Intelligent dynamic temperature control method and device for liquid cooling heat dissipation module, electronic equipment and medium

Abstract

The invention relates to the technical field of temperature adjustment, and provides an intelligent dynamic temperature control method and device for a liquid cooling heat dissipation module, electronic equipment and a medium. Real-time heterogeneous data fusion is carried out on a multi-source sensor network to generate a global thermal field dynamic matrix, and joint prediction processing is carried out through a TCN-BiGRU network according to the global thermal field dynamic matrix and equipment working condition parameters to obtain thermal load probability distribution data. Performing multi-target topological optimization according to the equipment flow constraint condition and the thermal load probability distribution data to obtain a cooling control vector, and performing variable gain feed-forward compensation on the cooling control vector according to the piezoelectric micropump characteristic data to generate driving voltage waveform data. According to the invention, through the multi-source sensor network, the depth sequential network, topological optimization and accurate driving of the piezoelectric micropump, more accurate, efficient and reliable liquid cooling heat dissipation control is realized.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of temperature regulation, in particular to an intelligent dynamic temperature control method and device for a liquid cooling heat dissipation module, an electronic device and a medium. BACKGROUND

[0002] In the field of hardware and plastic processing, a large amount of concentrated heat energy is generated on the surface of a workpiece and a mold during high-speed cutting, injection molding or die casting. If the heat cannot be quickly and uniformly removed, the temperature gradient of the mold will be too large, the dimensional accuracy of the part will decrease, and surface quality defects will occur. The liquid cooling heat dissipation module plays a key role in this scenario: it arranges cooling channels inside or on the surface of the mold and circulates temperature-controlled liquid, which not only efficiently absorbs and transfers processing heat but also maintains the mold temperature in an ideal range to ensure processing stability and product quality consistency.

[0003] The common liquid cooling temperature control method on the market currently relies on centralized cooling water tanks, thermostats or heat exchangers, which are connected to the mold through cooling water pipes and use traditional PID controllers to feedback and adjust the temperature of the loop. The system usually only arranges temperature sensors at the water tank or the outlet of the circulation, and the control strategy is based on fixed set values for on-off or linear adjustment. Moreover, the cooling channel design is mostly uniform and straight, lacking partition optimization for hot spot areas. SUMMARY

[0004] Therefore, the application provides an intelligent dynamic temperature control method and device for a liquid cooling heat dissipation module, an electronic device and a medium to solve the problem of slow response of heat dissipation.

[0005] The first aspect of the application provides an intelligent dynamic temperature control method for a liquid cooling heat dissipation module, which comprises: real-time heterogeneous data fusion of a preset multi-source sensor network to generate a global thermal field dynamic matrix; joint prediction processing of a preset TCN-BiGRU network according to the global thermal field dynamic matrix and real-time collected equipment working condition parameters to obtain thermal load probability distribution data; multi-objective topology optimization processing according to a preset equipment flow constraint condition and the thermal load probability distribution data to obtain a cooling control vector; variable gain feedforward compensation of the cooling control vector according to preset piezoelectric micropump characteristic data to generate driving voltage waveform data.

[0006] In an optional implementation, the real-time heterogeneous data fusion of the preset multi-source sensor network to generate the global thermal field dynamic matrix comprises: spatiotemporal synchronization alignment processing of the preset multi-source sensor network to obtain a multi-source synchronous heterogeneous detection data set; dimensionless normalization conversion processing is performed on the multi-source synchronous heterogeneous detection data set to obtain a temperature data set; three-dimensional temperature field reconstruction processing is performed according to a preset spatial distance weighted interpolation algorithm and the temperature data set to generate a global thermal field dynamic matrix.

[0007] In an optional implementation, the joint prediction processing performed by the preset TCN-BiGRU network according to the global thermal field dynamic matrix and the real-time collected device working condition parameters to obtain thermal load probability distribution data includes: feature splicing processing is performed on the global thermal field dynamic matrix and the real-time collected device working condition parameters to obtain a spatio-temporal feature tensor; local heat feature extraction processing is performed by the preset TCN-BiGRU network on the spatio-temporal feature tensor to obtain a primary heat feature vector; long-time dependence analysis processing is performed by the preset TCN-BiGRU network on the primary heat feature vector to obtain a thermal risk probability value; Monte Carlo random disturbance evaluation processing is performed on the thermal risk probability value to obtain thermal load probability distribution data.

[0008] In an optional implementation, the multi-objective topological optimization processing performed according to a preset device flow constraint condition and the thermal load probability distribution data to obtain a cooling control vector includes: a temperature balance optimization objective function is constructed according to a high-temperature region in the thermal load probability distribution data, and a cooling liquid consumption optimization objective function is constructed according to a flow threshold set in the preset device flow constraint condition; Pareto front solving processing is performed by a preset non-dominated sorting evolutionary algorithm on the temperature balance optimization objective function and the cooling liquid consumption optimization objective function to obtain a candidate flow distribution scheme set; decision screening processing is performed on the candidate flow distribution scheme set according to a preset energy consumption-performance trade-off rule to obtain a cooling control vector.

[0009] In an optional implementation, the piezoelectric micropump characteristic data includes response frequency-flow mapping data, dead-zone voltage threshold, and fluid transmission delay parameters, and the variable-gain feedforward compensation performed according to the preset piezoelectric micropump characteristic data on the cooling control vector to generate drive voltage waveform data includes: flow-response frequency conversion is performed on the cooling control vector according to the response frequency-flow mapping data to obtain a target response frequency table sequence; dead-zone compensation processing is performed on the target response frequency table sequence according to the dead-zone voltage threshold to obtain an actual drive frequency sequence; performing phase lead correction processing on the actual driving frequency sequence according to the fluid transmission delay parameter to obtain a delay-compensated frequency sequence; performing high-voltage sine wave synthesis processing according to the delay-compensated frequency sequence to generate driving voltage waveform data.

[0010] In an optional implementation, the method further includes: controlling a liquid cooling heat dissipation module according to the driving voltage waveform data and performing multi-source state acquisition processing on the liquid cooling heat dissipation module to obtain a device operation state data set; constructing a sensor node relationship topology graph according to the device operation state data set; performing abnormal mode recognition processing on the sensor node relationship topology graph through a preset graph convolutional network to obtain a device health index; when the device health index is lower than a preset risk threshold, generating an active early warning instruction according to a preset active early warning mode.

