Temperature monitoring method of main coolant pump in nuclear power plant based on improved multidimensional state estimation

Abstract

The present invention relates to the field of equipment reliability management, and specifically to a method for monitoring the temperature of a main coolant pump in a nuclear power plant based on improved multidimensional state estimation. It comprises the following steps: Step 1: constructing a monitoring model based on a multidimensional state estimation algorithm; Step 2: optimizing the monitoring model based on a genetic algorithm; Step 3: evaluating the equipment status based on the established model. The beneficial effect of the present invention is that multidimensional monitoring abandons the previous equipment status monitoring behavior based on a single measuring point, and now a multidimensional state estimation algorithm model is constructed to realize a monitoring method from multiple associated main coolant pump status measuring points, thereby realizing a more comprehensive and accurate equipment status monitoring method. The main coolant pump equipment suitable for nuclear power scenarios is in a stable operating state in most nuclear power scenarios, which provides rich samples for the modeling of this algorithm, and is consistent with the core idea of ​​this algorithm to determine the equipment status by comparing the difference between the estimated values ​​of the real-time observer of the equipment and the healthy sample.

Description

Technical Field

[0001] The present invention relates to the field of equipment reliability management, and in particular to a method for monitoring the temperature of a main coolant pump in a nuclear power plant based on improved multi-dimensional state estimation. Background Art

[0002] Pumps in nuclear power plants operate continuously under complex working conditions all year round. The units have very high requirements for equipment reliability and other aspects. Damage and failure of important components of the main coolant pump will directly or indirectly lead to power reduction or even shutdown of the nuclear power unit, resulting in unplanned replacement and maintenance costs.

[0003] Currently, nuclear power plants routinely monitor pumps using fixed threshold monitoring, which involves setting parameter alarm values ​​and generating corresponding alarms when these values ​​are reached. Fixed threshold monitoring has the following main problems:

[0004] 1. Repeated alarm: crossing the alarm threshold and repeatedly alarming.

[0005] 2. False positives and false negatives are a serious problem

[0006] The main reasons for the above problems are summarized as follows:

[0007] 1. Changes in process parameters cause repeated alarms or missed alarms;

[0008] 2. Improper alarm threshold settings result in false alarms or missed alarms;

[0009] 3. Improper selection of alarm features results in false alarms or missed alarms. Summary of the Invention

[0010] The purpose of the present invention is to provide a temperature monitoring method for the main coolant pump of a nuclear power plant based on improved multidimensional state estimation, which can realize real-time online monitoring and intelligent early warning of the main coolant pump equipment, improve the equipment reliability analysis and management capabilities, and thus assist enterprise managers in making maintenance decisions.

[0011] The technical solution of the present invention is as follows: A method for monitoring the temperature of a main coolant pump in a nuclear power plant based on improved multidimensional state estimation comprises the following steps:

[0012] Step 1: Build a monitoring model based on a multi-dimensional state estimation algorithm;

[0013] Step 2: Monitoring model optimization based on genetic algorithm;

[0014] Step 3: Evaluate the equipment status based on the established model.

[0015] The step 1 determines the valid and available historical operation data, selects the normal operation interval of the equipment and finally constructs the health sample matrix D.

[0016]

[0017] Among them, X(t i ) represents the measured value of group i, x n (t i ) represents the measured value of the nth sensor at the i-th moment.

[0018] X(t i )=[x1(t i ),x2(t i ),...,x n (t i )] T

[0019] After constructing the healthy sample matrix D, the weight calculation formula is introduced to generate the estimated value, where the weight is represented by W, Xest represents the estimated value, and Xobs represents the observed value.

[0020] X est =DW

[0021] W=(D T D) -1 D T X obs

[0022] In order to make (D T D) -1 Reversible, introducing nonlinear operators Make Reversible.

[0023]

[0024] Finally, the construction of the equipment monitoring model based on the multidimensional state estimation algorithm was completed.

[0025] In step 2, the genetic coding G is constructed for the existing weight W. e , introduce genetic algorithm to optimize the weights:

[0026] W opt =WG e

[0027]

[0028] Now we construct an evaluation function based on the estimated values ​​and observed values ​​generated by the initial multidimensional state estimation algorithm. Here, the minimum value of the evaluation function result is used as the training target:

[0029]

[0030] For the two selected parent chromosomes, perform the crossover operation according to the formula:

[0031] a kj =a kj (1-r)+a lj r

[0032] a lj =a lj (1-r)+a kj r

[0033] Here, the genetic method in biology is simulated to generate a new generation of chromosomes, and the individual kj Paragraph and a lj The crossover operation is performed on the segments, where r is a random constant in the range of [0,1];

[0034] Then, through the mutation operation, any individual is selected from the population to perform a mutation operation on a certain segment of the selected chromosome in order to produce a better individual. The jth gene a of the i-th individual is used. ij Perform mutation operation. The specific mutation method is as follows:

[0035] a ij =a ij +(a ij -a max )*f(g)r>0.5

[0036] a ij =a ij +(a min -a ij )*f(g)r≤0.5

[0037] In the formula, r is a random number between [0, 1], a max For the gene's parent, a min is the next term of the gene; f(g)=rand×(1-g / G max ) 2 , rand is a random number between [0, 1]; g is the current number of iterations; G max is the maximum number of evolutions, and finally the newly generated crossover and mutation individuals are used to generate a new population and calculate the evaluation function, and iterate multiple times until the optimal individual and the G required in the formula are generated. e .

