Methods and devices for predicting trends in nuclear measurement parameters, electronic equipment and storage media

Abstract

This disclosure provides a method, device, electronic equipment, and storage medium for predicting trends of nuclear measurement parameters, relating to the field of nuclear power unit technology. It obtains a predicted deviation sequence by employing a high-fidelity mechanistic model combined with a data deviation proxy model, then corrects the initial internal parameter set using a sensitivity analysis matrix and a hierarchical inversion algorithm. Finally, the corrected parameters and current unit process parameters are input into an ultra-fast approximate calculation module to complete the trend extrapolation of nuclear measurement parameters. Therefore, it solves the problems in existing technologies where key internal parameters are difficult to obtain accurately in real time, mechanistic models have initial state deviations that accumulate and amplify with increasing prediction time, and high-fidelity mechanistic model extrapolation calculations are inefficient and cannot meet the real-time monitoring requirements of nuclear power plants. This achieves high-precision and high-reliability trend prediction of nuclear measurement parameters for key equipment in nuclear power units, suppresses the cumulative error of long-term predictions, and meets the stringent requirements of real-time monitoring and decision-making in nuclear power plants.

Description

Technical Field

[0001] This disclosure relates to the field of nuclear power unit technology, and in particular to a method and apparatus for predicting the trend of nuclear measurement parameters, electronic equipment and storage medium. Background Technology

[0002] As a core pillar of the clean energy system, the safe and stable operation of nuclear power units relies on accurate trend prediction of nuclear measurement parameters such as neutron flux density. Among related technologies, a complete technical system has been constructed, encompassing core state perception and parameter extrapolation, through the coordinated operation of high-fidelity physical mechanism models, thermal-hydraulic equations, and neutron dynamics theory. Specifically, this system covers the entire process from initial internal parameter setting and boundary condition input to transient response calculation, including key aspects such as fuel burnup assessment, poison concentration distribution estimation, and control rod value correction.

[0003] However, related technologies directly use theoretical calculations or historical experience to set key internal state parameters such as the concentration of fission product poisons, without establishing an online dynamic correction mechanism based on real-time operational data. Because these parameters are difficult to measure directly and change drastically over time, the initial state of the model has inherent biases. As the prediction period increases, the errors accumulate and amplify during iterative calculations, severely impacting the accuracy of long-term trend predictions and the reliability of unit safety decisions. Summary of the Invention

[0004] This disclosure provides a method, apparatus, electronic device, and storage medium for predicting trends in nuclear measurement parameters. Its main objective is to at least partially address one of the technical problems in the related art.

[0005] According to a first aspect of this disclosure, a method for predicting trends in nuclear measurement parameters is provided, comprising:

[0006] The key parameter set of the unit is obtained, and the historical data in it is processed by a high-fidelity mechanism model to obtain the historical nuclear measurement prediction parameters. Based on the historical nuclear measurement prediction parameters and the corresponding historical nuclear measurement actual parameters, the historical deviation sequence is calculated. The key parameter set of the unit includes the unit process parameters, the initial internal parameter set and the actual nuclear measurement parameters. The training process involves using a set of key parameters of historical generating units as input features and a historical deviation sequence as the learning objective to obtain a trained data deviation proxy model. The current set of key parameters of the unit is input into the trained data deviation surrogate model to obtain the predicted deviation sequence. Based on the predicted deviation sequence and the sensitivity analysis matrix, the estimated deviation of the initial internal parameter set is inverted and corrected through the hierarchical inversion algorithm to obtain the corrected internal parameter set. The corrected internal parameter set and the current unit process parameters are input into the ultra-fast approximate calculation module for extrapolation and calculation to obtain the final trend prediction results of the nuclear measurement parameters.

[0007] According to a second aspect of this disclosure, a nuclear measurement parameter trend prediction device is provided, comprising: The acquisition unit is used to acquire the set of key parameters of the unit, process the historical data in it using a high-fidelity mechanism model to obtain historical nuclear measurement prediction parameters, and calculate the historical deviation sequence based on the historical nuclear measurement prediction parameters and the corresponding historical nuclear measurement actual parameters. The set of key parameters of the unit includes the unit process parameters, the initial internal parameter set and the actual nuclear measurement parameters. The training unit is used to train a data deviation proxy model by using the set of key parameters of historical units as input features and the historical deviation sequence as the learning target. The correction unit is used to input the current key parameter set of the unit into the trained data deviation surrogate model to obtain the predicted deviation sequence. Based on the predicted deviation sequence and the sensitivity analysis matrix, the estimated deviation of the initial internal parameter set is inverted and corrected through the hierarchical inversion algorithm to obtain the corrected internal parameter set. The calculation unit is used to input the corrected internal parameter set and the current unit process parameters into the ultra-fast approximate calculation module for extrapolation and calculation, and to obtain the final trend prediction results of the nuclear measurement parameters.

[0008] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.

[0009] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.

