A method for evaluating the safety status of sluices based on multi-source data fusion of mutual information

Abstract

The present invention discloses a method for evaluating the safety state of a sluice based on multi-source data fusion of mutual information, including: 1. Obtaining various safety monitoring time series data of the sluice monitoring area and performing standardized processing; 2. Screening influencing factors according to the principles of mutual information quantity, maximum correlation and minimum redundancy; 3. Constructing a Stacking ensemble learning model, where the first-layer training network is based on the RF, LSTM and SVM algorithms, and the second-layer training network integrates the output results of the first layer with the GBDT model to obtain the optimal model. 4. Predicting the deformation result of the sluice at the next moment through the constructed Stacking ensemble learning model; 5. Combining the historical deformation data of the sluice body and the predicted deformation data of the sluice to evaluate the safety state of the sluice. The present invention fuses various factors affecting the deformation of the sluice, predicts the future deformation situation, and evaluates the safety state of the sluice according to the deformation data of the sluice body.

Description

Technical Field

[0001] The present invention relates to a method for evaluating the safety state of a sluice based on multi-source data fusion of mutual information. Background Art

[0002] The deformation monitoring of a sluice reflects the overall deformation and stress state of the sluice, which is an important manifestation of the safety state of the sluice. Although certain progress has been made in the research and application of the theory and methods for analyzing the data of sluice safety monitoring, which has played a huge role in ensuring the safe operation of the sluice. However, there are still many problems and deficiencies in the analysis model and prediction method. The traditional physics-based method is difficult to achieve an ideal prediction effect when dealing with complex situations. And the data-driven method cannot integrate multiple factors, and there is a large error between the prediction result and the actual deformation. Summary of the Invention

[0003] In order to solve the above-mentioned deficiencies existing in the prior art, the present invention proposes a method for evaluating the safety state of a sluice based on multi-source data fusion of mutual information, in order to be able to use different types of monitoring data to predict the deformation of the sluice, so as to realize real-time monitoring and early warning of the deformation of the sluice, and ensure the accuracy and timeliness of the early warning, and further make a correct evaluation of the safety state of the sluice.

[0004] In order to achieve the above object of the invention, the following technical solutions are adopted:[[]]END]]

[0005] A method for evaluating the safety state of a sluice based on multi-source data fusion of mutual information according to the present invention is characterized in that it is carried out according to the following steps:[[]]END]]

[0006] Step 1: Obtain the time series set x of P sluice deformation factors in the monitoring area of the sluice, where the time series of the jth sluice deformation factor is denoted as x j =[x 1j ,x 2j ,…,x ij ,…,x nj T ; x ij represents the data at the ith moment of the jth sluice deformation factor, i = 1, 2, …, n, j = 1, 2, …, P, and n represents the time series length;

[0007] Perform standardization processing on x ij to obtain the data X ij at the ith moment of the standardized jth sluice deformation factor, so as to obtain the time series X j =[X 1j ,X 2j ,…,X ij ,…,X nj T ​​; Furthermore, the standardized set of P sluice deformation factors X is obtained;

[0008] Obtain the sluice body deformation data Y at the i-th moment i , so as to obtain the sluice body deformation sequence Y = [Y 1 , Y 2 , …, Y i , …, Y n T ;

[0009] Step 2: Screen the influencing factors according to the principle of mutual information, maximum correlation and minimum redundancy:

[0010] Step 2.1: Use Equation (1) to calculate the mutual information I(X j , Y) between the time series X of the j-th sluice deformation factor and the sluice body deformation sequence Y: j :

[0011]

[0012] In Equation (1), p is the probability density function; Y k is the sluice body deformation data at the k-th moment; k = 1, 2, …, n;

[0013] Step 2.2: Calculate the mutual information between the P sluice deformation factors and the sluice body deformation sequence Y in turn according to Equation (1), and select the time series of the influencing factor with the largest mutual information and denote it as X′ 1 ;

[0014] Define the influencing factor set as S and initialize S = X′ 1 ;

[0015] Step 2.3: After assigning X - X′ 1 to X, according to the judgment condition Θ shown in Equation (2), select the influencing factor X′ with the largest correlation with the sluice body deformation sequence Y and the smallest redundancy with S from X 2 ;

[0016]

[0017] In Equation (2), num(X) is the number of influencing factors in the sluice deformation factor set X, X h represents the time series of the h-th sluice deformation factor in S, and I(X u , X h ) is the mutual information between the time series X u of the u-th sluice deformation factor in X and the time series X h of the h-th sluice deformation factor in S;

