Risk prediction method and system for nuclear power equipment foundation under extreme working conditions

Abstract

The invention relates to a risk prediction method and system for a nuclear power equipment foundation under an extreme working condition, and the method comprises the steps: building an original state numerical model through obtaining soil layer parameters of the nuclear power equipment foundation, delimiting a risk region range according to the obtained original state numerical model, and arranging measurement points; deformation data corresponding to the measuring points and external environment factor data corresponding to the extreme working conditions are obtained, a predicted deformation value is obtained through the trained prediction model, and when the predicted deformation value exceeds the risk level set in the corresponding risk area range, risk early warning is given out. According to the method, numerical simulation and a machine learning model are combined, a numerical simulation result is used as a basis for initial modeling and monitoring point arrangement, meanwhile, a simulation model is combined with monitoring data to be used for training the machine learning model, a complex physical process can be considered, and implicit modes and rules are extracted from mass data; a large number of influence factors can be comprehensively considered, so that the accuracy and comprehensiveness of risk assessment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear energy equipment, and in particular to a risk prediction method and system for nuclear power equipment foundations under extreme working conditions. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the structure of a nuclear power plant, concrete plays a role in supporting the foundation of reactors, transmission towers and other key equipment. During the operation of a nuclear power plant, the concrete foundation bears hundreds of tons of structural weight, which places extremely high demands on the stability of the foundation structure. Some nuclear power plants are located in coastal areas with complex geological conditions. With the increase in operating time and the influence of environmental erosion, the foundation of nuclear power equipment will sink, age and suffer environmental damage. Generally, regular reliability analysis is performed to evaluate the health status of its foundation structure, which is a key link in ensuring the safe operation of the entire nuclear power system.

[0004] On the one hand, the remote location and inconvenient transportation conditions of nuclear power plants make it difficult to carry out monitoring and sampling work efficiently. On the other hand, in extreme natural environments such as heavy rain, the safety of the foundations of various equipment in nuclear power plants faces greater challenges. These extreme environmental factors often change the stress state of the foundation and the surrounding soil, making manual inspections and large-scale measurements time-consuming and laborious.

[0005] In response to this problem, although it is possible to predict the risks of various equipment base structures in nuclear power plants under extreme environments or natural disasters based on engineering experience and model data, such as using numerical models, empirical models or neural networks, the commonly used numerical models are simplified assumption models, which are difficult to accurately simulate the impact of extreme environments on complex nonlinear systems. Empirical models and neural networks have difficulties such as gradient explosion when dealing with multivariable coupling problems, and the prediction accuracy is not high. Summary of the invention

[0006] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a risk prediction method and system for the foundation of nuclear power equipment under extreme working conditions, which effectively improves the accuracy and practicality of the prediction results and realizes the dynamic deformation prediction of the transmission towers in nuclear power plants under extreme conditions such as heavy rain and landslides.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A first aspect of the present invention provides a risk prediction method for nuclear power equipment foundation under extreme working conditions, comprising the following steps:

[0009] Obtain soil parameters of the nuclear power equipment foundation, construct an original state numerical model, define the risk area based on the obtained original state numerical model, and arrange measurement points at set locations within the risk area;

[0010] Obtain the deformation data corresponding to the measuring point and the external environmental factor data corresponding to the extreme working condition, and use the trained prediction model to obtain the predicted deformation value. When the predicted deformation value exceeds the risk level set in the corresponding risk area, a risk warning is issued.

[0011] Furthermore, the deformation data corresponding to the measuring point include the overall vertical displacement Z of the foundation, the horizontal displacements S1 and S2, the strain ε and the inclination θ of the concrete; the external environmental factors include at least the rainfall Q1, the landslide load volume V1 and the measuring point coordinate position (Lx, Ly) and its coupling influence coefficient Q2.

[0012] Furthermore, the soil layer parameters of the nuclear power equipment foundation are obtained, and the original state numerical model is constructed. The risk area is delineated based on the obtained original state numerical model. Specifically, the soil layer parameters include at least cohesion c, internal friction angle φ, compression modulus Es and water supply degree μ. The stability, bearing capacity and settlement of the soil layer are evaluated based on the obtained soil layer parameters, and the original state numerical model is constructed. According to the soil layer parameters and the evaluation situation, the scope of the risk is predicted.