[0010] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0011] The nuclear measurement parameter trend prediction method, device, electronic equipment, and storage medium disclosed herein obtain a prediction deviation sequence by employing a high-fidelity mechanism model combined with a data deviation proxy model. Then, relying on a sensitivity analysis matrix and a hierarchical inversion algorithm, the initial internal parameter set is corrected. Finally, the corrected parameters and the current unit process parameters are input into an ultra-fast approximate calculation module to complete the nuclear measurement parameter trend extrapolation. Therefore, it can solve the problems in existing technologies where key internal parameters are difficult to obtain accurately in real time, mechanism models have initial state deviations that accumulate and amplify with increasing prediction time, and high-fidelity mechanism model extrapolation calculation efficiency is low and cannot meet the real-time monitoring requirements of nuclear power plants. This achieves high-precision and high-reliability trend prediction of nuclear measurement parameters for key equipment in nuclear power units, fundamentally suppressing the cumulative error of long-term predictions, while improving prediction calculation speed, thus meeting the stringent technical requirements of real-time monitoring and decision-making in nuclear power plants.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0013] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 A flowchart illustrating a method for predicting the trend of nuclear measurement parameters provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram of the structure of a nuclear measurement parameter trend prediction device provided in an embodiment of the present disclosure; Figure 3 A schematic block diagram of an example electronic device provided for embodiments of this disclosure. Detailed Implementation

[0014] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0015] The following description, with reference to the accompanying drawings, outlines a method and apparatus for predicting trends in nuclear measurement parameters, an electronic device, and a storage medium according to embodiments of the present disclosure.

[0016] Figure 1 This is a flowchart illustrating a method for predicting the trend of nuclear measurement parameters provided in an embodiment of this disclosure.

[0017] like Figure 1As shown, the method includes the following steps: Step 101: Obtain the set of key parameters of the unit, process the historical data in it using a high-fidelity mechanism model to obtain the historical nuclear measurement prediction parameters, and calculate the historical deviation sequence based on the historical nuclear measurement prediction parameters and the corresponding historical nuclear measurement actual parameters. The set of key parameters of the unit includes the unit process parameters, the initial internal parameter set and the nuclear measurement actual parameters.

[0018] In the embodiments of this disclosure, a historical deviation sequence is constructed to quantify the prediction error characteristics of the high-fidelity mechanism model under specific operating conditions, providing a data foundation for subsequent model correction. A set of key unit parameters reflecting the real-time operating state of the reactor is acquired synchronously, and the corresponding nuclear measurement prediction parameters are derived using a pre-trained high-fidelity mechanism model. Then, a comparative analysis is conducted to establish the deviation mapping relationship between historical prediction values ​​and actual measurement values. The set of key unit parameters, as a multi-dimensional state vector, encompasses unit process parameters describing macroscopic operating conditions, an initial set of internal parameters characterizing the microscopic physical state of the reactor core, and calibrated actual nuclear measurement parameters. During execution, firstly, multi-source heterogeneous data undergoes time alignment and preprocessing to eliminate dimensional differences and noise interference; subsequently, the process parameters and internal states at historical moments are input into the mechanism model to generate a theoretical prediction trajectory; finally, under a unified spatiotemporal reference, the difference sequence between the predicted trajectory and the actual observed trajectory is calculated, and the statistical or frequency domain features of this sequence can be further extracted to form a standardized historical deviation sequence. As one implementation method, the long-term series can be divided into data windows of fixed length. Instantaneous relative deviations can be calculated within each window, and statistical quantities such as mean and standard deviation can be extracted. These can then be bound to the corresponding average operating parameters to establish a correlation dataset between operating conditions and model deviation characteristics.

[0019] Through the above steps, the inherent biases of the mechanistic model caused by the uncertainty of its internal state parameters can be systematically mined and quantified, transforming abstract model errors into learnable and quantifiable data features. This not only provides high-quality supervised learning labels for subsequent training of biased surrogate models, but also fundamentally solves the problem of error accumulation and amplification caused by inaccurate initial states when relying solely on the mechanistic model, ensuring that the trend prediction method has the ability to self-correct online and a high-precision foundation.

[0020] Step 102: Train the model using the set of key parameters of the historical unit as input features and the historical deviation sequence as the learning target to obtain the trained data deviation proxy model.

[0021] In the embodiments of this disclosure, the training data deviation proxy model aims to establish a nonlinear mapping relationship between the unit's operating state and the prediction error of the mechanistic model. Its core lies in constructing a proxy model capable of generalizing prediction model deviations using a historical data-driven approach. By selecting a set of historical unit key parameters covering different operating conditions as the input feature space, and using the historical deviation sequence calculated at the corresponding time as the learning target for supervised learning, machine learning algorithms are employed to iteratively optimize the model parameters. This enables the trained data deviation proxy model to infer future prediction deviations based on current input features. In this process, the construction of input features may include physical enhancement processing of the original process parameters and internal state parameters to incorporate prior physical knowledge and improve feature representation capabilities. The choice of model architecture has broad adaptability, and various neural network structures or regression models capable of handling high-dimensional nonlinear relationships can be adopted. As a specific implementation, an enhanced input feature vector containing physically derived features and regionalized state features can be constructed based on the historical unit key parameter set. A data deviation proxy model is then constructed using a cascaded dual-path network architecture, which includes a shared base network, a scenario branch network, and a physical consistency verification branch. A composite loss function is used for optimization training. By performing this step, the systematic deviation patterns of the high-fidelity mechanism model under different operating scenarios can be effectively captured, providing a high-precision deviation prediction basis for subsequent online real-time correction of internal parameters. Thus, without changing the physical structure of the original mechanism model, the accuracy and robustness of the trend prediction of nuclear measurement parameters can be significantly improved.

