A method, device, equipment and medium for condition-based prediction and standardized residual construction for variable load conditions of a nuclear power plant water supply system

Abstract

The application provides a working condition conditioning prediction and standardized residual construction method, device, equipment and medium for variable load conditions of a nuclear power plant water supply system, and belongs to the technical field of nuclear power plant operation state monitoring and intelligent analysis. The method solves the problems of significant prediction error variance, incomparable cross-phase deviation and insufficient long-term drift adaptability of existing data-driven prediction methods under variable load transient conditions. It comprises data acquisition and preprocessing, working condition condition vector and phase division, working condition conditioning probability prediction model training, online standardized residual construction, intra-phase statistical calibration and online drift correction. It is mainly used for nuclear power plant operation state monitoring and intelligent analysis.

Description

Technical Field

[0001] This invention belongs to the field of nuclear power plant operation status monitoring and intelligent analysis technology, and in particular relates to a method, device, equipment and medium for condition-based prediction and standardized residual construction of nuclear power plant feedwater system under variable load conditions. Background Technology

[0002] As a key subsystem of the secondary loop, the feedwater system of a nuclear power plant directly affects the steam generator water level control and heat exchange efficiency. Under operating modes such as load tracking and peak shaving, nuclear power units experience transient processes with varying loads, resulting in highly time-varying, strongly coupled, and nonlinear dynamic changes in parameters such as feedwater pressure, flow rate, water level, and temperature.

[0003] Existing data-driven prediction methods mostly use point prediction outputs, which makes it difficult to characterize the heteroscedasticity of the error variance as it changes with the operating conditions during the variable load stage. This results in the residual amplitudes being incomparable at different phases, which can easily lead to false alarms and missed alarms. At the same time, the lack of phase-sensing modeling and online drift calibration mechanisms makes it difficult to cope with the problems of sensor zero drift, model aging, and slow drift of operating conditions during long-term operation.

[0004] Therefore, it is necessary to propose a probabilistic prediction and standardized residual construction technology for operating condition conditions under variable load transients, so as to output health deviations comparable across operating conditions and provide stable inputs for operation monitoring, health assessment and alarm. Summary of the Invention

[0005] In view of this, the present invention aims to propose a condition-based prediction and standardized residual construction method, device, equipment and medium for the variable load conditions of nuclear power plant feedwater systems, so as to solve the problems of significant variance in prediction error under variable load transients, incomparable cross-phase deviation, and insufficient long-term drift adaptability of existing data-driven prediction methods.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: According to a first aspect of the present invention, a method for condition-based prediction and standardized residual construction for variable load conditions in nuclear power plant feedwater systems is provided, comprising the following steps: Multivariable operating parameters of the nuclear power plant feedwater system under variable load conditions are collected, and the multivariable operating parameters are preprocessed to obtain a preprocessed parameter sequence. Based on the preprocessed parameter sequence, a working condition vector is constructed, and the parameter sequence is divided into transient phases according to the working condition vector to obtain a phase label sequence. The conditional probabilistic prediction model is trained based on historical operating parameters containing only normal variable load conditions. The conditional probabilistic prediction model can output the predicted mean and prediction uncertainty of the target parameters under the condition of historical parameter sequence, condition vector and phase label within a given preset time window. The preprocessed parameter sequence, operating condition vector, and phase label of the online running process are input into the operating condition-based probabilistic prediction model to obtain the predicted mean and prediction uncertainty of the target parameters, and the standardized residual sequence is calculated based on the predicted mean and prediction uncertainty. The standardized residual sequence is subjected to in-phase statistical calibration and online drift correction, and the output is a calibration residual sequence comparable across operating conditions, a prediction confidence interval, and / or a health index obtained by aggregating the calibration residual sequences.

[0007] Furthermore, the data preprocessing of the multivariate operating parameters specifically includes: performing time alignment, outlier handling, and normalization on the collected multivariate operating parameters, and constructing the input sample sequence using a sliding time window method.