[0013] Furthermore, the prediction model uses the deformation data corresponding to the measuring points and the external environmental factors when extreme working conditions occur as input features. The hidden layer adopts a bidirectional GRU unit and an attention mechanism, and the output layer is the predicted vertical displacement Z, horizontal displacements S1 and S2.

[0014] Furthermore, the bidirectional GRU unit has a forward propagation mechanism, which uses the reset gate to control the impact of forward information on the current state, uses the update gate to control the degree of integration of past information and current input to determine the impact on the current state, and uses the hidden state to capture long-term dependencies in sequence data.

[0015] Furthermore, the bidirectional GRU unit has a backward propagation mechanism, which first calculates the output error term, calculates the error term of the update gate, then calculates the error term and parameter gradient of the candidate hidden state, and then calculates the error term of the reset gate and the weight gradient of the update gate.

[0016] Furthermore, when the prediction model is trained, the thresholds of various soil layer parameters corresponding to risk scenarios are preset according to the design engineering specifications of different projects, and the risk range area corresponding to the occurrence of certain extreme working conditions is set.

[0017] A second aspect of the present invention provides a method for assembling a risk prediction method for a nuclear power equipment foundation under extreme working conditions, comprising the following steps:

[0018] The numerical simulation module is configured to: obtain soil layer parameters of the nuclear power equipment foundation, construct an original state numerical model, define the risk area range according to the obtained original state numerical model, and arrange measurement points at set positions within the risk area range;

[0019] The risk prediction module is configured to obtain the deformation data corresponding to the measuring point and the external environmental factor data corresponding to the extreme working condition, and obtain the predicted deformation value by using the trained prediction model. When the predicted deformation value exceeds the risk level set in the corresponding risk area, a risk warning is issued.

[0020] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the risk prediction method for nuclear power equipment foundations under extreme working conditions described above.

[0021] The fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the risk prediction method for the foundation of nuclear power equipment under the above-mentioned extreme working conditions are implemented.

[0022] Compared with the prior art, one or more of the above technical solutions have the following beneficial effects:

[0023] 1. Combine numerical simulation and machine learning models, build a numerical simulation model according to the soil parameters of the foundation of nuclear power equipment, use the risk area delineated by the numerical simulation results as the basis for initial modeling and monitoring point layout, and combine the simulation model with monitoring data to train the machine learning model. The machine learning model learns the deformation data obtained from the measuring points in different risk areas, and predicts all equipment with concrete bases in the nuclear power plant by outputting the predicted deformation values ​​of the measuring points, taking into account complex physical processes. The machine learning model extracts implicit deformation patterns and deformation laws from massive data, and can comprehensively consider a large number of influencing factors, thereby improving the accuracy and comprehensiveness of risk assessment.

[0024] 2. Traditional recurrent neural networks (RNNs) are prone to gradient vanishing or exploding problems, making it difficult to effectively learn long-term dependencies. The gating mechanism of the advanced GRU algorithm used in the machine learning model can better capture long-term dependencies. At the same time, the bidirectional structure can simultaneously utilize the forward and backward information of the sequence data to improve the learning and expression capabilities of the model.

[0025] 3. A real-time prediction and early warning mechanism has been implemented. After obtaining new expected or monitored data, the model can be directly input to update the prediction results and trigger risk warnings. This ability to respond in advance and in a timely manner is very important, which helps to discover potential risks as early as possible and take control measures in time to minimize losses. Traditional methods can often only issue warnings based on actual data and fixed threshold rules, and lack the flexibility of early perception and dynamic adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0027] Figure 1 is a schematic diagram of data and processing during risk prediction provided by one or more embodiments of the present invention;

[0028] Figure 2 is a schematic diagram of model training during risk prediction provided by one or more embodiments of the present invention;

[0029] Figure 3 is a schematic diagram of initial state numerical simulation modeling provided by one or more embodiments of the present invention;