[0022] By using a data-driven approach, the prediction errors caused by the uncertainty of internal state parameters in pure mechanistic models are compensated for, enabling rapid and accurate prediction of model biases and laying a solid foundation for subsequent parameter inversion and correction.

[0023] Step 103: Input the current key parameter set of the unit into the trained data deviation surrogate model to obtain the predicted deviation sequence. Based on the predicted deviation sequence and the sensitivity analysis matrix, the estimated deviation of the initial internal parameter set is inverted and corrected through the hierarchical inversion algorithm to obtain the corrected internal parameter set.

[0024] In the embodiments of this disclosure, the current set of key unit parameters is input into a trained data deviation proxy model. The model's internal feature mapping mechanism outputs a prediction deviation sequence characterizing the deviation trend between predicted and actual values ​​of nuclear measurement parameters at future moments. The core of this step lies in using the prediction deviation sequence to drive online correction of the internal state of the mechanistic model. Specifically, this is achieved by constructing a sensitivity analysis matrix describing the dynamic correlation between internal parameter disturbances and prediction deviation responses, combined with a hierarchical inversion algorithm to solve for the estimated deviation of the initial internal parameter set. Based on the characteristics of the prediction deviation sequence, the hierarchical inversion algorithm inversely deduces the root cause of model state mismatch in a multi-dimensional parameter space, calculates the correction amount for the initial internal parameter set, and then updates the initial internal parameter set to a corrected internal parameter set with more accurate physical meaning and better reflecting the real-time state of the reactor core. As a specific implementation method, the current set of key parameters of the unit can first be standardized and have its physical features enhanced in the same way as in the training phase. An enhanced input feature vector is constructed and input into the data deviation proxy model. After extracting features through a shared basic network, the predicted deviation sequence is obtained through weighted integration by a scenario branch network. Subsequently, a sensitivity analysis matrix is ​​constructed based on the similarity between historical typical operating conditions and current operating conditions. A hierarchical weighted inversion strategy with two stages, global coarse adjustment and local fine adjustment, is adopted to construct and solve the objective function with regularization terms to obtain the relative correction vector of key internal parameters. After boundary clipping, the corrected set of internal parameters is obtained.

[0025] By deeply integrating data-driven deviation prediction with physical mechanism inversion, online automatic correction of internal state parameters of key equipment in nuclear power units has been achieved, effectively eliminating model accumulation errors caused by inaccurate initial state estimation and significantly improving the long-term accuracy and physical consistency of nuclear measurement parameter trend prediction.

[0026] Step 104: Input the corrected internal parameter set and the current unit process parameters into the ultra-fast approximate calculation module for extrapolation and calculation to obtain the final trend prediction results of the nuclear measurement parameters.

[0027] In the embodiments of this disclosure, a fast computational model capable of high-fidelity mapping of the nonlinear relationship between input state and output response is driven by the system initial conditions after physical state correction, replacing the online iterative computation of traditional high-consumption mechanism models. This step constructs a data-driven approximate computational architecture that receives internal physical state variables corrected by a hierarchical inversion algorithm and real-time acquired unit operating boundary conditions, completing the deduction of the evolution trajectory of nuclear measurement parameters at multiple future time scales within an extremely short time scale. This fast computational module is essentially a function approximator trained with a specific sample library. The generation logic of its training samples maintains physical consistency with the aforementioned parameter inversion process, ensuring that the model can accurately respond to the dynamic impact of parameter corrections of different magnitudes on the system output. As a specific implementation, a deep feedforward neural network can be used to construct this ultra-fast approximate computational module. A gated feature fusion architecture is used to extract baseline state features and correction mode features respectively, and the feature fusion weights are dynamically adjusted according to the norm of the parameter correction amount, thereby mapping the corrected internal parameter set and the current unit process parameters into a future trend prediction sequence. This step significantly reduces computational complexity while ensuring that the prediction results have clear physical meaning and consistency. It compresses the mechanism model deduction process, which originally required minutes or even hours, to the millisecond level, effectively meeting the stringent timeliness requirements of real-time monitoring and rapid decision-making for nuclear power units, while avoiding the loss of accuracy caused by model simplification.

[0028] The nuclear measurement parameter trend prediction method disclosed herein obtains the prediction deviation sequence by using a high-fidelity mechanism model combined with a data deviation proxy model. Then, it corrects the initial internal parameter set by relying on a sensitivity analysis matrix and a hierarchical inversion algorithm. Finally, it inputs the corrected parameters and the current unit process parameters into an ultra-fast approximate calculation module to complete the nuclear measurement parameter trend extrapolation. Therefore, it can solve the problems in the prior art where key internal parameters are difficult to obtain accurately in real time, mechanism models have initial state deviations and errors accumulate and amplify with the increase of prediction time, and high-fidelity mechanism model extrapolation calculation efficiency cannot meet the real-time monitoring requirements of nuclear power plants. It achieves high-precision and high-reliability trend prediction of nuclear measurement parameters of key equipment in nuclear power units, fundamentally suppresses the cumulative error of long-term prediction, and improves the prediction calculation speed, thus meeting the stringent technical requirements of real-time monitoring and decision-making in nuclear power plants.

