CN116501104A - Temperature control device and method for liquid cooling heat dissipation system - Google Patents
Temperature control device and method for liquid cooling heat dissipation system Download PDFInfo
- Publication number
- CN116501104A CN116501104A CN202310448563.6A CN202310448563A CN116501104A CN 116501104 A CN116501104 A CN 116501104A CN 202310448563 A CN202310448563 A CN 202310448563A CN 116501104 A CN116501104 A CN 116501104A
- Authority
- CN
- China
- Prior art keywords
- data
- wolves
- temperature
- heat dissipation
- dsp
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000017525 heat dissipation Effects 0.000 title claims abstract description 64
- 238000000034 method Methods 0.000 title claims abstract description 48
- 239000007788 liquid Substances 0.000 title claims abstract description 42
- 238000001816 cooling Methods 0.000 title claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 claims abstract description 31
- 230000008569 process Effects 0.000 claims abstract description 19
- 238000004891 communication Methods 0.000 claims abstract description 7
- 238000011217 control strategy Methods 0.000 claims abstract description 7
- 241000282421 Canidae Species 0.000 claims description 83
- 241000282461 Canis lupus Species 0.000 claims description 43
- 230000015654 memory Effects 0.000 claims description 41
- 239000002245 particle Substances 0.000 claims description 31
- 238000012545 processing Methods 0.000 claims description 21
- 238000007781 pre-processing Methods 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000001514 detection method Methods 0.000 claims description 9
- 238000013500 data storage Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 5
- 238000005057 refrigeration Methods 0.000 claims description 5
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000007726 management method Methods 0.000 claims description 3
- 238000013178 mathematical model Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000007423 decrease Effects 0.000 claims description 2
- 238000005265 energy consumption Methods 0.000 abstract description 4
- 239000012530 fluid Substances 0.000 description 9
- 238000010586 diagram Methods 0.000 description 8
- 238000013461 design Methods 0.000 description 5
- 230000002431 foraging effect Effects 0.000 description 3
- 230000005855 radiation Effects 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 2
- 244000144992 flock Species 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000033228 biological regulation Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D23/00—Control of temperature
- G05D23/19—Control of temperature characterised by the use of electric means
- G05D23/20—Control of temperature characterised by the use of electric means with sensing elements having variation of electric or magnetic properties with change of temperature
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Control Of Temperature (AREA)
Abstract
The invention discloses a temperature control device and a method for a liquid cooling heat dissipation system, wherein the device comprises an ARM processor, a DSP processor, FLASH, SRAM, SDRAM, EEPROM and a plurality of external communication interfaces; the method comprises the following steps: the ARM processor receives various sensor data of the liquid cooling heat dissipation system and transmits the data to the DSP processor through a bus; the DSP processor stores, classifies and preprocesses the received data and trains a system temperature prediction model according to the PSO-GWO-RBF neural network theory; sending real-time data received by the DSP processor into a prediction model for operation to obtain a predicted temperature and sending the data back to the ARM processor; and finally, the ARM processor processes the obtained data, calculates a control strategy, controls the action of the actuator and controls the temperature of the heat dissipation system. The invention realizes the accurate temperature control of the system, improves the heat dissipation efficiency of the liquid cooling heat dissipation system and reduces the energy consumption of the system.
Description
Technical Field
The invention belongs to the technical field of heat dissipation system control, and relates to a liquid cooling heat dissipation system temperature prediction method and a control device.
Background
The liquid cooling heat dissipation system is a cooling system for dissipating heat through liquid. In the process of radiating heat generated by the system on equipment, the system is required to accurately control the radiating system because the equipment generates different heat under different working conditions, and particularly a large amount of heat can be generated when the equipment is abnormal.
The temperature of the liquid cooling heat radiation system liquid supply port is cooled by the refrigeration equipment and basically kept constant, so that the equipment heat consumption can be calculated by only detecting the temperature of the liquid return port, and then the actuator is controlled to act according to the heat consumption so as to accurately control the temperature, thereby achieving the heat radiation purpose.
At present, the temperature control of the liquid cooling heat dissipation system mainly comprises the following three modes:
the first temperature control mode is to radiate heat according to the maximum value of the heat radiation capacity of the designed equipment, mainly adopts the open-loop control of a water pump, and has constant fluid flow and no control of a valve. The defects of the design are energy consumption waste, temperature control cannot be realized, and energy conservation and emission reduction cannot be realized.
The second temperature control mode is to measure the temperature of the liquid return port as the main parameter and detect the temperature difference of the liquid at the inlet and outlet for heat dissipation control. The design has hysteresis in temperature detection caused by fluid heat exchange, hysteresis in heat dissipation control, and slow control response when abnormal conditions occur, so that heat dissipation of equipment is not facilitated, and equipment damage can be caused.
The third temperature control mode is to use a common neural network, a support vector machine and the like as models, adopt a common embedded single chip microcomputer as hardware design, use related data to predict the temperature of a liquid return port and regulate and control the system temperature. The design is low in temperature prediction precision and cannot achieve accurate temperature control and heat dissipation because the operation capability of the processor is not strong enough and the model structure is simple.
Disclosure of Invention
The invention aims to provide a liquid cooling heat dissipation system temperature control device and method with high temperature detection precision, strong real-time performance, accurate temperature control, high heat dissipation efficiency and low energy consumption.
