A multi-sensor fusion-based intelligent heat management method and system

Abstract

The application discloses a kind of intelligent heat management method and system based on multi-sensor fusion, its method includes the following steps: S1: multi-point temperature array data acquisition: in radiator, power device, busbar, electromagnetic shield area layout N point temperature sensor array, temperature data is collected by temperature sensor.The application is by in key area layout multi-point temperature sensor array and combines electrical parameter, environmental parameter and the full dimension acquisition of heat dissipation component state, realizes multi-source data fusion by means of extended Kalman filter to accurately inverse device real-time junction temperature, again through RBF network reconstruction global thermal field distribution to locate hot spot and thermal gradient, simultaneously utilize LSTM neural network to predict future hot spot temperature variation, finally according to junction temperature, thermal gradient and prediction result dynamic adjustment fan speed, liquid cooling pump flow and cooling mode.

Description

Technical Field

[0001] This invention relates to the field of thermal management technology, and in particular to an intelligent thermal management method and system based on multi-sensor fusion. Background Technology

[0002] Existing thermal management methods mainly rely on single-point temperature sensors or simple thermal resistance network models, which are insufficient for sensing the thermal state of power devices and cannot accurately reflect hot spot temperatures and dynamic heat distribution. In addition, traditional methods have limited utilization of multi-source data such as current, voltage, PWM status, and environmental parameters, and lack effective data fusion strategies, resulting in lag in the estimation of device junction temperature. The system cannot actively adjust to sudden load changes or environmental changes, affecting operational reliability. In view of the above, this application proposes an intelligent thermal management method and system based on multi-sensor fusion. Summary of the Invention

[0003] Based on the technical problems existing in the background technology, the present invention proposes an intelligent thermal management method and system based on multi-sensor fusion.

[0004] The present invention proposes an intelligent thermal management method based on multi-sensor fusion, comprising the following steps: S1: Multi-point temperature array data acquisition: An array of N temperature sensors is deployed in the areas of the heat sink, power devices, busbars, and electromagnetic shielding to acquire temperature data. ; S2: Electrical and Environmental Data Acquisition: Acquisition of device current I, bus voltage V, PWM frequency f, duty cycle D, ambient temperature T a And humidity H; S3: Heat dissipation component health monitoring: Collect operating status parameters such as fan speed ω, liquid cooling pump flow rate Q, inlet / outlet water temperature difference ΔT, and heat exchanger pressure difference ΔP to construct a heat dissipation health index H. cool ; S4: Multi-source data preprocessing: Normalize, remove outliers, and synchronize time data collected in S1, S2, and S3 for temperature, electrical, environmental, and heat dissipation data. S5: Multi-sensor data fusion and junction temperature feedback: An extended Kalman filter (EKF) is used to achieve fusion estimation of temperature, electrical, and heat dissipation parameters, and to construct the junction temperature... The inversion model; S6: Thermal Field Reconstruction and Hotspot Estimation: Spatial interpolation of multi-point temperatures using an RBF network to obtain the thermal field distribution matrix. Identify hotspot locations and thermal gradients; S7: Heat Trend Prediction and Risk Assessment: Based on LSTM neural network, predict the hot spot temperature change within the future Δt. When the predicted temperature exceeds the threshold, output early warning or active adjustment signal. S8: Intelligent Cooling Regulation: Based on estimated junction temperature, thermal gradient and prediction results, it dynamically adjusts fan speed, cooling pump flow and cooling mode to achieve active thermal management.

[0005] Preferably, in step S2, the current I is sampled using a Hall current sensor, the bus voltage V is sampled using an isolated voltage sampling module, the PWM frequency f is sampled using a timer of the central controller, the duty cycle D is calculated in real time using the PWM input capture module of the central controller, and the ambient temperature T is measured using a digital temperature sensor. a Humidity (H) is sampled using an integrated digital temperature and humidity sensor.

