A temperature prediction method for the host machine in the cryogenic liquid helium temperature zone of a tokamak

Abstract

The present invention relates to a temperature prediction method applicable to a tokamak cryogenic liquid helium temperature zone host, comprising the following steps: establishing a data set, performing normalization processing, and obtaining a pre-processed training set and test set; determining the structure of a BP neural network model; training the BP neural network model; optimizing the trained BP neural network model using a particle swarm algorithm; substituting the optimized weight threshold into a sparrow algorithm and calculating the fitness function f of the sparrow algorithm; using the sparrow population corresponding to the optimal value of the fitness function f as the optimal number of hidden layer nodes and weight threshold of the BP neural network model; training the BP neural network model using the optimal number of hidden layer nodes and weight threshold, and outputting a prediction result. The present invention optimizes the weights and thresholds of the BP neural network by hybridly utilizing the sparrow algorithm and the particle swarm algorithm, thereby improving prediction accuracy and stability in low-temperature environments and enabling simple and effective prediction of ultra-low-temperature environments.

Description

Technical Field

[0001] The present invention relates to the field of ultra-low temperature measurement technology, in particular to a temperature prediction method applicable to a host machine in a tokamak cryogenic liquid helium temperature zone. Background Art

[0002] With the rapid development of scientific exploration, fundamental physics research, aerospace technology, energy and chemical engineering, and other fields, the application of cryogenic technology is becoming increasingly widespread, placing higher demands and technical requirements on research in the field of cryogenic measurement. Research on related technologies is of great scientific and engineering significance. Temperature monitoring is crucial for maintaining the safety of the ultra-cold liquid helium temperature zone in nuclear fusion devices. In superconducting tokamaks, temperature is a key monitoring parameter for device operation safety, providing real-time information on the system status of the main engine in cryogenic environments. Regular monitoring of the operating status of large tokamaks requires multiple temperature measurement points to ensure an accuracy within 0.1K. However, the varying operating temperature ranges of various components in the main engine complicate precise measurement of temperature changes at ultra-low temperatures. In such environments, temperature changes are often difficult to accurately measure due to internal system influences, and even pose safety risks when the temperature falls below a critical value. To accurately monitor the operating status of the main engine in cryogenic environments and provide timely alarms when the device reaches critical temperatures, the challenge of accurately predicting temperature in ultra-low temperature environments urgently needs to be addressed.

[0003] Temperature monitoring in ultra-low temperature environments still suffers from low temperature accuracy. This is primarily due to the temperature measurement system being easily affected by factors such as external temperature, ambient air temperature, internal device temperature, and helium concentration, which can lead to errors and large fluctuations in measured temperature. Low-temperature measurements require cryogenic sensors, whose resistance changes with ambient temperature. By collecting this resistance signal, the current ambient temperature can be monitored. When measuring in ultra-low temperature environments, the current ambient temperature is monitored by collecting the resistance signal from the temperature sensor. Therefore, predicting the relationship between the measurement system's input resistance and output temperature is crucial for improving the system's temperature measurement accuracy and for monitoring the operating status of the tokamak's cryogenic liquid helium temperature zone mainframe.

[0004] Currently, when measuring temperature in the low-temperature liquid helium temperature zone of a tokamak, the existing method is to use the least squares method to predict the temperature value. This method has poor global search effect, is prone to obtaining local optimal solutions, and will produce ill-conditioned matrices that make the equation lose its validity. Summary of the Invention

[0005] In order to solve the problem of difficulty in accurately measuring temperature in ultra-low temperature environments, the purpose of the present invention is to provide a temperature prediction method suitable for the tokamak low-temperature liquid helium temperature zone host, which can fit the nonlinear relationship between sensor resistance and low-temperature temperature with high accuracy and simply and effectively predict temperature changes in ultra-low temperature environments.

