Dam leakage intelligent monitoring and prediction early warning method and system

Abstract

The invention discloses a dam leakage intelligent monitoring and prediction early warning method and system, and the method comprises the steps: collecting multi-aspect data of a dam in real time through a plurality of monitoring devices, and carrying out the processing and analysis of the data; a data training set is reconstructed through cross validation, in order to further improve the prediction precision, various base learners such as Transformer, BiGRU and LightGBM are effectively integrated, and an efficient prediction model is constructed through a stacking method. In addition, an improved squirrel optimization algorithm IPKO is adopted to optimize the KELM model so as to adapt to the complex environment and the real-time requirement of dam leakage prediction. Finally, based on the trained model, an online monitoring platform is constructed, data of the intelligent monitoring system can be received in real time and predicted and analyzed, the leakage condition is judged based on the seepage flow, the early warning level is determined through historical dam leakage data, and the monitoring capacity and the early warning response speed of the dam leakage condition are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy and hydropower engineering, and in particular to a method and system for intelligent monitoring, prediction and early warning of dam leakage. Background Art

[0002] In modern society, dams play an important role in water conservancy and hydropower. The existence of dams can play an important role in water conservancy regulation, flood protection, hydropower generation, water supply, etc., and is an indispensable and important part of modern engineering technology. However, due to the wide coverage of dams, once leakage occurs, it will cause great harm to people's lives and property, so dam leakage monitoring is extremely necessary.

[0003] With the development of modern science and technology, digital communication technology and electronic technology have also made breakthrough progress, and sensors, radars and remote sensing technologies have also provided sufficient conditions for dam monitoring. Most of the existing dams are gravity dams and concrete dams. After the impact of time, geological conditions, dam materials and structural deformation caused by temperature changes may cause leakage in the dam.

[0004] Deep learning has a more complex multi-layer structure than traditional models, with more information, better data extraction and stronger nonlinear feature expression capabilities, and can accurately map the complex mapping relationships hidden under the dam data. By taking advantage of the advantages of deep learning, data feature information can be fully extracted to improve the prediction accuracy of the model. So far, various deep learning models have been continuously developed, and many new deep models have been introduced. However, there are many problems with a single prediction model, such as local optimality and insufficient accuracy. In view of the large number of implicit data features contained in the data, integrating various algorithms and network models, combining their respective advantages, and improving the overall accuracy and generalization ability of the model has become a new research direction in predictive modeling. Summary of the invention

[0005] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0006] Therefore, in order to solve the shortcomings of the existing reservoir leakage monitoring methods, the present invention proposes a dam leakage intelligent monitoring and prediction and early warning method. First, various monitoring equipment is used to collect various data of the dam in real time, including temperature, pressure and other sensors, water meters, magnetostrictive water weir meters and immersion water level meters. The intelligent detection system collects data such as dam pressure, local temperature, dam displacement, precipitation, water level changes, and seepage, and processes and analyzes the data. Subsequently, cross-validation is used to reconstruct the data training set. In order to further improve the prediction accuracy, multiple base learners such as Transformer, BiGRU and LightGBM are effectively integrated, and an efficient prediction model is constructed through the stacking method. In addition, the improved Kingfisher optimization algorithm IPKO is innovatively used to optimize the meta-learner KELM model to adapt to the complex environment and real-time requirements of dam leakage prediction.

[0007] Another object of the present invention is to provide an intelligent monitoring and prediction and early warning system for dam leakage.

[0008] To achieve the above-mentioned purpose, the present invention provides a dam leakage intelligent monitoring and prediction and early warning method, comprising: Collect multivariate data from various monitoring devices of the reservoir dam; wherein the multivariate data includes dam pressure, dam displacement, dam cracks, local temperature, upstream and downstream water levels, seepage pressure, precipitation and seepage; Preprocess the multivariate data to obtain a training data set; Construct an integrated prediction model based on stacking, and use transformer, BiGRU, and LightGBM models as base learners of the integrated prediction model, use the KELM model as the meta-learner of the integrated prediction model, use the training data set to train the base learner to obtain the prediction result of the base learner, and use the prediction result of the base learner to train the KELM model to output the prediction result of the dam seepage; The PKO algorithm is improved by using good point set initialization and Gaussian random walk strategy, and the parameters of the KELM model are optimized by using the improved IPKO algorithm. The warning level is set based on the historical dam seepage data, and the final dam seepage prediction result is output based on the optimized KELM model. It is judged whether the final dam seepage prediction result is within the warning range. If it exceeds the prediction level, a warning is issued.