[0030] Figure 4 It is a schematic diagram of the layout of risk areas and monitoring points provided by one or more embodiments of the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0033] As introduced in the background technology, based on engineering experience and model data, the key to risk prediction of various equipment base structures in nuclear power plants under extreme environments or natural disasters lies in how to accurately evaluate the scope of the structural danger zone and how to accurately analyze the difficult-to-calculate factors such as displacement inclination based on easily available data. Although numerical modeling can analyze and predict specific environments at a certain moment, it is impossible for numerical models to be comprehensive. When faced with extreme environmental change factors such as heavy rain, it is difficult to accurately simulate the impact of extreme environments on complex nonlinear systems due to simplified assumptions and model limitations. Empirical models and conventional neural networks have difficulties such as gradient explosion when dealing with multivariable coupling problems, and the prediction accuracy is not high.

[0034] Therefore, the following embodiments provide a risk prediction method and system for the foundation of nuclear power equipment under extreme working conditions, which effectively improves the accuracy and practicality of the prediction results and realizes the dynamic deformation prediction of the transmission tower in the nuclear power plant under extreme conditions such as heavy rain and landslide.

[0035] Embodiment 1:

[0036] The risk prediction method of nuclear power equipment foundation under extreme conditions includes the following steps:

[0037] Obtain soil parameters of nuclear power equipment foundation and construct original state numerical model;

[0038] Delineate the risk area and arrange measurement points based on the original state numerical model;

[0039] Obtain the deformation data corresponding to the measuring point and the external environmental factor data corresponding to the extreme working conditions. The deformation data corresponding to the measuring point include the overall vertical displacement Z, horizontal displacement S1, S2, concrete strain ε and inclination θ of the foundation itself, as well as external environmental factors such as rainfall Q1, landslide load volume V1 and coordinate position (Lx, Ly) and its coupling influence Q2;

[0040] The trained prediction model is used to obtain the predicted deformation value. When the predicted deformation value exceeds the risk level set in the corresponding risk area, a risk warning is issued.

[0041] (1) Based on the exploration results, the soil parameters at the bottom of the foundation (such as cohesion c, internal friction angle φ, compression modulus Es, and water supply μ) are obtained for the case project. These parameters are of great significance to engineering design and construction, and can help evaluate the stability, bearing capacity, and settlement characteristics of the soil and rock layers. Figure 1 The original state numerical simulation modeling shown was used to preliminarily delineate the risk area and determine the on-site monitoring sites.

[0042] In this embodiment, the risk area is divided according to the evaluated numerical model to generate a preliminary risk area. The objective principle is to predict an approximate range of risk based on various parameters of the soil and rock layers, such as the deformation value breaking through the parameter design, which can timely reflect the real danger area. Since the basic characteristic values ​​of the soil and rock layers of each nuclear power plant are different, the range of the danger area simulated by training is also different, and different models need to be established, so it has little impact on the subsequent GRU model.

[0043] The sensors arranged at the on-site monitoring sites are used to obtain the overall vertical displacement Z, horizontal displacements S1, S2, concrete strain ε, inclination θ of the base itself, as well as external environmental factors such as rainfall Q1, landslide load volume V1 and coordinate position (Lx, Ly) and its coupling influence Q2, and a deformation basic database is formed based on the monitoring results.

[0044] (2) Process the basic database to form training sets and test sets, such as Figure 2 As shown, a GRU (Gated Recurrent Unit) neural network displacement prediction model is constructed. Parameters such as Z, S1, S2, ∈, θ, Q1, Q2, V1, (Lx, Ly) are used as input features. The hidden layer uses a bidirectional GRU unit and attention mechanism, and the output layer is the predicted S1 and S2 deformation values.

[0045] Since the design engineering specifications of each project will stipulate that the deformation value and other related parameters are within a certain range, danger may occur once the threshold is exceeded. Therefore, during training, according to the design engineering specifications of each project, the thresholds of each parameter corresponding to the risk scenario are preset, and by training the thresholds, a certain extreme working condition is set to occur, thereby obtaining a risk range area.

[0046] (3) Train the constructed bidirectional GRU model, use K-fold cross validation to evaluate the model and perform hyperparameter tuning.

[0047] (4) Apply the optimal model obtained through computational training to the actual prediction scenario, input new parameters such as Q1, V1, (Lx, Ly), predict the displacements of Z, S1, and S2 in the future, and output the results.