[0006] To achieve the above object, the present invention adopts the following technical solution: a temperature prediction method applicable to a tokamak cryogenic liquid helium temperature zone host, the method comprising the following steps in sequence:

[0007] (1) Using the collected temperature resistance data to establish a data set, randomly selecting elements from the data set to construct a training set and a test set, and then normalizing each element in the training set and the test set in turn to obtain the preprocessed training set and test set;

[0008] (2) Determine the structure of the BP neural network model: set the upper limit of the number of hidden layer nodes to 20 and the lower limit to 1; set the network training parameters and initialize the population; randomly generate the initial number of hidden layer nodes in the sparrow algorithm and determine the current number of hidden layer nodes;

[0009] (3) Using the preprocessed training set to train the BP neural network model to obtain a trained BP neural network model;

[0010] (4) Optimize the trained BP neural network model through the particle swarm algorithm: determine the particle dimension of the particle swarm algorithm, generate the initial particle swarm, randomly generate the velocity and position vectors of the particle swarm and perform iterative updates to search for the global optimal solution; calculate the fitness function F of the particle swarm algorithm to determine whether the number of iterations of the particle swarm algorithm is met. If so, output the optimized weight threshold; otherwise, continue iterative updates;

[0011] (5) Substitute the optimized weight threshold into the Sparrow algorithm, calculate the fitness function f of the Sparrow algorithm corresponding to the BP neural network model at this time, and use the fitness function f to search the number of hidden layer nodes generated by the Sparrow algorithm each time and the optimized weight threshold obtained by the particle swarm algorithm optimization;

[0012] (6) Return to step (2) to generate a new number of hidden layer nodes again, calculate the fitness function f through the sparrow search algorithm, iteratively update the positions of the discoverer, follower, and observer until the preset model training stop condition, i.e., the maximum number of iterations of the sparrow algorithm, is reached. The optimal position obtained at this time is used to determine the optimal individual of the current population, and the sparrow population corresponding to the optimal value of the fitness function f is used as the optimal number of hidden layer nodes and weight threshold of the BP neural network model;

[0013] (7) Use the optimal number of hidden layer nodes and weight threshold to train the BP neural network model and output the prediction results.

[0014] The step (1) specifically includes the following steps:

[0015] (1a) 420 sets of experimental data collected by the temperature measurement system were used as the data set, 315 sets of data were selected as the training set, and 105 sets of data were selected as the test set;

[0016] (1b) Normalize the training set and test set data. The normalization formula is:

[0017]

[0018] In the formula, y and y * Represent the data before and after normalization respectively; y max with y min represent the maximum and minimum values ​​before normalization.

[0019] The step (2) specifically refers to: establishing the topological structure of the BP neural network model, the learning rate of the BP algorithm is 0.01, the number of training times is 1000, and the expected error is 10 -6 , the input is resistance, the output is temperature, and when looking for the optimal number of nodes in a single hidden layer, the lower limit is set to 1 and the upper limit is set to 20. The number of nodes in a single hidden layer is determined by the sparrow algorithm; in the particle swarm algorithm, the population size is set to 30, the inertia weight is 0.7, and both learning factors are 2; in the sparrow algorithm, the population size is set to 30, the warning value is 0.8, and the proportion of discoverers accounts for 20% of the population.

[0020] The step (4) specifically includes the following steps:

[0021] (4a) At the beginning of the particle swarm algorithm, an initial particle swarm is randomly generated, i.e., the initial weight and threshold. The fitness value is used to measure the degree to which the particles find the optimal solution during the training process. The formula of the fitness function F is:

[0022]

[0023] Where N is the total number of samples, T i is the actual temperature value, y i is the expected output value;

[0024] (4b) Seek the global optimal solution through iterative operation, determine the individual extreme value data at the first position pbest in the particle swarm algorithm, and the global extreme value data at the second position gbest; calculate the individual fitness function, update the speed and position of the particle swarm, and after the iteration is completed, assign the position vector of the optimal particle to the weight and threshold of the BP neural network, that is, the optimized weight threshold, speed and position update equations are:

[0025]

[0026]

[0027] Where w is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers between (0, 1); is the velocity of particle i in the k-th iteration process, that is, the optimized weight; is the velocity of particle i in the (k + 1)-th iteration process, is the position of particle i in the k-th iteration process, that is, the optimized threshold; is the position of particle i in the (k + 1)-th iteration process, pbest i is the individual optimal position, gbest i is the global optimal position; and constitute the optimized weight threshold.