[0029] In the embodiments involved in this application, there are various feasible specific implementation methods. To clearly and completely illustrate the technical solutions of this disclosure, the implementation methods listed below are merely exemplary and do not constitute a limitation on the scope of protection of this disclosure. That is, in addition to the implementation methods described below, other implementation methods that can be obtained by those skilled in the art based on the technical content disclosed in this disclosure through reasonable logical analysis, reasoning, or limited experimentation should also be covered within the scope of protection of this disclosure. The following specifically describes some exemplary implementation methods: As a specific implementation of this disclosure, based on the basic scheme, a high-fidelity mechanism model is used to process the historical data to obtain historical nuclear measurement prediction parameters. Based on the historical nuclear measurement prediction parameters and the corresponding historical nuclear measurement actual parameters, a historical deviation sequence is calculated. This is further defined as follows: the unit process parameters and initial internal parameter set at each historical moment are input into the high-fidelity mechanism model to obtain nuclear measurement prediction parameters, and these parameters are aligned with the corresponding nuclear measurement actual parameters; the aligned sequence is divided into multiple continuous data windows, and deviation features are calculated in each window to generate a multi-dimensional deviation feature vector; the multi-dimensional deviation feature vector is associated with the operating status features of the corresponding window to form a historical deviation sequence.

[0030] Specifically, the unit process parameters (such as pressurizer water level and main pump speed) and initial internal parameter sets (such as the equivalent concentration distribution of fission product poisons) at each historical moment are first input into a pre-trained high-fidelity mechanism model (a three-dimensional neutron dynamics coupled thermal-hydraulic core physics model) to obtain the corresponding nuclear measurement prediction parameters. Then, the nuclear measurement prediction parameters are fully aligned with the corresponding actual nuclear measurement parameters in terms of physical meaning, data dimension, and timestamp to ensure data matching consistency. Next, the aligned time series is divided into multiple continuous data windows of fixed time length. Within each window, the instantaneous relative deviation is calculated for single-dimensional parameters. For multi-dimensional parameters such as the core axial height distribution, the deviation is calculated independently according to each spatial location to form a deviation distribution vector. Then, the mean, standard deviation, kurtosis and skewness, and specific frequency band energy are extracted from the deviation sequence / vector within the window to generate a multi-dimensional deviation feature vector. Finally, this vector is associated and bound with the average unit process parameters such as the average power level and average coolant temperature of the corresponding window, as well as the initial internal parameter set change characteristics such as fuel consumption increment, and integrated along the time axis to form a historical deviation sequence. Optionally, the data window can be set to a variable duration window, and the trend slope can also be added as a supplementary dimension for the deviation feature.

[0031] The triple alignment of physical meaning, dimension, and timestamp eliminates systematic errors in data matching. Fixed-window feature extraction avoids random interference from single-time-point deviations. The association and binding of deviation features with unit and core status features establishes a physical mapping relationship between operating status and model deviations. This provides physically meaningful samples for subsequent training of data deviation proxy models, improving the model's targeted learning of deviation patterns.

[0032] As a specific implementation of this disclosure, based on the basic scheme, a trained data deviation proxy model is obtained by using a set of historical unit key parameters as input features and a historical deviation sequence as the learning objective. This model is further defined as follows: an enhanced input feature vector is constructed based on the set of historical unit key parameters. This enhanced input feature vector includes features derived from physical knowledge and features after dimensionality reduction of high-dimensional parameters. A data-deviation proxy model is constructed through a multi-path network architecture, and the model is trained using shared feature extraction, scenario-adaptive weighting, and physical consistency constraints to obtain the trained data deviation proxy model.

[0033] Specifically, an enhanced input feature vector is first constructed based on the historical unit key parameter set: extract the historical unit key parameter set and nuclear measurement prediction parameters corresponding to each historical time window, and calculate the physical derived features from the original features by combining nuclear physics prior knowledge, such as the instantaneous deviation of the physical expected value of the nuclear measurement prediction parameters from the key values ​​of the unit process parameters; for high-dimensional spatial distribution parameters such as the toxic substance concentration field, divide the core into physical regions (upper / lower part, center / periphery), calculate the mean and variance of each region to generate low-dimensional regionalized state features, and concatenate the physical derived features, regionalized state features and standardized original features to form an enhanced input feature vector. The model is then constructed using a cascaded multi-path network architecture: the enhanced feature vectors are used to extract shared features through a shared base network, and a lightweight scenario discriminant meta-model outputs weight vectors for matching the physical scenarios. The prediction results of each scenario-specific branch are adaptively weighted. Simultaneously, the enhanced feature vectors and the predicted values ​​of the bias features are input into the physical consistency verification branch to calculate the virtual corrected physical quantity sequence. This sequence is then substituted into the built-in simplified physical model to derive the theoretical response value, which is compared with the actual measured value to obtain the physical residual vector. The model training is then completed using a composite loss function. Optionally, the physical derived features can include theoretical reactive change rates estimated based on power and temperature, and the shared base network can also be a lightweight convolutional neural network.