[0048] In step (1), the basic database includes parameters such as Z, S1, S2,∈, θ, Q1, Q2, V1, (Lx, Ly) and their coupling calculation relationships; it is stored in a data table in csv format in the form of time series.

[0049] In step (2), the input data is preprocessed by using pandas to read the csv format data, and the MinMaxScaler function in the scikit-learn library is used to standardize the input data and map each feature to the range of [0, 1]. The preprocessing step also includes outlier removal and dividing the training set and test set ratio. A bidirectional GRU model is constructed using the nn.Module class in PyTorch, including a bidirectional GRU layer and a fully connected layer.

[0050] The forward propagation mechanism of the GRU model is constructed as follows:

[0051] The first module is the reset gate, which controls the impact of forward information on the current state:

[0052] r t =σ(W r ·[h t-1 ,x t ]+b r );

[0053] The second module is the update gate, which controls the degree of integration of past information and current input and determines the impact on the current state:

[0054] z t =σ(W z ·[h t-1 ,x t ]+b z );

[0055] The third module is a hidden state, which is used to effectively capture long-term dependencies in sequence data:

[0056]

[0057] where h t-1 is the hidden state of the previous time step, x t is the input feature vector (feature vector in mathematical sense) of the t-th time step, σ is the Sigmoid function, W r ,W z ,W are the weight matrices of each module, b r ,b z , b is the bias term of each module. is a candidate hidden state, h t is the final hidden state, tanh is the hyperbolic tangent activation function, Represents an element-wise multiplication operation.

[0058] The back propagation mechanism of the GRU model is built as follows, where L is the loss function:

[0059] First, calculate the output error term δ t , calculate the update gate z t The error term δ zt :

[0060]

[0061] Second, calculate the candidate hidden states The error term And the parameter gradient:

[0062]

[0063] Again, calculate the error term of the reset gate and the weight gradient of the update gate:

[0064]

[0065] After obtaining the new prediction and remote sensing data Q1, V1, (Lx, Ly), this embodiment directly substitutes them into the GRU model to update the deformation prediction result.

[0066] According to the Nuclear Safety Guidelines HAD 102 / 02, the seismic design of nuclear power plants needs to consider the effects of deformation forced by the surrounding soil during and after an earthquake, as well as the displacement difference or load at the connection end with the building or other structure. This involves the deformation and crack control of concrete structures during long-term use. When the output deformation prediction value is close to the warning value, the warning risk location and risk level information are pushed, which helps to take corresponding deformation control measures in time. Different nuclear power plants should refer to different engineering specifications and input the maximum deformation value as a warning.

[0067] By combining numerical simulation and machine learning models, the numerical simulation results are used as the basis for initial modeling and monitoring point layout, and the simulation model is combined with monitoring data to train a composite neural network model. All equipment with a concrete base in a nuclear power plant can be predicted. It can take into account complex physical processes, and the GRU model can extract implicit patterns and laws from massive data. In contrast, traditional monitoring and early warning technologies often only focus on data of a single dimension, while this embodiment can comprehensively consider a large number of influencing factors, thereby improving the accuracy and comprehensiveness of risk assessment.

[0068] Traditional recurrent neural networks (RNNs) are prone to gradient vanishing or exploding problems, making it difficult to effectively learn long-term dependencies. The gating mechanism of the advanced GRU algorithm used in this embodiment can better capture long-term dependencies, and the bidirectional structure can simultaneously utilize the forward and backward information of sequence data to improve the learning and expression capabilities of the model.

[0069] This embodiment designs a real-time prediction and early warning mechanism, which can directly input the model after obtaining new expected or monitored data, update the prediction results and trigger risk warnings. This ability to respond in advance and in a timely manner is very important, which helps to discover potential risks as early as possible and take control measures in time, thereby minimizing losses. Traditional methods can often only issue warnings based on actual data that has occurred and fixed threshold rules, lacking the flexibility of early perception and dynamic adjustment.

[0070] An example is as follows: In a coastal area, the soil is bedrock inside and weathered rock outside, with two upper and lower slopes. A 30m high-voltage transmission tower is planned to be built. The size of the transmission tower concrete base is 2m×2m×1m, with a center spacing of 3m and a horizontal spacing of 16m.