The technical solution for realizing the purpose of the invention is as follows: a temperature control device of a liquid cooling heat dissipation system comprises an ARM processor, a DSP processor, FLASH, SRAM, SDRAM, EEPROM and a plurality of external communication interfaces;
the DSP processor is used for data processing and algorithm operation;
the ARM processor is used for data receiving and transmitting and system control;
the FLASH and the SRAM are used as memories required by the operation of the ARM processor and used for data storage and processing, and the EEPROM and the SDRAM are used as memories required by the DSP processor and used for data storage and processing;
the plurality of external communication interfaces comprise GPIO input/output interfaces, RS485/422 interfaces, ethernet control interfaces and DAC interfaces.
A temperature control method of a liquid cooling heat dissipation system comprises the following steps:
step 1, a data receiving module of a DSP processor receives system sensor data transmitted by an ARM processor, envelops pressure, temperature and flow data and stores the envelop pressure, temperature and flow data into a memory;
step 2, classifying and preprocessing the historical data stored in the memory by a data preprocessing module of the DSP processor to obtain a memory historical data sample;
step 3, a model building module of the DSP processor trains model parameters according to the theory of the PSO-GWO-RBF neural network by utilizing historical data samples of the memory to obtain a heat dissipation system temperature prediction model;
and 4, the DSP data output module sends the real-time data into a prediction model, and outputs the predicted temperature to the ARM processor through the calculation of the processor to control the temperature of the heat dissipation system.
Compared with the prior art, the invention has the remarkable advantages that: (1) The hardware platform based on ARM and DSP has strong operation and control capability, meets the system algorithm operation, data processing and control capability, and improves the temperature control efficiency of the liquid cooling heat dissipation system; (2) The system temperature is predicted based on the PSO-GWO-RBF neural network model, so that the accurate temperature of the liquid cooling heat dissipation system and the accurate heat dissipation of equipment are realized, and the method has the advantages of high convergence speed and high prediction precision; (3) The problems of lag temperature detection and low predicted temperature precision of the liquid cooling heat dissipation system are solved, the accurate temperature control of the liquid cooling heat dissipation system is realized, the heat dissipation efficiency of the liquid cooling heat dissipation system is improved, and the energy consumption of the system is reduced.
Drawings
Fig. 1 is a schematic diagram of a liquid-cooled heat dissipation system.
Fig. 2 is a schematic hardware diagram of a temperature control device of a liquid cooling heat dissipation system according to the present invention.
Fig. 3 is a software flow chart of a temperature control device of a liquid cooling heat dissipation system.
Fig. 4 is a schematic diagram of an RBF neural network according to an embodiment of the invention.
Fig. 5 is a flowchart of the GWO algorithm according to an embodiment of the invention.
Fig. 6 is a schematic flow chart of a PSO algorithm in an embodiment of the present invention.
FIG. 7 is a schematic flow chart of establishing a PSO-GWO-RBF neural network model according to an embodiment of the present invention.
Detailed Description
The invention relates to a temperature control device of a liquid cooling heat dissipation system, which comprises an ARM processor, a DSP processor, FLASH, SRAM, SDRAM, EEPROM and a plurality of external communication interfaces;
the DSP processor is used for data processing and algorithm operation;
the ARM processor is used for data receiving and transmitting and system control;
the FLASH and the SRAM are used as memories required by the operation of the ARM processor and used for data storage and processing, and the EEPROM and the SDRAM are used as memories required by the DSP processor and used for data storage and processing;
the plurality of external communication interfaces comprise GPIO input/output interfaces, RS485/422 interfaces, ethernet control interfaces and DAC interfaces.
As a specific example, the liquid-cooled heat dissipation system includes a heat dissipation system temperature control device, a sensor, a refrigeration device, a liquid supply pump, a heat exchanger, a filter, a valve, and a pipeline.
As a specific example, the software of the DSP processor includes a data receiving module, a data preprocessing module, a model building module, and a DSP data output module;
the data receiving module is used for receiving the sensor data transmitted by the ARM processor and storing the sensor data in the memory;
the data preprocessing module classifies and preprocesses received data;
the model building module is used for building a prediction model by using a PSO-GWO-RBF neural network theory;
and the DSP data output module sends the real-time data into the prediction model, calculates the predicted temperature, and then transmits the data to the ARM processor for controlling the system temperature.
As a specific example, the software of the ARM processor includes a data acquisition module, a data processing module, a driving executor module, an ARM data output module, and a detection module;
the data acquisition module acquires data and states of the sensor through a serial port, and receives predicted temperature data transmitted by the DSP through a data bus;
the data processing module sends the predicted temperature and the real-time temperature into the controller for processing, and calculates a control strategy;
the driving executor module converts the control error into digital quantity according to the control strategy, and sends the digital quantity to the DA chip to be converted into analog quantity for output, so as to drive the executor to work;
the ARM data output module is used for transmitting the acquired sensor data to the DSP on one hand; on the other hand, the obtained real-time data and state information are transmitted to the display equipment through a network;
the detection module detects various states of the equipment in real time and outputs level signals in real time through the IO pins; and input detection is carried out on the keys, so that the functions of real-time control and emergency stop are realized.
A temperature control method of a liquid cooling heat dissipation system comprises the following steps:
step 1, a data receiving module of a DSP processor receives system sensor data transmitted by an ARM processor, envelops pressure, temperature and flow data and stores the envelop pressure, temperature and flow data into a memory;
step 2, classifying and preprocessing the historical data stored in the memory by a data preprocessing module of the DSP processor to obtain a memory historical data sample;
step 3, a model building module of the DSP processor trains model parameters according to the theory of the PSO-GWO-RBF neural network by utilizing historical data samples of the memory to obtain a heat dissipation system temperature prediction model;
and 4, the DSP data output module sends the real-time data into a prediction model, and outputs the predicted temperature to the ARM processor through the calculation of the processor to control the temperature of the heat dissipation system.