[0009] The dam leakage intelligent monitoring and prediction and early warning method according to the embodiment of the present invention may also have the following additional technical features: In one embodiment of the present invention, the multivariate data of each monitoring device of the reservoir dam is collected, including: Collect dam pressure data through pressure sensors; Collect dam displacement data through displacement sensors; Collect dam crack data through video monitoring; Collect local temperature data through temperature sensors; Collect water level data upstream and downstream of the dam through submersible water level gauges; Seepage pressure data were collected via pore water pressure gauges; Collect precipitation data through rain gauges; Dam seepage data were collected using a magnetostrictive weir gauge.

[0010] In one embodiment of the present invention, preprocessing the multivariate data to obtain a training data set includes: Through dual-channel multi-variable data communication transmission, different networks are selected for data transmission in different situations, and after the data is transmitted to the detection platform, data elimination, mining and fusion operations are automatically performed on the data; The processed multivariate data were subjected to three-fold cross validation to obtain the training data set.

[0011] In one embodiment of the present invention, a transformer model is used as the first primary learner of an integrated prediction model; the transformer model consists of an encoder and a decoder, which are used to process input sequences and generate output sequences respectively; the encoder consists of a positional code, a self-attention mechanism, a feedforward neural network, a residual connection and a layer normalization, and the decoder consists of a self-attention mechanism, an encoder-decoder attention mechanism and a feedforward neural network; the transformer self-attention mechanism has three inputs, and the three matrices Q query, K key and V value all come from the same input, and the dot product between Q and K is calculated and divided by ,in is the dimension of the K vector; the result is normalized to a probability distribution using the Softmax operation, and then multiplied by the matrix V to obtain the representation of the weighted sum; the calculation process is expressed as: (1) BiGRU is used as the second primary learner of the integrated prediction model. The BiGRU model introduces a bidirectional structure based on the GRU model. GRU is a gated recurrent neural network unit, including an update gate and a reset gate. The relevant calculation formulas for the GRU update gate and reset gate are as follows: (2) (3) in is the update gate of GRU, is the reset gate of GRU, and are the weight matrices of the update gate and the reset gate, respectively. It is the Sigmoid function, which is used to convert the target data into a value in the range of 0 to 1. is the previous time step; The output of the GRU model at the current moment is related to the previous moment. The specific formula of the hidden state at the previous moment is as follows: (4) in It records the current moment. is the weight matrix of the candidate hidden state; The final output of the GRU model is related to the output at the current moment and the hidden state at the previous moment. The specific formula is as follows: (5) in For the final output result; LightGBM is used as the third primary learner of the integrated prediction model; LightGBM adopts a leaf-node-based decision tree algorithm and introduces mutually exclusive feature bundling algorithm and histogram bias algorithm; The KELM model is used as the second-layer meta-learner of the integrated model; ELM is a single hidden layer feedforward neural network, and the learning objective function F(x) is expressed as a matrix: (6) in is the input vector, , is the hidden layer node output, is the output weight, is the expected output; Transform network training into a linear system solution problem. according to Determine, among which for The generalized inverse matrix of and the identity matrix , then the least squares solution of the output weight is: (7) Introduce the kernel function into ELM, the kernel matrix is: (8) in , is the experimental input vector, then the objective function F(x) can be expressed as: (9) in For a given training sample, is the sample size, is the kernel function.

[0012] In one embodiment of the present invention, the Kingfisher Optimization Algorithm PKO is improved by using good point set initialization and Gaussian random walk strategy, and the parameters of the KELM model are optimized by using the improved Kingfisher Optimization Algorithm IPKO, including: The objective function of the Ban Kingfisher Optimization Algorithm PKO is set as the parameter regularization coefficient C and kernel function coefficient S of the KELM model. These two parameters are used as the input of the algorithm and the upper and lower limits are set. The fitness value is the training set error index. The original Kingfisher optimization algorithm is initialized randomly, and the good point set strategy is used to initialize it instead of the original random initialization: (10) Deviations meet: (11) in For the best point collection, For the best point, For points, Is only with The relevant constants, is an arbitrarily small positive number, yes The minimum prime number that satisfies Represents dimension, upper represents the upper bound, and lower represents the lower bound; The position update method of the Ban Kingfisher Optimization Algorithm PKO is divided into the exploration stage and the development stage. The Gaussian random walk strategy is introduced in the exploration stage and the development stage. The specific formula for the exploration phase is as follows: (12) in , Indicates individual kingfisher, is the position of individual i at time t+1; parameter , dim represents the dimension, represents a random number in a normal distribution; N represents the population size; , is an individual randomly selected from the dominant population; The value of parameter T is determined dynamically according to the current strategy. The exploration phase is divided into two strategies: perching and hovering; The parameter T in the habitat strategy is calculated as follows: (13) in is the maximum number of iterations, BF is set to 8, and rand is a random number between 0 and 1.