[0006] Preferably, in step S3, the fan speed ω is collected by a photoelectric speed sensor, the liquid cooling pump flow rate Q is collected by an impeller flow meter, the inlet / outlet water temperature difference ΔT is collected and calculated by an inlet / outlet pipe temperature sensor, and the heat exchanger pressure difference ΔP is collected and calculated by a pressure sensor. Heat dissipation health index H coo l For multiple normalized sub-indices m i According to weight w i The weighted sum is calculated using the following formula: Among them, the sub-index m i These are the operating parameters: fan speed ω, liquid cooling pump flow rate Q, inlet / outlet water temperature difference ΔT, and heat exchanger pressure difference ΔP.

[0007] Preferably, the specific logical steps of S5 are as follows: S501: Construct the state equation and observation equation for junction temperature estimation. The state equation is as follows: Where A is the state transition coefficient and B is the power gain coefficient. For the real-time power loss of the device, This is process noise; The observation equation is ,in S1 represents the measured values ​​of the multi-point temperature array collected by S1, and C represents the observation matrix. To observe noise; S502: Initialize the state estimate of the extended Kalman filter. Covariance Matrix ; S503: Update the time based on the state of the previous moment to obtain... The covariance matrix is ​​updated as follows: , where Q is the process noise covariance matrix; S504: Calculate Kalman gain , where R is the observation noise covariance matrix; S505: Combine measured data with measurements to update the estimated junction temperature. And update the covariance matrix. The final output junction temperature .

[0008] Preferably, the specific logical steps of S6 are as follows: S601: Define the thermal field spatial domain as a two-dimensional plane. The locations of the N temperature sensors in S1 are denoted as follows: The corresponding measured temperature is ; S602: Select Gaussian radial basis functions ,in Let be any point in the thermal field to be interpolated. The basis function width parameter is determined using cross-validation. ; S603: Constructing the interpolation matrix Solve for the weight vector ,in ; S604: Perform interpolation calculations on all grid points in the thermal field space to obtain the thermal field distribution matrix. ; S605: Traverse the thermal field distribution matrix, through... Determine the hot spot temperature by Calculate the thermal gradient and locate the hotspot coordinates. .

[0009] Preferably, the specific logical steps of S7 are as follows: S701: Construct the input feature vector for the LSTM neural network, including historical junction temperature sequences. Historical hotspot temperature series Real-time electrical parameters Environmental parameters and heat dissipation health , where m is the length of the time window, and its value ranges from 5 to 30; S702: Input the input feature vector from S1 into the LSTM network. The activation function of the LSTM network is calculated using the ReLU function, so that the output layer of the LSTM network is the hotspot temperature prediction sequence within the future Δt: The value of Δt ranges from 1 to 10 seconds. S703: Setting temperature warning threshold and emergency adjustment threshold , , This is the device's rated maximum junction temperature; S704: When any temperature in the prediction sequence When, output an emergency adjustment signal; when When, output a warning signal; when At this time, no signal is output.

[0010] Preferably, the specific logical steps of S8 are as follows: S801: Establish a multi-dimensional adjustment decision matrix, with the junction temperature estimate as the input parameter. thermal gradient Hotspot temperature forecast and heat dissipation health ; S802: When a warning signal is received, adjust the fan speed according to the linear adjustment strategy: Fan speed adjustment amount: Liquid cooling pump flow rate adjustment ,in, This is an adjustment coefficient, with a value range of 0.01-0.1; S803: When an emergency adjustment signal is received, non-linear enhanced adjustment is activated, the fan speed is directly increased to 80%-100% of the rated speed, the liquid cooling pump flow rate is increased to 90%-100% of the rated flow rate, and at the same time, the enhanced cooling mode is switched to activate the auxiliary heat dissipation function of the controllable thermal management actuator. S804: When there is no warning signal and , When the threshold for heat dissipation health is between 0.6 and 0.8, the current cooling power will be maintained first, and a maintenance prompt for the heat dissipation components will be output. S805: During the adjustment process, feedback data is collected in real time, and adjustment parameters are corrected to ensure that the junction temperature remains stable. Within the interval, This is the lower limit of the junction temperature for safe operation of the device.