As a specific example, the data preprocessing module of the DSP processor in step 2 classifies and preprocesses the history data stored in the memory to obtain a memory history data sample, which is specifically as follows:
step 2.1, classifying historical data by using an external memory of the DSP;
step 2.2, selecting an inlet temperature T1, an inlet pressure P1, an inlet flow L1, an outlet temperature T2, an outlet pressure P2 and an outlet flow L2 at the time T-1 as input variables of a solving model, and carrying out normalization processing on the input variables, wherein the expression is as follows:
wherein y represents a memory history data sample, and the value range is [0,1 ]]The method comprises the steps of carrying out a first treatment on the surface of the y represents sample data, y max Is the maximum value of the sample data, y min Is the minimum of the sample data;
and 2.3, taking the outlet temperature T3 at the next moment, namely the moment T, as an output variable of the solving model, training the solving model, and obtaining a memory historical data sample through the trained solving model.
As a specific example, the model building module of the DSP processor in step 3 trains model parameters according to the theory of the PSO-GWO-RBF neural network by using the memory historical data sample to obtain a heat dissipation system temperature prediction model, which is specifically as follows:
step 3.1, a DSP processor establishes a temperature prediction model based on an RBF neural network;
and 3.2, optimizing and training the neural network parameters by adopting a PSO-GWO algorithm by the DSP processor to obtain an optimized temperature prediction model.
As a specific example, the DSP processor in step 3.1 establishes a temperature prediction model based on the RBF neural network, specifically:
the RBF neural network adopting multiple inputs and single outputs consists of an input layer, an hidden layer and an output layer, wherein the relation between the output Y (X) and the input X is expressed as follows:
wherein Y (X) is temperature output, X is input vector, w 0 Is an offset for adjusting the output; w (w) i For connecting the weight, the weight between the i hidden node base function and the output node is represented; n is n c C is the number of hidden nodes i Field center, sigma, of the ith hidden node i The field width of the ith hidden node;
calculating the number n of hidden nodes of the neural network by using a clustering algorithm c Determining a connection weight w using a minimum gradient descent method i Field center C of ith hidden node i And the field width sigma of the ith hidden node i 。
As a specific example, the DSP processor described in step 3.2 uses the PSO-GWO algorithm to perform optimization training on the neural network parameters, which is specifically as follows:
in the process of one iteration operation, the GWO algorithm has three solutions of alpha, beta and delta, which are used for capturing the leading wolf group, wherein alpha is the head wolf in the wolf group, and the whole wolf group is managed; beta is the second grade of wolves, subject to alpha wolves, managing wolves lower than it; delta is the third grade wolf, obeys alpha, beta wolf, manages the low grade wolf group; omega is called the bottom layer of wolves, and listens to the management of alpha, beta and delta wolves, so as to surround, attack, acquire food and find the optimal solution;
for the theory above, the following mathematical model is proposed:
in the method, in the process of the invention,respectively representing the distances between the individual wolves and the alpha wolves, the beta wolves and the delta wolves; />Respectively representing the updated positions of the individual wolves under the guidance of alpha wolves, beta wolves and delta wolves; />Respectively represent alpha wolf and betaThe position of wolves, delta wolves;representing the position of individual gray wolves;
are all swinging factors->Are all convergence factors; the expression is as follows: c=2r 1 、A＝2ar 1 -a、/>Wherein C takes the value->A is a value->a is a control quantity parameter, and linearly decreases from 2 to 0; i is the current iteration number; i max The maximum iteration number; r is (r) 1 Is [0,1]A random vector between the two;
formulas (2) to (8) ensure that the whole population can search for more optimal positions in iteration, and the solutions of alpha, beta and delta provide multiple optimal positions;
in the PSO algorithm, the space for searching food is assumed to be P dimension, the total particle number is n, and the position of the ith particle in the P dimension space is x i ＝(x i1 ,x i2 ,…,x iP ) The flying speed is v i ＝(v i1 ,v i2 ,…,v iP ) Each particle has an adaptation value determined by an optimized objective function, and the following two factors need to be considered in searching: the particle searches for the best position P to date b The method comprises the steps of carrying out a first treatment on the surface of the All particles search for best P g ；
The position and speed update formula of the particle swarm algorithm is as follows:
wherein w is the inertial weight,for particle i in the mth iteration process speed f-th dimension component,/th>Representing the f-th dimensional component of the position of particle i during the mth iteration,/th>For particle i in all positions the optimal position, the f-th dimension component,>the f-th component of the optimal position in all particles; x is x 1 ,x 2 Is a weight factor, epsilon is [0,1 ]]Random numbers in between;
in the process of carrying out position operation, optimizing the position update of the gray wolves in the optimizing process by adopting a PSO algorithm; in the GWO algorithm, the best position calculated by individual iterative calculation of the wolf is added into a position updating formula by combining with the PSO algorithm, so that the best position information of the wolf is kept in the operation, the GWO algorithm is prevented from sinking into local optimum along with the increase of the iteration times, and the speed and the position of the wolf in the PSO-GWO algorithm are updated as follows:
v m,n (t+1)＝w 1 v m,n (t)+c 1 r 1 [p m,n -x m,n (t)]+c 2 r 2 [w 1 X 1 +w 2 X 2 +w 3 X 3 -x m,n (t)] (11)
x m,n (t+1)＝x m,n (t)+v m,n (t+1) (12)
wherein,,m＝1,2,…,N,n＝1,2,…,D；p m,n is the best position the individual of the wolf experiences in finding the prey, c 1 r 1 [p m,n -x m,n (t)]Experience of the position of the gray wolf individual; c 2 r 2 [w 1 X 1 +w 2 X 2 +w 3 X 3 -x m,n (t)]Learning positions for the alpha wolves, the beta wolves and the delta wolves for the gray wolves; w (w) 1 、w 2 、w 3 All are inertia weight coefficients; r is (r) 2 Is [0,1]A random vector between the two;