[0028] In step (5), the formula of the fitness function f of the sparrow algorithm is:

[0029]

[0030] Where N is the total number of samples, T i is the actual temperature value, y i is the expected output value, is the mean value of the actual temperature value.

[0031] In step (6), the specific steps for iteratively updating the positions of the discoverer, follower, and observer are as follows:

[0032] (6a) According to the sparrow search principle, the discoverer position update formula is:

[0033]

[0034] Where represents the coordinate position of the i-th sparrow optimized in the j-dimensional space at the t-th time; t represents the current iteration number; a is a uniform random number between (0, 1]; Q is a random number subject to a normal distribution; L is the identity matrix; r2 represents the alarm value within the interval [0, 1]; st is the safety threshold within the interval [0.5, 1]; when r2 < st, it means the surrounding is safe, and the discoverer will explore a larger area; when r2 ≥ st, it means danger appears, and the sparrow will quickly transfer its position;

[0035] (6b) The follower position update formula is:

[0036]

[0037] Where x worstis the current global worst position; Indicates the best position that sparrows search for during the t+1th optimization; A is a matrix whose elements are randomly assigned 1 or -1; n is the population size; when When , it indicates that the fitness of the i-th follower is poor and it needs to go to other areas to search for food;

[0038] (6c) The observer position update formula is:

[0039]

[0040] Where β is the step-size control random parameter that obeys the 0-1 normal distribution; is the current global optimal position; k represents the moving direction of the sparrow and is a random number between [-1,1]; f i is the fitness value of the current individual; f w and f g Represent the global worst and best fitness values ​​of the current individual respectively; ε is to avoid f i =f g When the denominator is 0, a very small constant is introduced.

[0041] It can be seen from the above technical solution that the beneficial effects of the present invention are: First, by combining the secondary optimization of the sparrow algorithm and the particle swarm algorithm, the BP neural network model most suitable for temperature prediction in an ultra-low temperature environment is determined, the optimal weights and thresholds are determined, and the BP neural network model is avoided from converging prematurely and falling into the local optimum, thereby improving the accuracy and stability of the temperature measurement system in predicting the temperature, which is of great significance for promoting the widespread application of the BP neural network model in predicting temperature in an ultra-low temperature environment; Second, the process of the present invention is simple and easy to operate, and the nonlinear relationship between the sensor resistance and the low temperature can be fitted with high accuracy, and the temperature change in the ultra-low temperature environment can be predicted simply and effectively, providing a method for monitoring the temperature change of the low-temperature liquid helium temperature zone in the large tokamak device; Third, the present invention provides technical references for data fitting, temperature monitoring and other aspects of the low-temperature measurement system, and expands research ideas for combining multiple algorithms to optimize the BP neural network model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of the method of the present invention;

[0043] Figure 2 This is a circuit block diagram of the temperature measurement system of the present invention;

[0044] Figure 3 This is a comparison chart of sample data predicted by the present invention and the existing algorithm. DETAILED DESCRIPTION

[0045] like Figure 1As shown, a temperature prediction method applicable to a host in a low-temperature liquid helium temperature zone of a tokamak comprises the following steps in sequence:

[0046] (1) Using the collected temperature resistance data to establish a data set, randomly selecting elements from the data set to construct a training set and a test set, and then normalizing each element in the training set and the test set in turn to obtain the preprocessed training set and test set;

[0047] (2) Determine the structure of the BP neural network model: set the upper limit of the number of hidden layer nodes to 20 and the lower limit to 1; set the network training parameters and initialize the population; randomly generate the initial number of hidden layer nodes in the sparrow algorithm and determine the current number of hidden layer nodes;

[0048] (3) Using the preprocessed training set to train the BP neural network model to obtain a trained BP neural network model;

[0049] (4) Optimize the trained BP neural network model through the particle swarm algorithm: determine the particle dimension of the particle swarm algorithm, generate the initial particle swarm, randomly generate the velocity and position vectors of the particle swarm and perform iterative updates to search for the global optimal solution; calculate the fitness function F of the particle swarm algorithm to determine whether the number of iterations of the particle swarm algorithm is met. If so, output the optimized weight threshold; otherwise, continue iterative updates;