[0008] Furthermore, the data preprocessing also includes denoising processing, which includes at least one of moving average filtering or low-pass filtering.

[0009] Furthermore, the operating condition vector includes at least the power P and the rate of change of power. Among them, the rate of change of power It is calculated by power difference between adjacent sampling times.

[0010] Furthermore, the transient phase division includes at least one of the following or a combination thereof: Based on the absolute value of the power change rate With preset threshold The parameter sequence is divided into steady-state phase and variable load phase; Based on power change rate The symbols subdivide the variable load phase into power-up phase and power-down phase; The phase boundary is corrected based on the power setpoint change point and / or control command change point.

[0011] Furthermore, the standardized residual sequence is composed of standardized residuals of multiple target parameter dimensions arranged in chronological order. For each time step and target parameter dimension, the expression for the standardized residual is: in, These are actual measured values. To predict the mean, To predict uncertainty The standard deviation.

[0012] Furthermore, the intra-phase statistical calibration specifically includes: maintaining the running mean and running variance of the standardized residuals for different phases, and normalizing the standardized residuals by phase to obtain cross-phase comparable calibration residuals; The online drift correction specifically includes: updating the drift bias term based on the sliding window statistics or exponential moving average of the calibration residual within a time period that meets the preset normality conditions, and compensating for the predicted mean and / or calibration residual to suppress systematic deviations caused by sensor zero drift, model aging, or slow operating condition drift. According to a second aspect of the present invention, a device for predicting operating conditions and standardizing residuals for variable load conditions in nuclear power plant feedwater systems is provided, comprising: The data acquisition and preprocessing module is used to acquire multivariable operating parameters of the nuclear power plant feedwater system under variable load conditions, and to preprocess the multivariable operating parameters to obtain a preprocessed parameter sequence. The working condition construction module is used to construct a working condition vector based on the preprocessed parameter sequence, and to perform transient phase division on the parameter sequence according to the working condition vector to obtain a phase label sequence. The conditional probability prediction module is used to train a conditional probability prediction model based on historical operating parameters containing only normal variable load conditions. This enables the conditional probability prediction model to output the predicted mean and prediction uncertainty of the target parameters within a given preset time window, given the historical parameter sequence, operating condition vector, and phase label. The standardized residual construction module is used to input the preprocessed parameter sequence, operating condition vector and phase label of the online operation into the operating condition probabilistic prediction model to obtain the predicted mean and prediction uncertainty of the target parameters, and to calculate the standardized residual sequence based on the predicted mean and prediction uncertainty. The drift calibration module is used to perform intra-phase statistical calibration and online drift correction on the standardized residual sequence, and output a calibration residual sequence comparable across operating conditions, a prediction confidence interval, and / or a health index obtained by aggregating the calibration residual sequences.

[0013] According to a third aspect of the present invention, an electronic device is provided, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method for condition-based prediction and standardized residual construction for variable load conditions of nuclear power plant feedwater systems.

[0014] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, the computer program being configured to cause the computer to perform the above-described method for conditional prediction and standardized residual construction for variable load conditions of nuclear power plant feedwater systems.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention provides a conditional prediction and standardized residual construction method for the variable load conditions of nuclear power plant feedwater systems. By introducing a combined mechanism of "conditional modeling + phase sensing + heteroscedastic probability prediction + standardized residual + phase calibration and drift correction", the residuals are comparable under different power levels and different variable load stages, thereby reducing the impact of normal dynamics of variable load on the deviation amplitude, reducing the impact of sensor zero drift and model drift on output stability during long-term operation, and providing stable input for operation monitoring, health assessment and alarm. 2. The present invention has strong cross-condition comparability. By predicting the uncertainty of the output through heteroscedastic probability and constructing standardized residuals, the deviation has consistent dimensions and statistical comparability under different power levels and different load stages. 3. The present invention has strong resistance to transient disturbances. Through phase division and intra-phase calibration mechanisms, it reduces the impact of cross-phase distribution differences on the magnitude and stability of deviation. 4. This invention suppresses sensor zero drift, model aging, and slow drift of operating conditions through an online drift correction mechanism, thereby improving long-term operational stability and enabling long-term deployment; 5. The present invention outputs calibration residuals and confidence intervals, which can be directly connected to operation monitoring, health assessment and alarm modules, facilitating engineering integration. Attached Figure Description