the process of performing optimization training is as follows:
(1) firstly, initializing parameters a, A and C in a software program; selecting parameter n=30, maximum iteration number C max Initial inertial weight w=1.5, initializing individual gray wolves positions Xi, and sequentially arranging individual gray wolves fitness;
(2) defining alpha, beta and delta as the gray wolves with the first three fitness values, wherein the positions corresponding to the alpha, beta and delta of the gray wolves are X alpha, X beta and X delta respectively;
(3) updating the values of a, A and C according to the wolf group surrounding formulas, namely formula (11) and formula (12);
(4) updating the position of the individual wolves according to the formulas (2) to (8);
(5) updating the speed V and the position X of the individual gray wolves according to the formulas (11) to (12);
if the iteration times are not reached, returning to the step (4) to continue to execute the operation; if the iteration times are reached, the gray wolf is considered to find the optimal position;
and establishing a heat dissipation system temperature prediction model according to the optimized parameters.
As a specific example, in step 4, the DSP data output module sends real-time data to the prediction model, and outputs the predicted temperature to the ARM processor after the calculation of the processor, so as to control the temperature of the heat dissipation system, specifically as follows:
and (3) the DSP data output module sends the real-time data into the heat dissipation system temperature prediction model calculated in the step (3), outputs a predicted temperature value, sends the predicted temperature value to the ARM processor, and the ARM calculates a control strategy to control the temperature of the heat dissipation system.
The invention will now be described in further detail with reference to the drawings and examples.
Example 1
The embodiment provides a temperature control device of a liquid cooling heat dissipation system, which is a control device based on ARM and DSP processors as cores and a temperature regulation and control method based on PSO-GWO-RBF neural network theory. The device and the method can be applied to temperature prediction and control of the liquid cooling heat dissipation system.
Fig. 1 is a schematic diagram of a liquid cooling system in this embodiment, which includes a temperature control device, a sensor, a refrigeration device, a liquid supply pump, a heat exchanger, a filter, a valve, and a pipeline. The fluid supply pump supplies fluid meeting the requirements to the load through pressure, flow, temperature and water quality parameters provided by the temperature control device; in the circulation process of the fluid, the temperature of the fluid is increased due to the heat of the power mechanical component and the friction heat of the fluid, the fluid returns to the liquid cooling system pipeline, the temperature of the fluid is reduced to a set value through the refrigerating equipment, the heat is taken away, and the circulation is performed. The temperature of the liquid supply port is reduced by the refrigeration equipment and kept constant, so that the heat consumption of the equipment can be calculated by only detecting the temperature of the liquid return port, and then the temperature is accurately controlled according to the heat consumption, such as the rotating speed of the actuator water pump, the opening of the valve and the like, so that the aim of heat dissipation of the equipment is achieved.
Fig. 2 is a schematic diagram of the temperature control device, which is the core of the whole liquid cooling system. The control device takes a DSP and an ARM processor as hardware circuits of the framework, and the DSP processor is used for data processing and algorithm operation due to excellent operation performance; the ARM processor is used for data receiving and system control functions. The DSP processor and the ARM processor are connected through a 16-bit parallel bus and used for exchanging original data and operation data. The SRAM and SDRAM are used as external expansion memories required by the operation of the DSP processor and the ARM processor respectively, and are used for storing and processing a large amount of data. According to the system requirements, ARM processors design a plurality of different types of external data interfaces for various system functions.
Example 2
Referring to fig. 3, the embodiment provides a temperature control method of a liquid cooling heat dissipation system, which includes the following steps:
step 1, a data receiving module of an ARM processor receives system sensor data transmitted by the ARM processor, envelops pressure, temperature and flow data and stores the envelop pressure, temperature and flow data into a memory;
step 2, classifying and preprocessing the historical data stored in the memory by a model building module of the DSP processor;
step 3, the DSP processor trains model parameters according to the theory of the PSO-GWO-RBF neural network by utilizing the historical data of the memory to obtain a heat dissipation system temperature prediction model;
and 4, the DSP data output module sends real-time data into the model, and outputs the predicted temperature to the ARM processor after calculation of the processor to control the temperature of the heat dissipation system.
Further, referring to fig. 7, the model building module of the DSP processor in step 2 classifies and pre-processes the history data stored in the memory,
step 2.1, classifying historical data by using an external memory of the DSP;
step 2.2, selecting an inlet temperature T1, an inlet pressure P1, an inlet flow L1, an outlet temperature T2, an outlet pressure P2 and an outlet flow L2 at the time T-1 as input variables of a solving model, and carrying out normalization processing on the input variables, wherein the expression is as follows:
wherein y represents a memory history data sample, and the value range is [0,1 ]]The method comprises the steps of carrying out a first treatment on the surface of the y represents sample data, y max Is the maximum value of the sample data, y min Is the minimum of the sample data;
and 2.3, taking the outlet temperature T3 at the next moment, namely the moment T, as an output variable of the solving model, training the solving model, and obtaining a memory historical data sample through the trained solving model.