[0011] The second aspect of the present application provides a liquid cooling heat dissipation module intelligent dynamic temperature control device, the device includes: a perception fusion module configured to perform real-time heterogeneous data fusion on a preset multi-source sensor network to generate a global thermal field dynamic matrix; a distribution prediction module configured to perform joint prediction processing on the global thermal field dynamic matrix and real-time collected device working condition parameters through a preset TCN-BiGRU network to obtain thermal load probability distribution data; a spraying control module configured to perform multi-objective topology optimization processing on a preset device flow constraint condition and the thermal load probability distribution data to obtain a cooling control vector; a feedforward compensation module configured to perform variable gain feedforward compensation on the cooling control vector according to preset piezoelectric micropump characteristic data to generate driving voltage waveform data.

[0012] The third aspect of the present application provides an electronic device including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the liquid cooling heat dissipation module intelligent dynamic temperature control method as described above when executing the computer program.

[0013] The fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the steps of the liquid cooling heat dissipation module intelligent dynamic temperature control method as described above.

[0014] In summary, the present application at least includes the following beneficial technical effects: 1. Through the TCN-BiGRU network, the future thermal load probability distribution is jointly predicted, so that the system can adjust the cooling strategy in advance before the thermal load really comes, to avoid the peak temperature exceeding caused by the response lag in passive cooling, and to quickly suppress local hot spots when the working condition changes.

[0015] 2. The optimal cooling control vector is generated by multi-objective topology optimization by comprehensively considering the equipment flow constraints. The global consideration of cooling channel and nozzle flow distribution is realized, which not only ensures sufficient heat dissipation capacity, but also maximally reduces unnecessary fluid circulation, thereby reducing the pumping power consumption and system energy consumption.

[0016] 3. By using the high bandwidth characteristics of the piezoelectric micropump, the cooling control vector obtained by topology optimization is converted into a driving voltage waveform through variable gain feedforward compensation, so that the flow regulation and on / off control can be realized in microseconds. The overshoot and oscillation are reduced, and the stability of temperature dynamic control is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 is a flow chart of the intelligent dynamic temperature control method of the liquid cooling heat dissipation module provided by the embodiments of the present application; Figure 2 is a functional module diagram of an intelligent dynamic temperature control device of a liquid cooling heat dissipation module provided by the embodiments of the present application; Figure 3 is a structural schematic diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] As shown in Figure 1 is a flow chart of the intelligent dynamic temperature control method of the liquid cooling heat dissipation module provided by the embodiments of the present application. The intelligent dynamic temperature control method of the liquid cooling heat dissipation module provided by the embodiments of the present application includes the following steps.

[0021] Step S1, real-time heterogeneous data fusion is performed on the preset multi-source sensor network to generate a global thermal field dynamic matrix.

[0022] To realize the detection of the running state of the target cooling equipment (for example, hardware machining and engraving equipment) and the liquid cooling heat dissipation module thereof, the embodiments of the present application deploy a multi-source sensor network around the target cooling equipment and the liquid cooling heat dissipation module thereof according to a predetermined layout. The multi-source sensor network includes but is not limited to a distributed temperature sensor, an infrared thermal imager, a cooling liquid flow meter, a cooling liquid pressure gauge and other sensors. The multi-source sensor network can capture temperature readings of contact sensors such as thermocouples every millisecond, and can also collect thermal images of tens of frames per second of the infrared thermal imager, and then record the flow and pressure changes of the cooling liquid in the pipeline. After obtaining the real-time sensor detection data set of the multi-source sensor network, the heterogeneous time sequence data of each sensor needs to be processed for space-time synchronization alignment. Therefore, the original time stamps of all collection terminals are unified to the level of microseconds through a high-precision clock synchronization mechanism, and the network time delay compensation algorithm is combined to re-label the time stamps of the video stream of the infrared thermal imager and the voltage signal of the temperature sensor, so as to accurately pair the collection time of the two types of data. For example, when the temperature sensor records 9 o'clock, the corresponding frame of the infrared thermal imager may be marked as 9 o'clock 0.2 seconds. After compensation, the data of both parties can be unified to 9 o'clock 0.1 seconds, thereby forming a multi-source synchronous heterogeneous detection data set with consistent time.

[0023] Subsequently, the multi-source synchronous heterogeneous detection data set is subjected to dimensionless normalization conversion processing to eliminate the output unit differences of different sensors, so that the voltage value, the pixel brightness value and the flow / pressure value can be mapped to the same temperature physical quantity. The mapping conversion of the embodiments of the present application adopts a linear or quadratic fitting formula based on a calibration curve, for example, a linear function relationship between the voltage of the thermocouple and the Celsius degree, and a quadratic function for mapping the infrared pixel brightness to the temperature reading by pre-calibration, so that a temperature data set with unified units and the same dimension can be obtained. For example, when the voltage of the thermocouple is 2.0 mV at a certain time and the pixel brightness at the same time point is 150, the temperature value of about 60℃ can be obtained after normalization and mapping, thereby ensuring that each data point in the subsequent thermal field analysis has the properties of comparability, additivity and interpolability. Next, based on the normalized temperature data and the corresponding physical space position distribution, spatial distance weighted interpolation reconstruction is implemented. By dividing the machining area into fixed-size three-dimensional grids (for example, 1 mm 3one grid unit), and the weight coefficients are calculated according to the Euclidean distance from the center of the grid to the positions of the temperature sensors. The weight coefficients can adopt an inverse square form to emphasize the contribution of the sensors in close proximity. The temperature of any grid point is estimated by weighted average of the temperature values of several adjacent sensors according to the weight. For example, if an internal grid point is 2 mm away from surface temperature sensor A and 3 mm away from sensor B, the corresponding weight is calculated according to the formula The calculation results are about 0.247 and 0.110, respectively, and the temperature of the grid point is obtained by weighted average of the temperature values corresponding to the ratio, i.e. 78.2℃. The global thermal field dynamic matrix covering the entire processing equipment and heat dissipation module can be finally generated by traversing the million-level grid.