[0050] (5) Substitute the optimized weight threshold into the Sparrow algorithm, calculate the fitness function f of the Sparrow algorithm corresponding to the BP neural network model at this time, and use the fitness function f to search the number of hidden layer nodes generated by the Sparrow algorithm each time and the optimized weight threshold obtained by the particle swarm algorithm optimization;

[0051] (6) Return to step (2) to generate a new number of hidden layer nodes again, calculate the fitness function f through the sparrow search algorithm, iteratively update the positions of the discoverer, follower, and observer until the preset model training stop condition, i.e., the maximum number of iterations of the sparrow algorithm, is reached. The optimal position obtained at this time is used to determine the optimal individual of the current population, and the sparrow population corresponding to the optimal value of the fitness function f is used as the optimal number of hidden layer nodes and weight threshold of the BP neural network model;

[0052] (7) Use the optimal number of hidden layer nodes and weight threshold to train the BP neural network model and output the prediction results.

[0053] Figure 1 The SSA in it is the sparrow algorithm.

[0054] The step (1) specifically includes the following steps:

[0055] (1a) 420 sets of experimental data collected by the temperature measurement system were used as the data set, 315 sets of data were selected as the training set, and 105 sets of data were selected as the test set;

[0056] (1b) Normalize the training set and test set data. The normalization formula is:

[0057]

[0058] In the formula, y and y * Represent the data before and after normalization respectively; y max with y min represent the maximum and minimum values ​​before normalization.

[0059] The step (2) specifically refers to: establishing the topological structure of the BP neural network model, the learning rate of the BP algorithm is 0.01, the number of training times is 1000, and the expected error is 10 -6 , the input is resistance, the output is temperature, and when looking for the optimal number of nodes in a single hidden layer, the lower limit is set to 1 and the upper limit is set to 20. The number of nodes in a single hidden layer is determined by the sparrow algorithm; in the particle swarm algorithm, the population size is set to 30, the inertia weight is 0.7, and both learning factors are 2; in the sparrow algorithm, the population size is set to 30, the warning value is 0.8, and the proportion of discoverers accounts for 20% of the population.

[0060] The step (4) specifically includes the following steps:

[0061] (4a) At the beginning of the particle swarm algorithm, an initial particle swarm is randomly generated, i.e., the initial weight and threshold. The fitness value is used to measure the degree to which the particles find the optimal solution during the training process. The formula of the fitness function F is:

[0062]

[0063] Where N is the total number of samples, T i is the actual temperature value, y i is the expected output value;

[0064] (4b) Seek the global optimal solution through iterative operation, determine the individual extreme value data at the first position pbest in the particle swarm algorithm, and the global extreme value data at the second position gbest; calculate the individual fitness function, update the speed and position of the particle swarm, and after the iteration is completed, assign the position vector of the optimal particle to the weight and threshold of the BP neural network, that is, the optimized weight threshold, speed and position update equations are:

[0065]

[0066]

[0067] Where, w is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers between (0, 1); is the velocity of particle i in the k-th iteration process, that is, the optimized weight; is the velocity of particle i in the (k + 1)-th iteration process, is the position of particle i in the k-th iteration process, that is, the optimized threshold; is the position of particle i in the (k + 1)-th iteration process, pbest i is the individual optimal position, gbest i is the global optimal position; and constitute the optimized weight threshold.

[0068] In step (5), the formula of the fitness function f of the sparrow algorithm is:

[0069]

[0070] Where, N is the total number of samples, T i is the actual temperature value, y i is the expected output value, is the mean value of the actual temperature value.

[0071] In step (6), the specific steps for iteratively updating the positions of the discoverer, follower, and observer are as follows:

[0072] (6a) According to the sparrow search principle, the formula for updating the position of the discoverer is: <0000​​​​​​​​​​​​​​​​​​Indicates the best position that sparrows search for during the t+1th optimization; A is a matrix whose elements are randomly assigned 1 or -1; n is the population size; when When , it indicates that the fitness of the i-th follower is poor and it needs to go to other areas to search for food;

[0078] (6c) The observer position update formula is:

[0079]

[0080] Where β is the step-size control random parameter that obeys the 0-1 normal distribution; is the current global optimal position; k represents the moving direction of the sparrow and is a random number between [-1,1]; f i is the fitness value of the current individual; f w and f g Represent the global worst and best fitness values ​​of the current individual respectively; ε is to avoid f i =f g When the denominator is 0, a very small constant is introduced.