[0013] The parameter T in the hover strategy is calculated as follows: (14) and are the fitness values ​​of the i-th and j-th kingfishers respectively; The position update formula in the development phase is as follows: (15) in is the optimal position of the kingfisher population at time t, ; , and Represents hunting ability, calculated as follows: (16) (17) (18) During the iteration process, the individual position of the kingfisher corresponding to the best fitness value is determined according to the training error index. The optimal position of the kingfisher is used as the output to update the parameter regularization coefficient C and kernel function coefficient S of the KELM prediction model.

[0014] To achieve the above object, the present invention further provides a dam leakage intelligent monitoring and prediction and early warning system, comprising: A data acquisition module is used to collect multivariate data from various monitoring devices of the reservoir dam; wherein the multivariate data includes: dam pressure, dam displacement, dam cracks, local temperature, upstream and downstream water levels, seepage pressure, precipitation and seepage; The data preprocessing module is used to transmit the multivariate data through dual-channel communication and store it in the local dam data history library, and to clean and eliminate the data and perform cross-validation reconstruction operations to obtain the training data set; The model algorithm module is used to construct an integrated prediction model based on stacking, and use the transformer, BiGRU, and LightGBM models as the base learners of the integrated prediction model, and use the KELM model as the meta-learner of the integrated prediction model. The base learners are trained with the training data set to obtain the prediction results of the base learners, and the prediction results of the base learners are used to train the KELM model to output the prediction results of the dam seepage; the Kingfisher Optimization Algorithm PKO is improved by using the good point set initialization and Gaussian random walk strategy, and the parameters of the KELM model are optimized by using the improved Kingfisher Optimization Algorithm IPKO; The early warning module is used to set the early warning level based on the historical dam seepage data, output the final dam seepage prediction result based on the optimized KELM model, and determine whether the final dam seepage prediction result is within the early warning range. If it exceeds the prediction level, an early warning will be issued.

[0015] The intelligent monitoring and prediction and early warning method and system for dam leakage of the embodiment of the present invention constructs an online monitoring platform based on the trained model, which can receive the data of the intelligent monitoring system in real time and perform predictive analysis, judge the leakage situation based on the seepage volume, and determine the early warning level through historical dam leakage data, thereby improving the monitoring capability of dam leakage and the early warning response speed.

[0016] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which: Figure 1 is a flow chart of a dam leakage intelligent monitoring and prediction and early warning method according to an embodiment of the present invention; Figure 2 is a schematic diagram of cross-validation of original training set data according to an embodiment of the present invention; Figure 3 is a framework diagram based on a stacking integration model according to an embodiment of the present invention; Figure 4 is a flow chart of an algorithm for optimizing model parameters according to an embodiment of the present invention; Figure 5 It is a structural diagram of the intelligent monitoring and prediction and early warning system for dam leakage according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0019] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0020] The following describes the intelligent monitoring and prediction and early warning method and system for dam leakage according to an embodiment of the present invention with reference to the accompanying drawings.

[0021] Figure 1 is a flow chart of a dam leakage intelligent monitoring and prediction and early warning method according to an embodiment of the present invention. Figure 1 As shown, the method includes: S1, collects multivariate data of various monitoring equipment of the reservoir dam; the multivariate data includes dam pressure, dam displacement, dam cracks, local temperature, upstream and downstream water levels, seepage pressure, precipitation and seepage volume.

[0022] It can be understood that the present invention collects multivariate data of various monitoring devices of the reservoir dam through the monitoring devices arranged around the reservoir.

