temperature rise distribution thermal map output by the multi-dimensional collaborative sensing module. The water-cooled actuator includes a microchannel heat exchange structure, a speed-regulating water pump, a heat exchange medium regulating component, and a heat exchanger. The microchannel heat exchange structure is attached to the bottom of the core component. The speed-regulating water pump adjusts the flow rate according to control parameters. The heat exchange medium regulating component stabilizes the water-cooled medium at a preset target temperature through a temperature control unit. The phase change cooling actuator includes a composite phase change material assembly and an electromagnetic drive mechanism. The composite phase change material assembly is made into a thin sheet and fills the gap between the components. The electromagnetic drive mechanism controls the adhesion and separation of the material and the components according to the command. The execution control unit adopts an industrial-grade microcontroller, which controls the fan speed and water pump flow rate through pulse width modulation signals, and controls the start and stop of the electromagnetic drive mechanism and water cooling system through switching signals; the integrated sampling module collects the cooling medium temperature and the operating current of each component in real time, and feeds back the data to the edge computing unit through the industrial Ethernet bus.

Images