[0018] Step 2.4: Assign X - X′ 2 ​Assign to X, and X' 1 +X' 2 After assigning to S, continuously update according to Equation (2), and screen out the time series of L groups of influencing factors that have a greater impact on the deformation of the sluice, denoted as X' 1 ,X' 2 ,…,X' l ,…,X' L , where X' l represents the time series of the l-th sluice deformation factor after screening, and X' l =[X' 1l ,X' 2l ,…,X' kl ,…,X' nl T ,X' kl represents the data at the k-th moment of the j-th sluice deformation factor after screening;

[0019] Step 3: Construct a Stacking ensemble learning model, including: the RF network, LSTM network, and SVM classification network in the first layer, and the GBDT model in the second layer;

[0020] Step 3.1: From the data X' k1 ,X' k2 ,…,X' kL of L groups of influencing factors at the k-th moment and the sluice body deformation data Y k at the k-th moment, form the k-th group of relationship data, and thus obtain the relationship matrix Z with a dimension of (n - 1)×(L + 1) using Equation (3), and split it into the training set Z tran with a dimension of M×(L + 1) and the test set Z test with a dimension of N×(L + 1), where M + N = n - 1;

[0021] Use Equation (4) to select the sluice body deformation data at the next moment in the relationship matrix Z as the prediction set O, and correspondingly divide it into the training prediction set O tran with a dimension of M×1 and the test prediction set O test ;

[0022]

[0023] O = [Y 2 ,…,Y k ,…,Y n T (4)

[0024] Step 3.2: The training set Z tran and the training prediction set O tran ​​They are respectively divided into A equal parts to obtain A equal parts of training subsets and A equal parts of training prediction subsets;

[0025] Step 3.3: Respectively take A - 1 equal parts of the training subsets as the inputs of the RF network, LSTM network, and SVM classification in the first layer, and take the corresponding A - 1 equal parts of the training prediction subsets as the outputs of the RF network, LSTM network, and SVM classification network in the first layer; thus, train the RF network, LSTM network, and SVM classification in the first layer respectively to obtain the trained RF model, LSTM network, and SVM classification network;

[0026] Take the other equal part of the training subset as the validation set and input it into the trained RF model, LSTM network, and SVM classification network respectively, so as to obtain a prediction result of the RF model validation set, a prediction result of the LSTM network validation set, and a prediction result of the SVM classification network validation set;

[0027] Step 3.4: Input the test set Z test into the trained RF model, LSTM network, and SVM classification network respectively, so as to obtain a prediction result of the RF model test set, a prediction result of the LSTM network test set, and a prediction result of the SVM classification network test set;

[0028] Step 3.5: According to the process of Step 3.3 - Step 3.4, use the cross - validation method to obtain A prediction results of the RF model validation set, A prediction results of the LSTM network validation set, A prediction results of the SVM classification network validation set, as well as A prediction results of the RF model test set, A prediction results of the LSTM network test set, and A prediction results of the SVM classification network test set;

[0029] Step 3.6: Combine the prediction results of the A RF model validation sets to obtain a new RF model training set;

[0030] Combine the prediction results of the A LSTM network validation sets to obtain a new LSTM network training set;

[0031] Combine the prediction results of the A SVM classification network validation sets to obtain a new SVM classification network training set;

[0032] Combine the new RF model training set, the new LSTM network training set, and the new LSTM network training set into a new training set Z′ tran ;

[0033] Take the average of the prediction results of the A RF model test sets to obtain a new RF model test set;

[0034] The prediction results of the test sets of A LSTM networks are averaged to obtain a new test set of the LSTM network;

[0035] The prediction results of the test sets of A SVM classification networks are averaged to obtain a new test set of the SVM classification network;

[0036] The new RF model test set, the new LSTM network test set, and the new LSTM network test set are combined into a new test set Z'; test ;

[0037] Step 3.7: Use the new training set Z' tran and the new test set Z' test as the input of the GBDT model in the second layer, and use the training prediction set O tran , the test prediction set O test as the output of the GBDT model, so as to train the GBDT model and obtain the optimal GBDT model;

[0038] The trained RF model, LSTM network, SVM classification network, and optimal GBDT model constitute a trained Stacking ensemble learning model;

[0039] Step 4: Input the data X' n1 , X' n2 , …, X' nL at the current nth moment and the gate deformation data Y n at the nth moment into the trained Stacking ensemble learning model, and output the predicted gate deformation data Y n+1 at the (n + 1)th moment;

[0040] Step 5: Combine the historical deformation data Y 1 , Y 2 , …, Y i , …, Y n and the predicted data Y n+1 , and calculate the gate deformation acceleration rate α. When α = 0, it means that the gate is in the uniform deformation stage and the current sluice is basically normal; when α < 1, it means that the gate is in the first stage of accelerated deformation and there are potential safety hazards in the current sluice; when α ≈ 1, it means that the gate is in the second stage of accelerated deformation and the current sluice is in an abnormal state; when α > 1, it means that the gate is in the third stage of accelerated deformation and the current sluice is in danger.