[0023] It can be understood that multivariate data include dam pressure, dam displacement, dam cracks, local temperature, upstream and downstream water levels, seepage pressure, precipitation and seepage volume.

[0024] Specifically, monitoring equipment is installed on the dam to collect various data of the dam. Dam pressure data is collected through pressure sensors, dam displacement data is collected through displacement sensors, dam crack data is collected through video monitoring, and local temperature data is collected through temperature sensors; upstream and downstream water level data of the dam are collected through immersion water level gauges, seepage pressure data are collected through pore water pressure gauges, precipitation data are collected through rain gauges, and finally, dam seepage data are collected through magnetostrictive water weir meters.

[0025] S2, preprocess the multivariate data to obtain the training data set.

[0026] It is understandable that the transmitted data is stored, cleaned and eliminated, and the processed training set data is reconstructed using cross-validation. Figure 2 shown.

[0027] Specifically, the multivariable data of the dam collected in step S1 is transmitted through the 5G-Beidou dual-channel communication to ensure the reliability of data transmission. In normal weather and when the 5G network is available, the monitoring terminal automatically selects the 5G network for data transmission; when the network is interrupted due to bad weather, the monitoring terminal will switch to the Beidou system for data transmission; after the data is transmitted to the detection platform, the system will automatically eliminate, mine and fuse the data; The present invention adopts three-fold cross validation for the processed data, and trains the model through multiple data divisions to improve the generalization ability, and reduces the risk of overfitting by processing the data. The steps of three-fold cross validation are as follows: first, the data is evenly divided into three parts, one of which is taken as a test set and the other two as training sets, and a model is trained using this data; then, different data is used as test set data in the same way, and the model is trained in sequence, and a total of three trainings are performed.

[0028] S3, construct an integrated prediction model based on stacking, and use the transformer, BiGRU, and LightGBM models as the base learners of the integrated prediction model, and the KELM model as the meta-learner of the integrated prediction model. The base learners are trained with the training data set to obtain the prediction results of the base learners, and the prediction results of the base learners are used to train the KELM model to output the dam seepage prediction results.

[0029] The embodiment of the present invention constructs an integrated model based on stacking, uses transformer, BiGRU, and LightGBM models as base learners of the integrated model, and uses the KELM model as the meta learner of the integrated model. Figure 3 shown.

[0030] Specifically, in the field of ensemble learning, Stacking is an efficient ensemble method that enhances prediction performance by building a two-layer model architecture. In the first layer, multiple base models independently predict the data as base learners and generate prediction results using the data of S2. The prediction results generated by the first layer are used as the input of the second-layer meta-learner of the stacking ensemble model. The task of the meta-model is to analyze and synthesize the output of the first-layer model to produce the final prediction.

[0031] In the present invention, three different models Transformer, BiGRU and LightGBM are used as the base learners of the first layer. These models each predict the cross-validated data set, and their outputs are integrated to build the model of the second layer. In the second layer, the present invention uses the KELM model as a meta-learner, which further analyzes and learns the prediction results of the first layer model to generate the final prediction output. This method not only improves the accuracy of the model, but also enhances the generalization ability of the model.

[0032] In one embodiment of the present invention, a transformer model is used as the first primary learner of the integrated model, and a self-attention mechanism module is adopted internally to achieve the effect of parallel computing.

[0033] The Transformer model consists of two parts: the encoder and the decoder, which are used to process the input sequence and generate the output sequence respectively. The encoder consists of position encoding, self-attention mechanism, feedforward neural network, residual connection and layer normalization, and the decoder consists of self-attention mechanism, encoder-decoder attention mechanism and feedforward neural network.

[0034] The most important module of Transformer is the self-attention mechanism, which has three inputs: Q (query), K (key), and V (value). For Self Attention, the three matrices Q (query), K (key), and V (value) all come from the same input. First, the dot product between Q and K is calculated. In order to prevent the result from being too large, it is divided by ,in is the dimension of the K vector; then the Softmax operation is used to normalize the result into a probability distribution, and then multiplied by the matrix V to obtain the weighted summation representation. The calculation process can be expressed as: (1) At the same time, the Multi-Head Attention mechanism is introduced in Transformer, allowing the model to perform attention calculations in multiple different subspaces, increasing the expressive power of the model.

[0035] In one embodiment of the present invention, BiGRU is used as the second primary learner of the integrated model. The BiGRU model introduces a bidirectional structure based on the GRU model. GRU is a gated recurrent neural network unit including an update gate and a reset gate.