[0024] In step S2, the TCN-BiGRU network is used to jointly predict the thermal load probability distribution data according to the global thermal field dynamic matrix and the real-time collected equipment working condition parameters.

[0025] It should be understood that the equipment working condition parameters include but are not limited to the spindle speed of the processing equipment, the feed speed of the processing equipment, and the type code of the processed material, etc. The global thermal field dynamic matrix and the real-time collected equipment working condition parameters jointly reveal the running strength of the processing load and the material properties. The spatial distribution information of the temperature field and the mechanical motion state information are integrated by splicing the global thermal field dynamic matrix and the equipment working condition parameters. The spliced space-time feature tensor enables the subsequent prediction model to capture the diffusion and aggregation of heat in space and perceive the influence of sudden changes in load conditions on temperature evolution. In the splicing process, first, the two types of data are aligned according to the unified timestamp with the time axis as the reference benchmark, and then the global thermal field dynamic matrix and the corresponding equipment working condition parameters are stacked in the channel dimension of the tensor in sequence, so that the generated space-time feature tensor contains not only continuous multiple frames of thermal field snapshots, but also speed, speed and material code information under different working conditions, laying a foundation for the joint processing of the subsequent convolution and recurrent network.

[0026] The TCN-BiGRU network is composed of a pre-trained and fixed-weight temporal convolutional network (TCN) and a bidirectional gated recurrent unit (BiGRU). The TCN is used for local heat feature extraction processing on the spatiotemporal feature tensor to obtain a primary heat feature vector; the BiGRU is used for long-time dependence analysis processing on the feature sequence composed of the primary heat feature vector to obtain a heat risk probability value at each sampling time. The TCN has the advantages of scalable receptive field and residual connection, and captures the temperature change trend and working condition fluctuation features of adjacent regions at different times by sliding a one-dimensional convolution kernel in the time dimension. Specifically, each convolution kernel length corresponds to, for example, temperature and speed information of the past 10 frames, and the convolution output is calculated continuously during the sliding window process and high response features are extracted through an activation function, thereby forming a feature map containing multiple spatiotemporal receptive fields. At the same time, multiple convolution kernels with different dilation rates work in parallel to combine long and short time distance change patterns, so that both subtle temperature mutations within 10 ms and slow trends within hundreds of milliseconds can be focused on during the same forward propagation process, thereby obtaining a primary heat feature vector and providing more rich local pattern information for subsequent time series modeling. For example, when the temperature at a certain position in the processing area continuously rises for 0.2 seconds with a sudden increase in speed, the receptive field of the temporal convolutional network can generate a high-amplitude response in the convolution output, indicating that the region may be in the initial stage of heat load accumulation.

[0027] The BiGRU is composed of two gate recurrent units with the same structure. One of them processes the feature sequence from the past to the present, determines how much historical information to retain in the hidden state through the update gate, and controls the contribution of the current input to the hidden state through the reset gate, thereby realizing the memory and forgetting of the cumulative effect in the sequence. The other one processes the same sequence from the future to the present in the reverse direction, corrects the deviation of the forward prediction through the opposite time sequence of information flow, and supplements the possible lag error. The hidden states of the two directions are spliced or weighted and fused at each time to form a comprehensive deep hidden representation, which can more accurately reflect the potential risk of the system under the dual action of the spatial heat field and the mechanical working condition at that time. If a certain region shows a significant temperature drop or speed drop in the subsequent frames, the reverse GRU can feed back this information to the representation at the current time, thereby avoiding excessive warning; on the contrary, if the temperature continues to rise or the speed continues to climb at the future time, the hidden state of the forward GRU can strengthen the accumulation effect of the heat risk. After the BiGRU processing is completed, a fully connected layer and a Sigmoid activation function are used to map the fused hidden representation to a heat risk probability value between 0 and 1, which can measure the possibility of overheating at a certain spatial position at the current sampling time.

[0028] After obtaining the single forward propagation thermal risk probability value, in order to be robust to random perturbation and unknown noise, a Monte Carlo random perturbation evaluation process needs to be carried out on the value to obtain more robust thermal load probability distribution data. Specifically, in the inference stage, a small random noise is applied to the weight or neuron connection (for example, about 10% of the GRU hidden units are randomly masked in each prediction, or a small variance Gaussian noise is added to the convolution kernel weight), and multiple forward inferences are performed on the same spatio-temporal feature tensor, typically 20 times, and then the output probability values of all rounds are averaged as the final report value, and the variance or confidence interval of each round result is calculated as an uncertainty indicator, so as to be able to make a more reliable cooling strategy according to the distribution data with higher confidence. For example, if the thermal risk probability value distribution obtained in 20 times of perturbation evaluation is between 0.82 and 0.88, then 0.85 will be taken as the thermal load probability of the region, and the variance range of ±0.03 will be attached to prompt the potential fluctuation, thereby effectively avoiding the misjudgment risk caused by the instability of single prediction.

[0029] The joint prediction processing flow seamlessly integrates the global thermal field and the device working condition, uses the efficient capture of the space-time local mode by the TCN and the bidirectional modeling of the long and short term dependence by the BiGRU, and finally ensures the reliability of the result through the Monte Carlo perturbation evaluation, so as to be able to accurately predict the overheating risk of each region hundreds of milliseconds in advance in the processing process, and provide high-precision and low-latency decision basis for the dynamic allocation of cooling resources.

[0030] Step S3, performing multi-objective topology optimization processing according to a preset device flow constraint condition and the thermal load probability distribution data to obtain a cooling control vector.