[0034] The construction of physical derivative features and regionalized state features not only enriches the physical representation capability of input features but also eliminates redundant information from high-dimensional parameters. The shared extraction of multi-path networks, scenario weighting, and physical consistency constraint design allow the model to adapt to the deviation patterns of different operating conditions of the unit, while avoiding the prediction results from deviating from the basic laws of nuclear physics, which greatly improves the accuracy and physical reliability of the deviation prediction of the data-deviation proxy model.

[0035] As a specific implementation of this disclosure, based on the basic scheme, the estimated deviation of the initial internal parameter set is inverted and corrected by combining the prediction deviation sequence with the sensitivity analysis matrix through a hierarchical inversion algorithm. The method is further defined as follows: constructing a sensitivity analysis matrix, which is used to characterize the degree of influence of changes in each internal parameter on the prediction deviation at future time, and weighting and fusing the sensitivity data of historical working conditions according to the current working conditions; executing a hierarchical inversion algorithm based on the prediction deviation sequence and the sensitivity analysis matrix, first performing coarse adjustment of global parameters, and then performing local fine adjustment of key parameters to obtain the corrected internal parameter set.

[0036] Specifically, a sensitivity analysis matrix is ​​constructed: M key internal parameters, such as fission product poison concentration and core burnup depth, are selected, along with L typical historical operating conditions covering reactor startup, full power, and power reduction. A small relative perturbation of ε=0.01 is applied to each internal parameter under each operating condition, and the change in prediction deviation at each future time point is calculated to obtain the single-condition sensitivity. Then, the similarity weight between the current operating condition and each historical operating condition is calculated using a Gaussian kernel function. All single-condition sensitivity data are then weighted and fused to obtain a sensitivity analysis matrix adapted to the current operating state, whose elements... This characterizes the degree of influence of the unit relative change of the j-th internal parameter on the prediction bias at time i. Subsequently, a hierarchical inversion algorithm is executed: in the global coarse-tuning stage, a diagonal weight matrix is ​​constructed. The objective function, with a regularization coefficient α (α can take values ​​but is not limited to 0.05), is used to solve the linear inversion model via the conjugate gradient iteration method. This yields a preliminary parameter correction vector, identifying the K key parameters with the largest absolute values. In the local refinement stage, the corresponding simplified sensitivity sub-matrix is ​​extracted, constructing a second objective function containing a physical smoothing constraint matrix Γ and a regularization coefficient β (β can take values ​​but is not limited to 0.02). The Cholesky decomposition method is used to obtain the refinement correction amount, which is assigned to the corresponding positions in the complete correction vector, and the remaining positions are set to zero. After calculating the absolute correction amount, physical boundary pruning is performed on the parameters to obtain the corrected internal parameter set. Optionally, the relative perturbation ε can be adjusted within the range of 0.005-0.02, and the operating condition similarity weight can also be calculated using a cosine similarity algorithm.

[0037] Weighted fusion based on operating condition similarity enables the sensitivity analysis matrix to adapt to real-time operating conditions, avoiding inversion bias caused by a fixed matrix; the global coarse adjustment and local fine adjustment of the hierarchical inversion not only ensure the global rationality of parameter correction, but also focus on key parameters to improve inversion accuracy, while physical boundary pruning ensures that the corrected internal parameters conform to the basic laws of nuclear physics and eliminates the appearance of parameter values ​​without physical meaning.

[0038] As a specific implementation of this disclosure, based on the basic scheme, a sensitivity analysis matrix is ​​constructed, which is further defined as follows: for selected key internal parameters and typical working conditions, the dynamic sensitivity under each working condition is calculated by applying parameter perturbation; the weights are determined according to the similarity between the current working condition and each historical typical working condition, and the dynamic sensitivity under each working condition is weighted and summed to obtain a sensitivity analysis matrix with working condition adaptability.

[0039] Specifically, firstly, M key internal parameters to be inverted and L historical typical operating conditions covering the typical operating range of the units are selected. Each operating condition is defined by a specific initial set of internal parameters and boundary conditions; for each operating condition, OP... l and each intrinsic parameter θ j First, run a high-fidelity mechanism model to obtain the baseline prediction value P for the next N time points. mb (t i ), and then for θ j Applying a small relative perturbation ε yields θ j =θ j (1+ε), with other parameters remaining unchanged, the model is run under the same operating conditions to obtain the predicted value P after disturbance. mp (t i ), through the formula Δδ l (t i ,θ j )=[(P mp (t i ) P mb (t i )) / P mb (t i The dynamic sensitivity under this operating condition is calculated using )] / ε. Then, the current operating condition OP is extracted. now The key process parameters of each historical typical operating condition, after standardization, are combined with the parameter importance weight V. k The comprehensive difference measure between operating conditions is calculated by using a Gaussian kernel function to solve for the similarity weight ω(OP) of the current operating condition to each historical operating condition. l |OP now ); finally, according to the formula The dynamic sensitivity of each operating condition is weighted and summed to obtain a sensitivity analysis matrix with operating condition adaptability. Optionally, the small relative disturbance ε can be selected as a fixed value in the range of 0.005-0.02, and the operating condition similarity can also be calculated using the cosine similarity algorithm.