[0036] The relevant calculation formulas for the GRU update gate and reset gate are as follows: (2) (3) in is the update gate of GRU, is the reset gate of GRU, and are the weight matrices of the update gate and the reset gate, respectively. It is the Sigmoid function, which is used to convert the target data into a value in the range of 0 to 1. is the previous time step.

[0037] The output of the GRU model at the current moment is related to the previous moment. The specific formula of the hidden state at the previous moment is as follows: (4) in It records the current moment. is the weight matrix of the candidate hidden states.

[0038] The final output of the GRU model is related to the output at the current moment and the hidden state at the previous moment. The specific formula is as follows: (5) in For the final output result, when The closer it is to 1, the more memory is stored about the candidate hidden state of the current node.

[0039] BiGRU consists of two independent GRU units, one processes data forward in time series, and the other processes data in reverse time series. Through this bidirectional structure, the BiGRU model can capture both forward and backward information of sequence data, thereby better understanding and predicting patterns in the sequence.

[0040] In one embodiment of the present invention, LightGBM is used as the third primary learner of the integrated model. The core principle of LightGBM is based on the gradient boosting framework, which continuously approaches the optimal value of the objective function by iteratively training the decision tree model. Compared with the traditional gradient boosting decision tree (GBDT), LightGBM introduces a histogram-based algorithm, which reduces the complexity of the algorithm and improves the training speed by discretizing the eigenvalues.

[0041] LightGBM uses a leaf-node-based decision tree algorithm and effectively builds and splits decision trees through a histogram algorithm. In addition, LightGBM also introduces a mutually exclusive feature bundling algorithm and a histogram bias algorithm to further improve the accuracy and generalization ability of the model.

[0042] In one embodiment of the present invention, the present invention uses the KELM model as the second-layer meta-learner of the integrated model. The kernel extreme learning machine KELM is an improved algorithm based on the extreme learning machine ELM and combined with the kernel function. KELM can improve the prediction performance of the model while retaining the advantages of ELM.

[0043] ELM is a single hidden layer feedforward neural network, and its learning objective function F(x) can be expressed as a matrix: (6) in is the input vector, , is the hidden layer node output, is the output weight, is the expected output.

[0044] Transform network training into a linear system solution problem. according to Determine, among which for The generalized inverse matrix of . In order to enhance the stability of the neural network, the regularization coefficient is introduced and the identity matrix , then the least squares solution of the output weight is: (7) Introduce the kernel function into ELM, the kernel matrix is: (8) in , is the experimental input vector, then the objective function F(x) can be expressed as: (9) in For a given training sample, is the sample size, is the kernel function.

[0045] S4, the Kingfisher Optimization Algorithm PKO is improved by using good point set initialization and Gaussian random walk strategy, and the parameters of the KELM model are optimized by using the improved Kingfisher Optimization Algorithm IPKO.

[0046] Specifically, the good point set initialization and Gaussian random walk strategy are used to improve the Ban Kingfisher optimization algorithm PKO, and the improved Ban Kingfisher optimization algorithm IPKO is used to optimize the KELM model parameters, the parameters including: regularization coefficient C, kernel function coefficient S; Figure 4 shown.

[0047] The objective function of the Ban Kingfisher Optimization Algorithm PKO is set as the parameter regularization coefficient C and kernel function coefficient S of the KELMM model. These two parameters are used as the input of the algorithm and the upper and lower limits are set. The fitness value is the training set error index. The original Kingfisher optimization algorithm is initialized randomly, so the population distribution is random, which easily leads to PKO falling into local optimality. The present invention uses the good point set strategy initialization instead of the original random initialization: (10) Its deviation satisfies: (11) in For the best point collection, For the best point, For points, Is only with The relevant constants, is an arbitrarily small positive number, yes The minimum prime number that satisfies Represents dimension, upper represents the upper bound, and lower represents the lower bound.

[0048] The position update method of the Ban Kingfisher Optimization Algorithm PKO is divided into the exploration phase and the development phase. The Gaussian random walk strategy is introduced in the exploration phase and the development phase. The introduction of the Gaussian random walk strategy can achieve a large disturbance in the early stage of iteration, and the disturbance is rapidly reduced in the later stage of iteration, thereby balancing the exploration and development capabilities of the algorithm.