[0038] Step 3 evaluates the residuals between the generated estimated values ​​and the actual values, and constructs the validation sample set D vobs Input the improved multidimensional state estimation algorithm to obtain the estimated value set D vest , through D vobs -D vest After obtaining the residual set E, the monitoring threshold bandwidth v is obtained based on the three sigma rule. u and the bandwidth v under the monitoring threshold dIn addition, a sliding window is further constructed to count abnormal measurement points in the sliding window and compare them with the set value S t For comparison, in the actual application of the model, if the difference between the estimated value and the observed value in the sliding window is greater than v d And the number of occurrences is greater than S t If the condition is met, it means that the relevant measurement point is in a lower healthy state than before. On the contrary, if the difference between the observed value and the estimated value in the sliding window is greater than v u And the number of occurrences is greater than S t If the condition is met, it proves that the real-time measurement points related to the equipment have deviated from the healthy state, indicating that the equipment may be in an abnormal state.

[0039] The beneficial effects of the present invention are:

[0040] (1) Realize multi-dimensional monitoring

[0041] Abandoning the previous equipment status monitoring behavior based on a single measuring point, a multi-dimensional state estimation algorithm model is now constructed to realize the monitoring method from multiple associated main coolant pump status measuring points, realizing a more comprehensive and accurate equipment status monitoring method.

[0042] (2) Main coolant pump equipment suitable for nuclear power scenarios

[0043] In nuclear power scenarios, most equipment is in a stable operating state, which provides rich samples for modeling of this algorithm and is consistent with the core idea of ​​this algorithm to determine the equipment status by comparing the difference between the real-time observer of the equipment and the estimated value of the healthy sample.

[0044] (3) Improved multidimensional state estimation algorithm

[0045] The multidimensional state estimation algorithm is optimized based on genetic algorithms to increase the redundancy of the algorithm model, making it more suitable for actual industrial application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Modeling a flow chart for the improved multi-dimensional state estimation algorithm;

[0047] Figure 2 This is a flow chart of the method for monitoring the temperature of the main coolant pump of a nuclear power plant based on improved multi-dimensional state estimation provided by the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] The present invention provides a method for monitoring the temperature of a nuclear power plant's main coolant pump based on improved multidimensional state estimation. This method uses a multidimensional state estimation algorithm improved by a genetic algorithm (GA) to estimate the operating state of the nuclear power plant's main coolant pump. An algorithm model is established based on samples of the main coolant pump's normal state to derive estimated values ​​for corresponding measuring points of the main coolant pump. A sliding window residual statistics method is then used to analyze the residuals between the estimated and actual values ​​of the measuring points. The residual values ​​can be used to intuitively characterize the current operating state of the main coolant pump. When the residual value curve exceeds a set dynamic threshold range, a residual statistics window is constructed to analyze the equipment status. Results demonstrate that the nuclear power plant main coolant pump early warning system based on the improved multidimensional state estimation algorithm is capable of monitoring the operating state of the unit's main coolant pump in real time.

[0050] The main purpose of the present invention is as follows:

[0051] (1) Analyze the equipment's online and offline parameters to obtain the equipment health index. The present invention constructs a health index based on the main coolant pump's historical data to characterize the equipment's health status.

[0052] (2) Establish an online monitoring and early warning model to monitor the health of the main coolant pump in real time and issue early warnings for equipment abnormalities. Based on the equipment's historical normal samples, abnormal equipment status monitoring and early warning can be performed without abnormal sample data. The constructed model has stronger redundancy and accuracy, making it very suitable for nuclear power-related equipment status monitoring. The main coolant pump status is evaluated by adopting a multidimensional state estimation algorithm, and the weight parameters in the multidimensional state estimation algorithm are modified by a genetic algorithm to obtain the optimal estimated value. The sliding window residual statistics method is then used to analyze the residuals between the estimated value and the actual value. The threshold interval of the residual mean is defined according to the residual change trend during training.