[0071] The soil parameters (c, φ, E, μ) of the construction site are obtained based on the exploration results of the case project. Figure 1 and Figure 3 The original state numerical simulation modeling shown in the figure was used to preliminarily define the risk area and arrange on-site monitoring sites, such as Figure 4As shown in the figure, No. 1-4 is the monitoring point of the concrete base of the transmission tower, No. 5 is the monitoring point of the lower slope deformation, No. 6 and No. 7 are the monitoring points of the lateral risk area, and No. 8 is the monitoring point of the upper slope deformation. The reliability of the model is verified based on the monitoring result data obtained from each monitoring point.

[0072] By simulating the precipitation parameters and landslide bodies, the basic model is simulated for multimodal risk simulation calculation (the input value is the warning value of simulated collapse or critical deformation, that is, the surface model data results of various places, and prediction is achieved through continuous training. The measured value is input to predict whether it reaches the critical point of danger). According to the multimodal calculation results, the overall vertical displacement Z, horizontal displacement S1, S2, concrete strain ε, inclination angle θ of the base itself, and external environmental factors such as rainfall Q1, landslide body load volume V1 and coordinate position (Lx, Ly) and its coupling influence Q2 are obtained, and the basic database is formed in conjunction with the monitoring data. According to the input data set, advance training is carried out in the following model algorithm. If a critical danger occurs, the dangerous area can be divided to achieve the purpose of risk prediction.

[0073] Construct a bidirectional GRU model. The gating mechanism is described above and will not be repeated here:

[0074] First, use pandas to read the data in CSV format and preprocess the data, including outlier removal and feature standardization. Use scikit-learn's MinMaxScaler to map the features to the range [0, 1].

[0075] Divide the training set and test set by random sampling with a ratio of 8:2. Use PyTorch's DataLoader to load the dataset and set the batch size to 32.

[0076] Define the BiGRU model class, which inherits from nn.Module. The model contains a bidirectional GRU layer and a fully connected layer. The number of hidden layer units in the bidirectional GRU layer is set to 64, the number of input features is 10, and the output is 2.

[0077] Set the hyperparameters to 100 training rounds and 0.001 learning rate.

[0078] Instantiate the model, use mean square error loss as the loss function, and use Adam optimizer to optimize the model.

[0079] Run a training loop on the constructed model, and for each epoch, iterate over the training data loader. Compute the model output and loss, perform backpropagation and optimizer updates. Use K-fold cross validation to evaluate the model and perform hyperparameter tuning. Shuffle the data randomly and divide it into 5 folds (k=5) using KFold. Initialize best_model and best_val_loss variables to store the model with the lowest validation loss and the corresponding loss value. Iterate over the training set and validation set indexes of each fold. Construct the training data loader train_loader and the validation data loader val_loader from the original data based on the index of the current fold.

[0080] After obtaining the new prediction and remote sensing data Q1, V1, (Lx, Ly), substitute it into this bidirectional GRU model and update the deformation prediction result. When the output deformation prediction value is close to the warning value, the warning risk location and risk level information are pushed.

[0081] Embodiment 2:

[0082] The risk prediction method system for nuclear power equipment foundation under extreme conditions includes:

[0083] The numerical simulation module is configured to: obtain soil layer parameters of the nuclear power equipment foundation and construct an original state numerical model;

[0084] The measuring point arrangement module is configured to: define the risk area range and arrange the measuring points according to the original state numerical model;

[0085] The risk prediction module is configured to: obtain deformation data corresponding to the measuring point and external environmental factor data corresponding to the extreme working conditions, the deformation data corresponding to the measuring point includes the overall vertical displacement Z, horizontal displacement S1, S2, concrete strain ε and inclination θ of the foundation itself, and external environmental factors such as rainfall Q1, landslide load volume V1 and coordinate position (Lx, Ly) and its coupling influence Q2;

[0086] The risk prediction module is also configured to: use the trained prediction model to obtain a predicted deformation value, and issue a risk warning when the predicted deformation value exceeds the risk level set in the corresponding risk area range.

[0087] Embodiment three:

[0088] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps in the risk prediction method for nuclear power equipment foundation under extreme working conditions as described in the above-mentioned embodiment 2 are implemented.