[0016] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating the working condition-based prediction and standardized residual construction method for variable load conditions in nuclear power plant feedwater systems, as described in this invention. Figure 2 This is a flowchart of step S1 in an embodiment of the present invention; Figure 3 This is a flowchart of step S2 in an embodiment of the present invention; Figure 4 This is a flowchart of step S3 in an embodiment of the present invention; Figure 5 This is a flowchart of step S4 in an embodiment of the present invention; Figure 6This is a flowchart of step S5 in an embodiment of the present invention; Figure 7 This is a block diagram of a device for predicting operating conditions and standardizing residuals for variable load conditions in nuclear power plant feedwater systems, as described in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other, and the described embodiments are only some embodiments of the present invention, not all embodiments.

[0018] See Figure 1-7 This implementation method is described below.

[0019] Figure 1 This is a flowchart illustrating the method for predicting operating conditions and constructing standardized residuals for variable load conditions in nuclear power plant feedwater systems, as described in this invention.

[0020] like Figure 1 As shown, the method for condition-based prediction and standardized residual construction for variable load conditions in nuclear power plant feedwater systems includes the following steps: In step S1, multivariable operating parameters of the nuclear power plant feedwater system under variable load conditions are collected, and the multivariable operating parameters are preprocessed to obtain a preprocessed parameter sequence.

[0021] In some embodiments, the data preprocessing of the multivariate operating parameters specifically includes: performing time alignment, outlier handling and normalization on the collected multivariate operating parameters, and constructing the input sample sequence using a sliding time window method.

[0022] In some embodiments, the data preprocessing further includes denoising, which includes at least one of moving average filtering or low-pass filtering.

[0023] Figure 2 This is a flowchart of step S1.

[0024] like Figure 2 As shown, step S1 is executed as follows: Collect multivariable operating parameters of the feedwater system under variable load conditions in nuclear power plant simulators or actual equipment, including nuclear power or electrical power, feedwater flow rate, feedwater pressure, feedwater temperature, steam generator water level and temperature, pressurizer pressure and water level, reactor coolant inlet and outlet temperatures and average temperature, etc. The collected data is time-aligned and resampled to ensure that the timestamps of all parameters are consistent; Perform outlier handling, including filling in missing values ​​in the data, removing spike outliers, and correcting out-of-bounds data that exceeds the reasonable operating range; Denoising is performed using moving average filtering or low-pass filtering to remove high-frequency noise from the data and smooth the parameter sequence. Normalize or standardize the filtered parameter data to eliminate differences in the units of different parameters and improve the stability of subsequent model training. A sliding time window of fixed length L is set, and parameter sequences are extracted in chronological order to construct a continuous input sample sequence. This provides input for subsequent construction of working conditions and model training. The output is a window of data from time t-L+1 to time t (where t represents the time step). .

[0025] The final output is the preprocessed parameter sequence / window sample. .

[0026] In step S2, a working condition vector is constructed based on the preprocessed parameter sequence, and the parameter sequence is divided into transient phases according to the working condition vector to obtain a phase label sequence.

[0027] Constructing operating condition vectors based on power sequences At least includes power and power change rate ;in It is calculated from the power difference between adjacent sampling times and can be smoothed. Based on... With threshold The sequence is divided into steady-state phase and variable load phase, and can be based on... The symbols are further subdivided into power-up phase and power-down phase; furthermore, the phase boundaries can be corrected by combining the power setpoint change points or control command change points to obtain a phase tag sequence. .

[0028] In some embodiments, the operating condition vector includes at least power P and the rate of change of power. Among them, the rate of change of power It is calculated by power difference between adjacent sampling times.