[0011] This invention also proposes an intelligent thermal management system based on multi-sensor fusion, including a temperature acquisition module, an electrical and environmental acquisition module, a heat dissipation status monitoring module, a central controller, and an intelligent control execution module. The central controller includes a data fusion processing module, a thermal field reconstruction and hot spot estimation module, and a thermal trend prediction module. The intelligent control execution module includes a fan controller, a liquid cooling pump controller, and a controllable thermal management actuator. The temperature acquisition module collects the temperature of the heat sink and power device areas through a multi-point temperature sensor array, generating temperature data. ; The electrical and environmental data acquisition module is used to acquire data such as device current I, bus voltage V, PWM frequency f, duty cycle D, and ambient temperature T. aIn addition to humidity (H), it provides heat-related operating condition data; The heat dissipation status monitoring module is used to collect operating status parameters such as fan speed ω, liquid cooling pump flow rate Q, inlet / outlet water temperature difference ΔT, and heat exchanger pressure difference ΔP, and calculate heat dissipation health index. ; The data fusion processing module is used to receive preprocessed data, invert the junction temperature of the device using an extended Kalman filter, and eliminate single sensor errors. The thermal field reconstruction and hot spot estimation module is used to generate a thermal field matrix based on the junction temperature through RBF network interpolation, locate hot spots and calculate thermal gradients. The heat trend prediction module uses an LSTM network to predict future hotspot temperatures based on historical and real-time parameters, and outputs early warning or adjustment signals. The fan controller is used to receive instructions to adjust the fan speed using a PWM signal and to feed back the actual speed to the monitoring module. The liquid cooling pump controller adjusts the liquid cooling pump flow rate according to instructions and feeds back the actual flow rate to the heat dissipation status monitoring module. The controllable thermal management actuator is used to control the start / stop or speed of the auxiliary heat dissipation component, and feeds back the execution status to the central controller to control the operation of the auxiliary heat dissipation component and ensure that the junction temperature is stable within a safe range.

[0012] Compared with existing technologies, the beneficial effects of this invention are: 1. By combining multi-point temperature array acquisition and RBF network thermal field reconstruction, N temperature sensors are deployed in key areas such as heat sinks and power devices. Spatial interpolation is achieved by combining junction temperature inversion results to generate a complete thermal field distribution matrix. This can accurately identify hot spot coordinates and thermal gradient changes, upgrading thermal state perception from "single-point local" to "global dynamic", avoiding device overheating damage caused by missing hot spots. 2. By incorporating temperature, electrical parameters, environmental parameters, and heat dissipation health parameters into a unified fusion framework, and eliminating single-sensor errors through filtering algorithms, the system achieves complementarity and verification of multi-dimensional data, making the junction temperature inversion results more consistent with the actual heat loss state of the device, thus solving the judgment bias problem caused by the "data silos" of traditional methods. 3. Through real-time data preprocessing and dynamic junction temperature inversion, the collected data is synchronized and anomalies are eliminated. Combined with the real-time power loss of the device, a state equation is constructed to achieve millisecond-level real-time inversion of junction temperature. This can quickly respond to load and environmental changes and avoid untimely heat dissipation adjustment due to the lag in junction temperature estimation. 4. Based on the prediction of hot spot temperature within the future Δt using LSTM neural network, a graded adjustment strategy is formulated in combination with heat dissipation health index: when a warning is issued, the fan speed / liquid pump flow rate is linearly fine-tuned, and when an emergency is triggered, an enhanced cooling mode is activated, forming a "prediction-adjustment-feedback" closed loop, so that the heat dissipation capacity is dynamically matched with the heat load demand, significantly improving the system's operational reliability under load fluctuations and environmental changes, and reducing the risk of downtime caused by passive heat dissipation. 5. By collecting parameters such as fan speed, liquid cooling pump flow rate, and heat exchanger differential pressure, a weighted health index is constructed, which can assess the performance degradation of heat dissipation components in real time, output maintenance reminders in advance, avoid system paralysis caused by sudden component failures, and reduce unplanned downtime maintenance costs. This invention deploys a multi-point temperature sensor array in key areas and combines it with comprehensive data acquisition of electrical parameters, environmental parameters, and the status of heat dissipation components. It utilizes extended Kalman filtering to achieve multi-source data fusion for accurate inversion of real-time device junction temperature. Then, an RBF network reconstructs the global thermal field distribution to locate hotspots and thermal gradients. Simultaneously, an LSTM neural network predicts future hotspot temperature changes. Finally, based on the junction temperature, thermal gradient, and prediction results, the fan speed, liquid cooling pump flow rate, and cooling mode are dynamically adjusted. This not only upgrades thermal state perception from "single-point local" to "global dynamic," solving the problems of data silos and delayed junction temperature estimation in traditional methods, but also forms a proactive thermal management closed loop of "prediction-adjustment-feedback." This allows for rapid response to load and environmental changes, preventing device damage caused by missed hotspots and untimely heat dissipation. Furthermore, by monitoring thermal health, it provides early warnings of component aging, reducing maintenance costs and significantly improving system reliability. Attached Figure Description