[0031] Firstly, two objective functions which are mutually restricted but must be optimized coordinately are constructed by combining the high temperature regions in the thermal load probability distribution data with the equipment flow constraints, so as to provide a clear optimization direction for the cooling liquid spraying strategy. The high temperature regions in the thermal load probability distribution data refer to the spatial distribution blocks in which each three-dimensional grid element is evaluated as a medium or high risk state, obtained through joint prediction in the previous step. Each block not only contains the corresponding three-dimensional coordinate boundary, but also carries the overheat probability value and the corresponding area or volume information. The flow threshold set in the equipment flow constraints is the minimum and maximum flow limit values that each spray head can provide according to the physical characteristics of the piezoelectric micropump and the pipeline system. These limit values are related to whether the pipeline pressure exceeds the safe range, whether the cooling liquid distribution is uniform, and whether the system can remain stable in dynamic adjustment. Combining the high temperature regions with the flow threshold set can construct the temperature balance optimization objective function and the cooling liquid consumption optimization objective function, respectively. The core intention of the temperature balance optimization objective function is to make the temperature in all space blocks marked as overheated or potentially overheated as close to the safety threshold as possible, so as to avoid the continuous persistence or re-accumulation of local hot spots. In mathematics, the difference between the temperature deviation and the target temperature in each high temperature region is squared and weighted to sum up the over-limit degree of the region. The greater the penalty value means the greater the temperature difference of the region away from the preset safety temperature, and thus a higher priority is obtained in the optimization process. The cooling liquid consumption optimization objective function takes the sum of the actual distribution flow of each spray head as a measurement index to depict the overall liquid consumption level of the system. The greater the liquid consumption means the lower the resource utilization efficiency, and the pipeline load and operation cost will also be correspondingly increased. Therefore, this objective function pursues to compress the total liquid consumption as much as possible under the premise of meeting the cooling demand of the high temperature regions.

[0032] After the objective functions are established, the non-dominated sorting evolutionary algorithm is used to solve the Pareto front of the two objectives of temperature uniformity and minimum liquid consumption. Specifically, the first step of the non-dominated sorting evolutionary algorithm is to generate several groups of initial flow rate schemes. Each group of schemes is randomly or based on historical experience to assign a flow rate value to each nozzle, while ensuring that each flow rate value falls between the minimum and maximum thresholds set in advance, forming an initial curve floating in the optimization space. Then, the two objective function values of each group of schemes are calculated. The temperature uniformity function value is calculated by simulating the application of the scheme corresponding flow rate vector to the high temperature area in the heat load probability distribution, and estimating the final temperature distribution deviation based on the pre-set flow rate-temperature mapping relationship. The cooling liquid consumption function value is directly equal to the algebraic sum of all elements in the vector. After completing the objective value evaluation, the non-dominated sorting step is entered. By comparing the objective value relationship of any two groups of schemes, the schemes that are not simultaneously superior to any other schemes in the two objectives are classified as the first level, and vice versa. Then, the elite preservation is performed on the high-level schemes using the tournament selection strategy. The flow rate vectors of two schemes are combined in proportion in each dimension according to the random crossover probability to generate offspring schemes, and a small random mutation is applied to each dimension of the offspring vector to explore new search areas. The above selection, crossover, mutation and non-dominated sorting steps are repeated in each iteration, and the entire evolutionary process is constantly approaching the ideal Pareto front in the high-dimensional search space until the preset iteration number is exhausted or the Pareto set converges. The final candidate flow rate allocation scheme set represents a variety of trade-off solutions that are balanced between different energy consumption levels and different cooling effects, and can be regarded as the optimal boundary curve between temperature uniformity and resource consumption.

[0033] After obtaining the candidate flow allocation scheme set, it is necessary to screen the candidate flow allocation scheme set according to the preset energy consumption-performance trade-off rule to finally determine the cooling control vector that can be implemented. Among them, the energy consumption-performance trade-off rule is a strategy formulated in advance according to the actual application energy consumption budget and safe cooling requirements, for example, it can be stipulated that when the liquid consumption increases by no more than 10% compared with the minimum liquid consumption, the scheme with the lowest temperature balance optimization index is preferentially selected, or when the cooling rate increases by less than a certain threshold, the scheme with the minimum liquid consumption is preferentially selected. Specifically, first, the energy consumption increase of each scheme in the candidate flow allocation scheme set is calculated, that is, the ratio of the difference between the total liquid consumption value and the minimum liquid consumption value in the candidate set to the minimum liquid consumption value, and the schemes with the ratio less than the preset threshold are screened into the candidate set; then compare the temperature balance index of each scheme in the candidate set, select the scheme with the minimum index as the final output, to ensure that the best high-temperature area cooling effect is obtained under the premise of controlled energy consumption. For example, if the minimum liquid consumption scheme is 1000 mL / s, a candidate scheme is 1080 mL / s, its increase is 8%, and the cooling index is reduced by 15% compared with the minimum scheme, then the scheme meets the condition that the increase is not more than 10%, and the cooling effect is significantly improved, and can be finally selected.

[0034] Finally, the generated cooling control vector is a flow value list corresponding to the number of all nozzles, each value represents the required cooling liquid output flow of the nozzle in the next control period, and the vector will be issued to the flow allocation module of the execution layer. According to the vector value, the module adjusts the driving signal and opening degree of each nozzle in real time, so that the system can accurately carry out cooling work according to the optimization result in the next period. The generation of this control vector not only integrates the global temperature risk distribution and equipment dynamic constraints, but also balances the energy saving and emission reduction and safety protection two goals, and provides a solid algorithm guarantee for the efficient, accurate and reliable dynamic temperature control of the equipment under complex load and environmental conditions.

[0035] Step S4, according to the preset piezoelectric micropump characteristic data, the cooling control vector is gain-variable feedforward compensation to generate driving voltage waveform data.

[0036] It should be understood that piezoelectric micropump characteristic data are key indicators obtained through experiments or manufacturer manuals, including response frequency-flow rate mapping data, dead zone voltage threshold, and fluid transmission delay parameters. The response frequency-flow rate mapping data reveals the liquid volume that can be output by the nozzle per unit time at different excitation frequencies; the dead zone voltage threshold represents the lower limit of the voltage at which the valve port has not opened or the flow rate is negligible under low voltage driving; and the fluid transmission delay parameter reflects the time delay required for the liquid to reach the nozzle nozzle from the valve port opening. Through the comprehensive use of the above-mentioned key indicators, not only can the response lag caused by hose damping, micropump inertia, and pipe rebound be overcome, but also the non-linear characteristics of flow control can be compensated at the output signal level, thereby realizing fine adjustment of the flow rate within one second or even shorter time scales.

