Optical system temperature field acquisition method and system

Abstract

The invention discloses an optical system temperature field acquisition method and system, and the method comprises the steps: obtaining the temperature data of a preset position of an optical system in a temperature changing process, obtaining a current time period temperature sequence, and enabling the current time period temperature sequence to be formed through the time arrangement of the temperature data of the preset position in the current time period along with the process; the current time period temperature sequence is input into a temperature prediction model, a next time period temperature sequence is output, and the next time period temperature sequence is formed by conducting time arrangement on predicted temperature data of a preset position in a next time period along with the process; and obtaining the temperature field of the optical system in the latter time period according to the temperature sequence of the latter time period. The step of obtaining the temperature prediction model comprises obtaining temperature data of a preset position of the optical system in the temperature changing process, the step of obtaining the training data comprises an input temperature sequence and an output temperature sequence, and the training data is used for training to obtain the temperature prediction model. The temperature field of the optical system can be obtained through prediction, so that the temperature condition of the optical system can be monitored.

Description

Technical Field

[0001] The present invention relates to the field of optical systems, and in particular to a method and system for acquiring a temperature field of an optical system. Background Art

[0002] The design and research of infrared optical systems in low-temperature environments are becoming increasingly important in the fields of space exploration, astronomical observation, etc. Whether in space or on the ground, multi-physical field load impacts such as temperature, vibration, gravity, etc. will affect the optical performance of the optical system. Among them, temperature is the biggest external factor affecting low-temperature infrared optical systems, mainly due to large deformation and thermal stress between optical elements and mechanical structures due to temperature gradients, which may cause potential damage. Therefore, how to predict and monitor the temperature changes of optical systems in real time, and predict and analyze the various performances of optical systems, is the key to improving the efficiency and life of optical systems. Summary of the invention

[0003] Based on the above, the purpose of the present invention is to provide a method and system for acquiring the temperature field of an optical system, which can predict the temperature field of the optical system, so that the temperature condition of the optical system can be monitored, which has a reference significance for predicting and analyzing various performances of the optical system.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A method for acquiring a temperature field of an optical system, comprising:

[0006] During the temperature change process, the temperature data of the preset position of the optical system is acquired to obtain a temperature sequence of the current period, wherein the temperature sequence of the current period is formed by arranging the temperature data of the preset position in the current period as the process progresses;

[0007] Input the temperature sequence of the current period into a temperature prediction model, and the temperature prediction model outputs a temperature sequence of a subsequent period, wherein the temperature sequence of the subsequent period is formed by arranging the temperature data of the preset position in the subsequent period as the process progresses, and the subsequent period is the period after the current period;

[0008] According to the temperature sequence of the latter time period, a temperature field of the optical system in the latter time period is obtained, wherein the temperature field of the optical system includes temperature data of the preset position of the optical system and temperature data of other positions of the optical system except the preset position;

[0009] Obtaining the temperature prediction model includes:

[0010] During the temperature change process, the temperature data of the preset position of the optical system is obtained, and training data is obtained based on the obtained temperature data, the training data including an input temperature sequence and an output temperature sequence, the input temperature sequence is formed by arranging the temperature data of the preset position in a previous time period as the process progresses, and the output temperature sequence is formed by arranging the temperature data of the preset position in a later time period as the process progresses, and the constructed temperature prediction model is trained using the training data to obtain the temperature prediction model.

[0011] Optionally, in the temperature changing process, acquiring temperature data of the preset position of the optical system, and acquiring training data according to the acquired temperature data comprises:

[0012] During the cooling process, obtaining temperature data of the preset position of the optical system, and obtaining the training data according to the obtained temperature data;

[0013] Alternatively, in the process of returning to a temperature after cooling down, the temperature data of the preset position of the optical system is obtained, and the training data is obtained according to the obtained temperature data;

[0014] Alternatively, during the cooling process and during the warming process after cooling, the temperature data of the preset position of the optical system is respectively acquired, and the training data is acquired according to the acquired temperature data.

[0015] Optionally, obtaining training data according to the obtained temperature data includes:

[0016] The obtained temperature data is processed into the input temperature sequence and the output temperature sequence through the time window. The input temperature sequence is expressed as , the output temperature sequence is expressed as , represents a time window of length m starting from time t, express Then the time window of length n, Represents X t is a vector formed by the temperature data obtained for i preset positions.

[0017] Optionally, using the training data to train the constructed temperature prediction model includes:

[0018] Establishing a loss function, the loss function describes the loss of the predicted value of the output temperature sequence relative to the true value, and derivatizing the loss function with respect to the temperature prediction model parameters to obtain a gradient function of the loss function derivatized with respect to the temperature prediction model parameters;

[0019] In any iteration, the current temperature prediction model parameter value is substituted into the gradient function to obtain the current gradient value, and the first-order moment estimate is updated according to the current first-order moment estimate and the current gradient value, the second-order moment estimate is updated according to the current second-order moment estimate and the current gradient value, and the temperature prediction model parameter value is updated according to the updated first-order moment estimate and the updated second-order moment estimate, and the next iteration is performed with the updated temperature prediction model parameter value;

[0020] When an iteration stop condition is met, the iteration is stopped, and the temperature prediction model is obtained according to the temperature prediction model parameter values ​​obtained in the last iteration;

[0021] Among them, the first-order moment estimate represents the estimate of the first-order moment of the gradient of the loss function derived from the temperature prediction model parameters, and the second-order moment estimate represents the estimate of the second-order moment of the gradient of the loss function derived from the temperature prediction model parameters.

[0022] Optionally, updating the first-order moment estimate according to the current first-order moment estimate and the current gradient value includes:

[0023] Obtain a first product of a first decay rate and the current first-order moment estimate, obtain a difference between 1 and the first decay rate, and obtain a second product of the difference and the current gradient value, and use the sum of the first product and the second product as an updated first-order moment estimate;

[0024] Updating the second-order moment estimate based on the current second-order moment estimate and the current gradient value includes:

[0025] Obtain a third product of the second attenuation rate and the current second-order moment estimate, obtain a difference between 1 and the second attenuation rate, and obtain a fourth product of the square of the current gradient value, and use the sum of the third product and the fourth product as the updated second-order moment estimate.