[0041] An electronic device according to the present invention includes a memory and a processor, characterized in that the memory is used to store a program for supporting the processor to execute the sluice safety status evaluation method, and the processor is configured to execute the program stored in the memory.

[0042] A computer-readable storage medium of the present invention stores a computer program thereon. The feature is that when the computer program is run by a processor, it executes the steps of the water gate safety status evaluation method.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] 1. According to the mutual information between the deformation of the water gate and various influencing factors, the present invention selects several main influencing factors with relatively large correlations for fusion, and then predicts the future data situation, solving the limitation that the previous detection equipment can only monitor in real time and cannot predict data. It can not only monitor in real time with high accuracy but also has strong timeliness.

[0045] 2. The present invention introduces the Stacking integration model. By selecting different models as the base learners, diversity is introduced. At the same time, using the meta-learner as the integration method can better learn the relationship between the prediction results of multiple base learners and the deformation of the water gate.

[0046] 3. The present invention has a wide range of applications: without adding additional auxiliary information, this method can be widely used for the deformation monitoring of water gates and dams, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is the architecture diagram of the Stacking integration learning model proposed by the present invention;

[0048] Figure 2 It is the first-layer network architecture diagram of the Stacking integration learning model proposed by the present invention;

[0049] Figure 3 It is the process of the water gate safety monitoring and early warning model proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] In this embodiment, a water gate safety status evaluation method based on mutual information multi-source data fusion includes the following steps:

[0051] Step 1: Obtain the time series of P water gate deformation factors in the water gate monitoring area. Among them, the time series of the jth water gate deformation factor is denoted as x j =[x 1j ,x 2j ,…,x ij ,…,x nj T ; x ij represents the data of the i-th moment of the jth water gate deformation factor, i = 1, 2,..., n, j = 1, 2,..., P, and n represents the time series length;

[0052] For x ij ​Perform standardization processing to obtain the data \(X\) of the \(i\)-th moment of the \(j\)-th sluice deformation factor after standardization ij , so as to obtain the time series \(X\) of the \(j\)-th sluice deformation factor after standardization j =\([X 1j ,X 2j ,…,X ij ,…,X nj T ; Furthermore, obtain the set \(X\) of \(P\) sluice deformation factors after standardization;

[0053] Obtain the sluice body deformation data \(Y\) at the \(k\)-th moment k , so as to obtain the sluice body deformation sequence \(Y = [Y 1 ,Y 2 ,…,Y k ,…,Y n T , \(k = 1, 2, …, n\);

[0054] Step 2: Screen influencing factors according to the principle of mutual information, maximum correlation and minimum redundancy:

[0055] Step 2.1: Use Equation (1) to calculate the mutual information \(I(X j ,Y)\) between the time series \(X\) of the \(j\)-th sluice deformation factor and the sluice body deformation sequence \(Y\): j :

[0056]

[0057] In Equation (1), \(p\) is the probability density function; \(Y k is the sluice body deformation data at the \(k\)-th moment; \(k = 1, 2, …, n\);

[0058] Step 2.2: Calculate the mutual information between the \(P\) sluice deformation factors and the sluice body deformation sequence \(Y\) in turn according to Equation (1), and select the time series of the influencing factor with the largest mutual information and denote it as \(X'\) 1 ;

[0059] Define the influencing factor set as \(S\), and initialize \(S = X'\) 1 ;

[0060] Step 2.3: After assigning \(X - X'\) 1 to \(X\), according to the judgment condition \(\Theta\) shown in Equation (2), select the influencing factor \(X'\) from \(X\) that has the largest correlation with the sluice body deformation sequence \(Y\) and the smallest redundancy with \(S\) 2 ;

[0061]

[0062] In Equation (2), \(num(X)\) is the number of influencing factors in the sluice deformation factor set \(X\), \(X​​h Denote the time series of the h-th sluice deformation factor in S as \(I(X u , X h ), where \(X u is the time series of the u-th sluice deformation factor in X, and \(X h is the mutual information between the time series of the h-th sluice deformation factor in S and \(X

[0063] Step 2.4: Assign \(X - X'\) to \(X\), and assign \(X' + X'\) to \(S\). Then, continuously update according to Equation (2), and screen out the time series of the L groups of influencing factors that have a greater impact on the sluice deformation, denoted as \(X' 2 , \(X' 1 , \(\cdots\), \(X' 2 , \(\cdots\), \(X' 1 , \(\cdots\), \(X' 2 , \(\cdots\), \(X' l , \(\cdots\), \(X' L , where \(X' l represents the time series of the l-th sluice deformation factor after screening, and \(X' l = [X' 1l , \(X' 2l , \(\cdots\), \(X' kl , \(\cdots\), \(X' nl T , and \(X' kl represents the data of the k-th moment of the j-th sluice deformation factor after screening. Specifically for this experiment, refer to Table 1. In this embodiment, the three influencing factors of aging, water pressure, and cracks are selected.