[0029] In some embodiments, the transient phase division includes at least one or a combination of the following: Based on the absolute value of the power change rate With preset threshold The parameter sequence is divided into steady-state phase and variable load phase; Based on power change rate The symbols subdivide the variable load phase into power-up phase and power-down phase; The phase boundary is corrected based on the power setpoint change point and / or control command change point.

[0030] Figure 3 This is a flowchart of step S2.

[0031] like Figure 3 As shown, step S2 is executed as follows: Input preprocessed power sequence Optional input control command u(t) or set value data; For the power sequence By performing differentiation, the rate of change of power at each time step can be obtained. ; Constructing the working condition vector The operating condition vector Include at least the power value at the current moment. and power change rate It can also incorporate other operating parameters such as valve position and target water level; Based on a preset rate of change threshold Regarding the power change rate Perform operating condition phase determination and classify three operating condition phases: When the power change rate satisfies When the current operating condition is determined to be a steady-state phase, ; When the power change rate satisfies When the current operating condition is determined to be a power-up phase, ; When the power change rate satisfies When the current operating condition is determined to be a power reduction phase, .

[0032] Based on the change points of set values ​​or the abrupt changes of control commands, the boundary of phase switching is corrected to optimize the judgment accuracy of phase switching and avoid misjudgment of operating conditions. Output the operating condition vector and the corrected operating condition phase label sequence .

[0033] In step S3, a conditional probabilistic prediction model is trained based on historical operating parameters containing only normal variable load conditions. This model is able to output the predicted mean and prediction uncertainty of the target parameters within a given preset time window, given the historical parameter sequence, condition vector, and phase label.

[0034] Training and validation sets were constructed using data containing only normal variable load conditions. The parameter sequence, load condition vector, and phase labels within the historical window were used as input to train the load condition-conditioned probabilistic prediction model, which outputs the predicted mean. With prediction uncertainty .

[0035] The training objective function may include probabilistic prediction loss (for simultaneous optimization) and ) and conditional invariance regularization terms (used to constrain standardized residuals to have zero mean, unit variance, or low correlation with operating conditions within the phase), and for To improve stability, upper and lower bounds can be set, or log-variance parameterization can be used. The model structure can include conditional modulation units and in-phase attention units to achieve adaptive modeling for different power levels and different load rates.

[0036] In some embodiments, the conditional probabilistic prediction model includes a conditional modulation unit and an intra-phase attention unit. The conditional modulation unit gates, scales, and / or biases the intermediate features of the model based on the conditional vector to achieve adaptive modeling for different power levels and different variable load rates. The intra-phase attention unit calculates and normalizes attention weights within the time step set corresponding to the same phase label to suppress the interference of cross-phase distribution differences on attention allocation.

[0037] Figure 4 This is a flowchart of step S3.

[0038] like Figure 4 As shown, step S3 is executed as follows: The input contains only historical datasets of the normal variable load conditions of the nuclear power plant feedwater system. Model training samples are constructed based on the dataset, and each sample contains: time-series window data. Operating condition vector Operating phase and corresponding prediction labels .

[0039] The constructed samples are input into the operational condition probabilistic prediction model. The model completes feature processing and modeling through three core units: Main Temporal Modeling / Feature Extraction Unit: For temporal window data Perform time series modeling to extract the time series features of the data; Conditional modulation unit: composed of operating condition vector The driver injects operating condition information into the main timing features through gating, scaling, and biasing operations. In-phase attention unit: based on operating condition phase The samples are grouped, and the feature weights are normalized within the phase to strengthen the feature correlation under the same phase condition and improve the model's adaptability to different phase conditions.

[0040] Model output layer outputs future The probability prediction results of the step, including the prediction mean. and prediction uncertainty ; Based on prediction uncertainty Calculate the standard deviation The standardized residuals are obtained and used for subsequent constraint term calculations. Constructing a multi-objective loss function Joint optimization of the model: Probabilistic prediction loss Joint optimization prediction mean With prediction uncertainty This improves the accuracy and reliability of probability prediction; Conditional invariance regularization term : Constrain the model to be robust to non-operating condition-related disturbances, ensuring that the model only learns effective features related to operating conditions; Numerical stability term To avoid gradient explosion / vanishing during training and ensure the numerical stability of model training; The total loss function is: in, , These are the weight coefficients for the conditional invariance regularization term and the numerical stability term, respectively.