The DSP processor in the step 3 trains model parameters according to the theory of the PSO-GWO-RBF neural network by utilizing the historical data of the memory to obtain a heat dissipation system temperature prediction model, and the method is concretely as follows:
step 3.1, a DSP processor establishes a temperature prediction model based on an RBF neural network, and the temperature prediction model is specifically as follows:
fig. 4 is a schematic structural diagram of an RBF neural network, which is a basic model of system temperature prediction. The RBF neural network is a feed-forward network with a single hidden layer, has a simple structure and high learning speed, and can approximate any continuous function with any precision. In FIG. 4, the RBF neural network with multiple inputs and single output is composed of three layers of an input layer, an hidden layer and an output layer, and the relationship between the output Y (X) and the input X can be expressed as
Wherein Y (X) is temperature output, X is input vector, w 0 Is an offset for adjusting the output; w (w) i For connecting the weight, the weight between the i hidden node base function and the output node is represented; n is n c C is the number of hidden nodes i Field center, sigma, of the ith hidden node i The field width of the i-th hidden node.
Calculating the number n of hidden nodes of the neural network by using a clustering algorithm c Determining a connection weight w using a minimum gradient descent method i Hidden node field center C i And a hidden node field width sigma i ；
And 3.2, optimizing and training the neural network parameters by adopting a PSO-GWO algorithm to obtain a heat dissipation system temperature prediction model, wherein the method comprises the following steps of:
fig. 5 is a schematic diagram of the GWO algorithm, GWO is a population intelligent optimization algorithm, and three solutions, called α, β, δ respectively, are presented during one iteration operation for capturing behavior of the main lead wolf population. Wherein alpha is called the head wolf in the wolf group, and can manage the whole wolf group; beta is called the second class of wolves, which obeys alpha wolves, managing wolves lower than it; delta is called third class wolf, which is subject to alpha, beta wolf, managing low class wolves; omega is called the bottom-most wolf, listening to the management of alpha, beta, delta wolves, thus surrounding, attacking, getting food, and finding the optimal solution. For the theory above, the following mathematical model is proposed:
respectively representing the distances between the individual wolves and the alpha wolves, the beta wolves and the delta wolves; />Respectively representing the updated positions of the individual wolves under the guidance of alpha wolves, beta wolves and delta wolves; />Respectively representing the positions of alpha wolves, beta wolves and delta wolves; />Representing the position of the individual gray wolves. />Are all swinging factors->Are all convergence factors, expressed as follows:
C＝2r 1
A＝2ar 1 -a
wherein a is a control quantity parameter, and is linearly reduced from 2 to 0; i is the current iteration number; i max The maximum iteration number; r is (r) 1 、r 2 Is [0,1]Random vector between.
Equations (2) through (8) can ensure that the entire population can find a better position in the iteration, and the solutions of alpha, beta and delta can provide a plurality of optimal positions.
FIG. 6 is a schematic diagram of a PSO algorithm that originates from the process of foraging a flock of birds, equivalent to simulating the foraging of a flock of birds. Each bird is a particle in the algorithm, i.e. we solve the possible solutions of the model. The bird can continuously change its position and speed during foraging, and is used for being closer to food. Assuming that the space for searching food is P dimension, the total particle number is n, and the position of the ith particle in the P dimension space is x i ＝(x i1 ,x i2 ,…,x iP ) The flying speed is v i ＝(v i1 ,v i2 ,…,v iP ). Each particle has an adaptation value determined by an optimized objective function, and the following two factors need to be considered in searching:
1. the particle searches for the best position P to date b
2. All particles search for best P g
The position and speed update formula of the particle swarm algorithm is as follows:
wherein w is the inertial weight,show the f-th dimension component of the velocity of particle i during the mth iteration,/th>Representing the f-th dimensional component of the position of particle i during the mth iteration,/th>For particle i in all positions the optimal position, the f-th dimension component,>the f-th dimension component of the optimal position in all particles. X is x 1 ,x 2 Is a weight factor, epsilon is [0,1 ]]Random numbers in between.
In the process of carrying out position operation, a PSO algorithm is adopted to optimize the position update of the gray wolves in the optimizing process. In the GWO algorithm, the best position of iterative computation of the gray wolf individuals is added into a position updating formula by combining with the PSO algorithm, so that the best position information of the gray wolf individuals can be reserved in the operation, and the problem that the accuracy is low due to the fact that the GWO algorithm is in local optimum along with the increase of the iteration times is avoided. The speed and position of the wolf in the PSO-GWO algorithm are updated as follows:
v m,n (t+1)＝w 1 v m,n (t)+c 1 r 1 [p m,n -x m,n (t)]+c 2 r 2 [w 1 X 1 +w 2 X 2 +w 3 X 3 -x m,n (t)] (11)
x m,n (t+1)＝x m,n (t)+v m,n (t+1) (12)
where m=1, 2, …, N, n=1, 2, …, D; p is p m,n Is the best position the individual of the wolf experiences in finding the prey, c 1 r 1 [p m,n -x m,n (t)]Experience of the position of the gray wolf individual; c 2 r 2 [w 1 X 1 +w 2 X 2 +w 3 X 3 -x m,n (t)]Learning positions for the alpha wolves, the beta wolves and the delta wolves for the gray wolves; w (w) 1 、w 2 、w 3 Are all inertial weight coefficients.