[0013] Figure 1 This is a flowchart of an intelligent thermal management method based on multi-sensor fusion proposed in this invention; Figure 2 This is a block diagram of an intelligent thermal management system based on multi-sensor fusion proposed in this invention. Detailed Implementation

[0014] The present invention will be further explained below with reference to specific embodiments.

[0015] Example 1 Reference Figure 1 This embodiment proposes an intelligent thermal management method based on multi-sensor fusion, including the following steps: S1: Multi-point temperature array data acquisition: An array of N temperature sensors is deployed in the areas of the heat sink, power devices, busbars, and electromagnetic shielding to acquire temperature data. ; S2: Electrical and Environmental Data Acquisition: Acquisition of device current I, bus voltage V, PWM frequency f, duty cycle D, ambient temperature T a And humidity H; The system utilizes a Hall effect current sensor for current I sampling, an isolated voltage sampling module for bus voltage V sampling, a central controller timer for PWM frequency f sampling, a central controller PWM input capture module for real-time duty cycle D calculation, and a digital temperature sensor for ambient temperature T. a Humidity (H) is sampled using an integrated digital temperature and humidity sensor. S3: Heat dissipation component health monitoring: Collect operating status parameters such as fan speed ω, liquid cooling pump flow rate Q, inlet / outlet water temperature difference ΔT, and heat exchanger pressure difference ΔP to construct a heat dissipation health index H. cool ; The fan speed ω is collected by a photoelectric speed sensor, the liquid cooling pump flow rate Q is collected by an impeller flow meter, the inlet / outlet water temperature difference ΔT is collected and calculated by inlet / outlet pipe temperature sensors, and the heat exchanger pressure difference ΔP is collected and calculated by a pressure sensor. Heat dissipation health index H coo l For multiple normalized sub-indices m i According to weight w i The weighted sum is calculated using the following formula: Among them, the sub-index m i These are the operating parameters: fan speed ω, liquid cooling pump flow rate Q, inlet / outlet water temperature difference ΔT, and heat exchanger pressure difference ΔP. S4: Multi-source data preprocessing: Normalize, remove outliers, and synchronize time data collected in S1, S2, and S3 for temperature, electrical, environmental, and heat dissipation data. S5: Multi-sensor data fusion and junction temperature feedback: An extended Kalman filter (EKF) is used to achieve fusion estimation of temperature, electrical, and heat dissipation parameters, and to construct the junction temperature... The inversion model; The specific logical steps are as follows: S501: Construct the state equation and observation equation for junction temperature estimation. The state equation is as follows: Where A is the state transition coefficient and B is the power gain coefficient. For the real-time power loss of the device, This is process noise; The observation equation is ,in S1 represents the measured values ​​of the multi-point temperature array collected by S1, and C represents the observation matrix. To observe noise; S502: Initialize the state estimate of the extended Kalman filter. Covariance Matrix ; S503: Update the time based on the state of the previous moment to obtain... The covariance matrix is ​​updated as follows: , where Q is the process noise covariance matrix; S504: Calculate Kalman gain , where R is the observation noise covariance matrix; S505: Combine measured data with measurements to update the estimated junction temperature. And update the covariance matrix. The final output junction temperature ; S6: Thermal Field Reconstruction and Hotspot Estimation: Spatial interpolation of multi-point temperatures using an RBF network to obtain the thermal field distribution matrix. Identify hotspot locations and thermal gradients; The specific logical steps are as follows: S601: Define the thermal field spatial domain as a two-dimensional plane. The locations of the N temperature sensors in S1 are denoted as follows: The corresponding measured temperature is ; S602: Select Gaussian radial basis functions ,in Let be any point in the thermal field to be interpolated. The basis function width parameter is determined using cross-validation. ; S603: Constructing the interpolation matrix Solve for the weight vector ,in ; S604: Perform interpolation calculations on all grid points in the thermal field space to obtain the thermal field distribution matrix. ; S605: Traverse the thermal field distribution matrix, through... Determine the hot spot temperature by Calculate the thermal gradient and locate the hotspot coordinates. ; S7: Heat Trend Prediction and Risk Assessment: Based on LSTM neural network, predict the hot spot temperature change within the future Δt. When the predicted temperature exceeds the threshold, output early warning or active adjustment signal. The specific logical steps are as follows: S701: Construct the input feature vector for the LSTM neural network, including historical junction temperature sequences. Historical hotspot temperature series Real-time electrical parameters Environmental parameters and heat dissipation health , where m is the length of the time window, and its value ranges from 5 to 30; S702: Input the input feature vector from S1 into the LSTM network. The activation function of the LSTM network is calculated using the ReLU function, so that the output layer of the LSTM network is the hotspot temperature prediction sequence within the future Δt: The value of Δt ranges from 1 to 10 seconds. S703: Setting temperature warning threshold and emergency adjustment threshold , , This is the device's rated maximum junction temperature; S704: When any temperature in the prediction sequence When, output an emergency adjustment signal; when When, output a warning signal; when At this time, no signal is output; S8: Intelligent cooling regulation: Based on estimated junction temperature, thermal gradient and prediction results, it dynamically adjusts fan speed, cooling pump flow and cooling mode to achieve active thermal management; The specific logical steps are as follows: S801: Establish a multi-dimensional adjustment decision matrix, with the junction temperature estimate as the input parameter. thermal gradient Hotspot temperature forecast and heat dissipation health ; S802: When a warning signal is received, adjust the fan speed according to the linear adjustment strategy: Fan speed adjustment amount: Liquid cooling pump flow rate adjustment ,in, This is an adjustment coefficient, with a value range of 0.01-0.1; S803: When an emergency adjustment signal is received, non-linear enhanced adjustment is activated, the fan speed is directly increased to 80%-100% of the rated speed, the liquid cooling pump flow rate is increased to 90%-100% of the rated flow rate, and at the same time, the enhanced cooling mode is switched to activate the auxiliary heat dissipation function of the controllable thermal management actuator. S804: When there is no warning signal and , When the threshold for heat dissipation health is between 0.6 and 0.8, the current cooling power will be maintained first, and a maintenance prompt for the heat dissipation components will be output. S805: During the adjustment process, feedback data is collected in real time, and adjustment parameters are corrected to ensure that the junction temperature remains stable. Within the interval, This is the lower limit of the junction temperature for safe operation of the device. This embodiment deploys a multi-point temperature sensor array in key areas and combines it with comprehensive data acquisition of electrical parameters, environmental parameters, and the status of heat dissipation components. Extended Kalman filtering is used to achieve multi-source data fusion for accurate inversion of real-time device junction temperature. An RBF network is then used to reconstruct the global thermal field distribution to locate hotspots and thermal gradients. Simultaneously, an LSTM neural network is used to predict future hotspot temperature changes. Finally, based on the junction temperature, thermal gradient, and prediction results, the fan speed, liquid pump flow rate, and cooling mode are dynamically adjusted. This not only upgrades thermal state perception from "single-point local" to "global dynamic," solving the problems of data silos and delayed junction temperature estimation in traditional methods, but also forms a proactive thermal management closed loop of "prediction-adjustment-feedback." This allows for rapid response to load and environmental changes, preventing hotspot omissions and device damage caused by untimely heat dissipation. Furthermore, monitoring thermal health provides early warnings of component aging, reducing maintenance costs and significantly improving system reliability.