[0037] When the cooling control vector is combined with the response frequency-flow rate mapping data, the flow rate value first needs to be converted to the target response frequency. Specifically, each nozzle flow rate value stored in the vector is input into the mapping function module according to the pre-established empirical curve or lookup table function. For example, a flow rate of 15 mL / s is mapped to an excitation frequency of about 32 Hz, in other words, the piezoelectric micropump can continuously and stably output the corresponding flow rate at this frequency. The establishment of the mapping function can use the multi-point calibration method, that is, the corresponding flow rate values are measured at a series of known frequencies, and a one-dimensional frequency-flow rate curve is formed using curve fitting or interpolation algorithm. The generation of the frequency output table sequence converts the continuous amplitude control problem of flow control into a frequency control problem, so that the subsequent voltage waveform synthesis can be directly preprocessed with frequency as the input, thereby avoiding the dead zone and non-linear distortion caused by directly adjusting the high voltage amplitude.

[0038] After obtaining the target response frequency table sequence, the influence of the voltage dead zone on the opening of the micropump must also be considered. It should be understood that the voltage dead zone threshold refers to the situation that when the excitation voltage is below a certain level, the valve cannot be opened due to the spring pre-tightening force or material rigidity, resulting in a flow rate of zero or far below the target value. In order to solve the control blind area that occurs at low flow rate demand, dead zone compensation processing is required for the target response frequency sequence, that is, the driving frequency of the nozzle is artificially increased above the dead zone threshold frequency when the frequency is too low, and a high-frequency micro-vibration component with a small amplitude is superimposed in the signal to ensure that the micropump maintains linear response in the entire working range. Specifically, the driving frequency of the nozzle can be forcibly increased to above the dead zone threshold frequency when the target frequency is below a certain threshold, and a high-frequency dithering component with a small amplitude is superimposed in the signal to ensure that the valve core generates enough energy to overcome the spring pre-tightening in the static state. For example, when the target frequency calculation value is 12 Hz and the lowest controllable frequency corresponding to the dead zone threshold is 15 Hz, the system will increase the 12 Hz signal to 15 Hz and superimpose a 50 Hz sine dithering with an amplitude of ± 5 Hz on it to ensure that a small amplitude oscillation is generated inside the micropump, prompting the valve port to open and release a small amount of cooling liquid.

[0039] After completing the dead-time compensation, the driving frequency sequence of each nozzle needs to be further corrected for phase advance to compensate for the time delay of the fluid in the pipeline transmission. When the pipe length, liquid viscosity, and pump structure interact, the actual liquid from the valve to the nozzle to form a stable water mist often lags behind the electrical signal to the driving end by several milliseconds, which will cause the system to delay the flow and misalign the peak value in the high-frequency control mode if not corrected. The principle of phase advance correction is to shift the input signal time axis forward by a certain amount of time, which can be determined by the fluid transmission delay parameter. For example, under the condition of a 2m pipeline length and a 2m / s flow rate, if the measured liquid lag time is 0.2s, then all driving frequency signals will be sent 0.2s in advance to ensure that the actual output time of the nozzle is almost synchronized with the required time of the cooling decision. In practice, a timestamp field can be added to each frequency value in the frequency sequence data structure, and the delay time is uniformly subtracted before being sent to the driving hardware, so that the system can still achieve millisecond-level or even microsecond-level timing accuracy under high-frequency oscillation.

[0040] When the target frequency is subjected to dead-time compensation and phase advance correction, the final delay-compensated frequency sequence represents the real-time driving frequency required by all nozzles in each sampling period. At this time, the frequency sequence needs to be converted into a high-voltage sinusoidal wave voltage signal that can be directly driven by the piezoelectric micropump. The sinusoidal wave synthesis processing link relies on a power amplifier and a digital-to-analog converter, where each channel generates a sinusoidal voltage of 180 volts peak or other preset peak value according to the corresponding frequency value, and maintains high precision in phase, amplitude, and frequency. Specifically, a digital signal processor or microcontroller can output a pulse width modulation signal with variable frequency according to the clock beat of the same sampling rate, and then smooth it into a standard sinusoidal wave through a filtering and amplifying circuit. During synthesis, phase continuity needs to be ensured, i.e. the phase difference between adjacent two sampling points is strictly equal to "sampling period x feedforward frequency", otherwise phase jump or frequency jitter will occur, affecting the stability of the sprinkling flow. For example, when the delay-compensated frequency sequence value at a certain time is 32Hz, the system will output the corresponding high-voltage sinusoidal wave at 32Hz, and when the next sampling period jumps to 28Hz, the phase will not start from zero again, but naturally transition based on the previous phase, ensuring smooth transformation of the flow output and precise matching of the cooling efficiency.

[0041] The technical solution converts the originally intended cooling control vector of the order of milliliter per second into voltage waveform data that can directly drive the piezoelectric pump, thereby achieving high precision and low latency of flow control. The voltage waveform data is a sequence of high- timing electrical signals that can be directly issued to the piezoelectric valve or micro-pump driver, converting the digitized cooling liquid flow demand into executable driving signals at the physical level, so that the piezoelectric element generates corresponding mechanical vibration and valve core displacement to control the injection amount and timing of the cooling liquid. In the application case of shell die-casting in the experimental stage, the combination of high-speed switching frequency and dead zone compensation shortens the response time of the spraying flow from the original 20 ms to less than 5 ms, while ensuring stable output in low flow mode, thereby improving the production rhythm and saving about 38% of the cooling liquid consumption while maintaining the uniform surface temperature of the microstructure mold.

[0042] In an optional embodiment, to realize fault early warning of the equipment through multi-dimensional operating state features, the method further comprises: While issuing the driving voltage waveform data to the execution unit of the liquid cooling heat dissipation module, the multi-source sensors arranged inside and outside the liquid cooling heat dissipation module automatically carry out state acquisition processing within the same time window to obtain the equipment operating state data set. The equipment operating state data set includes but is not limited to the data collected by various heterogeneous sensors such as vibration acceleration sensors, pipeline pressure sensors, cooling liquid pH sensors, flow meters, and temperature sensors at the same time, including time domain information (e.g., amplitude, instantaneous frequency), numerical range (e.g., pressure fluctuation ± 0.1 MPa, pH value 6.0-8.0), and actual output value of flow, etc., to fully reflect the current working condition of the liquid cooling system and the chemical and physical properties of the cooling medium.