[0053] like Figure 1 As shown, the method for monitoring the temperature of the main coolant pump of a nuclear power plant based on improved multidimensional state estimation includes the following steps:

[0054] Step 1: Build a monitoring model based on a multi-dimensional state estimation algorithm

[0055] First, based on the experience of power plant experts (equipment engineers), equipment status-related measurement points are selected and the historical operation data of the monitored objects are managed through the experience of engineers to determine the effective and available historical operation data, select the normal operating range of the equipment, and finally construct the health sample matrix D.

[0056]

[0057] Among them, X(t i ) represents the measured value of group i, x n (ti ) represents the measured value of the nth sensor at the i-th moment.

[0058] X(t i )=[x1(t i ),x2(t i ),...,x n (t i )] T

[0059] After constructing the healthy sample matrix D, the weight calculation formula is introduced to generate the estimated value, where the weight is represented by W, Xest represents the estimated value, and Xobs represents the observed value.

[0060] X est =DW

[0061] W=(D T D) -1 D T X obs

[0062] In order to make (D T D) -1 Reversible, introducing nonlinear operators Make Reversible.

[0063]

[0064] Finally, the construction of the equipment monitoring model based on the multidimensional state estimation algorithm was completed.

[0065] Step 2: Monitoring model optimization based on genetic algorithm

[0066] Genetic algorithm is an optimization search algorithm based on the principles of natural selection and genetics, which often plays a relatively ideal effect in multi-parameter optimization. e , introduce genetic algorithm to optimize the weights:

[0067] W opt =WG e

[0068]

[0069] W opt Based on the weight vector optimized by genetic algorithm, the weight vector W is the weight assigned to each sample. The weight W is based on the "genetic coding G calculated by genetic algorithm and training samples" e Further optimize the selected weights to obtain the optimized weight vector W opt .

[0070] Now the genetic code Ge The calculation is explained:

[0071] First, a plurality of first-generation individuals are randomly generated to form an initial population. Usually, binary coding is used to encode the initial population, but this encoding method is too redundant and difficult to encode in many contexts. Therefore, real number coding is used for encoding in the embodiment of the present invention.

[0072] Now we construct an evaluation function based on the estimated values ​​and observed values ​​generated by the initial multidimensional state estimation algorithm. Here, the minimum value of the evaluation function result is used as the training target:

[0073]

[0074] The crossover operation generates new individuals, which plays an important role in promoting the iterative upgrade of the genetic algorithm. Its essence is to generate the next generation of individuals with a certain probability from the parent population through the selection operation. The probability of an individual being selected is related to the fitness f. The better the fitness value, the greater the probability of being selected. For the two selected parent chromosomes, the crossover operation is performed according to the formula:

[0075] a kj =a kj (1-r)+a lj r

[0076] a lj =a lj (1-r)+a kj r

[0077] Here, the genetic method in biology is simulated to generate a new generation of chromosomes, and the individual kj Paragraph and a lj The crossover operation is performed on the segments, where r is a random constant in the range [0,1].

[0078] Mutation operation is the core step of genetic algorithm. Then, through mutation operation, we select individuals from the population and perform mutation operation on a certain segment of the selected chromosome in order to produce better individuals. For example, we use the jth gene a of the i-th individual. ij Perform mutation operation. The specific mutation method is as follows:

[0079] a ij =a ij +(a ij -a max )*f(g)r>0.5

[0080] a ij =a ij +(a min -a ij )*f(g)r≤0.5

[0081] In the formula, r is a random number between [0, 1], a max For the gene's parent, a min is the next term of the gene; f(g)=rand×(1-g / G max ) 2 , rand is a random number between [0, 1]; g is the current number of iterations; G max is the maximum number of evolutions.

[0082] Finally, the newly generated crossover and mutation individuals are used to generate a new population and the evaluation function is calculated. Multiple iterations are performed until the optimal individual and the G required in the formula are generated. e .

[0083] Step 3: Evaluate the equipment status based on the established model

[0084] Now evaluate the residuals between the generated estimated values ​​and the actual values ​​and construct the validation sample set D vobs Input the improved multidimensional state estimation algorithm to obtain the estimated value set D vest By D vobs -D vest After obtaining the residual set E, the monitoring threshold bandwidth v is obtained based on the three sigma rule. u and the bandwidth v under the monitoring threshold d In addition, a sliding window is constructed to count abnormal measurement points in the sliding window and compare them with the set value S t In the actual application of the model, if the difference between the estimated value and the observed value in the sliding window is greater than v d And the number of occurrences is greater than S t If the condition is , it means that the relevant measurement point is in a lower healthy state than before. On the contrary, if the difference between the observed value and the estimated value in the sliding window is greater than v u And the number of occurrences is greater than S t If the condition is met, it proves that the real-time measurement points related to the equipment have deviated from the healthy state, indicating that the equipment may be in an abnormal state.