[0040] The dynamic sensitivity is calculated by applying small perturbations to a single parameter independently, ensuring the linearity and accuracy of the sensitivity value and eliminating the calculation error caused by multi-parameter coupling. The weighted summation based on the similarity of operating conditions allows the sensitivity analysis matrix to accurately match the real-time operating status of the unit, avoiding the adaptation deviation of fixed sensitivity data when the operating conditions change, and providing a realistic quantitative basis for the parameter influence of the hierarchical inversion algorithm.

[0041] As a specific implementation of this disclosure, based on the basic scheme, a hierarchical inversion algorithm is executed based on the prediction deviation sequence and the sensitivity analysis matrix, which is further defined as follows: in the global coarse adjustment stage, the preliminary parameter correction vector is solved based on the sensitivity analysis matrix; the key components in the preliminary parameter correction vector are identified, and a simplified sensitivity submatrix is ​​constructed; in the local fine adjustment stage, the fine correction amount of the key parameters is solved based on the simplified sensitivity submatrix, and the corrected internal parameter set is obtained by combining it with the current initial internal parameter set.

[0042] Specifically, the first step is to enter the global coarse adjustment phase to predict the bias sequence δ. pred Using the sensitivity analysis matrix S as the core input, a diagonal weight matrix W is constructed. δ The first objective function J1(Δθ) consists of the prior constraint matrix Λ of the parameters and the first regularization coefficient α. rel )=(δ pred S Δθ rel ) T W δ (δ pred S Δθ rel )+α ΔθrelT Λ Δθrel, the first normal equation derived by solving the objective function with zero partial derivatives is obtained by using the conjugate gradient iteration method. T W δ S+αΛ)Δθ rels =STW δ δ pred This yields the initial parameter relative correction vector Δθ in M ​​dimensions. rel。Next, calculate the absolute values of the components of the preliminary parameter relative correction vector and sort them, automatically identify the top K (K << M) components with the largest absolute values, record their index set Ω, and extract the columns corresponding to the index set Ω from the original sensitivity analysis matrix S to construct a reduced sensitivity sub-matrix Ssub of dimension N×K. Finally, enter the local refinement stage, using the reduced sensitivity sub-matrix S sub , the prediction deviation sequence δ pred as the input, construct a second objective function that includes a physical smoothing constraint matrix Γ and a second regularization coefficient β , use the Cholesky decomposition method to solve the derived second normal equation to obtain a K-dimensional key parameter fine relative correction vector; assign this fine correction vector to the positions corresponding to the indices Ω of the complete correction vector, and set the remaining positions to zero to obtain the final complete parameter relative correction vector. Combine it with the current initial internal parameter set θ init to calculate the parameter absolute correction amount Δθ = θ init θ relz , and then according to θ corrected = θ init +Δθ to obtain the preliminary corrected parameters, and after physically bounding θ correctedj = max(θ minj , min(θ correctedj , θ maxj )), obtain the corrected internal parameter set. Optionally, the quasi-Newton method can be used to solve the normal equation in the global rough adjustment stage, and the QR decomposition method can also be used to complete the solution of the fine correction amount in the local refinement stage.

[0043] As a specific implementation form of the present disclosure, based on the basic scheme, input the corrected internal parameter set and the current unit process parameters into the ultra-fast approximate calculation module for deduction calculation, which is further limited to: construct an ultra-fast approximate calculation module, the ultra-fast approximate calculation module adopts a deep neural network architecture, adaptively fuse the reference state features and correction mode features through a gated fusion mechanism, and output the predicted value of the nuclear measurement parameter at the future moment; input the corrected internal parameter set and the current unit process parameters into the ultra-fast approximate calculation module to obtain the final trend prediction result of the nuclear measurement parameter.

[0044] Specifically, an ultra-fast approximation calculation module is constructed. This module adopts a deep feedforward neural network architecture. The input layer receives the normalized current unit process parameters, nominal internal parameter set, and absolute parameter corrections in three paths. The module has a dual-branch feature extraction structure. One path extracts baseline state features through two fully connected layers (with 128 and 64 neurons respectively), and the other path extracts correction mode features through three fully connected layers (with 256, 128, and 64 neurons respectively). At the same time, a gated fusion mechanism is introduced. The L2 norm of the absolute parameter correction is calculated to represent the correction magnitude. Based on this magnitude, a fusion weight in the 0-1 interval is dynamically generated using the Sigmoid function, and the two types of features are adaptively weighted and fused. The fused features are then processed by a nonlinear mapping calculation through four fully connected layers to directly output the predicted sequence of kernel measurement parameters for multiple time steps in the next 10 / 30 / 60 minutes, thus completing the module construction. Subsequently, the absolute correction amount between the corrected internal parameter set and the initial internal parameter set is calculated. The normalized current unit process parameters, nominal internal parameter set, and this absolute correction amount are input into the trained module. After forward propagation through the network, a sequence of predicted values ​​is obtained. Then, denormalization is performed to restore the values ​​to physical units, yielding the final trend prediction result of the nuclear measurement parameters. Optionally, a lightweight convolutional neural network architecture can be used for the deep neural network, and the weight generation for gated fusion can use the Softmax function. The time step length can be flexibly configured according to the monitoring requirements of the nuclear power plant.