[0016] Example 2 Reference Figure 2 This embodiment proposes an intelligent thermal management system based on multi-sensor fusion, including a temperature acquisition module, an electrical and environmental acquisition module, a heat dissipation status monitoring module, a central controller, and an intelligent control execution module. The central controller includes a data fusion processing module, a thermal field reconstruction and hot spot estimation module, and a thermal trend prediction module. The intelligent control execution module includes a fan controller, a liquid cooling pump controller, and a controllable thermal management actuator. The temperature acquisition module collects the temperature of the heat sink and power device areas through a multi-point temperature sensor array and generates temperature data. ; The electrical and environmental data acquisition module is used to collect data on device current (I), bus voltage (V), PWM frequency (f), duty cycle (D), and ambient temperature (T). a In addition to humidity (H), it provides heat-related operating condition data; The heat dissipation status monitoring module is used to collect operating status parameters such as fan speed ω, liquid cooling pump flow rate Q, inlet / outlet water temperature difference ΔT, and heat exchanger pressure difference ΔP, and calculate heat dissipation health indicators. ; The data fusion processing module is used to receive preprocessed data, invert the junction temperature of the device using an extended Kalman filter, and eliminate single sensor errors. The thermal field reconstruction and hot spot estimation module is used to generate a thermal field matrix based on the junction temperature through RBF network interpolation, locate hot spots and calculate thermal gradients. The heat trend prediction module uses an LSTM network to predict future hotspot temperatures based on historical and real-time parameters, and outputs early warning or adjustment signals. The fan controller receives commands to adjust the fan speed using PWM signals and feeds back the actual speed to the monitoring module. The liquid cooling pump controller adjusts the liquid cooling pump flow rate according to instructions and feeds back the actual flow rate to the heat dissipation status monitoring module. The controllable thermal management actuator is used to control the start, stop or speed of the auxiliary heat dissipation component, and feeds back the execution status to the central controller to control the operation of the auxiliary heat dissipation component and ensure that the junction temperature is stable within a safe range.

[0017] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart thermal management method based on multi-sensor fusion, characterized in that, Includes the following steps: S1: Multi-point temperature array data acquisition: An array of N temperature sensors is deployed in the areas of the heat sink, power devices, busbars, and electromagnetic shielding to acquire temperature data. ; S2: Electrical and Environmental Data Acquisition: Acquisition of device current I, bus voltage V, PWM frequency f, duty cycle D, ambient temperature T a And humidity H; S3: Heat dissipation component health monitoring: Collect operating status parameters such as fan speed ω, liquid cooling pump flow rate Q, inlet / outlet water temperature difference ΔT, and heat exchanger pressure difference ΔP to construct a heat dissipation health index H. cool ; S4: Multi-source data preprocessing: Normalize, remove outliers, and synchronize time data collected in S1, S2, and S3 for temperature, electrical, environmental, and heat dissipation data. S5: Multi-sensor data fusion and junction temperature feedback: An extended Kalman filter (EKF) is used to achieve fusion estimation of temperature, electrical, and heat dissipation parameters, constructing a junction temperature... The inversion model; S6: Thermal Field Reconstruction and Hotspot Estimation: Spatial interpolation of multi-point temperatures using an RBF network to obtain the thermal field distribution matrix. Identify hotspot locations and thermal gradients; S7: Heat Trend Prediction and Risk Assessment: Based on LSTM neural network, predict the hot spot temperature change within the future Δt. When the predicted temperature exceeds the threshold, output early warning or active adjustment signal. S8: Intelligent Cooling Regulation: Based on estimated junction temperature, thermal gradient and prediction results, it dynamically adjusts fan speed, cooling pump flow and cooling mode to achieve active thermal management.