[0081] like Figure 2 As shown in the figure, during actual measurement, a 220V AC voltage source is connected, which is converted into a 24V constant current source through an AC / DC converter. The temperature sensor connected to the ultra-low temperature environment in the signal processing circuit is excited by the constant current source. The output analog voltage is amplified, filtered, and A / D converted, and then output as a digital signal. The signal is transmitted to the MCU controller, and the resistance signal is transmitted to the host computer through host computer communication, and the host computer obtains the predicted temperature value.

[0082] like Figure 3 As shown in the figure, the prediction results of the BP algorithm before and after optimization are compared with the actual values ​​of the sample data when T < 10 K. Analysis of the temperature values ​​predicted by different algorithms shows that when the ambient temperature is low, T < 10 K, the deviation between the predicted temperature and the actual temperature is large for the BP neural network algorithm and the PSO-BP neural network algorithm, and the fitting effect of the present invention is better. Figure 3 Where T is temperature and K is the unit of Fahrenheit.

[0083] In summary, the present invention optimizes the weights and thresholds of the BP neural network by adopting a hybrid of the sparrow algorithm and the particle swarm algorithm, so that the optimized BP neural network can achieve better training results during training, thereby improving the prediction accuracy and stability in low-temperature environments. The process of the present invention is simple and easy to operate. It can fit the nonlinear relationship between sensor resistance and low-temperature temperature with high accuracy, and can easily and effectively predict temperature changes in ultra-low temperature environments, providing a method for monitoring temperature changes in the low-temperature liquid helium temperature zone in large tokamak devices. The present invention greatly reduces the probability of the network falling into a local minimum, avoiding the situation in which the traditional method of determining the number of hidden layer nodes is time-consuming and prone to overfitting.

Claims

1. A temperature prediction method applicable to a tokamak cryogenic liquid helium temperature zone host, characterized by: The method comprises the following steps in sequence: (1) Using the collected temperature resistance data to establish a data set, randomly selecting elements from the data set to construct a training set and a test set, and then normalizing each element in the training set and the test set in turn to obtain the preprocessed training set and test set; (2) Determine the structure of the BP neural network model: set the upper limit of the number of hidden layer nodes to 20 and the lower limit to 1; Set network training parameters and initialize the population; randomly generate the initial number of hidden layer nodes in the sparrow algorithm and determine the current number of hidden layer nodes; (3) Using the preprocessed training set to train the BP neural network model to obtain a trained BP neural network model; (4) Optimize the trained BP neural network model through the particle swarm algorithm: determine the particle dimension of the particle swarm algorithm, generate the initial particle swarm, randomly generate the velocity and position vectors of the particle swarm and perform iterative updates to search for the global optimal solution; calculate the fitness function F of the particle swarm algorithm to determine whether the number of iterations of the particle swarm algorithm is met. If so, output the optimized weight threshold; otherwise, continue iterative updates; (5) Substitute the optimized weight threshold into the Sparrow algorithm, calculate the fitness function f of the Sparrow algorithm corresponding to the BP neural network model at this time, and use the fitness function f to search the number of hidden layer nodes generated by the Sparrow algorithm each time and the optimized weight threshold obtained by the particle swarm algorithm optimization; (6) Return to step (2) to generate a new number of hidden layer nodes again, calculate the fitness function f through the sparrow search algorithm, iteratively update the positions of the discoverer, follower, and observer until the preset model training stop condition, i.e., the maximum number of iterations of the sparrow algorithm, is reached. The optimal position obtained at this time is used to determine the optimal individual of the current population, and the sparrow population corresponding to the optimal value of the fitness function f is used as the optimal number of hidden layer nodes and weight threshold of the BP neural network model; (7) Use the optimal number of hidden layer nodes and weight threshold to train the BP neural network model and output the prediction results.