[0041] Based on total loss Backpropagation to update the conditional probabilistic prediction model parameters .

[0042] Determine whether the model training meets the preset stopping conditions (such as loss function convergence, the number of iterations reaching the upper limit, validation set performance no longer improving, etc.): If the stopping conditions are not met, return and continue iterative training; if the stopping conditions are met, terminate the training process.

[0043] Save the parameters of the trained model The final working condition-based probabilistic prediction model is obtained. .

[0044] In step S4, the preprocessed parameter sequence, operating condition vector, and phase label of the online running process are input into the operating condition-based probabilistic prediction model to obtain the predicted mean and prediction uncertainty of the target parameters, and the standardized residual sequence is calculated based on the predicted mean and prediction uncertainty.

[0045] During online execution, the real-time parameter sequence is acquired and preprocessing and sliding window construction are performed to generate the operating condition vector and phase label according to Example 2. These are then input into the trained prediction model to obtain... and ,calculate Press again Construct standardized residual sequences; and simultaneously output prediction confidence intervals (e.g. (For operation monitoring or alarm purposes)

[0046] In some embodiments, the standardized residual sequence consists of standardized residuals of multiple target parameter dimensions arranged in chronological order. For each time step and target parameter dimension, the expression for the standardized residual is: in, These are actual measured values. To predict the mean, To predict uncertainty The standard deviation.

[0047] Figure 5 This is a flowchart of step S4.

[0048] like Figure 5 As shown, step S4 is executed as follows: The input data obtained from online preprocessing consists of three types: time-series window data. Operating condition vector Operating condition phase sequence , where L is the preset sliding window length; Call the conditional probabilistic prediction model trained in step S3 The three types of input data mentioned above are then input into the model for forward inference; Model outputs future The probability prediction results for each step include: Predict the mean sequence : Reflects the central trend of the forecast; Prediction uncertainty (variance) sequence : Reflects the confidence interval range of the prediction result.

[0049] Based on the prediction uncertainty sequence, calculate the corresponding standard deviation sequence: Obtaining and predicting the timestamp of the mean series Aligned sequence of actual measurements This ensures that the predicted values ​​are completely synchronized with the actual values ​​in terms of time dimension.

[0050] By combining the predicted mean, standard deviation, and actual measured values, a standardized residual series is constructed: , This sequence eliminates the influence of dimensions and prediction uncertainty, and can be directly used for subsequent anomaly detection.

[0051] Output the constructed standardized residual sequence .

[0052] In step S5, the standardized residual sequence is subjected to in-phase statistical calibration and online drift correction, and a calibration residual sequence comparable across operating conditions, a predicted confidence interval, and / or a health index obtained by aggregating the calibration residual sequences is output.

[0053] For each phase, the operating mean and variance of the standardized residuals are maintained. The residuals are then normalized by removing the mean and variance according to phase to obtain the calibration residuals. Within a time period that meets preset normality conditions, the drift bias term is updated using sliding window statistics or exponential moving average, and the predicted mean or calibration residuals are compensated to suppress systematic biases caused by sensor zero drift, model aging, or slow operating condition drift. The final output includes comparable calibration residuals, predicted confidence intervals, and / or health indices across operating conditions.

[0054] In some embodiments, the intra-phase statistical calibration specifically includes: maintaining the running mean and running variance of the standardized residuals for different phases, and normalizing the standardized residuals by phase to obtain cross-phase comparable calibration residuals; The online drift correction specifically includes: updating the drift bias term based on the sliding window statistics or exponential moving average of the calibration residual within a time period that meets the preset normality conditions, and compensating for the predicted mean and / or calibration residual to suppress systematic deviations caused by sensor zero drift, model aging, or slow operating condition drift. Figure 6 This is a flowchart of step S5.