2. The intelligent thermal management method based on multi-sensor fusion according to claim 1, characterized in that, In step S2, current I is sampled using a Hall current sensor, bus voltage V is sampled using an isolated voltage sampling module, PWM frequency f is sampled using a timer of the central controller, the duty cycle D is calculated in real time using the PWM input capture module of the central controller, and ambient temperature T is measured using a digital temperature sensor. a Humidity (H) is sampled using an integrated digital temperature and humidity sensor.

3. The intelligent thermal management method based on multi-sensor fusion according to claim 1, characterized in that, In S3, the fan speed ω is collected by a photoelectric speed sensor, the liquid cooling pump flow rate Q is collected by an impeller flow meter, the inlet / outlet water temperature difference ΔT is collected and calculated by an inlet / outlet pipe temperature sensor, and the heat exchanger pressure difference ΔP is collected and calculated by a pressure sensor. Heat dissipation health index H coo l For multiple normalized sub-indices m i According to weight w i The weighted sum is calculated using the following formula: Among them, sub-index m i These are the operating parameters: fan speed ω, liquid cooling pump flow rate Q, inlet / outlet water temperature difference ΔT, and heat exchanger pressure difference ΔP.

4. The intelligent thermal management method based on multi-sensor fusion according to claim 1, characterized in that, The specific logical steps of S5 are as follows: S501: Construct the state equation and observation equation for junction temperature estimation. The state equation is as follows: Where A is the state transition coefficient and B is the power gain coefficient. For the real-time power loss of the device, This is process noise; The observation equation is ,in S1 represents the measured values ​​of the multi-point temperature array collected by S1, and C represents the observation matrix. To observe noise; S502: Initialize the state estimate of the extended Kalman filter. Covariance Matrix ; S503: Update the time based on the state of the previous moment to obtain... The covariance matrix is ​​updated as follows: , where Q is the process noise covariance matrix; S504: Calculate Kalman gain , where R is the observation noise covariance matrix; S505: Combine measured data with measurements to update the estimated junction temperature. And update the covariance matrix. The final output junction temperature .

5. The intelligent thermal management method based on multi-sensor fusion according to claim 1, characterized in that, The specific logical steps of S6 are as follows: S601: Define the thermal field spatial domain as a two-dimensional plane. The locations of the N temperature sensors in S1 are denoted as follows: The corresponding measured temperature is ; S602: Select Gaussian radial basis functions ,in Let be any point in the thermal field to be interpolated. The basis function width parameter is determined using cross-validation. ; S603: Constructing the interpolation matrix Solve for the weight vector ,in ; S604: Perform interpolation calculations on all grid points in the thermal field space to obtain the thermal field distribution matrix. ; S605: Traverse the thermal field distribution matrix, through... Determine the hot spot temperature by Calculate the thermal gradient and locate the hotspot coordinates. .