2. The temperature prediction method for a tokamak cryogenic liquid helium temperature zone host according to claim 1, characterized in that: The step (1) specifically includes the following steps: (1a) 420 sets of experimental data collected by the temperature measurement system were used as the data set, 315 sets of data were selected as the training set, and 105 sets of data were selected as the test set; (1b) Normalize the training set and test set data. The normalization formula is: In the formula, y and y * Represent the data before and after normalization respectively; y max with y min represent the maximum and minimum values ​​before normalization.

3. The temperature prediction method for a tokamak cryogenic liquid helium temperature zone host according to claim 1, characterized in that: The step (2) specifically refers to: establishing the topological structure of the BP neural network model, the learning rate of the BP algorithm is 0.01, the number of training times is 1000, and the expected error is 10 -6 , the input is resistance, the output is temperature, and when looking for the optimal number of nodes in a single hidden layer, the lower limit is set to 1 and the upper limit is set to 20. The number of nodes in a single hidden layer is determined by the sparrow algorithm; in the particle swarm algorithm, the population size is set to 30, the inertia weight is 0.7, and both learning factors are 2; in the sparrow algorithm, the population size is set to 30, the warning value is 0.8, and the proportion of discoverers accounts for 20% of the population.

4. The temperature prediction method for a tokamak cryogenic liquid helium temperature zone host according to claim 1, characterized in that: The step (4) specifically includes the following steps: (4a) At the beginning of the particle swarm algorithm, an initial particle swarm is randomly generated, i.e., the initial weight and threshold. The fitness value is used to measure the degree to which the particles find the optimal solution during the training process. The formula of the fitness function F is: Where N is the total number of samples, T i is the actual temperature value, y i is the expected output value; (4b) Seek the global optimal solution through iterative operation, determine the individual extreme value data at the first position pbest in the particle swarm algorithm, and the global extreme value data at the second position gbest; calculate the individual fitness function, update the speed and position of the particle swarm, and after the iteration is completed, assign the position vector of the optimal particle to the weight and threshold of the BP neural network, that is, the optimized weight threshold, speed and position update equations are: Where w is the inertia weight, c1 and c2 are learning factors, and r1 and r2 are random numbers between (0, 1); is the velocity of particle i during the k-th iteration, i.e., the optimized weight; is the velocity of particle i during the k+1th iteration, is the position of particle i during the k-th iteration, that is, the optimized threshold; is the position of particle i during the k+1th iteration, pbest i is the individual optimal position, gbest i is the optimal position of the group; and The weight threshold after composition optimization.

5. The temperature prediction method for a tokamak cryogenic liquid helium temperature zone host according to claim 1, characterized in that: In step (5), the fitness function f of the sparrow algorithm is expressed as follows: Where N is the total number of samples, T i is the actual temperature value, y i is the expected output value, is the mean of the actual temperature values.

6. The temperature prediction method for a tokamak cryogenic liquid helium temperature zone host according to claim 1, characterized in that: In step (6), the iterative updating of the positions of the discoverer, follower, and observer specifically includes the following steps: (6a) According to the sparrow search principle, the formula for updating the discoverer's position is: In the formula, represents the coordinate position of the i-th sparrow optimized in the j-dimensional space at the t-th time; t represents the current iteration number; a is a uniform random number between (0, 1]; Q is a random number subject to a normal distribution; L is an identity matrix; r2 represents the alarm value within the interval [0, 1]; st is the safety threshold within the interval [0.5, 1]; when r2 < st, it indicates that the surrounding is safe, and the discoverer will explore a larger area; when r2 ≥ st, it indicates that danger appears, and the sparrow quickly transfers its position; (6b) The follower position update formula is: Where x worst is the current global worst position; Indicates the best position that sparrows search for during the t+1th optimization; A is a matrix whose elements are randomly assigned 1 or -1; n is the population size; when When , it indicates that the fitness of the i-th follower is poor and it needs to go to other areas to search for food; (6c) The observer position update formula is: Where β is the step-size control random parameter that obeys the 0-1 normal distribution; is the current global optimal position; k represents the moving direction of the sparrow and is a random number between [-1,1]; f i is the fitness value of the current individual; f w and f g Represent the global worst and best fitness values ​​of the current individual respectively; ε is to avoid f i =f g When the denominator is 0, a very small constant is introduced.

Images