[0026] Optionally, updating the temperature prediction model parameter value according to the updated first-order moment estimate and the updated second-order moment estimate includes:

[0027] The updated first-order moment estimate is corrected for deviation, and the updated second-order moment estimate is corrected for deviation, and the temperature prediction model parameter value is updated according to the corrected first-order moment estimate and the corrected second-order moment estimate, wherein the ratio of the updated first-order moment estimate to the difference between 1 and the first attenuation factor is the corrected first-order moment estimate, and the ratio of the updated second-order moment estimate to the difference between 1 and the second attenuation factor is the corrected second-order moment estimate.

[0028] Optionally, using the training data to train the constructed temperature prediction model includes:

[0029] A loss function is established and iterative operations are performed based on the loss function to obtain temperature prediction model parameter values ​​that meet conditions, so as to obtain the temperature prediction model determined by the temperature prediction model parameter values ​​that meet the conditions, wherein the loss function expresses the average of multiple powers of the difference between the true value and the predicted value of the output temperature sequence of the training data.

[0030] Optionally, the temperature prediction model includes an encoder and a decoder;

[0031] The encoder includes a plurality of layers in parallel, the layers including a first network and a second network, the first network is used to learn the dependency relationship between the input temperature sequence and the temperature data in the output temperature sequence, and generate a representation expressing the correlation of the temperature data at the previous and next moments, and the second network is used to perform a nonlinear transformation on the representation generated by the first network;

[0032] The decoder includes an internal memory unit, an input gate, a forget gate and an output gate. The internal memory unit is used to allow sequence information to be transmitted along the internal memory unit. The forget gate is used to determine whether the sequence information transmitted along the internal memory unit is discarded. The input gate is used to determine whether the sequence information is stored in the internal memory unit. The output gate is used to determine whether the sequence information in the internal memory unit is output.

[0033] Optionally, it also includes:

[0034] The temperature sequence of the next period is input into the temperature prediction model, and the temperature prediction model outputs the temperature sequence of the period after the next period, wherein the temperature sequence of the period after the next period is formed by arranging the predicted temperature data of the preset position in the period after the next period as the process progresses;

[0035] According to the temperature sequence of a time period after the latter time period, the temperature field of the optical system in a time period after the latter time period is obtained.

[0036] An optical system temperature field acquisition system, comprising:

[0037] Memory for storing computer programs;

[0038] A processor is used to implement the steps of the method for acquiring the temperature field of an optical system as described in any one of the above items when executing the computer program.

[0039] It can be seen from the above technical scheme that the optical system temperature field acquisition method and system provided by the present invention include: in the temperature change process, obtaining the temperature data of the preset position of the optical system, obtaining the temperature sequence of the current period, the temperature sequence of the current period is formed by the temperature data of the preset position in the current period being arranged at the same time as the process progresses; inputting the temperature sequence of the current period into the temperature prediction model, and the temperature prediction model outputs the temperature sequence of the next period, the temperature sequence of the next period is formed by the predicted temperature data of the preset position in the next period being arranged at the same time as the process progresses, and the next period is the period after the current period; according to the temperature sequence of the next period, the temperature field of the optical system in the next period is obtained. Obtaining the temperature prediction model includes: in the temperature change process, obtaining the temperature data of the preset position of the optical system, obtaining training data according to the obtained temperature data, the training data includes an input temperature sequence and an output temperature sequence, the input temperature sequence is formed by the temperature data of the preset position in the previous period being arranged at the same time as the process progresses, and the output temperature sequence is formed by the temperature data of the preset position in the next period being arranged at the same time as the process progresses, and the constructed temperature prediction model is trained using the training data to obtain the temperature prediction model.

[0040] The optical system temperature field acquisition method and system of the present invention obtains the temperature data of the preset position of the optical system during the temperature change process, and obtains the training data including the input temperature sequence and the output temperature sequence, so as to obtain the temperature prediction model of the optical system, so that according to the temperature sequence of the current period acquired in real time from the optical system, the temperature prediction model can be used to predict the temperature sequence of the optical system in the next period, and further obtain the temperature field of the optical system in the next period, thereby realizing the prediction of the temperature field of the optical system. Therefore, the optical system temperature field acquisition method and system of the present invention can predict the temperature field of the optical system, so that the temperature condition of the optical system can be monitored, which has a reference significance for predicting and analyzing various performances of the optical system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A flow chart of a method for acquiring a temperature field of an optical system provided by an embodiment of the present invention;

[0043] Figure 2-1 A temperature data diagram obtained for an optical system in a method for obtaining a temperature field of an optical system according to an embodiment of the present invention;

[0044] Figure 2-2It is a schematic diagram of processing the temperature data of an optical system through a time window in a method for acquiring a temperature field of an optical system according to an embodiment of the present invention;

[0045] Figure 3 A schematic diagram of a temperature prediction model in a method for acquiring a temperature field of an optical system according to an embodiment of the present invention;

[0046] Figure 4-1 A structural model of an optical system in a method for acquiring a temperature field of an optical system according to an embodiment of the present invention;

[0047] Figure 4-2 The optical system temperature field obtained in the optical system temperature field acquisition method according to an embodiment of the present invention;

[0048] Figure 5 It is the change of the predicted value and the real value of the temperature measurement point of the primary mirror, the secondary mirror, the tertiary mirror and the connection platform under the cooling condition in the optical system temperature field acquisition method according to one embodiment of the present invention;

[0049] Figure 6 The changes in the predicted values ​​and the actual values ​​of the temperature measurement points of the primary mirror, the secondary mirror, the tertiary mirror and the connection platform in the temperature recovery condition in the optical system temperature field acquisition method according to one embodiment of the present invention;

[0050] Figure 7 This is a comparison chart of the mean absolute percentage error and the root mean square error of the temperature prediction model constructed using five networks in one embodiment of the present invention.