6. The intelligent thermal management method based on multi-sensor fusion according to claim 1, characterized in that, The specific logical steps of S7 are as follows: S701: Construct the input feature vector for the LSTM neural network, including historical junction temperature sequences. Historical hotspot temperature series Real-time electrical parameters Environmental parameters and heat dissipation health , where m is the length of the time window, and its value ranges from 5 to 30; S702: Input the input feature vector from S1 into the LSTM network. The activation function of the LSTM network is calculated using the ReLU function, so that the output layer of the LSTM network is the hotspot temperature prediction sequence within the future Δt: The value of Δt ranges from 1 to 10 seconds. S703: Setting temperature warning threshold and emergency adjustment threshold , , This is the device's rated maximum junction temperature; S704: When any temperature in the prediction sequence When, output an emergency adjustment signal; when When, output a warning signal; when At this time, no signal is output.

7. The intelligent thermal management method based on multi-sensor fusion according to claim 1, characterized in that, The specific logical steps of S8 are as follows: S801: Establish a multi-dimensional adjustment decision matrix, with the junction temperature estimate as the input parameter. thermal gradient Hotspot temperature forecast and heat dissipation health ; S802: When a warning signal is received, adjust the fan speed according to the linear adjustment strategy: Fan speed adjustment amount: Liquid cooling pump flow rate adjustment ,in, This is an adjustment coefficient, with a value range of 0.01-0.1; S803: When an emergency adjustment signal is received, non-linear enhanced adjustment is activated, the fan speed is directly increased to 80%-100% of the rated speed, the liquid cooling pump flow rate is increased to 90%-100% of the rated flow rate, and at the same time, the enhanced cooling mode is switched to activate the auxiliary heat dissipation function of the controllable thermal management actuator. S804: When there is no warning signal and , When the threshold for heat dissipation health is between 0.6 and 0.8, the current cooling power will be maintained first, and a maintenance prompt for the heat dissipation components will be output. S805: During the adjustment process, feedback data is collected in real time, and adjustment parameters are corrected to ensure that the junction temperature remains stable. Within the interval, This is the lower limit of the junction temperature for safe operation of the device.

8. A smart thermal management system based on multi-sensor fusion, used to implement the method described in any one of claims 1-7, characterized in that, It includes a temperature acquisition module, an electrical and environmental acquisition module, a heat dissipation status monitoring module, a central controller, and an intelligent control execution module. The central controller includes a data fusion processing module, a thermal field reconstruction and hot spot estimation module, and a thermal trend prediction module. The intelligent control execution module includes a fan controller, a liquid cooling pump controller, and a controllable thermal management actuator. The temperature acquisition module collects the temperature of the heat sink and power device areas through a multi-point temperature sensor array, generating temperature data. ; The electrical and environmental data acquisition module is used to acquire data such as device current I, bus voltage V, PWM frequency f, duty cycle D, and ambient temperature T. a In addition to humidity (H), it provides heat-related operating condition data; The heat dissipation status monitoring module is used to collect operating status parameters such as fan speed ω, liquid cooling pump flow rate Q, inlet / outlet water temperature difference ΔT, and heat exchanger pressure difference ΔP, and calculate heat dissipation health index. ; The data fusion processing module is used to receive preprocessed data, invert the junction temperature of the device using an extended Kalman filter, and eliminate single sensor errors. The thermal field reconstruction and hot spot estimation module is used to generate a thermal field matrix based on the junction temperature through RBF network interpolation, locate hot spots and calculate thermal gradients. The heat trend prediction module uses an LSTM network to predict future hotspot temperatures based on historical and real-time parameters, and outputs early warning or adjustment signals. The fan controller is used to receive instructions to adjust the fan speed using a PWM signal and to feed back the actual speed to the monitoring module. The liquid cooling pump controller adjusts the liquid cooling pump flow rate according to instructions and feeds back the actual flow rate to the heat dissipation status monitoring module. The controllable thermal management actuator is used to control the start / stop or speed of the auxiliary heat dissipation component, and feeds back the execution status to the central controller to control the operation of the auxiliary heat dissipation component and ensure that the junction temperature is stable within a safe range.

Images