[0043] After fully collecting the multi-source equipment operating state data set, the real-time readings of each sensor need to be mapped to the node features in the graph structure, and the sensor node relationship topology graph needs to be constructed according to the physical layout and signal correlation between sensors. The sensor node represents the vertex of a single sensor, and the node feature vector contains the value captured by the sensor at a specific sampling time and necessary identification information, such as amplitude energy, pressure value, pH value, flow value, and position coordinates, etc. The connection between nodes represents the connection relationship between nodes, and the weight can be calculated from the actual physical distance between two sensors, the signal correlation coefficient, or the pipeline fluid coupling degree, such as the vibration sensor and the pressure sensor with a distance of 0.3 m and a measured signal correlation coefficient of 0.85, which has an edge weight of 0.85. Constructing the sensor node relationship topology graph defined in this way can consider both the local signal strength and the spatial structure of the sensor layout when the graph convolution network transfers and fuses node features, thereby providing efficient and interpretable topology information for subsequent anomaly identification.

[0044] After constructing the sensor node relationship topology graph, the preset and optimized graph convolution network is used for abnormal mode recognition processing of the graph to obtain the device health index. The graph convolution network is a model capable of deep learning on graph structure data, which extracts higher order node representations and captures local and global topology information by performing weighted summation on the feature vectors of the node and its neighbor nodes in each iteration layer and combining an activation function. Specifically, the initialized node feature matrix and adjacency matrix are first input into the first layer of graph convolution layer, and the abnormal score of the whole liquid cooling system is obtained by calculating where X represents the node feature matrix, represents the matrix after adding a self-loop to the adjacency matrix, is a degree matrix, is a learnable weight matrix of the first layer, and sigma is a nonlinear activation function. After several layers of graph convolution propagation, each node finally obtains a deep representation vector that integrates multi-hop neighbor information. These vectors are then combined into an abnormal score representing the health status of the whole liquid cooling system in the output layer. After normalization processing, the abnormal score can be mapped to the device health index, and the device health index range is usually set between 0 and 100, which is used to measure the deviation between the system “physiological state” and the “health benchmark”.

[0045] After obtaining the device health index, it is compared with the preset risk threshold to determine whether to trigger the active warning instruction. When the device health index is greater than or equal to the risk threshold, it means that the system operating state is in an acceptable range and no additional intervention is needed. When the device health index is less than the risk threshold, it indicates that the system may have potential failure risks such as vibration abnormalities, pressure abnormalities, deterioration of cooling liquid chemical properties, or pipe blockage. The active warning mode can be divided into multiple modes, such as sending a short message or email to notify the maintenance personnel, triggering the alarm output in the industrial field bus, popping up a red warning window on the monitoring interface, and automatically generating a maintenance work order and submitting it to the fault management system. The choice of mode depends on the real-time and reliability requirements of the application scenario, for example, in a highly automated production line, the mode of automatic shutdown and switching to a redundant system can be preferred, while in a laboratory environment, short message and interface warning can be preferred. For example, in the trial run of a certain motor housing die casting line, once the system health index drops from 75 to below 60, the monitoring platform immediately outputs a circuit breaking signal through the industrial Ethernet, accompanied by a short message to notify the maintenance engineer, and at the same time, the fault node coordinates and the type of the sensor belonging to the SCADA system interface are highlighted, thereby realizing the whole process closed loop from data acquisition to fault response, effectively avoiding the “sudden death” failure of the equipment in the process of high load and long period operation.

[0046] The application belongs to the technical field of temperature regulation, and is applied to real-time heterogeneous data fusion of a multi-source sensor network to generate a global thermal field dynamic matrix, joint prediction processing of the global thermal field dynamic matrix and device working condition parameters by a TCN-BiGRU network to obtain thermal load probability distribution data, multi-objective topological optimization of the device flow constraint condition and the thermal load probability distribution data to obtain a cooling control vector, and variable gain feedforward compensation of the cooling control vector according to piezoelectric micropump characteristic data to generate driving voltage waveform data. The application realizes more accurate, efficient and reliable liquid cooling heat dissipation control through the multi-source sensor network, deep time sequence network, topological optimization and accurate driving of the piezoelectric micropump, and provides an important technical solution for heat dissipation of high-power and high-density electronic devices.

[0047] As shown in Figure 2 , it is a functional module diagram of the liquid cooling heat dissipation module intelligent dynamic temperature control device provided by the embodiment of the application.

[0048] In some embodiments, the liquid cooling heat dissipation module intelligent dynamic temperature control device 2 can include a plurality of functional modules composed of computer program segments. The computer programs of each program segment in the liquid cooling heat dissipation module intelligent dynamic temperature control device 2 can be stored in the memory of the server and executed by at least one processor to perform the functions of the liquid cooling heat dissipation module intelligent dynamic temperature control method (see Figure 1 Description) for details.

[0049] In this embodiment, the liquid cooling heat dissipation module intelligent dynamic temperature control device 2 can be divided into a plurality of functional modules according to the functions it performs. The functional modules can include a perception fusion module 21, a distribution prediction module 22, a spraying control module 23, a feedforward compensation module 24 and a heat dissipation early warning module 25. The module referred to by the application refers to a series of computer program segments that can be executed by at least one processor and can complete a fixed function, which are stored in the memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0050] The perception fusion module 21 is configured to perform real-time heterogeneous data fusion on a preset multi-source sensor network to generate a global thermal field dynamic matrix.

[0051] In an optional implementation, the perception fusion module 21 is specifically configured to: perform spatio-temporal synchronization alignment processing on the preset multi-source sensor network to obtain a multi-source synchronous heterogeneous detection data set; perform dimensionless normalization conversion processing on the multi-source synchronous heterogeneous detection data set to obtain a temperature data set; perform three-dimensional temperature field reconstruction processing according to a preset spatial distance weighted interpolation algorithm and the temperature data set to generate a global thermal field dynamic matrix.