[0051] The reference numerals in the drawings of the specification include:

[0052] 100-connection platform, 101-primary mirror, 102-secondary mirror, 103-tertiary mirror. DETAILED DESCRIPTION

[0053] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0054] This embodiment provides a method for obtaining the temperature field of an optical system. Figure 1 , Figure 1 A flow chart of a method for acquiring a temperature field of an optical system provided by an embodiment includes the following steps:

[0055] S11: During the temperature change process, the temperature data of the preset position of the optical system is acquired to obtain a temperature sequence of the current period, wherein the temperature sequence of the current period is formed by arranging the temperature data of the preset position in the current period as the process progresses.

[0056] For any preset position of the optical system, the temperature data of the preset position of the optical system is obtained. The temperature sequence of the current period is formed by arranging the temperature data of the preset position of the optical system at the time of each process in the current period.

[0057] S12: Input the temperature sequence of the current time period into a temperature prediction model, and the temperature prediction model outputs a temperature sequence of a subsequent time period, wherein the temperature sequence of the subsequent time period is formed by arranging the temperature data of the preset position in the subsequent time period as the process progresses, and the subsequent time period is the time period after the current time period.

[0058] The temperature sequence of the next period is formed by the predicted temperature data of the preset position of the optical system at the time of each process in the next period, which is arranged along with the time of the process.

[0059] S13: obtaining a temperature field of the optical system in the latter time period according to the temperature sequence of the latter time period, wherein the temperature field of the optical system includes temperature data of the preset position of the optical system and temperature data of other positions of the optical system except the preset position.

[0060] The temperature sequence of the next period includes the predicted temperature data of the preset position of the optical system. For any process in the next period, the temperature field of the optical system at the time of the process can be obtained according to the temperature data of the preset position of the optical system at the time of the process.

[0061] Correspondingly, obtaining the temperature prediction model includes the following process: in the temperature changing process, obtaining the temperature data of the preset position of the optical system, obtaining training data based on the obtained temperature data, the training data including an input temperature sequence and an output temperature sequence, the input temperature sequence is formed by arranging the temperature data of the preset position in the previous time period as the process progresses, the output temperature sequence is formed by arranging the temperature data of the preset position in the next time period as the process progresses, and the constructed temperature prediction model is trained using the training data to obtain the temperature prediction model.

[0062] As the temperature changing process proceeds, temperature data is collected at a preset position of the optical system, and the temperature data corresponds to the time of the process.

[0063] A set of training data includes an input temperature sequence and an output temperature sequence, wherein the latter period corresponding to the output temperature sequence is the period after the former period corresponding to the input temperature sequence. The input temperature sequence is formed by arranging the temperature data obtained at the preset position of the optical system at each process time in the former period as the process time is progressed. The output temperature sequence is formed by arranging the temperature data obtained at the preset position of the optical system at each process time in the latter period as the process time is progressed.

[0064] The optical system temperature field acquisition method of this embodiment obtains the temperature data of the preset position of the optical system during the temperature change process, and obtains the training data including the input temperature sequence and the output temperature sequence to obtain the temperature prediction model of the optical system, so that according to the current time period temperature sequence obtained in real time from the optical system, the temperature prediction model can be used to predict the temperature sequence of the optical system in the next time period, and further obtain the temperature field of the optical system in the next time period, thereby realizing the prediction of the optical system temperature field. Therefore, the optical system temperature field acquisition method of this embodiment can predict the optical system temperature field, so that the temperature condition of the optical system can be monitored, which has a reference significance for predicting and analyzing various performances of the optical system.

[0065] In some embodiments, the temperature change process is a process in which the temperature changes with a single change trend, that is, the temperature change process is a cooling process, or the temperature change process is a heating process. For some optical systems, it is necessary to perform a low temperature environment test on the optical system, and the test process includes a cooling test process and a warming test process after cooling. In view of this, in some embodiments, the process of obtaining a temperature prediction model may include: in the cooling process, obtaining the temperature data of the preset position of the optical system, and obtaining training data based on the obtained temperature data. In some embodiments, the process of obtaining a temperature prediction model may include: in the warming process after cooling, obtaining the temperature data of the preset position of the optical system, and obtaining training data based on the obtained temperature data.

[0066] In some embodiments, the process of obtaining the temperature prediction model may include: obtaining temperature data of the preset position of the optical system during the cooling process and during the warming process after the cooling process, respectively, and obtaining training data based on the obtained temperature data. The temperature data obtained from the optical system can be divided into cooling conditions and heating conditions, which can be expressed as:

[0067] t≤t(T min ), cooling condition;

[0068] t>t(T min ), reheating condition;

[0069] Among them, t represents the time corresponding to the temperature data, T minIndicates the lowest temperature reached by the optical system.

[0070] In some embodiments, a temperature sensor may be provided at a preset position of the optical system to collect temperature data, for example, the temperature sensor continuously collects temperature data during a cooling test and continuously collects temperature data during a warming test.

[0071] In some embodiments, obtaining training data according to the obtained temperature data includes: processing the obtained temperature data into the input temperature sequence and the output temperature sequence through a time window, wherein the input temperature sequence is represented as , the output temperature sequence is expressed as , represents a time window of length m starting from time t, express Then the time window of length n, Represents X t is a vector formed by temperature data obtained at i preset positions. i is a positive integer greater than or equal to 1.

[0072] In this embodiment, the length m of the input temperature sequence and the length n of the output temperature sequence are not limited, and in practical applications, they can be set according to the requirements for temperature prediction accuracy or the computational efficiency of the temperature prediction model. Figure 2-1 and Figure 2-2 , Figure 2-1 is a temperature data diagram obtained from an optical system in a method for obtaining a temperature field of an optical system according to an embodiment of the present invention, Figure 2-2 This is a schematic diagram of processing the temperature data of an optical system through a time window in a method for acquiring the temperature field of an optical system in an embodiment. As shown in the figure, the temperature data of the optical system includes a cooling condition and a heating condition as time goes by. The time window is slid along the time axis, and the temperature data arranged over time are processed to obtain a group of input temperature sequences and output temperature sequences.

[0073] In some embodiments, training the constructed temperature prediction model using training data includes:

[0074] Establishing a loss function, the loss function describes the loss of the predicted value of the output temperature sequence relative to the true value, and derivatizing the loss function with respect to the temperature prediction model parameters to obtain a gradient function of the loss function derivatized with respect to the temperature prediction model parameters;

[0075] In any iteration, the current temperature prediction model parameter value is substituted into the gradient function to obtain the current gradient value, and the first-order moment estimate is updated according to the current first-order moment estimate and the current gradient value, the second-order moment estimate is updated according to the current second-order moment estimate and the current gradient value, and the temperature prediction model parameter value is updated according to the updated first-order moment estimate and the updated second-order moment estimate, and the next iteration is performed with the updated temperature prediction model parameter value;

[0076] When an iteration stop condition is met, the iteration is stopped, and the temperature prediction model is obtained according to the temperature prediction model parameter values ​​obtained in the last iteration;

[0077] Among them, the first-order moment estimate represents the estimate of the first-order moment of the gradient of the loss function derived from the temperature prediction model parameters, and the second-order moment estimate represents the estimate of the second-order moment of the gradient of the loss function derived from the temperature prediction model parameters.

[0078] If it is the first iteration, the current temperature prediction model parameter value refers to the initial value of the temperature prediction model parameter, the current first-order moment estimate value refers to the initial value of the first-order moment estimate, and the current second-order moment estimate value refers to the initial value of the second-order moment estimate. If it is any iteration after the first iteration, the current temperature prediction model parameter value is the temperature prediction model parameter value updated in the previous iteration, the current first-order moment estimate value refers to the first-order moment estimate value updated in the previous iteration, and the current second-order moment estimate value refers to the second-order moment estimate value updated in the previous iteration. In any iteration, the current temperature prediction model parameter value is substituted into the gradient function, and the value of the gradient function is the current gradient value.

[0079] In any iteration, the first-order moment estimate is updated according to the current first-order moment estimate and the current gradient value, the second-order moment estimate is updated according to the current second-order moment estimate and the current gradient value, and the temperature prediction model parameter value is updated according to the updated first-order moment estimate and the updated second-order moment estimate. In this embodiment, the temperature prediction model parameter value is updated using the first-order moment estimate and the second-order moment estimate of the gradient derived from the loss function of the temperature prediction model parameter, and the temperature prediction model parameters are optimized and updated using adaptive moment estimation. Adaptive learning makes parameter adjustment more flexible and efficient, and can meet the needs of less adjustment of learning rate. This optimization and update method is more suitable for processing large data sets and high-dimensional space, and is suitable for situations where the objective function changes over time or there are many local minima.

[0080] In some embodiments, updating the first-order moment estimate based on the current first-order moment estimate and the current gradient value includes: obtaining a first product of a first decay rate and the current first-order moment estimate, obtaining a difference between 1 and the first decay rate, and a second product with the current gradient value, and taking the sum of the first product and the second product as the updated first-order moment estimate.

[0081] In some embodiments, updating the second-order moment estimate based on the current second-order moment estimate and the current gradient value includes: obtaining a third product of a second attenuation rate and the current second-order moment estimate, obtaining a difference between 1 and the second attenuation rate, and a fourth product of the square of the current gradient value, and taking the sum of the third product and the fourth product as the updated second-order moment estimate.

[0082] In some embodiments, updating the temperature prediction model parameter value according to the updated first-order moment estimate and the updated second-order moment estimate includes: obtaining a first ratio of the updated first-order moment estimate to the square root of the updated second-order moment estimate, obtaining a fifth product of the learning rate and the first ratio, and using the difference between the current temperature prediction model parameter value and the fifth product as the updated temperature prediction model parameter value. Obtaining the ratio of the updated first-order moment estimate to the square root of the updated second-order moment estimate, i.e., the first ratio. Obtaining the product of the learning rate and the first ratio, i.e., the fifth product. Using the difference between the current temperature prediction model parameter value and the fifth product as the updated temperature prediction model parameter value.

[0083] Exemplarily, the process of optimizing and updating the temperature prediction model parameter values ​​can be expressed as:

[0084] ; (1)

[0085] ; (2)

[0086] ; (3)

[0087] ; (4)

[0088] Among them, g t Represents the gradient function of the loss function with respect to the temperature prediction model parameters, θ t Represents the current temperature prediction model parameter value, m t represents the updated first-order moment estimate, m t-1 represents the current first-order moment estimate, β1 represents the first decay rate, that is, the decay rate of the first-order moment estimate, v t represents the updated second-order moment estimate, v t-1 represents the current second-order moment estimate, β2 represents the second decay rate, that is, the decay rate of the second-order moment estimate, θt+1 represents the updated temperature prediction model parameter value, and η represents the learning rate. ε is used to avoid division by zero errors in calculations, and ε is usually set to a very small value, such as ε≤1e-5, such as ε=1e-8. The above formulas (1)-(4) can be considered as an adaptive moment estimation optimizer, which is used to optimize and update the temperature prediction model parameters.

[0089] m t or m t-1 It can represent the first-order moment estimation of the gradient of the loss function for the temperature prediction model parameters in momentum form. t or v t-1 It can represent the second-order moment estimation of the gradient of the loss function with respect to the parameters of the temperature prediction model in the form of momentum.

[0090] In some embodiments, updating the temperature prediction model parameter value according to the updated first-order moment estimate and the updated second-order moment estimate includes: performing bias correction on the updated first-order moment estimate, performing bias correction on the updated second-order moment estimate, and updating the temperature prediction model parameter value according to the corrected first-order moment estimate and the corrected second-order moment estimate, wherein the ratio of the updated first-order moment estimate to the difference between 1 and the first attenuation factor is the corrected first-order moment estimate, and the ratio of the updated second-order moment estimate to the difference between 1 and the second attenuation factor is the corrected second-order moment estimate.

[0091] In this implementation, the current first-order moment estimate is corrected according to the first attenuation factor, and the current second-order moment estimate is corrected according to the second attenuation factor. Since the first-order moment estimate and the second-order moment estimate of the gradient derived from the loss function with respect to the temperature prediction model parameters are updated in an exponential decay manner, in the early stage of iterative optimization, the first-order moment estimate and the second-order moment estimate are relatively small. If no correction is made, the first-order moment estimate and the second-order moment estimate will always be small values, which is not conducive to model training. By correcting the deviation of the first-order moment estimate and the second-order moment estimate, it helps the optimizer to find the correct update direction more quickly, accelerates model convergence, and improves stability during the optimization process. It can also better adapt to different learning rates, accelerate model convergence, and improve training efficiency and performance.

[0092] Exemplarily, the process of optimizing and updating the temperature prediction model parameter values ​​can be expressed as:

[0093] ; (1)

[0094] ; (2)

[0095] ; (3)

[0096] ; (5)

[0097] ; (6)

[0098] ; (7)

[0099] Among them, g t Represents the gradient function of the loss function with respect to the temperature prediction model parameters, θ t Represents the current temperature prediction model parameter value, m t represents the updated first-order moment estimate, m t-1 represents the current first-order moment estimate, β1 represents the first decay rate, that is, the decay rate of the first-order moment estimate, v t represents the updated second-order moment estimate, v t-1 represents the current second-order moment estimate, β2 represents the second decay rate, that is, the decay rate of the second-order moment estimate, θ t+1 represents the updated temperature prediction model parameter value, and η represents the learning rate. represents the corrected first-order moment estimate, represents the corrected second-order moment estimate, β t 1 represents the first attenuation factor, that is, the attenuation factor of the first-order moment estimate, β t 2 represents the second attenuation factor, that is, the attenuation factor of the second-order moment estimation. ε is used to avoid the error of dividing by zero in the calculation, and ε is usually set to a very small value, such as ε≤1e-5, such as ε=1e-8. The above formulas (1)-(7) can be considered as an adaptive moment estimation optimizer, which is used to optimize and update the parameters of the temperature prediction model.

[0100] In some embodiments, the initial values ​​of the temperature prediction model parameters can be randomly set. The initial value of the first-order moment estimation can be set to 0, and the initial value of the second-order moment estimation can be set to 0. The first attenuation factor β t 1 can be taken as 0.9, the second attenuation factor β t 2 can be taken as 0.99.

[0101] In some embodiments, a maximum number of iterations may be set, and the iteration stop condition may be that the number of iterations reaches the maximum number of iterations, and when the iteration is stopped, the temperature prediction model is obtained according to the temperature prediction model parameter value obtained in the last iteration. In some embodiments, the iteration stop condition may be that the prediction accuracy of the temperature prediction model meets the requirements, and illustratively, the prediction accuracy of the temperature prediction model reaches a prediction accuracy of ±1K.

[0102] In this embodiment, the loss function is not limited, as long as it can express the loss of the predicted value of the output temperature sequence relative to the true value. In some embodiments, using the training data to train the constructed temperature prediction model includes: establishing a loss function and iterating based on the loss function to obtain the temperature prediction model parameter value that meets the conditions, so as to obtain the temperature prediction model determined by the temperature prediction model parameter value that meets the conditions, and the loss function expresses the average of the multiple powers of the difference between the true value and the predicted value of the output temperature sequence of the training data. The predicted value of the output temperature sequence refers to the output temperature sequence predicted by the temperature prediction model determined by the current temperature prediction model parameter value.

[0103] The loss function can be expressed as the average of the quadratic difference between the true value and the predicted value of the output temperature sequence of the training data, that is, the loss function can adopt the mean squared error loss function (Mean Squared Error Loss), which can be expressed as:

[0104] ; (8)

[0105] Where N represents the total number of training data. represents the true value of the jth training data, It represents the predicted value obtained by the temperature prediction model. The number of training data samples can be adjusted appropriately according to the time situation.

[0106] In some embodiments, the temperature prediction model may include an encoder and a decoder; the encoder includes multiple parallel layers, the layers include a first network and a second network, the first network is used to learn the dependency relationship between the input temperature sequence and the output temperature sequence, and generate a representation that expresses the correlation between the temperature data at previous and next moments, and the second network is used to perform a nonlinear transformation on the representation generated by the first network; the decoder includes an internal memory unit, an input gate, a forget gate and an output gate, the internal memory unit is used to allow sequence information to be transmitted along the internal memory unit, the forget gate is used to determine whether the sequence information transmitted along the internal memory unit is discarded, the input gate is used to determine whether the sequence information is stored in the internal memory unit, and the output gate is used to determine whether the sequence information in the internal memory unit is output.

[0107] The encoder includes multiple layers in parallel, each layer includes a first network and a second network, and each set of input temperature sequences and output temperature sequences are processed by each layer. The first network can be a self-attention (Transformer) mechanism network, in which the encoder can simultaneously consider the dependency of all temperature data in the input temperature sequence and the output temperature sequence, and generate a representation expressing the correlation of the temperature data before and after.

[0108] For example, reference may be made to Figure 3 , Figure 3 The schematic diagram of the temperature prediction model in the optical system temperature field acquisition method of an embodiment is shown in the figure. As shown in the figure, the temperature prediction model includes an encoder and a decoder. In the encoder, Q (query), K (key), and V (value) are three important links of the self-attention mechanism. The Query (Q) query matrix is ​​used to extract the information of the current point to be paid attention to. In the self-attention mechanism, the Query query matrix is ​​used to perform similarity calculations with all Key key matrices to determine the contribution of each input element to the current output. The Key (K) key matrix is ​​used to represent the information of each element in the input sequence. It is used to match the Query query matrix to calculate the correlation between the Query query matrix and all Key key matrices. The correlation between the Query query matrix and the Key key matrix is ​​then used as a weight for weighting the Value value matrix. The Value (V) value matrix contains the actual information in the input sequence, which will be used in the weighting process of the self-attention mechanism. Specifically, each element of the Value value matrix is ​​weighted and summed according to the correlation (weight) between its corresponding Key value matrix and the Query query matrix to generate the final output.

[0109] The second network can be a feed-forward neural network, and the representation generated by the first network is passed to the feed-forward neural network for further nonlinear transformation. In some embodiments, the layers of the encoder are connected by a residual network or a layer normalization network, which can improve training stability and performance. The encoder can be considered as a structure composed of multiple identical layers stacked together. This encoder allows processing of variable-length sequences and effectively captures complex relationships within the sequence.

[0110] The internal memory unit is the core of the decoder, which allows sequence information to be transmitted along the internal memory unit. The network structure of the internal memory unit is linearly interactive, so that the sequence information can be stored in the internal memory unit for a long time. The internal memory unit can also be called a cell state, which has only slight linear interactions, which allows information to be stored in the network for a long time.

[0111] The Forget Gate is used to decide which sequence information should be discarded from the internal memory unit. It generates a vector between 0 and 1 by looking at the previous hidden state and the current input, where 1 means "completely keep this information" and 0 means "completely discard this information".

[0112] The input gate is used to determine which new sequence information is stored in the internal memory unit. In some embodiments, the input gate includes a first activation function and a second activation function, the first activation function is used to determine whether the sequence information needs to be updated, and the second activation function is used to generate a new candidate value based on the sequence information. The first activation function can map the sequence information to a range of 0 to 1, and the second activation function can map the sequence information to a range of -1 to 1. Whether the sequence information is added to the internal memory unit is determined based on the output value of the first activation function and the output value of the second activation function. The first activation function can be a sigmoid function, and the second activation function can be a tanh function, which is used to increase nonlinearity.

[0113] The output gate is used to determine which sequence information of the internal memory unit should be included in the updated hidden state. The output gate determines which sequence information of the internal memory unit will be output through the third activation function, and then passes the internal memory unit through the fourth activation function, and further multiplies the output of the internal memory unit after passing through the fourth activation function with the output of the third activation function to obtain the final output. The third activation function can map the sequence information to a range of 0 to 1, and the fourth activation function can map the sequence information to a range of -1 to 1. The third activation function can be a sigmoid function, and the fourth activation function can be a tanh function. The sigmoid function and the tanh function are used to increase nonlinearity.

[0114] The decoder can be a Long Short-Term Memory (LSTM) network, which is a special recurrent neural network that is used to solve the gradient vanishing or gradient exploding problems encountered by recurrent neural networks when dealing with long-distance dependency problems. This goal is achieved by introducing three gate controllers (input gate, forget gate, and output gate) and internal memory units.

[0115] For example, reference may be made to Figure 3 As shown, in the decoder, H´ t-1 Represents the output of the encoder and the current hidden state of the decoder H t-1 The concatenated value, H t Denotes the updated hidden state of the decoder. The decoder’s hidden state H is used to capture long-term dependencies in the sequence, and it is updated at each time step. t-1 Indicates the current cell state, C t Indicates the cell state after update. t Represents the forget gate, which determines which sequence information should be discarded from the cell state. It generates a vector between 0 and 1 by looking at the current hidden state and the current input, where 1 means "completely keep this information" and 0 means "completely discard this information".t represents the input gate, which determines which new sequence information is stored in the cell state. t Represents the output gate, which is used to determine which parts of the cell state should be included in the next hidden state. The output gate determines which parts of the cell state will be output through the third activation function, and then passes the cell state through the fourth activation function, and further multiplies the output of the cell state after passing through the fourth activation function with the output of the third activation function to obtain the final output.

[0116] In this embodiment, the temperature prediction model includes an encoder and a decoder. The encoder adopts a self-attention mechanism network and a feedforward neural network, and the decoder adopts a long short-term memory network. Therefore, the temperature prediction model adopts a hybrid neural network, which can improve the accuracy and prediction precision of the optical system temperature prediction.

[0117] In some embodiments, the constructed temperature prediction model is trained using training data, and obtaining the temperature prediction model may include: the training data includes test data, the input temperature sequence of the test data is input into the obtained temperature prediction model, and the predicted output temperature sequence is obtained, and the mean absolute percentage error (MAPE) or / and the root mean square error (RMSE) are calculated according to the actual output temperature sequence and the predicted output temperature sequence to test the obtained temperature prediction model. For example, the mean absolute percentage error MAPE and the root mean square error RMSE may be calculated according to the following formula:

[0118] ;

[0119] Among them, J represents the total number of samples in the test set, represents the true value of the test set, that is, the true output temperature sequence, Represents the predicted value, that is, the predicted output temperature sequence.

[0120] In some embodiments, according to the temperature sequence of the subsequent period, obtaining the temperature field of the optical system in the subsequent period includes: interpolating the temperature data of other positions except the preset positions on the structural model of the optical system according to the temperature data of each preset position included in the temperature sequence of the subsequent period, and obtaining the temperature field of the optical system in the subsequent period. It can be that for any process in the subsequent period, according to the temperature data of the present process in the temperature sequence of the subsequent period, according to the temperature data of each preset position at the time of the present process, interpolating the temperature data of other positions except the preset position on the structural model of the optical system at the time of the present process, obtaining the temperature data of the preset position of the optical system at the time of the present process in the subsequent period and the temperature data of other positions of the optical system except the preset position, and obtaining the temperature field of the optical system at the time of the present process. The constructed temperature field can be considered as a digital twin model of the temperature field of the optical system.

[0121] In some embodiments, the optical system temperature field acquisition method may further include the following process: inputting the temperature sequence of the next period into the temperature prediction model, and the temperature prediction model outputs the temperature sequence of the next period after the next period, and the temperature sequence of the next period after the next period is formed by the predicted temperature data of the preset position in the next period after the next period arranged as the process progresses; according to the temperature sequence of the next period after the next period, the temperature field of the optical system in the next period after the next period is obtained. According to the temperature sequence of the current period acquired from the optical system, after inputting the temperature prediction model to predict the temperature sequence of the optical system in the next period after the current period, that is, the temperature sequence of the next period, the predicted temperature sequence of the next period can also be input into the temperature prediction model to predict the temperature sequence of the optical system in the next period after the next period, and in this way, the temperature sequence of the optical system in each period after the current period can be predicted using the temperature prediction model. Further, for each period of each period after the current period, the temperature field of the optical system in the current period can be obtained according to the predicted temperature sequence of the current period. In this way, the temperature field of the entire temperature change process of the optical system can be predicted and constructed. Based on the predicted and constructed temperature field, potential problems in the use of the optical system can be predicted, thereby improving the use efficiency and life of the optical system. For any process in any period of time, the temperature field of the optical system at the time of the process in the temperature sequence of this period can be obtained based on the temperature data of the process in the temperature sequence of this period.

[0122] In this embodiment, the preset position of the optical system is not limited, and the preset position is the temperature measurement point of the optical system. The preset position of the optical system can be, but is not limited to, the center of the mirror surface of the optical element, the edge of the mirror surface, the position on the support rod of the optical element, or the connection platform. Temperature sensors can be set at these positions to collect temperature, such as setting a temperature sensor at the center of the back side of the mirror surface of the optical element or the edge of the back side of the mirror surface of the optical element, setting a temperature sensor on the support rod of the optical element, and setting a temperature sensor on the connection platform. For example, reference can be made to Figure 4-1 and Figure 4-2 , Figure 4-1 is a structural model of an optical system in a method for acquiring a temperature field of an optical system according to an embodiment of the present invention, Figure 4-2 The optical system temperature field obtained in the optical system temperature field acquisition method of an embodiment. Figure 4-1 As shown, in this example, the optical system includes a primary mirror 101, a secondary mirror 102 and a third mirror 103, each mirror being mounted on a connecting platform 100. Figure 4-2 As shown, the temperature field of the optical system reflects the temperature distribution of the optical system, including the temperature distribution of the primary mirror 101 , the temperature distribution of the secondary mirror 102 , the temperature distribution of the tertiary mirror 103 , and the temperature distribution of the connection platform 100 .

[0123] In a specific example, the optical system includes a primary mirror 101, a secondary mirror 102, and a third mirror 103, and also includes a connecting platform 100. The primary mirror 101, the secondary mirror 102, and the third mirror 103 each include three temperature measuring points, and the connecting platform 100 includes one temperature measuring point. The difference between the predicted value and the actual value of the cooling condition of the optical system and the difference between the predicted value and the actual value of the warming condition are shown in Table 1. It can be seen that the temperature prediction errors of each temperature measuring point of the primary mirror 101, the secondary mirror 102, and the third mirror 103 of the optical system are all less than 0.5, indicating that the method is feasible. Figure 5 is the change between the predicted value and the actual value of the temperature measurement points of the primary mirror, the secondary mirror, the tertiary mirror and the connection platform under the cooling condition in the optical system temperature field acquisition method of one embodiment, Figure 6 The diagram shows changes in predicted values ​​and actual values ​​of temperature measurement points of a primary mirror, a secondary mirror, a tertiary mirror and a connection platform under a rewarming condition in a method for acquiring a temperature field of an optical system in one embodiment.

[0124] Table 1

[0125]

[0126] In addition, in order to verify that the temperature prediction model constructed by the hybrid neural network has the best prediction accuracy, this example uses the self-attention network (Transformer), convolutional neural network (CNN), long short-term memory network LSTM and multilayer perceptron (MLP) for comparison, and the mean absolute percentage error (MAPE) and root mean square error (RMSE) of the model on the test set are used as evaluation indicators. The comparison results are shown in Table 2 and Figure 7 As shown, Figure 7 This is a comparison chart of the mean absolute percentage error and root mean square error of the temperature prediction model constructed using five networks in one embodiment, where Hybrid represents the temperature prediction model constructed using a hybrid neural network in this embodiment, LSTM represents the temperature prediction model constructed using a long short-term memory network, MLP represents the temperature prediction model constructed using a multi-layer perceptron, CNN represents the temperature prediction model constructed using a convolutional neural network, and Transformer represents the temperature prediction model constructed using a self-attention network. It can be seen that the MAPE and RMSE evaluation indicators of the temperature prediction model constructed using a hybrid neural network (Hybrid) are lower, and the model constructed using a hybrid neural network (Hybrid) has the best prediction effect.

[0127] Table 2

[0128]

[0129] The optical system temperature field acquisition method of this embodiment can effectively predict the temperature field of the optical system throughout its life cycle, can be used to monitor the temperature distribution of the optical system in real time at various time periods, can be used to predict potential problems that exist during the use of the optical system, predict potential factors that may cause damage to the optical system, and improve the use efficiency and life of the optical system.

[0130] This embodiment also provides an optical system temperature field acquisition system, including:

[0131] Memory for storing computer programs;

[0132] A processor is used to implement the steps of the method for acquiring the temperature field of an optical system as described in any one of the above embodiments when executing the computer program.

[0133] The optical system temperature field acquisition system of this embodiment obtains the temperature data of the preset position of the optical system during the temperature change process, and obtains the training data including the input temperature sequence and the output temperature sequence, so as to obtain the temperature prediction model of the optical system, so that according to the temperature sequence of the current period obtained in real time from the optical system, the temperature prediction model can be used to predict the temperature sequence of the optical system in the next period, and further obtain the temperature field of the optical system in the next period, thereby realizing the prediction of the temperature field of the optical system. Therefore, the optical system temperature field acquisition system of this embodiment can predict the temperature field of the optical system, so that the temperature condition of the optical system can be monitored, which has a reference significance for predicting and analyzing various performances of the optical system.

[0134] The above is a detailed introduction to the optical system temperature field acquisition method and system provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and core ideas of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A method for acquiring a temperature field of an optical system, characterized in that: include: During the temperature change process, the temperature data of the preset position of the optical system is acquired to obtain a temperature sequence of the current period, wherein the temperature sequence of the current period is formed by arranging the temperature data of the preset position in the current period as the process progresses; Input the temperature sequence of the current period into a temperature prediction model, and the temperature prediction model outputs a temperature sequence of a subsequent period, wherein the temperature sequence of the subsequent period is formed by arranging the temperature data of the preset position in the subsequent period as the process progresses, and the subsequent period is the period after the current period; According to the temperature sequence of the latter time period, a temperature field of the optical system in the latter time period is obtained, wherein the temperature field of the optical system includes temperature data of the preset position of the optical system and temperature data of other positions of the optical system except the preset position; Obtaining the temperature prediction model includes: During the temperature change process, the temperature data of the preset position of the optical system is obtained, and training data is obtained based on the obtained temperature data, the training data including an input temperature sequence and an output temperature sequence, the input temperature sequence is formed by arranging the temperature data of the preset position in a previous time period as the process progresses, and the output temperature sequence is formed by arranging the temperature data of the preset position in a later time period as the process progresses, and the constructed temperature prediction model is trained using the training data to obtain the temperature prediction model.

2. The method for acquiring the temperature field of an optical system according to claim 1, characterized in that: During the temperature change process, obtaining temperature data of the preset position of the optical system, and obtaining training data according to the obtained temperature data comprises: During the cooling process, obtaining temperature data of the preset position of the optical system, and obtaining the training data according to the obtained temperature data; Alternatively, in the process of returning to a temperature after cooling down, the temperature data of the preset position of the optical system is obtained, and the training data is obtained according to the obtained temperature data; Alternatively, during the cooling process and during the warming process after cooling, the temperature data of the preset position of the optical system is respectively acquired, and the training data is acquired according to the acquired temperature data.

3. The method for acquiring the temperature field of an optical system according to claim 1, characterized in that: The training data obtained based on the obtained temperature data include: The obtained temperature data is processed into the input temperature sequence and the output temperature sequence through the time window. The input temperature sequence is expressed as , the output temperature sequence is expressed as , represents a time window of length m starting from time t, express Then the time window of length n, Represents X t is a vector formed by the temperature data obtained for i preset positions.

4. The method for acquiring the temperature field of an optical system according to claim 1, characterized in that: Training the constructed temperature prediction model using the training data includes: Establishing a loss function, the loss function describes the loss of the predicted value of the output temperature sequence relative to the true value, and derivatizing the loss function with respect to the temperature prediction model parameters to obtain a gradient function of the loss function derivatized with respect to the temperature prediction model parameters; In any iteration, the current temperature prediction model parameter value is substituted into the gradient function to obtain the current gradient value, and the first-order moment estimate is updated according to the current first-order moment estimate and the current gradient value, the second-order moment estimate is updated according to the current second-order moment estimate and the current gradient value, and the temperature prediction model parameter value is updated according to the updated first-order moment estimate and the updated second-order moment estimate, and the next iteration is performed with the updated temperature prediction model parameter value; When an iteration stop condition is met, the iteration is stopped, and the temperature prediction model is obtained according to the temperature prediction model parameter values ​​obtained in the last iteration; Among them, the first-order moment estimate represents the estimate of the first-order moment of the gradient of the loss function derived from the temperature prediction model parameters, and the second-order moment estimate represents the estimate of the second-order moment of the gradient of the loss function derived from the temperature prediction model parameters.

5. The method for acquiring the temperature field of an optical system according to claim 4, characterized in that: Updating the first-order moment estimate based on the current first-order moment estimate and the current gradient value includes: Obtain a first product of a first decay rate and the current first-order moment estimate, obtain a difference between 1 and the first decay rate, and obtain a second product of the difference and the current gradient value, and use the sum of the first product and the second product as an updated first-order moment estimate; Updating the second-order moment estimate based on the current second-order moment estimate and the current gradient value includes: Obtain a third product of the second attenuation rate and the current second-order moment estimate, obtain a difference between 1 and the second attenuation rate, and obtain a fourth product of the square of the current gradient value, and use the sum of the third product and the fourth product as the updated second-order moment estimate.

6. The method for acquiring the temperature field of an optical system according to claim 4, characterized in that: Updating the temperature prediction model parameter values ​​according to the updated first-order moment estimate and the updated second-order moment estimate includes: The updated first-order moment estimate is corrected for deviation, and the updated second-order moment estimate is corrected for deviation, and the temperature prediction model parameter value is updated according to the corrected first-order moment estimate and the corrected second-order moment estimate, wherein the ratio of the updated first-order moment estimate to the difference between 1 and the first attenuation factor is the corrected first-order moment estimate, and the ratio of the updated second-order moment estimate to the difference between 1 and the second attenuation factor is the corrected second-order moment estimate.

7. The method for acquiring the temperature field of an optical system according to claim 1, characterized in that: Training the constructed temperature prediction model using the training data includes: A loss function is established and iterative operations are performed based on the loss function to obtain temperature prediction model parameter values ​​that meet conditions, so as to obtain the temperature prediction model determined by the temperature prediction model parameter values ​​that meet the conditions, wherein the loss function expresses the average of multiple powers of the difference between the true value and the predicted value of the output temperature sequence of the training data.

8. The method for acquiring the temperature field of an optical system according to claim 1, characterized in that: The temperature prediction model includes an encoder and a decoder; The encoder includes a plurality of layers in parallel, the layers including a first network and a second network, the first network is used to learn the dependency relationship between the input temperature sequence and the temperature data in the output temperature sequence, and generate a representation expressing the correlation of the temperature data at the previous and next moments, and the second network is used to perform a nonlinear transformation on the representation generated by the first network; The decoder includes an internal memory unit, an input gate, a forget gate and an output gate. The internal memory unit is used to allow sequence information to be transmitted along the internal memory unit. The forget gate is used to determine whether the sequence information transmitted along the internal memory unit is discarded. The input gate is used to determine whether the sequence information is stored in the internal memory unit. The output gate is used to determine whether the sequence information in the internal memory unit is output.

9. The method for acquiring the temperature field of an optical system according to any one of claims 1 to 8, characterized in that: Also includes: The temperature sequence of the next period is input into the temperature prediction model, and the temperature prediction model outputs the temperature sequence of the period after the next period, wherein the temperature sequence of the period after the next period is formed by arranging the predicted temperature data of the preset position in the period after the next period as the process progresses; According to the temperature sequence of a time period after the latter time period, the temperature field of the optical system in a time period after the latter time period is obtained.

10. An optical system temperature field acquisition system, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the method for acquiring the temperature field of an optical system as described in any one of claims 1 to 9 when executing the computer program.

Images