Selecting proper parameters a, A and C in a software program for initialization; selecting parameter n=30, maximum iteration number C max Initial inertial weight w=1.5, initializing individual gray wolves positions Xi, and sequentially arranging individual gray wolves fitness;
defining alpha, beta and delta as the gray wolves with the first three fitness values, wherein the corresponding positions of the gray wolves are X alpha, X beta and X delta respectively;
updating the values of a, A, C according to the wolf crowd surrounding formula;
updating the position of the individual gray wolves according to formulas (2) to (8); updating the speed V and the position X of the individual gray wolves according to the formulas (11) - (12); if the iteration times are not reached, returning to continue to execute the operation of the step; if the iteration times are reached, the gray wolf is considered to find the optimal position;
and establishing a heat dissipation system temperature prediction model according to the optimized parameters.
Further, the DSP data output module in step 4 sends the real-time data into the model, and outputs the predicted temperature to the ARM processor after the calculation of the processor, so as to control the temperature of the heat dissipation system.
It is easy to understand that various embodiments of the present invention can be envisioned by those of ordinary skill in the art without altering the true spirit of the present invention in light of the present teachings. Accordingly, the following detailed description and drawings are merely illustrative of the invention and are not intended to be exhaustive or to limit or restrict the invention.
Claims (10)
1. The temperature control device of the liquid cooling heat dissipation system is characterized by comprising an ARM processor, a DSP processor, FLASH, SRAM, SDRAM, EEPROM and a plurality of external communication interfaces;
the DSP processor is used for data processing and algorithm operation;
the ARM processor is used for data receiving and transmitting and system control;
the FLASH and the SRAM are used as memories required by the operation of the ARM processor and used for data storage and processing, and the EEPROM and the SDRAM are used as memories required by the DSP processor and used for data storage and processing;
the plurality of external communication interfaces comprise GPIO input/output interfaces, RS485/422 interfaces, ethernet control interfaces and DAC interfaces.
2. The liquid-cooled heat sink system temperature control device of claim 1, wherein the liquid-cooled heat sink system comprises a heat sink system temperature control device, a sensor, a refrigeration device, a liquid supply pump, a heat exchanger, a filter, a valve, and a pipe.
3. The device according to claim 1, wherein the software of the DSP processor includes a data receiving module, a data preprocessing module, a model building module, and a DSP data output module;
the data receiving module is used for receiving the sensor data transmitted by the ARM processor and storing the sensor data in the memory;
the data preprocessing module classifies and preprocesses received data;
the model building module is used for building a prediction model by using a PSO-GWO-RBF neural network theory;
and the DSP data output module sends the real-time data into the prediction model, calculates the predicted temperature, and then transmits the data to the ARM processor for controlling the system temperature.
4. The liquid cooling heat dissipation system temperature control device according to claim 1, wherein the software of the ARM processor comprises a data acquisition module, a data processing module, a driving actuator module, an ARM data output module and a detection module;
the data acquisition module acquires data and states of the sensor through a serial port, and receives predicted temperature data transmitted by the DSP through a data bus;
the data processing module sends the predicted temperature and the real-time temperature into the controller for processing, and calculates a control strategy;
the driving executor module converts the control error into digital quantity according to the control strategy, and sends the digital quantity to the DA chip to be converted into analog quantity for output, so as to drive the executor to work;
the ARM data output module is used for transmitting the acquired sensor data to the DSP on one hand; on the other hand, the obtained real-time data and state information are transmitted to the display equipment through a network;
the detection module detects various states of the equipment in real time and outputs level signals in real time through the IO pins; and input detection is carried out on the keys, so that the functions of real-time control and emergency stop are realized.
5. The temperature control method of the liquid cooling heat dissipation system is characterized by comprising the following steps of:
step 1, a data receiving module of a DSP processor receives system sensor data transmitted by an ARM processor, envelops pressure, temperature and flow data and stores the envelop pressure, temperature and flow data into a memory;
step 2, classifying and preprocessing the historical data stored in the memory by a data preprocessing module of the DSP processor to obtain a memory historical data sample;
step 3, a model building module of the DSP processor trains model parameters according to the theory of the PSO-GWO-RBF neural network by utilizing historical data samples of the memory to obtain a heat dissipation system temperature prediction model;
and 4, the DSP data output module sends the real-time data into a prediction model, and outputs the predicted temperature to the ARM processor through the calculation of the processor to control the temperature of the heat dissipation system.
6. The method of claim 5, wherein the data preprocessing module of the DSP processor in step 2 classifies and preprocesses the history data stored in the memory to obtain a memory history data sample, and the method specifically comprises the following steps:
step 2.1, classifying historical data by using an external memory of the DSP;
step 2.2, selecting an inlet temperature T1, an inlet pressure P1, an inlet flow L1, an outlet temperature T2, an outlet pressure P2 and an outlet flow L2 at the time T-1 as input variables of a solving model, and carrying out normalization processing on the input variables, wherein the expression is as follows:
wherein y represents a memory history data sample, and the value range is [0,1 ]]The method comprises the steps of carrying out a first treatment on the surface of the y represents sample data, y max Is the maximum value of the sample data, y min Is the minimum of the sample data;
and 2.3, taking the outlet temperature T3 at the next moment, namely the moment T, as an output variable of the solving model, training the solving model, and obtaining a memory historical data sample through the trained solving model.