[0052] a distribution prediction module 22 configured to perform joint prediction processing on the global thermal field dynamic matrix and the real-time collected device operating condition parameters according to a preset TCN-BiGRU network to obtain thermal load probability distribution data.

[0053] In an optional implementation, the distribution prediction module 22 is specifically configured to: perform feature splicing processing on the global thermal field dynamic matrix and the real-time collected device operating condition parameters to obtain a spatio-temporal feature tensor; perform local thermal feature extraction processing on the spatio-temporal feature tensor according to a preset TCN-BiGRU network to obtain a primary thermal feature vector; perform long-time dependence analysis processing on the primary thermal feature vector according to the preset TCN-BiGRU network to obtain a thermal risk probability value; perform Monte Carlo random disturbance evaluation processing on the thermal risk probability value to obtain thermal load probability distribution data.

[0054] a spray control module 23 configured to perform multi-objective topological optimization processing on the thermal load probability distribution data according to preset device flow constraint conditions to obtain a cooling control vector.

[0055] In an optional implementation, the spray control module 23 is specifically configured to: construct a temperature balance optimization objective function according to a high-temperature region in the thermal load probability distribution data, and construct a coolant consumption optimization objective function according to a flow threshold set in the preset device flow constraint conditions; perform Pareto frontier solving processing on the temperature balance optimization objective function and the coolant consumption optimization objective function according to a preset non-dominated sorting evolutionary algorithm to obtain a candidate flow distribution scheme set; perform decision screening processing on the candidate flow distribution scheme set according to a preset energy consumption-performance trade-off rule to obtain a cooling control vector.

[0056] a feedforward compensation module 24 configured to perform variable-gain feedforward compensation on the cooling control vector according to preset piezoelectric micropump characteristic data to generate driving voltage waveform data.

[0057] In an optional implementation, the feedforward compensation module 24 is specifically configured to: perform flow-response frequency conversion on the cooling control vector according to the response frequency-flow mapping data to obtain a target response frequency table sequence; perform dead-zone compensation processing on the target response frequency table sequence according to the dead-zone voltage threshold to obtain an actual driving frequency sequence; The actual driving frequency sequence is phase-advanced according to the fluid transmission delay parameter to obtain a delay-compensated frequency sequence; High-voltage sine wave synthesis processing is performed according to the delay-compensated frequency sequence to generate driving voltage waveform data.

[0058] In an optional embodiment, the liquid cooling heat dissipation module intelligent dynamic temperature control device 2 further comprises a heat dissipation early warning module 25, and the heat dissipation early warning module 25 is specifically used for: The liquid cooling heat dissipation module is controlled according to the driving voltage waveform data, and multi-source state acquisition processing is performed on the liquid cooling heat dissipation module to obtain a device operation state data set; A sensor node relationship topology graph is constructed according to the device operation state data set; An abnormal mode recognition processing is performed on the sensor node relationship topology graph through a preset graph convolution network to obtain a device health index; When the device health index is lower than a preset risk threshold, an active early warning instruction is generated according to a preset active early warning mode.

[0059] It should be understood that various changes and specific embodiments in the method provided by the above embodiments are also applicable to the liquid cooling heat dissipation module intelligent dynamic temperature control device of the present embodiment. Through the foregoing detailed description of the liquid cooling heat dissipation module intelligent dynamic temperature control method, those skilled in the art can clearly understand the implementation method of the liquid cooling heat dissipation module intelligent dynamic temperature control device in the present embodiment. For the sake of brevity of the description, no further description is given here.

[0060] As shown in FIG. 1, it is a structural schematic diagram of an electronic device provided by an embodiment of the present application. Figure 3 As shown in FIG. 1, it is a structural schematic diagram of an electronic device provided by an embodiment of the present application.

[0061] In the preferred embodiment of the present application, the electronic device 3 can include, but is not limited to, a memory 31, at least one processor 32, and at least one communication bus 33.

[0062] Those skilled in the art should understand that, Figure 3 The structure of the electronic device 3 shown in the figure is not a limitation of the embodiments of the present application. The electronic device 3 can also include more or less other hardware or software, or different component arrangements.

[0063] In some embodiments, the electronic device 3 is a device capable of automatically performing numerical calculation and / or information processing according to pre-set or stored instructions. The hardware thereof includes, but is not limited to, a microprocessor, an application specific integrated circuit, a programmable gate array, a digital processor, and an embedded device, etc.

[0064] It should be noted that the electronic device 3 is only an example, and other existing or future electronic products can also be applicable to the present application and should be included in the protection scope of the present application.

[0065] In some embodiments, the memory 31 stores a computer program which, when executed by the at least one processor 32, implements all or part of the steps of the intelligent dynamic temperature control method of the liquid cooling heat dissipation module as described. The memory 31 includes a Read-Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), a One-time Programmable Read-Only Memory (OTPROM), an Electrically-Erasable Programmable Read-Only Memory (EEPROM), a Compact Disc Read-Only Memory (CD-ROM) or other optical disk memory, a magnetic disk memory, a magnetic tape memory, or any other computer readable medium capable of carrying or storing data. Further, the computer readable storage medium can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, etc.

[0066] In some embodiments, the at least one processor 32 is a control unit of the electronic device 3, which connects various components of the entire electronic device 3 through various interfaces and lines, and performs various functions of the electronic device 3 and processes data by running or executing programs or modules stored in the memory 31 and calling data stored in the memory 31. For example, the at least one processor 32 executes the computer program stored in the memory 31 to implement all or part of the steps of the intelligent dynamic temperature control method of the liquid cooling heat dissipation module as described in the embodiments of the present application, or to implement all or part of the functions of the intelligent dynamic temperature control device of the liquid cooling heat dissipation module. The at least one processor 32 can be composed of integrated circuits, for example, can be composed of a single packaged integrated circuit, or can be composed of multiple packaged integrated circuits with the same function or different functions, including one or more combinations of central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips, etc.

[0067] In some embodiments, the at least one communication bus 33 is configured to enable connection communication between the memory 31 and the at least one processor 32. Although not shown, the electronic device 3 can further include a power supply (such as a battery) for powering the various components. Preferably, the power supply is logically connected to the at least one processor 32 via a power management device, which can enable management of charging, discharging, and power consumption management, etc. The power supply can also include one or more AC or DC power sources, recharging circuits, power failure detection circuitry, power conversion or inverter circuits, power status reporting circuits, and any other components associated with the generation, management and distribution of power for the electronic device 3. The electronic device 3 can further include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which are not described herein.