[0055] like Figure 6 As shown, step S5 is executed as follows: Input the standardized residual sequence generated in step S401 and the corresponding operating condition phase label ; According to the phase of the working condition Grouping, and maintaining corresponding operational statistics for each phase: running average : The real-time mean of the standardized residuals at this phase; Operating variance : The real-time variance of the standardized residuals at this phase; Using the sliding window method or the exponential moving average (EMA) method, for , Perform online updates to adapt to changes in statistical characteristics under different operating conditions and phases.

[0056] Based on the operational statistics of each phase, the standardized residual z(t) is subjected to intra-phase secondary calibration, and the phase calibration residual is calculated. : This step eliminates the systematic deviation of the residual distribution under different operating conditions and phases, so that the calibrated residuals follow the standard normal distribution characteristics in each phase.

[0057] Based on preset normality conditions, determine whether the current residual sequence is in a stable normal operating condition (preset conditions are as follows). And this condition is continuous (All time windows are valid) If the normality condition is not met: determine that the current operating condition may have an abnormal trend, and freeze the drift bias. That is, to maintain To avoid abnormal data interfering with drift estimation; If the normality condition is met: determine that the current condition is stable and normal, and update the drift bias using the exponential moving average (EMA) method. : Where α∈(0,1) is the smoothing coefficient, which controls the update speed of the drift bias and enables tracking of the drift of the slowly varying system.

[0058] Based on the updated or frozen drift bias, the phase calibration residuals Drift correction is performed to obtain the final calibration residual. : Output drift-corrected calibration residual sequence .

[0059] The method for condition-based prediction and standardized residual construction for variable load conditions in nuclear power plant feedwater systems, proposed according to embodiments of the present invention, has the following beneficial effects: (1) This method introduces a combination mechanism of “working condition modeling + phase sensing + heteroscedastic probability prediction + standardized residual + phase calibration and drift correction” to make the residuals comparable under different power levels and different load stages, thereby weakening the impact of normal dynamic load on the deviation amplitude, reducing the impact of sensor zero drift and model drift on output stability during long-term operation, and providing stable input for operation monitoring, health assessment and alarm. (2) This method has strong comparability across working conditions. By predicting the output uncertainty through heteroscedastic probability and constructing standardized residuals, the deviation has consistent dimensions and statistical comparability at different power levels and different load stages. (3) This method has strong resistance to transient disturbances. Through phase division and intra-phase calibration mechanism, it reduces the influence of cross-phase distribution differences on the magnitude and stability of deviation. (4) This method suppresses sensor zero drift, model aging and slow drift of operating conditions through online drift correction mechanism, improves long-term operation stability and can be deployed for a long time; (5) This method outputs calibration residuals and confidence intervals, which can be directly connected to operation monitoring, health assessment and alarm modules, making it easy to integrate into engineering projects.

[0060] Figure 7 This is a block diagram of a device for predicting operating conditions and standardizing residuals for variable load conditions in nuclear power plant feedwater systems, as described in this invention.

[0061] like Figure 7 As shown, the condition-based prediction and standardized residual construction device 70 for variable load conditions of nuclear power plant feedwater systems includes: The data acquisition and preprocessing module 701 is used to acquire multivariable operating parameters of the nuclear power plant feedwater system under variable load conditions, and to preprocess the multivariable operating parameters to obtain a preprocessed parameter sequence. The working condition construction module 702 is used to construct a working condition vector based on the preprocessed parameter sequence, and to perform transient phase division on the parameter sequence according to the working condition vector to obtain a phase label sequence. The conditional probability prediction module 703 is used to train a conditional probability prediction model based on historical operating parameters containing only normal variable load conditions, so that the conditional probability prediction model can output the predicted mean and prediction uncertainty of the target parameters under the condition of historical parameter sequence, condition vector and phase label within a given preset time window. The standardized residual construction module 704 is used to input the preprocessed parameter sequence, operating condition vector and phase label of the online running into the operating condition probabilistic prediction model to obtain the predicted mean and prediction uncertainty of the target parameters, and to calculate the standardized residual sequence based on the predicted mean and prediction uncertainty. The drift calibration module 705 is used to perform intra-phase statistical calibration and online drift correction on the standardized residual sequence, and output a calibration residual sequence comparable across operating conditions, a prediction confidence interval, and / or a health index obtained by aggregating the calibration residual sequences.