7. The method according to claim 5, wherein the model building module of the DSP processor in step 3 trains model parameters according to the theory of the PSO-GWO-RBF neural network by using the memory history data samples to obtain a heat dissipation system temperature prediction model, and the method comprises the following steps:
step 3.1, a DSP processor establishes a temperature prediction model based on an RBF neural network;
and 3.2, optimizing and training the neural network parameters by adopting a PSO-GWO algorithm by the DSP processor to obtain an optimized temperature prediction model.
8. The method for controlling temperature of a liquid cooling heat dissipation system according to claim 7, wherein the DSP processor in step 3.1 establishes a temperature prediction model based on an RBF neural network, specifically:
the RBF neural network adopting multiple inputs and single outputs consists of an input layer, an hidden layer and an output layer, wherein the relation between the output Y (X) and the input X is expressed as follows:
wherein Y (X) is temperature output, X is input vector, w 0 Is an offset for adjusting the output; w (w) i For connecting the weight, the weight between the i hidden node base function and the output node is represented; n is n c C is the number of hidden nodes i Field center, sigma, of the ith hidden node i The field width of the ith hidden node;
calculating the number n of hidden nodes of the neural network by using a clustering algorithm c Determining a connection weight w using a minimum gradient descent method i Field center C of ith hidden node i And the field width sigma of the ith hidden node i 。
9. The method for controlling the temperature of a liquid-cooled heat dissipation system according to claim 7, wherein the DSP processor in step 3.2 performs optimization training on the parameters of the neural network by using a PSO-GWO algorithm, specifically as follows:
in the process of one iteration operation, the GWO algorithm has three solutions of alpha, beta and delta, which are used for capturing the leading wolf group, wherein alpha is the head wolf in the wolf group, and the whole wolf group is managed; beta is the second grade of wolves, subject to alpha wolves, managing wolves lower than it; delta is the third grade wolf, obeys alpha, beta wolf, manages the low grade wolf group; omega is called the bottom layer of wolves, and listens to the management of alpha, beta and delta wolves, so as to surround, attack, acquire food and find the optimal solution;
for the theory above, the following mathematical model is proposed:
in the method, in the process of the invention,respectively representing the distances between the individual wolves and the alpha wolves, the beta wolves and the delta wolves; />Respectively representing the updated positions of the individual wolves under the guidance of alpha wolves, beta wolves and delta wolves; />Respectively representing the positions of alpha wolves, beta wolves and delta wolves; />Representing the position of individual gray wolves;
are all swinging factors->Are all convergence factors; the expression is as follows: c=2r 1 、A＝2ar 1 a、Wherein C takes the value->A is a value->a is a control quantity parameter, and linearly decreases from 2 to 0; i is the current iteration number; i max The maximum iteration number; r is (r) 1 Is [0,1]A random vector between the two;
formulas (2) to (8) ensure that the whole population can search for more optimal positions in iteration, and the solutions of alpha, beta and delta provide multiple optimal positions;
in the PSO algorithm, the space for searching food is assumed to be P dimension, the total particle number is n, and the position of the ith particle in the P dimension space is x i ＝(x i1 ,x i2 ,…,x iP ) The flying speed is v i ＝(v i1 ,v i2 ,…,v iP ) Each particle has an optimized objective functionAdaptation value, and the following two factors need to be considered in searching: the particle searches for the best position P to date b The method comprises the steps of carrying out a first treatment on the surface of the All particles search for best P g ；
The position and speed update formula of the particle swarm algorithm is as follows:
wherein w is the inertial weight,for particle i in the mth iteration process speed f-th dimension component,/th>Representing the f-th dimensional component of the position of particle i during the mth iteration,/th>For particle i in all positions the optimal position, the f-th dimension component,>the f-th component of the optimal position in all particles; x is x 1 ,x 2 Is a weight factor, epsilon is [0,1 ]]Random numbers in between;
in the process of carrying out position operation, optimizing the position update of the gray wolves in the optimizing process by adopting a PSO algorithm; in the GWO algorithm, the best position calculated by individual iterative calculation of the wolf is added into a position updating formula by combining with the PSO algorithm, so that the best position information of the wolf is kept in the operation, the GWO algorithm is prevented from sinking into local optimum along with the increase of the iteration times, and the speed and the position of the wolf in the PSO-GWO algorithm are updated as follows:
v m,n (t+1)＝w 1 v m,n (t)+c 1 r 1 [p m,n -x m,n (t)]+c 2 r 2 [w 1 X 1 +w 2 X 2 +w 3 X 3 -x m,n (t)] (11)
x m,n (t+1)＝x m,n (t)+v m,n (t+1) (12)
wherein m=1, 2, …, N, n=1, 2, …, D; p is p m,n Is the best position the individual of the wolf experiences in finding the prey, c 1 r 1 [p m,n -x m,n (t)]Experience of the position of the gray wolf individual; c 2 r 2 [w 1 X 1 +w 2 X 2 +w 3 X 3 -x m,n (t)]Learning positions for the alpha wolves, the beta wolves and the delta wolves for the gray wolves; w (w) 1 、w 2 、w 3 All are inertia weight coefficients; r is (r) 2 Is [0,1]A random vector between the two;
the process of performing optimization training is as follows:
(1) firstly, initializing parameters a, A and C in a software program; selecting parameter n=30, maximum iteration number C max Initial inertial weight w=1.5, initializing individual gray wolves positions Xi, and sequentially arranging individual gray wolves fitness;
(2) defining alpha, beta and delta as the gray wolves with the first three fitness values, wherein the positions corresponding to the alpha, beta and delta of the gray wolves are X alpha, X beta and X delta respectively;
(3) updating the values of a, A and C according to the wolf group surrounding formulas, namely formula (11) and formula (12);
(4) updating the position of the individual wolves according to the formulas (2) to (8);
(5) updating the speed V and the position X of the individual gray wolves according to the formulas (11) to (12);
if the iteration times are not reached, returning to the step (4) to continue to execute the operation; if the iteration times are reached, the gray wolf is considered to find the optimal position;
and establishing a heat dissipation system temperature prediction model according to the optimized parameters.