[0068] The integrated units in the form of software function modules described above can be stored in a computer readable storage medium. The software function modules described above are stored in a storage medium, and include a plurality of instructions for causing an electronic device (which can be a personal computer, an electronic device, or a network device, etc.) or a processor to execute part of the method described in each embodiment of the present application.

[0069] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented by other manners. For example, the apparatus embodiments described above are merely illustrative, for example, the division of the modules is merely a logical function division, and another division manner can be used in actual implementation.

[0070] The modules described as separate components can or can not be physically separated, and the components shown as modules can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0071] The above are the preferred embodiments of the present application, which do not limit the protection scope of the present application, therefore: any equivalent changes made on the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A liquid cooling module intelligent dynamic temperature control method, characterized in that: The method comprises: Perform real-time heterogeneous data fusion on the preset multi-source sensor network to generate a global thermal field dynamic matrix; A preset TCN-BiGRU network is used to perform joint prediction processing based on the global thermal field dynamic matrix and the equipment operating condition parameters collected in real time to obtain heat load probability distribution data; Performing multi-objective topology optimization processing according to preset equipment flow constraints and the heat load probability distribution data to obtain a cooling control vector; The cooling control vector is subjected to variable gain feedforward compensation according to preset piezoelectric micropump characteristic data to generate driving voltage waveform data.

2. The intelligent dynamic temperature control method for a liquid cooling module according to claim 1, characterized in that: The real-time heterogeneous data fusion of the preset multi-source sensor network to generate a global thermal field dynamic matrix includes: Perform spatiotemporal synchronization alignment on the preset multi-source sensor network to obtain a multi-source synchronous heterogeneous detection dataset; Performing dimension normalization conversion processing on the multi-source synchronous heterogeneous detection data set to obtain a temperature data set; A three-dimensional temperature field reconstruction process is performed according to a preset spatial distance weighted interpolation algorithm and the temperature data set to generate a global thermal field dynamic matrix.

3. The intelligent dynamic temperature control method for a liquid cooling module according to claim 1, characterized in that: The method of performing joint prediction processing based on the global thermal field dynamic matrix and the real-time collected equipment operating parameters through the preset TCN-BiGRU network to obtain heat load probability distribution data includes: Performing feature splicing processing on the global thermal field dynamic matrix and the equipment operating parameters collected in real time to obtain a spatiotemporal feature tensor; Performing local thermal feature extraction processing on the spatiotemporal feature tensor through a preset TCN-BiGRU network to obtain a primary thermal feature vector; Performing long-term dependency analysis on the primary thermal feature vector through a preset TCN-BiGRU network to obtain a thermal risk probability value; The thermal risk probability value is subjected to Monte Carlo random perturbation evaluation processing to obtain thermal load probability distribution data.

4. The intelligent dynamic temperature control method for a liquid cooling module according to claim 1, characterized in that: The performing of multi-objective topology optimization processing according to the preset equipment flow constraint conditions and the heat load probability distribution data to obtain the cooling control vector includes: Constructing a temperature balance optimization objective function based on the high temperature area in the heat load probability distribution data, and constructing a coolant consumption optimization objective function based on the flow threshold set in the preset equipment flow constraint condition; Performing Pareto front solving processing on the temperature balance optimization objective function and the coolant consumption optimization objective function by using a preset non-dominated sorting evolutionary algorithm to obtain a set of candidate flow distribution schemes; The candidate flow distribution scheme set is subjected to decision screening processing according to a preset energy consumption-performance trade-off rule to obtain a cooling control vector.

5. The intelligent dynamic temperature control method for a liquid cooling module according to claim 1, characterized in that: The piezoelectric micropump characteristic data includes response frequency-flow mapping data, dead zone voltage threshold, and fluid transmission delay parameter. The performing variable gain feedforward compensation on the cooling control vector according to the preset piezoelectric micropump characteristic data to generate driving voltage waveform data includes: Performing flow-response frequency conversion on the cooling control vector according to the response frequency-flow mapping data to obtain a target response frequency table sequence; Performing dead zone compensation processing on the target response frequency table sequence according to the dead zone voltage threshold to obtain an actual driving frequency sequence; Performing phase advance correction processing on the actual driving frequency sequence according to the fluid transmission delay parameter to obtain a delay compensation frequency sequence; A high-voltage sine wave synthesis process is performed according to the delay compensation frequency sequence to generate driving voltage waveform data.

6. The intelligent dynamic temperature control method for a liquid cooling module according to claim 1, characterized in that: The method further comprises: Controlling a liquid cooling and heat dissipation module according to the driving voltage waveform data and performing multi-source state acquisition processing on the liquid cooling and heat dissipation module to obtain a device operation status data set; Constructing a sensor node relationship topology diagram based on the device operation status data set; Performing abnormal pattern recognition processing on the sensor node relationship topology graph through a preset graph convolutional network to obtain a device health index; When the equipment health index is lower than a preset risk threshold, an active warning instruction is generated according to a preset active warning method.

7. An intelligent dynamic temperature control device for a liquid cooling module, characterized in that: The device comprises: The perception fusion module is used to perform real-time heterogeneous data fusion on the preset multi-source sensor network to generate a global thermal field dynamic matrix; A distribution prediction module is used to perform joint prediction processing based on the global thermal field dynamic matrix and the equipment operating condition parameters collected in real time through a preset TCN-BiGRU network to obtain heat load probability distribution data; A spray control module is used to perform multi-objective topology optimization processing based on preset equipment flow constraints and the heat load probability distribution data to obtain a cooling control vector; The feedforward compensation module is used to perform variable gain feedforward compensation on the cooling control vector according to preset piezoelectric micropump characteristic data to generate driving voltage waveform data.

8. An electronic device, characterized in that: The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the intelligent dynamic temperature control method for the liquid cooling module according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent dynamic temperature control method for a liquid cooling heat dissipation module according to any one of claims 1 to 6 are implemented.

Images