[0089] Embodiment 4:

[0090] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the risk prediction method for the foundation of nuclear power equipment under extreme working conditions as described in the above-mentioned embodiment 2 are implemented.

[0091] The steps involved in the above embodiments 2 to 4 correspond to those in embodiment 1. For the specific implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0092] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A risk prediction method for nuclear power equipment foundation under extreme conditions, characterized in that: The following steps are involved: Obtain soil parameters of the nuclear power equipment foundation, construct an original state numerical model, define the risk area based on the obtained original state numerical model, and arrange measurement points at set locations within the risk area; Obtain the deformation data corresponding to the measuring point and the external environmental factor data corresponding to the extreme working condition, and use the trained prediction model to obtain the predicted deformation value. When the predicted deformation value exceeds the risk level set in the corresponding risk area, a risk warning is issued.

2. The risk prediction method for nuclear power equipment foundation under extreme working conditions according to claim 1 is characterized in that: The deformation data corresponding to the measuring point include the overall vertical displacement Z of the foundation, the horizontal displacements S1 and S2, the strain ε and inclination θ of the concrete; the external environmental factors include at least the rainfall Q1, the landslide load volume V1 and the coordinate position of the measuring point (Lx, Ly) and its coupling influence coefficient Q2.

3. The risk prediction method for nuclear power equipment foundation under extreme working conditions according to claim 1 is characterized in that: The soil parameters of the nuclear power equipment foundation are obtained, and the original state numerical model is constructed. The risk area is delineated based on the obtained original state numerical model. Specifically, the soil layer parameters include at least cohesion c, internal friction angle φ, compression modulus Es and water supply degree μ. The stability, bearing capacity and settlement of the soil layer are evaluated based on the obtained soil layer parameters, and the original state numerical model is constructed. According to the soil layer parameters and the evaluation situation, the scope of the risk is predicted.

4. The risk prediction method for nuclear power equipment foundation under extreme working conditions according to claim 1 is characterized in that: The prediction model uses the deformation data corresponding to the measuring points and the external environmental factors when extreme working conditions occur as input features. The hidden layer adopts a bidirectional GRU unit and an attention mechanism. The output layer is the predicted vertical displacement Z, horizontal displacements S1 and S2.

5. The risk prediction method for nuclear power equipment foundation under extreme working conditions according to claim 4 is characterized in that: The bidirectional GRU unit has a forward propagation mechanism, which uses the reset gate to control the impact of forward information on the current state, uses the update gate to control the degree of integration of past information and current input to determine the impact on the current state, and uses the hidden state to capture long-term dependencies in sequence data.

6. The risk prediction method for nuclear power equipment foundation under extreme working conditions according to claim 4 is characterized in that: The bidirectional GRU unit has a backward propagation mechanism, which first calculates the output error term, calculates the error term of the update gate, then calculates the error term and parameter gradient of the candidate hidden state, and then calculates the error term of the reset gate and the weight gradient of the update gate.

7. The risk prediction method for nuclear power equipment foundation under extreme working conditions according to claim 1, characterized in that: When training the prediction model, the thresholds of various soil layer parameters corresponding to risk scenarios are preset according to the design engineering specifications of different projects, and the risk range area corresponding to the occurrence of certain extreme working conditions is set.

8. Risk prediction system for nuclear power equipment foundation under extreme conditions, characterized by: include: The numerical simulation module is configured to: obtain soil layer parameters of the nuclear power equipment foundation, construct an original state numerical model, define the risk area range according to the obtained original state numerical model, and arrange measurement points at set positions within the risk area range; The risk prediction module is configured to obtain the deformation data corresponding to the measuring point and the external environmental factor data corresponding to the extreme working condition, and use the trained prediction model to obtain the predicted deformation value. When the predicted deformation value exceeds the risk level set in the corresponding risk area, a risk warning is issued.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, which, when executed by a processor, implements the steps in the risk prediction method for the foundation of nuclear power equipment under extreme working conditions as described in any one of claims 1 to 7.

10. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the risk prediction method for the foundation of nuclear power equipment under extreme working conditions as described in any one of claims 1 to 7 are implemented.

Images