[0062] It should be noted that the explanation of the above-mentioned embodiment of the condition-based prediction and standardized residual construction method for variable load conditions of nuclear power plant feedwater system also applies to the condition-based prediction and standardized residual construction device for variable load conditions of nuclear power plant feedwater system in this embodiment, and will not be repeated here.

[0063] The condition-based prediction and standardized residual construction device for variable load conditions of nuclear power plant feedwater systems proposed in this embodiment of the invention has the following beneficial effects: (1) This device introduces a combination mechanism of “working condition modeling + phase sensing + heteroscedastic probability prediction + standardized residual + phase calibration and drift correction” to make the residuals comparable under different power levels and different load stages, thereby weakening the impact of normal dynamic load on the deviation amplitude, reducing the impact of sensor zero drift and model drift on output stability during long-term operation, and providing stable input for operation monitoring, health assessment and alarm. (2) This device has strong comparability across operating conditions. By predicting the output uncertainty through heteroscedastic probability and constructing standardized residuals, the deviation has consistent dimensions and statistical comparability under different power levels and different load stages. (3) This device has strong resistance to transient disturbances. Through phase division and intra-phase calibration mechanism, it reduces the impact of cross-phase distribution differences on deviation amplitude and stability. (4) This device suppresses sensor zero drift, model aging and slow drift of operating conditions through online drift correction mechanism, improves long-term operation stability and can be deployed for a long time; (5) The device outputs calibration residuals and confidence intervals, which can be directly connected to operation monitoring, health assessment and alarm modules, making it easy to integrate into the project.

[0064] This invention proposes an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the condition-based prediction and standardized residual construction method for the variable load conditions of the nuclear power plant feedwater system.

[0065] This invention proposes a computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the steps of the condition-based prediction and standardized residual construction method for variable load conditions of nuclear power plant feedwater systems.

[0066] The memory in this application embodiment can be volatile memory or non-volatile memory, or it can include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DRRAM). It should be noted that the memory used in the methods described in this invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0067] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state drives (SSDs)).

[0068] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0069] It should be noted that the processor in the embodiments of this application can be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method embodiments can be completed by the integrated logic circuitry in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads the information in the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0070] The above provides a detailed description of the condition-based prediction and standardized residual construction for variable load conditions in nuclear power plant feedwater systems proposed in this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for condition-based prediction and standardized residual construction for variable load conditions in nuclear power plant feedwater systems, characterized in that: Includes the following steps: Multivariable operating parameters of the nuclear power plant feedwater system under variable load conditions are collected, and the multivariable operating parameters are preprocessed to obtain a preprocessed parameter sequence. Based on the preprocessed parameter sequence, a working condition vector is constructed, and the parameter sequence is divided into transient phases according to the working condition vector to obtain a phase label sequence. The conditional probabilistic prediction model is trained based on historical operating parameters containing only normal variable load conditions. The conditional probabilistic prediction model can output the predicted mean and prediction uncertainty of the target parameters under the condition of historical parameter sequence, condition vector and phase label within a given preset time window. The preprocessed parameter sequence, operating condition vector, and phase label of the online running process are input into the operating condition-based probabilistic prediction model to obtain the predicted mean and prediction uncertainty of the target parameters, and the standardized residual sequence is calculated based on the predicted mean and prediction uncertainty. The standardized residual sequence is subjected to in-phase statistical calibration and online drift correction, and the output is a calibration residual sequence comparable across operating conditions, a prediction confidence interval, and / or a health index obtained by aggregating the calibration residual sequences.

2. The method for condition-based prediction and standardized residual construction of a nuclear power plant feedwater system under varying load conditions, as described in claim 1, is characterized in that: The data preprocessing of the multivariate operating parameters specifically includes: performing time alignment, outlier handling and normalization on the collected multivariate operating parameters, and constructing the input sample sequence using a sliding time window method.

3. The method for condition-based prediction and standardized residual construction of a nuclear power plant feedwater system under varying load conditions, as described in claim 2, is characterized in that: The data preprocessing also includes denoising, which includes at least one of moving average filtering or low-pass filtering.

4. The method for condition-based prediction and standardized residual construction of a nuclear power plant feedwater system under varying load conditions, as described in claim 1, is characterized in that: The operating condition vector includes at least power P and the rate of change of power. Among them, the rate of change of power It is calculated by power difference between adjacent sampling times.

5. The method for condition-based prediction and standardized residual construction of a nuclear power plant feedwater system under varying load conditions, as described in claim 4, is characterized in that: The transient phase division includes at least one of the following or a combination thereof: Based on the absolute value of the power change rate With preset threshold The parameter sequence is divided into steady-state phase and variable load phase; Based on power change rate The symbols subdivide the variable load phase into power-up phase and power-down phase; The phase boundary is corrected based on the power setpoint change point and / or control command change point.

6. The method for condition-based prediction and standardized residual construction of a nuclear power plant feedwater system under varying load conditions, as described in claim 1, is characterized in that: The standardized residual sequence consists of standardized residuals from multiple target parameter dimensions arranged in chronological order. For each time step and target parameter dimension, the expression for the standardized residual is: in, These are actual measured values. To predict the mean, To predict uncertainty The standard deviation.

7. The method for condition-based prediction and standardized residual construction of a nuclear power plant feedwater system under varying load conditions, as described in claim 1, is characterized in that: The intra-phase statistical calibration specifically includes: maintaining the running mean and running variance of the standardized residuals for different phases, and normalizing the standardized residuals by phase to obtain cross-phase comparable calibration residuals; The online drift correction specifically includes: updating the drift bias term based on the sliding window statistics or exponential moving average of the calibration residual within a time period that meets the preset normality conditions, and compensating for the predicted mean and / or calibration residual to suppress systematic deviations caused by sensor zero drift, model aging, or slow operating condition drift.

8. A device for predicting operating conditions and standardizing residuals for variable load conditions in nuclear power plant feedwater systems, characterized in that: include: The data acquisition and preprocessing module is used to acquire multivariable operating parameters of the nuclear power plant feedwater system under variable load conditions, and to preprocess the multivariable operating parameters to obtain a preprocessed parameter sequence. The working condition construction module is used to construct a working condition vector based on the preprocessed parameter sequence, and to perform transient phase division on the parameter sequence according to the working condition vector to obtain a phase label sequence. The conditional probability prediction module is used to train a conditional probability prediction model based on historical operating parameters containing only normal variable load conditions. This enables the conditional probability prediction model to output the predicted mean and prediction uncertainty of the target parameters within a given preset time window, given the historical parameter sequence, operating condition vector, and phase label. The standardized residual construction module is used to input the preprocessed parameter sequence, operating condition vector and phase label of the online operation into the operating condition probabilistic prediction model to obtain the predicted mean and prediction uncertainty of the target parameters, and to calculate the standardized residual sequence based on the predicted mean and prediction uncertainty. The drift calibration module is used to perform intra-phase statistical calibration and online drift correction on the standardized residual sequence, and output a calibration residual sequence comparable across operating conditions, a prediction confidence interval, and / or a health index obtained by aggregating the calibration residual sequences.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a condition-based prediction and standardized residual construction method for variable load conditions of nuclear power plant feedwater systems as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that enables the computer to execute a condition-based prediction and standardized residual construction method for variable load conditions of nuclear power plant feedwater systems, as described in any one of claims 1-7.

Images