[0049] The specific formula for the exploration phase is as follows: (12) in , Indicates individual kingfisher, is the position of individual i at time t+1; parameter , dim represents the dimension, represents a random number in a normal distribution; N represents the population size; , is an individual randomly selected from the dominant population.

[0050] The value of parameter T is dynamically determined according to the current strategy. The exploration phase is divided into two strategies: perching and hovering.

[0051] The parameter T in the habitat strategy is calculated as follows: (13) in is the maximum number of iterations, BF is set to 8, and rand is a random number between 0 and 1.

[0052] The parameter T in the hover strategy is calculated as follows: (14) and are the fitness values ​​of the i-th and j-th kingfishers respectively.

[0053] The position update formula in the development phase is as follows: (15) in is the optimal position of the kingfisher population at time t, ; , and Represents hunting ability, calculated as follows: (16) (17) (18) During the iteration process, the individual position of the kingfisher corresponding to the best fitness value is determined according to the training error index. The optimal position of the kingfisher is used as the output to update the parameter regularization coefficient C and kernel function coefficient S of the KELM prediction model.

[0054] S5, set the warning level based on the historical dam seepage data, output the final dam seepage prediction result based on the optimized KELM model, and judge whether the final dam seepage prediction result is within the warning range. If it exceeds the prediction level, issue a warning.

[0055] Specifically, historical dam seepage data is obtained and the warning level is set based on expert experience. After the seepage flow is predicted by the optimized integrated model, it is determined whether the seepage flow is within the warning range. If it exceeds the predicted level, the speaker will issue a warning.

[0056] Specifically, by retrieving the historical seepage data of local dams and the leakage of dams, experts have established a mature set of leakage warning levels through experience. The leakage warning level is specifically divided into three levels. When the seepage volume is at level one, no treatment is done; when the seepage volume reaches level two, professional personnel are required to conduct on-site surveys and timely repairs; when the dam seepage volume reaches level three standards, an escape warning is required for the dam and surrounding areas to protect the lives and property of the people, and timely reports are made to the local water conservancy department. After the integrated model outputs the predicted seepage volume value, different responses are made according to the seepage volume situation.

[0057] According to the intelligent monitoring and prediction and early warning method of dam leakage of the embodiment of the present invention, multiple dam data are used as input to predict the dam seepage, which ensures the accuracy of dam seepage prediction and plays an important role in dam early warning; the prediction model based on stacking integration proposed by the present invention uses transformer, BiGRU, and LightGBM as base learners, KELM as meta-learner, and uses the improved Kingfisher optimization algorithm IPKO to optimize the parameters of the meta-learner KELM model. Different base learners extract different features of data based on model characteristics. This method takes into account the differences in training principles of different algorithms and gives full play to the advantages of each model in the prediction process. The stronger the learning ability of the base learner, the smaller the degree of correlation between them, and the better the final prediction effect. Compared with the traditional single model prediction, the dam seepage prediction model based on stacking integration of the present invention has higher accuracy and generalization ability.

[0058] In order to implement the above embodiment, Figure 5 As shown, this embodiment also provides a dam leakage intelligent monitoring and prediction and early warning system, including: The data acquisition module 100 is used to collect multivariate data of various monitoring devices of the reservoir dam; wherein the multivariate data includes: dam pressure, dam displacement, dam cracks, local temperature, upstream and downstream water levels, seepage pressure, precipitation and seepage; The data preprocessing module 200 is used to transmit the multivariate data through dual-channel communication and store it in the local dam data history library, and to clean and eliminate the data and perform cross-validation reconstruction operations to obtain a training data set; The model algorithm module 300 is used to construct an integrated prediction model based on stacking, and use the transformer, BiGRU, and LightGBM models as base learners of the integrated prediction model, use the KELM model as the meta-learner of the integrated prediction model, use the training data set to train the base learner to obtain the prediction result of the base learner, and use the prediction result of the base learner to train the KELM model to output the dam seepage prediction result; use the good point set initialization and Gaussian random walk strategy to improve the Ban Kingfisher optimization algorithm PKO, and use the improved Ban Kingfisher optimization algorithm IPKO to optimize the parameters of the KELM model; The early warning module 400 is used to set the early warning level based on the historical dam seepage data, output the final dam seepage prediction result based on the optimized KELM model, and determine whether the final dam seepage prediction result is within the early warning range. If it exceeds the prediction level, an early warning is issued.