[0045] The gating fusion mechanism dynamically adjusts the feature fusion ratio according to the parameter correction magnitude, enabling the module to have accurate modeling capabilities for both small and large correction scenarios, avoiding prediction bias caused by a single fusion method; the lightweight architecture design of the deep neural network enables millisecond-level rapid extrapolation of nuclear measurement parameters, significantly improving computational efficiency while ensuring prediction accuracy, and fully matching the stringent timing requirements of real-time monitoring of nuclear power plants.

[0046] It should be noted that the embodiments of this disclosure may include multiple steps. For ease of description, these steps are numbered, but these numbers are not a limitation on the execution time slots or execution order between the steps; these steps can be implemented in any order, and the embodiments of this disclosure do not limit this.

[0047] Corresponding to the aforementioned method for predicting trends in nuclear measurement parameters, this disclosure also proposes a device for predicting trends in nuclear measurement parameters. Since the device embodiments of this disclosure correspond to the method embodiments described above, details not disclosed in the device embodiments can be referred to the method embodiments described above, and will not be repeated here.

[0048] Figure 2 This is a schematic diagram of the structure of a nuclear measurement parameter trend prediction device provided in an embodiment of the present disclosure, as shown below. Figure 2 As shown, it includes: The acquisition unit 21 is used to acquire the set of key parameters of the unit, process the historical data therein using a high-fidelity mechanism model to obtain historical nuclear measurement prediction parameters, and calculate the historical deviation sequence based on the historical nuclear measurement prediction parameters and the corresponding historical nuclear measurement actual parameters. The set of key parameters of the unit includes the unit process parameters, the initial internal parameter set and the nuclear measurement actual parameters. Training unit 22 is used to train a data deviation proxy model by using the set of key parameters of historical units as input features and the historical deviation sequence as the learning target. Correction unit 23 is used to input the current key parameter set of the unit into the trained data deviation proxy model to obtain the predicted deviation sequence. Based on the predicted deviation sequence and the sensitivity analysis matrix, the estimated deviation of the initial internal parameter set is inverted and corrected through the hierarchical inversion algorithm to obtain the corrected internal parameter set. The calculation unit 24 is used to input the corrected internal parameter set and the current unit process parameters into the ultra-fast approximate calculation module for extrapolation calculation to obtain the final trend prediction results of the nuclear measurement parameters.

[0049] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of this embodiment, and the principle is the same, so it is not limited in this embodiment.

[0050] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0051] Figure 3 A schematic block diagram of an example electronic device 300 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0052] like Figure 3As shown, the electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 302 or a computer program loaded from storage unit 308 into RAM (Random Access Memory) 303. The RAM 303 may also store various programs and data required for the operation of the electronic device 300. The computing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An I / O (Input / Output) interface 305 is also connected to the bus 304.

[0053] Multiple components in electronic device 300 are connected to I / O interface 305, including: input unit 306, such as keyboard, mouse, etc.; output unit 307, such as various types of displays, speakers, etc.; storage unit 308, such as disk, optical disk, etc.; and communication unit 309, such as network card, modem, wireless transceiver, etc. Communication unit 309 allows electronic device 300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0054] The computing unit 301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above, such as the nuclear measurement parameter trend prediction method. For example, in some embodiments, the nuclear measurement parameter trend prediction method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 300 via ROM 302 and / or communication unit 309. When the computer program is loaded into RAM 303 and executed by the computing unit 301, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 301 may be configured to perform the aforementioned nuclear measurement parameter trend prediction method by any other suitable means (e.g., by means of firmware).

[0055] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0056] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0057] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0058] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0059] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0060] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0061] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0062] The various numerical designations such as "first," "second," etc., used in this disclosure are merely for ease of description and are not intended to limit the scope of the embodiments of this disclosure, nor do they indicate a sequential order.

[0063] At least one of the features described in this disclosure can also be described as one or more, and multiple features can be two, three, four or more, and this disclosure does not impose any limitations. In the embodiments of this disclosure, for a technical feature, the technical features in that technical feature are distinguished by "first", "second", "third", "A", "B", "C" and "D", etc., and there is no sequential order or size order among the technical features described by "first", "second", "third", "A", "B", "C" and "D".