10. The method according to claim 5, wherein in step 4, the DSP data output module sends real-time data to the prediction model, and outputs the predicted temperature to the ARM processor after the calculation of the processor, so as to control the temperature of the heat dissipation system, specifically as follows:
and (3) the DSP data output module sends the real-time data into the heat dissipation system temperature prediction model calculated in the step (3), outputs a predicted temperature value, sends the predicted temperature value to the ARM processor, and the ARM calculates a control strategy to control the temperature of the heat dissipation system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310448563.6A CN116501104A (en) | 2023-04-24 | 2023-04-24 | Temperature control device and method for liquid cooling heat dissipation system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310448563.6A CN116501104A (en) | 2023-04-24 | 2023-04-24 | Temperature control device and method for liquid cooling heat dissipation system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116501104A true CN116501104A (en) | 2023-07-28 |
Family
ID=87325972
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310448563.6A Pending CN116501104A (en) | 2023-04-24 | 2023-04-24 | Temperature control device and method for liquid cooling heat dissipation system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116501104A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117156828A (en) * | 2023-10-31 | 2023-12-01 | 联通(广东)产业互联网有限公司 | Data center heat dissipation system and method based on liquid cooling |
| CN117369604A (en) * | 2023-12-05 | 2024-01-09 | 苏州元脑智能科技有限公司 | Flow control method and device for server liquid cooling system |
| CN118111098A (en) * | 2023-12-01 | 2024-05-31 | 翰沃思(浙江)流体技术有限公司 | Construction method of optimized simulation model of fresh air conditioner heat recovery control system |
-
2023
- 2023-04-24 CN CN202310448563.6A patent/CN116501104A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117156828A (en) * | 2023-10-31 | 2023-12-01 | 联通(广东)产业互联网有限公司 | Data center heat dissipation system and method based on liquid cooling |
| CN117156828B (en) * | 2023-10-31 | 2024-02-02 | 联通(广东)产业互联网有限公司 | Data center heat dissipation system and method based on liquid cooling |
| CN118111098A (en) * | 2023-12-01 | 2024-05-31 | 翰沃思(浙江)流体技术有限公司 | Construction method of optimized simulation model of fresh air conditioner heat recovery control system |
| CN117369604A (en) * | 2023-12-05 | 2024-01-09 | 苏州元脑智能科技有限公司 | Flow control method and device for server liquid cooling system |
| CN117369604B (en) * | 2023-12-05 | 2024-02-20 | 苏州元脑智能科技有限公司 | Flow control method and device for server liquid cooling system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116501104A (en) | Temperature control device and method for liquid cooling heat dissipation system | |
| Li et al. | Thermal error modeling of machine tool spindle based on the improved algorithm optimized BP neural network | |
| CN111795484B (en) | Intelligent air conditioner control method and system | |
| CN111126605B (en) | Data center machine room control method and device based on reinforcement learning algorithm | |
| CN110410942A (en) | A kind of Cooling and Heat Source machine room energy-saving optimal control method and system | |
| CN109375994A (en) | Data center's task temperature prediction and dispatching method based on RBF neural | |
| CN114202111A (en) | Electronic expansion valve flow characteristic prediction based on particle swarm optimization BP neural network | |
| EP2965019A1 (en) | Energy management systems and methods of use | |
| CN113825356B (en) | Energy-saving control method and device for cold source system, electronic equipment and storage medium | |
| CN112379589B (en) | Worm wheel machining shape controllable process | |
| CN111582604A (en) | Data processing method and device, electronic device, and medium | |
| CN114625229A (en) | Method and device for optimizing regulation and control of immersion liquid cooling heat dissipation and data center | |
| CN116989442A (en) | Central air conditioner load prediction method and system | |
| JP2022105276A (en) | Power management method and device | |
| CN113885609A (en) | Box body temperature control method and device of vehicle-mounted refrigerator and vehicle-mounted refrigerator | |
| CN116205148A (en) | Data center cooling load prediction method based on Wide & Deep model | |
| CN117366799A (en) | Control method of building cold source system | |
| CN116431439A (en) | Energy consumption monitoring and management system for green data center | |
| CN113503749B (en) | Intelligent water temperature control method for indirect air cooling system | |
| CN114679885A (en) | Cold plate liquid cooling intelligent heat dissipation regulation and control system, regulation and control optimization method and device | |
| CN116187196A (en) | Neural network-based full-process working condition turbine disk cavity temperature estimation and regulation method | |
| CN116017936A (en) | Control method and device for air conditioner room, electronic equipment and storage medium | |
| CN115630331A (en) | Air-conditioning refrigeration station global energy-saving optimization and regulation and control method based on mathematical physical model | |
| CN111637614B (en) | Intelligent control method for data center active ventilation floor | |
| CN115775038A (en) | Short-term load prediction method based on IGWO optimization LSSVM |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |