Measurement error-considered nuclear reactor main pipeline coolant temperature estimation model

Abstract

The invention discloses a nuclear reactor main pipeline coolant temperature estimation model considering a measurement error, which belongs to the technical field of nuclear power plant temperature measurement, and comprises the following steps: obtaining actual temperature measurement data of a power plant, and establishing a data set corresponding to the measurement data and an agreed true value; by analyzing the actual temperature measurement process, the measurement error is classified; considering different error distribution characteristics, combining a data driving algorithm with a random model, and establishing a temperature estimation model; designing a solving method for estimating prior information in the temperature estimation model and calculating a model coefficient; and completing the optimization and solution of the temperature estimation model by using the manufactured data set, and verifying the accuracy and reliability of the temperature estimation model. The problems that existing measurement errors are difficult to refine and research and real distribution of nuclear pipeline coolant temperature is difficult to solve are solved.

Description

Technical Field

[0001] The invention belongs to the technical field of temperature measurement in nuclear power plants, and in particular relates to the establishment of a nuclear reactor main pipeline coolant temperature estimation model taking measurement errors into consideration. Background Art

[0002] A nuclear power reactor is a fast and efficient conversion device of the three energies of "nuclear, thermal, and dynamic" with high power density, high operating parameters, and high safety requirements. Therefore, timely and accurate grasp of the nuclear energy release and thermal energy transmission status of the nuclear reactor core is an important basis for ensuring the safe and reliable operation of the reactor. According to the structure and working principle of the pressurized water reactor, the temperature of the coolant in the hot section of the main pipeline of the nuclear reactor system can directly reflect the nuclear power and the core heat exchange state, and is the core parameter for nuclear reactor power control and safety protection. If the temperature of the primary circuit is too low, it cannot meet the power generation needs of the nuclear power plant; if the temperature of the primary circuit is too high, it may cause damage to the fuel cladding or even melting of the fuel core, endangering the safety of the power plant. Therefore, how to correctly estimate the true temperature of the coolant in the main pipeline of the nuclear reactor is of great practical significance, and relevant research is urgently needed.

[0003] In order to ensure the safe and reliable operation of the reactor, nuclear power plants now use measurement uncertainty analysis methods to measure the closeness between the measured value and the true value. This method traces the source of the uncertainty of the measurement parameter, synthesizes all the uncertainties to obtain the overall uncertainty of the parameter, and simply adds the calculated overall uncertainty to the design value of the nuclear power plant to obtain a numerical range. It is believed that the measured value plus this range can obtain the distribution range of the true temperature. In traditional measurement uncertainty analysis methods, it is believed that the true value of the measurement parameter cannot be obtained, and the difference between the measured value and the true value can only be compensated by tracing the error term. There are problems such as difficulty in tracing the source, over-reliance on engineering experience, and lack of accurate evaluation of measurement parameters.

[0004] The present invention hopes to use a data-driven model to directly invert and construct the actual measured temperature field based on the actual measured data, so as to more accurately estimate the actual temperature of the coolant in the main pipeline of the nuclear reactor. However, when processing the actual measured data, the existing research does not carefully consider and analyze the measurement errors carried in the measured values. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a nuclear reactor main pipe coolant temperature estimation model taking measurement errors into account. Based on the actual temperature measurement data of the power plant obtained, a data set corresponding to the measurement data and the agreed true value is established. The actual temperature measurement process is analyzed and the measurement errors are classified and processed. Then, considering different error distribution characteristics, the data-driven algorithm is combined with the random model to establish a temperature estimation model and design a solution method, thereby realizing the inversion of the true temperature value and solving the problem that the true value of the coolant temperature of the main pipe of the nuclear reactor is difficult to estimate and the accuracy assessment of the measurement value is relatively conservative.

[0006] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0007] The present invention provides a nuclear reactor main pipeline coolant temperature estimation model taking into account measurement errors, comprising the following steps:

[0008] S1. Obtain the actual temperature measurement data of the power plant and establish a data set corresponding to the measured data and the agreed true value;

[0009] S2. By analyzing the actual temperature measurement process, the measurement errors are classified and processed;

[0010] S3, considering different error distribution characteristics, combining data-driven algorithm with stochastic model to establish temperature estimation model;

[0011] S4. Estimation of prior information in the temperature estimation model and calculation and design of solution methods for model coefficients;

[0012] S5. Use the data set in S1 to optimize and solve the temperature estimation model, and verify its accuracy and reliability.

[0013] The beneficial effects of the present invention are as follows: a nuclear reactor main pipe coolant temperature estimation model taking measurement errors into account is provided by the present invention. Different from the traditional method of regarding measurement errors as normally distributed, the measurement errors existing in the measurement process are classified into detailed categories, and assumptions are made for the different statistical characteristics of each error, thereby alleviating the problem of conservative estimation that is prone to occur in traditional uncertainty assessment methods. The input used by the present invention is the actual measurement data of the nuclear power plant, and the probability model improves the physical interpretability of the data-driven model, and reasonably explains the "many-to-many" functional relationship between the measurement value and the true value, which is effectively combined with actual engineering applications.

[0014] Furthermore, the step of preparing the data set in S1 comprises the following steps:

[0015] S11, the actual measurement error is the difference between the measured value and the design value (theoretical calculation value), obtain the working condition The design value of the temperature under ;

[0016] S12, the measured value sequence of a period of time in the power plant With design value Subtract and get the measured error sequence , analyze the statistical characteristics of the error size and obtain the measured error distribution ;

[0017] S13. Determine the fluctuation range of the temperature design value based on the fluctuation of the operating conditions during the actual operation of the nuclear reactor ;

[0018] S14. Based on the temperature range in S13 and the continuity of temperature change, generate a time length of Temperature Convention True Value Sequence ;

[0019] S15, based on the measured error distribution obtained in S12 , Second-rate The error sequence is sampled and then compared with the sequence described in S14 Adding together, we can get A set of measurement data sequences, each set of measurement data sequence can be expressed as .

[0020] The beneficial effect of adopting the above further scheme is that the present invention establishes a data set corresponding to the temperature measurement value and the true value based on the actual temperature measurement data of the power station and the design value of the temperature in the design manual. In the data set preparation, the present invention uses the actual measured data of the power plant, and fully considers the randomness of the error, and establishes a "many-to-many" functional relationship between the true value of the temperature and the measured value, so that the present method has high credibility and accuracy.

[0021] Furthermore, the error analysis and classification processing method of S2 includes the following contents:

[0022] S21. The measurement data itself consists of two parts: the true value and the measurement error.

[0023] The measurement error is not a fixed value, but has a certain randomness. The statistical characteristics of non-homologous errors are different, so it is difficult to give a unified quantitative description.

[0024] The overall measurement error is divided into two categories, namely steady-state error and transient error. Among them, the steady-state error has stable statistical characteristics, and its distribution is independent of time. According to the uncertainty calculation standard of the nuclear reactor main pipeline, the error characteristics such as instrument error and process error can be described by white noise, so they can be regarded as steady-state errors; the transient error is defined as the error whose distribution is related to time. By observing the entire measurement process, it can be found that the temperature change of the nuclear reactor main pipeline is also related to time. Therefore, the present invention believes that the transient error comes from the change of the real temperature. Therefore, the temperature at the current moment can be preliminarily estimated, and then the transient error at the current moment can be extracted through this prior information.

[0025] The beneficial effect of adopting the above further scheme is that in the process of modeling using the data-driven method, since the actual measurement data used as input carries errors, the errors will cause the model to have pathological predictions. In existing research, there are few studies on this issue, and most of them simply regard the error as a normal distribution. The present invention classifies the errors existing in the measurement in detail, and describes the error properties differently according to different statistical characteristics, which is helpful to better realize the accurate estimation of the coolant temperature of the main pipeline of the nuclear reactor in the future.

[0026] Furthermore, the temperature estimation model of S3 has the following characteristics:

[0027] S31. Since the measured data itself is not only random but also has strong time correlation, the commonly used model for processing random time series is the Autoregressive Moving Average Model (ARMA). This model assumes that the label value fluctuates around the general trend of time, where the trend is affected by historical labels, the fluctuation is affected by accidental events within a period of time, and the general trend itself is not necessarily stable. The temperature estimation model (NARMA) proposed in this invention draws on the basic idea of ​​this model;

[0028] The temperature estimation model described in the present invention contains three items. The first item assumes that the agreed true value at the current moment is affected by the measured value within a period of time (including the past, current, and future time), which makes up for the fact that the traditional ARMA does not consider the impact of future data on the current value; the second item is the transient error information, which can be obtained through the prior information of the temperature field; the third item is the steady-state error information, which is consistent with the white noise information described in ARMA. The mathematical expression of the model can be written as:

[0029] in, is The agreed true value estimated by the model at a certain moment may have different values ​​under the same input due to input uncertainty; It is in the time period The sequence of measurements within ; is the measurement coefficient matrix, which represents the influence of the measured value on the agreed true value; It is in the time period The prior information in The preliminary estimate of the temperature value at each moment in Function of is the transient error coefficient matrix, which is used to measure the impact of transient errors on the agreed true value; It is the steady-state error in the measurement process, which is consistent with the statistical characteristics of white noise; is the random error coefficient matrix, which is used to measure the influence of steady-state error on the agreed true value; since the error is accidental, and The elements in are not fixed values, but random variables that obey a certain distribution.

[0030] The beneficial effect of adopting the above further scheme is: in order to realize the inversion of the true temperature field based on the actual temperature measurement data, combined with the ARMA idea, the components of the true temperature value are divided into the contribution of the measured value and the contribution of the error term, and the basic framework of the temperature estimation model is built, which lays the foundation for the establishment and solution of the subsequent model.

[0031] Furthermore, the temperature estimation model solving process of S4 is divided into the following steps:

[0032] S41. Use the Attention-LSTM model to estimate prior information;

[0033] S42. Use the FCNN-BN model to solve the model coefficients.

[0034] Furthermore, the Attention-LSTM model in S41 estimates the prior information including the following steps:

[0035] S411, the temperature measurement data of n thermal resistors at time t Send it to the attention mechanism module (Attention) to obtain data carrying the location information of other measurement points ;

[0036] S412: n updated inputs By sending it into the LSTM neural network structure, we can get the prior estimation information of the temperature at the n measuring points at time t;

[0037] S413, the time length is Repeat the above steps with the measured data to obtain the prior estimation information of temperature .

[0038] The beneficial effects of adopting the above further scheme are as follows: It is found through experiments that the spatial correlation in the data set carries a large amount of information. In the Attention-LSTM model designed by the present invention, the spatial correlation between several measuring points is first calculated through the self-attention mechanism, the original input is replaced, and then the new input is passed to the LSTM unit. When the LSTM module calculates its temporal correlation, since the input itself carries the spatial information of the positions of other measuring points, the secondary extraction of spatial characteristics is achieved, ensuring that the model fully mines the spatiotemporal data features.

[0039] Furthermore, the FCNN-BN model in S42 solves the model coefficients including the following steps:

[0040] S421, the temperature measurement data of n thermal resistors at time t Send it to the fully connected neural network (FCNN), through the weights , calculated ;

[0041] Will is sent to the Bayesian neural network (BN) and The characteristics are assumed (usually it is assumed to follow a normal distribution with a mean of 0 and a variance of 1), and the weights are , , calculated ;

[0042] Will and Adding together gives the final result , put the final result into the loss function, train and optimize the model parameters , , .

[0043] The beneficial effects of adopting the above further scheme are: for the constant coefficient W1 and the variable coefficients W2 and W3, according to their different characteristics, the fully connected neural network and the Bayesian neural network are used to solve them respectively, thus ensuring the accuracy of the final result; at the same time, the introduction of the probability model (Bayesian network) also provides a more reasonable explanation for the randomness of the error.

[0044] Furthermore, the accuracy and reliability verification in S5 comprises the following steps:

[0045] S51. Due to the introduction of the probability model, for the same input, the output is different each time. The present invention uses the commonly used absolute mean error, mean square error, and root mean square error to perform accuracy analysis on each experimental result;

[0046] S52. Considering that the result of the probability model is not a constant, the evaluation index in S51 cannot evaluate the accuracy of the distribution estimation, and it is also necessary to use continuous graded probability scores and explained variance to evaluate the accuracy of the model estimated distribution. The beneficial effect of adopting the above further scheme is: the present invention provides a method for calculating the expectation and variance of the measured temperature field, and the randomness of the measured temperature field itself is statistically analyzed by probability method, which can more accurately represent its internal law.

[0047] The beneficial effect of adopting the above further scheme is: based on the characteristics of the temperature estimation model itself, the present invention verifies the accuracy of the model from two aspects: single-point prediction and distribution prediction, thereby ensuring the accuracy of the final solved model.

[0048] Other advantages of the present invention will be analyzed in more detail in subsequent embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 The present invention is a flowchart of the steps of a nuclear reactor main pipe coolant temperature estimation model taking into account measurement errors in an embodiment of the present invention.

[0051] Figure 2 A flowchart of the steps for preparing a data set in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] 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 a part of the embodiments of the present invention, rather than all the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0054] All other embodiments belong to the protection scope of the present invention.

[0055] like Figure 1As shown, in one embodiment of the present invention, the present invention provides a nuclear reactor main pipe coolant temperature estimation model taking into account measurement errors, comprising the following steps:

[0056] S1. Obtain the actual temperature measurement data of the power plant and establish a data set corresponding to the measured data and the agreed true value;

[0057] The preparation of the data set in S1 includes the following steps:

[0058] S11, the actual measurement error is the difference between the measured value and the design value (theoretical calculation value), obtain the working condition The design value of the temperature under ;

[0059] S12, the measured value sequence of a period of time in the power plant With design value Subtract and get the measured error sequence , analyze the statistical characteristics of the error size and obtain the measured error distribution ;

[0060] S13. Determine the fluctuation range of the temperature design value based on the fluctuation of the operating conditions during the actual operation of the nuclear reactor ;

[0061] S14. Based on the temperature range in S13 and the continuity of temperature change, generate a time length of Temperature Convention True Value Sequence ;

[0062] S15, based on the measured error distribution obtained in S12 , Second-rate The error sequence is sampled and then compared with the sequence described in S14 Adding together, we can get A set of measurement data sequences, each set of measurement data sequence can be expressed as .

[0063] The data set preparation method flow chart is as follows Figure 2 shown.

[0064] By analyzing the actual temperature measurement process, the measurement errors are classified and processed;

[0065] The S2 comprises the following steps:

[0066] S21. The measurement data itself consists of two parts: the true value and the measurement error.

[0067] S22. There are many reasons for measurement errors, including instrument factors, environmental factors, human factors, etc.

[0068] S23. The measurement error is not a fixed value but has a certain randomness. The statistical characteristics of non-homologous errors are different, so it is difficult to give a unified quantitative description.

[0069] S24. The overall measurement error is divided into two categories, namely steady-state error and transient error. Among them, the steady-state error has stable statistical characteristics, and its distribution is independent of time. According to the uncertainty calculation standard of the nuclear reactor main pipeline, the error characteristics such as instrument error and process error can be described by white noise, so they can be regarded as steady-state errors; the transient error is defined as the error whose distribution is related to time. By observing the entire measurement process, it can be found that the change in the temperature of the nuclear reactor main pipeline is also related to time, that is, the transient error comes from the change in the real temperature. Therefore, the temperature at the current moment can be preliminarily estimated, and then the transient error at the current moment can be extracted through this prior information.

[0070] S3, considering different error distribution characteristics, combining data-driven algorithm with stochastic model to establish temperature estimation model;

[0071] The S3 comprises the following steps:

[0072] S31. Since the measured data itself is not only random but also has strong time correlation, the commonly used model for processing random time series is the Autoregressive Moving Average Model (ARMA). This model assumes that the label value fluctuates around the general trend of time, where the trend is affected by historical labels, the fluctuation is affected by accidental events within a period of time, and the general trend itself is not necessarily stable. The temperature estimation model (NARMA) proposed in this invention draws on the basic idea of ​​this model;

[0073] S32. The temperature estimation model of the present invention includes three items. The first item assumes that the agreed true value at the current moment is affected by the measured value within a period of time (including the past, current, and future time), which makes up for the fact that the traditional ARMA does not consider the influence of future data on the current value; the second item is the transient error information, which can be obtained through the prior information of the temperature field; the third item is the steady-state error information, which is consistent with the white noise information described in ARMA. The mathematical expression of this model can be written as:

[0074] in, is The agreed true value estimated by the model at a certain moment may have different values ​​under the same input due to input uncertainty; It is in the time period The sequence of measurements within ; is the measurement coefficient matrix, which represents the influence of the measured value on the agreed true value; It is in the time period The prior information in The preliminary estimate of the temperature value at each moment in Function of is the transient error coefficient matrix, which is used to measure the impact of transient errors on the agreed true value; It is the steady-state error in the measurement process, which is consistent with the statistical characteristics of white noise; is the random error coefficient matrix, which is used to measure the influence of steady-state error on the agreed true value; since the error is accidental, and The elements in are not fixed values, but random variables that obey a certain distribution.

[0075] S4. Estimation of prior information in the temperature estimation model and calculation and design of solution methods for model coefficients;

[0076] The S4 comprises the following steps:

[0077] S41. Use the Attention-LSTM model to estimate prior information;

[0078] S42. Use the FCNN-BN model to solve the model coefficients.

[0079] S5. Optimize and solve the temperature estimation model using the data set in S1, and verify its accuracy and reliability;

[0080] The S5 comprises the following steps:

[0081] S51. Due to the introduction of the probability model, for the same input, the output is different each time. The present invention uses the commonly used mean absolute error (MAE), mean square error (MSE), and root mean square error (RMSE) to perform accuracy analysis on each experimental result. The evaluation index calculation method is as follows:

[0082]

[0083]

[0084] in, is the true value, is the model prediction value, is the total number of results;

[0085] S52. Considering that the probability model result is not a constant value, the evaluation indicators in S51 cannot evaluate the accuracy of the distribution estimation. It is also necessary to use the continuous graded probability score (CRPS) and explained variance (EVS) to evaluate the accuracy of the model estimated distribution. The calculation method of the evaluation indicators is as follows:

[0086] in, is the cumulative distribution function.

[0087] The beneficial effect of adopting the above further scheme is: the present invention provides a method for calculating the expectation and variance of the measured temperature field, and uses a probability method to statistically analyze the randomness of the measured temperature field itself, which can more accurately represent its internal laws.

[0088] Aiming at the problem of evaluating the accuracy of temperature measurement data of nuclear reactor main pipe, the present invention proposes a nuclear reactor main pipe temperature estimation model that takes into account measurement errors. The model can input the actual measurement data of a certain period of time. , the Attention-LSTM module is used to obtain a preliminary estimate of the temperature during this time period , and then input both into the FCNN-BNN model to solve the coefficients of the temperature estimation model , and finally output the predicted temperature value of the corresponding measuring point position within the time period. The model prediction value can be used to understand the actual temperature distribution of the nuclear reactor main pipeline, thereby evaluating the reliability and accuracy of the measurement value;

[0089] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A nuclear reactor main pipe coolant temperature estimation model considering measurement errors, characterized in that: The steps include: S1. Obtain the actual temperature measurement data of the power plant and establish a data set corresponding to the measured data and the agreed true value; S2. By analyzing the actual temperature measurement process, the measurement errors are classified and processed; S3, considering different error distribution characteristics, combining data-driven algorithm with stochastic model to establish temperature estimation model; S4. Estimation of prior information in the temperature estimation model and calculation and design of solution methods for model coefficients; S5. Use the data set in S1 to optimize and solve the temperature estimation model, and verify its accuracy and reliability.

2. The nuclear pipeline coolant temperature estimation model considering measurement errors according to claim 1, characterized in that: The preparation of the data set in S1 includes the following steps: S11. Define the actual measurement error as the difference between the measured value and the design value (theoretical calculation value) to obtain the working condition The design value of the temperature under ; S12, the measured value sequence of a period of time in the power plant With design value Subtract and get the measured error sequence , analyze the statistical characteristics of the error size and obtain the measured error distribution ; S13. Determine the range of temperature design value fluctuations based on the fluctuations during actual operation of the nuclear reactor ; S14. Based on the temperature range in S13 and the continuity of temperature change, generate a time length of Temperature Convention True Value Sequence ; S15, based on the measured error distribution obtained in S12 , Second-rate The error sequence is sampled and then compared with the sequence described in S14 Adding together, we can get A set of measurement data sequences, each set of measurement data sequence can be expressed as ; S16. Repeat S14-S15 to produce a large number of data sets.

3. The data set corresponding to the measured data and the agreed true value according to claim 2, characterized in that: The measured error used in S15 is not a constant value, but follows a certain probability distribution. Therefore, a set of agreed true value sequences corresponds to A set of measurement data series.

4. The nuclear pipeline coolant temperature estimation model considering measurement errors according to claim 1, characterized in that: The error characteristics and classification processing methods of S2 include the following: S21. The measurement data itself consists of two parts: the true value and the measurement error. S22. There are many reasons for measurement errors, including instrument factors, environmental factors, human factors, etc. S23. The measurement error is not a fixed value but has a certain randomness. The statistical characteristics of non-homologous errors are different, so it is difficult to give a unified quantitative description. S24. The present invention divides the overall measurement error into two categories, namely steady-state error and transient error; wherein, the steady-state error has stable statistical characteristics, and its distribution is independent of time. According to the uncertainty calculation standard of the nuclear reactor main pipeline, it can be known that the error characteristics such as instrument error and process error can be described by white noise, so they can be regarded as steady-state errors; the transient error is defined as an error whose distribution is related to time. By observing the entire measurement process, it can be found that the change in the temperature of the nuclear reactor main pipeline is also related to time. The present invention believes that the transient error comes from the change in the real temperature. Therefore, a preliminary estimate of the temperature at the current moment can be made, and then the transient error at the current moment can be extracted through the prior information.

5. The nuclear pipeline coolant temperature estimation model considering measurement errors according to claim 1, characterized in that: The temperature estimation model of S3 has the following characteristics: S31. Since the measured data itself has strong time correlation in addition to randomness, the commonly used model for processing random time series is the Autoregressive Moving Average Model (ARMA). This model assumes that the label value fluctuates around the general trend of time, where the trend is affected by historical labels, and the fluctuation is affected by accidental events within a period of time. Moreover, the general trend itself is not necessarily stable. The temperature estimation model (NARMA) proposed in this invention draws on the basic idea of ​​this model. S32. The temperature estimation model of the present invention includes three items. The first item assumes that the agreed true value at the current moment is affected by the measured value within a period of time (including the past, current, and future time), which makes up for the fact that the traditional ARMA does not consider the influence of future data on the current value; the second item is the transient error information, which can be obtained through the prior information of the temperature field; the third item is the steady-state error information, which is consistent with the white noise information described in ARMA. The mathematical expression of the model can be written as: ； in, is The agreed true value estimated by the model at a certain moment may have different values ​​under the same input due to input uncertainty; It is in the time period The sequence of measurements within ; is the measurement coefficient matrix, which represents the influence of the measured value on the agreed true value; It is in the time period The prior information in The preliminary estimate of the temperature value at each moment in Function of is the transient error coefficient matrix, which is used to measure the impact of transient errors on the agreed true value; It is the steady-state error in the measurement process, which is consistent with the statistical characteristics of white noise; is the random error coefficient matrix, which is used to measure the influence of steady-state error on the agreed true value; since the error is accidental, and The elements in are not fixed values, but random variables that obey a certain distribution.

6. The nuclear pipeline coolant temperature estimation model considering measurement errors according to claim 1, characterized in that: The temperature estimation model solving process of S4 is divided into the following steps: S41. Use the Attention-LSTM model to estimate prior information; S42. Use the FCNN-BN model to solve the model coefficients.

7. The nuclear pipeline coolant temperature estimation model considering measurement errors according to claim 6, characterized in that: The Attention-LSTM model in S41 estimates the prior information including the following steps: S411, the temperature measurement data of n thermal resistors at time t Send it to the attention mechanism module (Attention) to obtain data carrying the location information of other measurement points ; S412: n updated inputs By sending it into the LSTM neural network structure, we can get the prior estimation information of the temperature at the n measuring points at time t; S413, the time length is Repeat the above steps with the measured data to obtain the prior estimation information of temperature .

8. The nuclear pipeline coolant temperature estimation model considering measurement errors according to claim 6, characterized in that: The FCNN-BN model in S42 solves the model coefficients including the following steps: S421, the temperature measurement data of n thermal resistors at time t Send it to the fully connected neural network (FCNN), through the weights , calculated ; S422, will is sent to the Bayesian neural network (BN) and The characteristics are assumed (usually it is assumed to follow a normal distribution with a mean of 0 and a variance of 1), and the weights are , , calculated ; S423, will and Adding together gives the final result , put the final result into the loss function, train and optimize the model parameters , , .

9. The nuclear pipeline coolant temperature estimation model considering measurement errors according to claim 1, characterized in that: The accuracy and reliability verification in S5 includes the following steps: S51. Due to the introduction of the probability model, for the same input, the output is different each time. The present invention uses the commonly used absolute mean error, mean square error, and root mean square error to perform accuracy analysis on each experimental result; S52. Considering that the results of the probability model are not constant, the evaluation indicators in S51 cannot evaluate the accuracy of the distribution estimation. It is also necessary to use continuous graded probability scores and explained variance to evaluate the accuracy of the model estimated distribution.

Images