[0064] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0065] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for predicting the trend of nuclear measurement parameters, characterized in that, include: A set of key parameters of the unit is obtained, and historical data in the set is processed using a high-fidelity mechanism model to obtain historical nuclear measurement prediction parameters. Based on the historical nuclear measurement prediction parameters and the corresponding historical nuclear measurement actual parameters, a historical deviation sequence is calculated. The set of key parameters of the unit includes unit process parameters, initial internal parameter set and nuclear measurement actual parameters. The historical unit key parameter set is used as input features and the historical deviation sequence is used as the learning target to train a data deviation proxy model that has been trained. The current set of key parameters of the unit is input into the trained data deviation proxy model to obtain the predicted deviation sequence. Based on the predicted deviation sequence and the sensitivity analysis matrix, the estimated deviation of the initial internal parameter set is inverted and corrected through the hierarchical inversion algorithm to obtain the corrected internal parameter set. The modified internal parameter set and the current unit process parameters are input into the ultra-fast approximate calculation module for extrapolation and calculation to obtain the final trend prediction results of the nuclear measurement parameters.

2. The method according to claim 1, characterized in that, The process of acquiring the key parameter set of the unit, processing historical data within it using a high-fidelity mechanism model to obtain historical nuclear measurement prediction parameters, and calculating the historical deviation sequence based on the historical nuclear measurement prediction parameters and the corresponding historical nuclear measurement actual parameters includes: The unit process parameters and initial internal parameter set at each historical moment are input into the high-fidelity mechanism model to obtain the nuclear measurement prediction parameters, and then aligned with the corresponding actual nuclear measurement parameters. The aligned sequence is divided into multiple consecutive data windows, and the deviation features are calculated in each window to generate a multidimensional deviation feature vector. The multidimensional deviation feature vector is associated with the running status features of the corresponding window to form a historical deviation sequence.

3. The method according to claim 1, characterized in that, The process of training a data deviation proxy model using the historical unit key parameter set as input features and the historical deviation sequence as the learning target, resulting in the trained model, includes: An enhanced input feature vector is constructed based on a set of key parameters of historical generating units. The enhanced input feature vector includes features derived from physical knowledge and features after dimensionality reduction of high-dimensional parameters. A data-bias proxy model is constructed using a multi-path network architecture. The model is trained by using shared feature extraction, context-adaptive weighting, and physical consistency constraints.

4. The method according to claim 1, characterized in that, The step of inverting and correcting the estimation bias of the initial internal parameter set based on the predicted bias sequence combined with the sensitivity analysis matrix using a hierarchical inversion algorithm includes: A sensitivity analysis matrix is ​​constructed to characterize the degree of influence of changes in each internal parameter on the prediction deviation at future times, and the sensitivity data of historical working conditions are weighted and fused according to the current working conditions. Based on the predicted deviation sequence and sensitivity analysis matrix, a hierarchical inversion algorithm is executed. First, global parameters are coarsely adjusted, and then key parameters are finely adjusted locally to obtain the corrected internal parameter set.

5. The method according to claim 4, characterized in that, The construction of the sensitivity analysis matrix includes: For the selected key internal parameters and typical operating conditions, the dynamic sensitivity under each operating condition is calculated by applying parameter perturbations. The weights are determined based on the similarity between the current operating conditions and typical historical operating conditions. The dynamic sensitivity under each operating condition is then weighted and summed to obtain a sensitivity analysis matrix that is adaptable to different operating conditions.

6. The method according to claim 4, characterized in that, The hierarchical inversion algorithm based on the predicted deviation sequence and sensitivity analysis matrix includes: In the global coarse-tuning stage, the preliminary parameter correction vector is solved based on the sensitivity analysis matrix; Identify the key components in the preliminary parameter correction vector and construct a simplified sensitivity submatrix; During the local refinement stage, the fine correction amount of the key parameters is solved based on the simplified sensitivity submatrix, and the corrected internal parameter set is obtained by combining the current initial internal parameter set.

7. The method according to claim 1, characterized in that, The step of inputting the corrected internal parameter set and the current unit process parameters into the ultra-fast approximate calculation module for extrapolation calculation includes: An ultrafast approximation calculation module is constructed. The ultrafast approximation calculation module adopts a deep neural network architecture and adaptively fuses the baseline state features and the correction mode features through a gating fusion mechanism to output the predicted values ​​of the kernel measurement parameters at future times. The corrected internal parameter set and the current unit process parameters are input into the ultra-fast approximate calculation module to obtain the final trend prediction results of the nuclear measurement parameters.

8. A device for predicting trends in nuclear measurement parameters, characterized in that, include: The acquisition unit is used to acquire the set of key parameters of the unit, process the historical data therein using a high-fidelity mechanism model to obtain historical nuclear measurement prediction parameters, and calculate the historical deviation sequence based on the historical nuclear measurement prediction parameters and the corresponding historical nuclear measurement actual parameters. The set of key parameters of the unit includes the unit process parameters, the initial internal parameter set and the actual nuclear measurement parameters. The training unit is used to train the historical unit key parameter set as input features and the historical deviation sequence as the learning target to obtain the trained data deviation proxy model. The correction unit is used to input the current key parameter set of the unit into the trained data deviation proxy model to obtain the predicted deviation sequence. Based on the predicted deviation sequence and the sensitivity analysis matrix, the estimated deviation of the initial internal parameter set is inverted and corrected through the hierarchical inversion algorithm to obtain the corrected internal parameter set. The calculation unit is used to input the corrected internal parameter set and the current unit process parameters into the ultra-fast approximate calculation module for extrapolation and calculation, so as to obtain the final trend prediction result of the nuclear measurement parameters.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images