[0085] The implementation method of the present invention is algorithm development, and the specific steps are as follows:

[0086] The present invention first establishes a set of temperature index measurement points for the main coolant pump of a nuclear power plant. The main temperature indexes selected are as follows:

[0087]

[0088] Collect historical data (2 years) of the temperature index measurement points of the main coolant pumps of nuclear power plants, remove abnormal data values ​​from the historical data, and supplement missing values;

[0089] Construct a monitoring model based on a multi-dimensional state estimation algorithm;

[0090] Use genetic algorithms for model optimization;

[0091] Based on data analysis and expert experience, a dynamic threshold table of the main coolant pump temperature model is determined, in which the upper and lower limits of the dynamic thresholds of each measuring point are given.

[0092]

[0093]

[0094] Implement equipment status assessment based on the established model.

[0095] This invention is applicable to systems where sensors are installed on key components critical to equipment health and capable of signal monitoring. The algorithm can be used independently or integrated into existing equipment monitoring systems. The resulting status assessment can provide engineers with effective information for developing appropriate operation and maintenance strategies.

Claims

1. A method for monitoring the temperature of a main coolant pump in a nuclear power plant based on improved multidimensional state estimation, characterized in that: The steps include: Step 1: Build a monitoring model based on a multi-dimensional state estimation algorithm; The step 1 determines the valid and available historical operation data, selects the normal operation interval of the equipment and finally constructs the health sample matrix D. Among them, X(t i ) represents the measured value of group i, x n (t i ) represents the measured value of the nth sensor at the i-th moment, X(t i )=[x1(t i ),x2(t i ),...,x n (t i )] T ; After constructing the healthy sample matrix D in step 1, the weight calculation formula is introduced to generate the estimated value, where the weight is represented by W, Xest represents the estimated value, and Xobs represents the observed value. X est =DW W=(D T D) -1 D T X obs In order to make (D T D) -1 Reversible, introducing nonlinear operators Make reversible, Finally, the construction of the equipment monitoring model based on the multi-dimensional state estimation algorithm was completed; Step 2: Monitoring model optimization based on genetic algorithm; In step 2, the genetic coding G is constructed for the existing weight W. e , introduce genetic algorithm to optimize the weights: IN opt =WG e Now we construct an evaluation function based on the estimated values ​​and observed values ​​generated by the initial multidimensional state estimation algorithm. Here, the minimum value of the evaluation function result is used as the training target: For the two selected parent chromosomes, perform the crossover operation according to the formula: a kj =a kj (1-r)+a lj r a lj =a lj (1-r)+a kj r Here, the genetic method in biology is simulated to generate a new generation of chromosomes, and the individual kj Paragraph and a lj The crossover operation is performed on the segments, where r is a random constant in the range of [0,1]; Step 3: Evaluate the equipment status based on the established model; Step 3 evaluates the residuals between the generated estimated values ​​and the actual values, and constructs the validation sample set D vobs Input the improved multidimensional state estimation algorithm to obtain the estimated value set D vest , through D vobs -D vest After obtaining the residual set E, the monitoring threshold bandwidth v is obtained based on the three sigma rule. u and the bandwidth v under the monitoring threshold d .

2. The method for monitoring the temperature of a main coolant pump in a nuclear power plant based on improved multidimensional state estimation according to claim 1, wherein: In step 2, an individual is randomly selected from the population through a mutation operation to perform a mutation operation on a certain segment of the selected chromosome in order to produce a better individual. The jth gene a of the i-th individual is used. ij Perform mutation operation. The specific mutation method is as follows: a ij =a ij +(a ij -a max )*f(g)r>0.5 a ij =a ij +(a min -a ij )*f(g)r≤0.5 In the formula, r is a random number between [0, 1], a max For the gene's parent, a min is the next term of the gene; f(g)=rand×(1-g / G max ) 2 , rand is a random number between [0, 1]; g is the current number of iterations; G max is the maximum number of evolutions, and finally the newly generated crossover and mutation individuals are used to generate a new population and calculate the evaluation function, and iterate multiple times until the optimal individual and the G required in the formula are generated. e .

3. The method for monitoring the temperature of a main coolant pump in a nuclear power plant based on improved multidimensional state estimation according to claim 1, wherein: The step 3 further constructs a sliding window, and counts the abnormal measurement points in the sliding window and compares them with the set value S t For comparison, in the actual application of the model, if the difference between the estimated value and the observed value in the sliding window is greater than v d And the number of occurrences is greater than S t If the condition is met, it means that the relevant measurement point is in a lower healthy state than before. On the contrary, if the difference between the observed value and the estimated value in the sliding window is greater than v u And the number of occurrences is greater than S t If the condition is met, it proves that the real-time measurement points related to the equipment have deviated from the healthy state, indicating that the equipment may be in an abnormal state.

Images