[0059] According to the intelligent monitoring and prediction and early warning system for dam leakage of the embodiment of the present invention, multiple dam data are used as input to predict the dam seepage, which ensures the accuracy of dam seepage prediction and plays an important role in dam early warning. The prediction model based on stacking integration proposed by the present invention uses transformer, BiGRU, and LightGBM as base learners, KELM as meta-learner, and uses the improved Kingfisher optimization algorithm IPKO to optimize the parameters of the meta-learner KELM model. Different base learners extract different features of data based on model characteristics. This method takes into account the differences in training principles of different algorithms and gives full play to the advantages of each model in the prediction process. The stronger the learning ability of the base learners, the smaller the degree of correlation between them, and the better the final prediction effect. Compared with the traditional single model prediction, the dam seepage prediction model based on stacking integration of the present invention has higher accuracy and generalization ability.

[0060] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0061] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

Claims

1. A dam leakage intelligent monitoring and prediction and early warning method, characterized in that: include: Collect multivariate data from various monitoring devices of the reservoir dam; wherein the multivariate data includes dam pressure, dam displacement, dam cracks, local temperature, upstream and downstream water levels, seepage pressure, precipitation and seepage; Preprocess the multivariate data to obtain a training data set; Construct an integrated prediction model based on stacking, and use transformer, BiGRU, and LightGBM models as base learners of the integrated prediction model, use the KELM model as the meta-learner of the integrated prediction model, use the training data set to train the base learner to obtain the prediction result of the base learner, and use the prediction result of the base learner to train the KELM model to output the prediction result of the dam seepage; The PKO algorithm is improved by using good point set initialization and Gaussian random walk strategy, and the parameters of the KELM model are optimized by using the improved IPKO algorithm. The warning level is set based on the historical dam seepage data, and the final dam seepage prediction result is output based on the optimized KELM model. It is judged whether the final dam seepage prediction result is within the warning range. If it exceeds the prediction level, a warning is issued.

2. The method according to claim 1, characterized in that Collect multivariate data from various monitoring devices at reservoirs and dams, including: Collect dam pressure data through pressure sensors; Collect dam displacement data through displacement sensors; Collect dam crack data through video monitoring; Collect local temperature data through temperature sensors; Collect water level data upstream and downstream of the dam through submersible water level gauges; Seepage pressure data were collected via pore water pressure gauges; Collect precipitation data through rain gauges; Dam seepage data were collected using a magnetostrictive weir gauge.

3. The method according to claim 2, characterized in that Preprocess the multivariate data to obtain the training data set, including: Through dual-channel multi-variable data communication transmission, different networks are selected for data transmission in different situations, and after the data is transmitted to the detection platform, data elimination, mining and fusion operations are automatically performed on the data; The processed multivariate data were subjected to three-fold cross validation to obtain the training data set.

4. The method according to claim 3, characterized in that The transformer model is used as the first primary learner of the integrated prediction model; the transformer model consists of two parts: the encoder and the decoder, which are used to process the input sequence and generate the output sequence respectively; The encoder consists of position encoding, self-attention mechanism, feedforward neural network, residual connection and layer normalization, and the decoder consists of self-attention mechanism, encoder-decoder attention mechanism and feedforward neural network; Transformer self-attention mechanism has three inputs, Q query, K key, V value three matrices all come from the same input, calculate the dot product between Q and K, and divide by ,in is the dimension of the K vector; the result is normalized to a probability distribution using the Softmax operation, and then multiplied by the matrix V to obtain the representation of the weighted sum; The calculation process is expressed as: (1) BiGRU is used as the second primary learner of the integrated prediction model. The BiGRU model introduces a bidirectional structure based on the GRU model. GRU is a gated recurrent neural network unit, including an update gate and a reset gate. The relevant calculation formulas for the GRU update gate and reset gate are as follows: (2) (3) in is the update gate of GRU, is the reset gate of GRU, and are the weight matrices of the update gate and the reset gate, respectively. It is the Sigmoid function, which is used to convert the target data into a value in the range of 0 to 1. is the previous time step; The output of the GRU model at the current moment is related to the previous moment. The specific formula of the hidden state at the previous moment is as follows: (4) in It records the current moment. is the weight matrix of the candidate hidden state; The final output of the GRU model is related to the output at the current moment and the hidden state at the previous moment. The specific formula is as follows: (5) in For the final output result; LightGBM is used as the third primary learner of the integrated prediction model; LightGBM adopts a leaf-node-based decision tree algorithm and introduces mutually exclusive feature bundling algorithm and histogram bias algorithm; The KELM model is used as the second-layer meta-learner of the integrated model; ELM is a single hidden layer feedforward neural network, and the learning objective function F(x) is expressed as a matrix: (6) in is the input vector, , is the hidden layer node output, is the output weight, is the expected output; Transform network training into a linear system solution problem. according to Determine, among which for The generalized inverse matrix of and the identity matrix , then the least squares solution of the output weight is: (7) Introduce the kernel function into ELM, the kernel matrix is: (8) in , is the experimental input vector, then the objective function F(x) can be expressed as: (9) in For a given training sample, is the sample size, is the kernel function.

5. The method according to claim 4, characterized in that The good point set initialization and Gaussian random walk strategy are used to improve the Ban Kingfisher optimization algorithm PKO, and the improved Ban Kingfisher optimization algorithm IPKO is used to optimize the parameters of the KELM model, including: The objective function of the Ban Kingfisher Optimization Algorithm PKO is set as the parameter regularization coefficient C and kernel function coefficient S of the KELM model. These two parameters are used as the input of the algorithm and the upper and lower limits are set. The fitness value is the training set error index. The original Kingfisher optimization algorithm is initialized randomly, and the good point set strategy is used to initialize it instead of the original random initialization: (10) Deviations meet: (11) in For the best point collection, For the best point, For points, Is only with The relevant constants, is an arbitrarily small positive number, yes The minimum prime number that satisfies Represents dimension, upper represents the upper bound, and lower represents the lower bound; The position update method of the Ban Kingfisher Optimization Algorithm PKO is divided into the exploration stage and the development stage. The Gaussian random walk strategy is introduced in the exploration stage and the development stage. The specific formula for the exploration phase is as follows: (12) in , Indicates individual kingfisher, is the position of individual i at time t+1; parameter , dim represents the dimension, represents a random number in a normal distribution; N represents the population size; , is an individual randomly selected from the dominant population; The value of parameter T is determined dynamically according to the current strategy. The exploration phase is divided into two strategies: perching and hovering; The parameter T in the habitat strategy is calculated as follows: (13) in is the maximum number of iterations, BF is set to 8, and rand is a random number between 0 and 1. The parameter T in the hover strategy is calculated as follows: (14) and are the fitness values ​​of the i-th and j-th kingfishers respectively; The position update formula in the development phase is as follows: (15) in is the optimal position of the kingfisher population at time t, ; , and Represents hunting ability, calculated as follows: (16) (17) (18) During the iteration process, the individual position of the kingfisher corresponding to the best fitness value is determined according to the training error index. The optimal position of the kingfisher is used as the output to update the parameter regularization coefficient C and kernel function coefficient S of the KELM prediction model.

6. An intelligent monitoring and prediction and early warning system for dam leakage, characterized in that: include: A data acquisition module is used to collect multivariate data from various monitoring devices of the reservoir dam; wherein the multivariate data includes: dam pressure, dam displacement, dam cracks, local temperature, upstream and downstream water levels, seepage pressure, precipitation and seepage; The data preprocessing module is used to transmit the multivariate data through dual-channel communication and store it in the local dam data history library, and to clean and eliminate the data and perform cross-validation reconstruction operations to obtain the training data set; The model algorithm module is used to construct an integrated prediction model based on stacking, and use the transformer, BiGRU, and LightGBM models as the base learners of the integrated prediction model, and use the KELM model as the meta-learner of the integrated prediction model. The base learners are trained with the training data set to obtain the prediction results of the base learners, and the prediction results of the base learners are used to train the KELM model to output the prediction results of the dam seepage; the Kingfisher Optimization Algorithm PKO is improved by using the good point set initialization and Gaussian random walk strategy, and the parameters of the KELM model are optimized by using the improved Kingfisher Optimization Algorithm IPKO; The early warning module is used to set the early warning level based on the historical dam seepage data, output the final dam seepage prediction result based on the optimized KELM model, and determine whether the final dam seepage prediction result is within the early warning range. If it exceeds the prediction level, an early warning will be issued.

Images