[0064] Table 1

[0065]

[0066] Step 3: Construct a Stacking ensemble learning model, as shown in Figure 1 , which includes: the RF network, LSTM network, and SVM classification network in the first layer, and the GBDT model in the second layer;

[0067] Step 3.1: Use the data \(X' k1 , \(X' k2 , \(\cdots\), \(X' kL of the L groups of influencing factors at the k-th moment and the sluice deformation data \(Y k at the k-th moment to form the k-th group of relationship data. Then, use Equation (3) to obtain a relationship matrix \(Z\) with a dimension of \((n - 1)\times(L + 1)\), and split it into a training set \(Z tran with a dimension of \(M\times(L + 1)\) and a test set \(Z test with a dimension of \(N\times(L + 1)\), where \(M + N = n - 1\);

[0068] ​Select the gate deformation data at the next moment in the relationship matrix Z using Equation (4) as the prediction set O, and correspondingly divide it into a training prediction set O with a dimension of M×1 tran , a test prediction set O with a dimension of N×1 test ;

[0069]

[0070] O = [Y 2 ,…,Y k ,…,Y n T (4)

[0071] Step 3.2: Divide the training set Z tran and the training prediction set O tran into A equal parts respectively to obtain A equal parts of training subsets and A equal parts of training prediction subsets; as Figure 2 shown, take A = 5;

[0072] Step 3.3: Take A - 1 equal parts of the training subsets as the inputs of the RF network, LSTM network and SVM classification in the first layer respectively, and take the corresponding A - 1 equal parts of the training prediction subsets as the outputs of the RF network, LSTM network and SVM classification network in the first layer respectively; thus, train the RF network, LSTM network and SVM classification in the first layer respectively to obtain the trained RF model, LSTM network and SVM classification network;

[0073] Take the other equal part of the training subset as the validation set and input it into the trained RF model, LSTM network and SVM classification network respectively, so as to obtain the prediction results of an RF model validation set, a prediction result of an LSTM network validation set, and a prediction result of an SVM classification network validation set;

[0074] Step 3.4: Input the test set Z test into the trained RF model, LSTM network and SVM classification network respectively, so as to obtain the prediction results of an RF model test set, a prediction result of an LSTM network test set, and a prediction result of an SVM classification network test set;

[0075] Step 3.5: According to the process of Step 3.3 - Step 3.4, use the cross - validation method to obtain A prediction results of the RF model validation set, A prediction results of the LSTM network validation set, A prediction results of the SVM classification network validation set, as well as A prediction results of the RF model test set, A prediction results of the LSTM network test set, and A prediction results of the SVM classification network test set;

[0076] ​Step 3.6: Combine the prediction results of the A RF model validation sets to obtain a new RF model training set;

[0077] Combine the prediction results of the A LSTM network validation sets to obtain a new LSTM network training set;

[0078] Combine the prediction results of the A SVM classification network validation sets to obtain a new SVM classification network training set;

[0079] Combine the new RF model training set, the new LSTM network training set, and the new LSTM network training set into a new training set Z′ tran ;

[0080] Average the prediction results of the A RF model test sets to obtain a new RF model test set;

[0081] Average the prediction results of the A LSTM network test sets to obtain a new LSTM network test set;

[0082] Average the prediction results of the A SVM classification network test sets to obtain a new SVM classification network test set;

[0083] Combine the new RF model test set, the new LSTM network test set, and the new LSTM network test set into a new test set Z′ test ;

[0084] Step 3.7: Use the new training set Z′ tran and the new test set Z′ test as the input of the GBDT model in the second layer, and use the training prediction set O tran , the test prediction set O test as the output of the GBDT model, so as to train the GBDT model and obtain the optimal GBDT model;

[0085] The trained RF model, LSTM network, SVM classification network, and optimal GBDT model constitute a trained Stacking ensemble learning model;

[0086] Step 4: Input the data X′ n1 , X′ n2 , …, X′ nL at the current nth moment and the gate deformation data Y n at the nth moment into the trained Stacking ensemble learning model, and output the predicted gate deformation data Y n+1 at the (n + 1)th moment;

[0087] Meanwhile, different influencing factors are selected based on the same sample set, and after fusion, they are brought into the Stacking model for prediction. Through the comparative analysis of the experimental results in Table 2, it can be seen that the fusion data of water pressure + aging + cracks has better prediction effect. Thus, it can be seen that the fusion method proposed in this patent has superiority compared with ordinary data fusion.

[0088] Table 2

[0089]

[0090] Step 5: Figure 3 is the flow chart of the safety monitoring and early warning model for the sluice. Combining the historical deformation data Y of the deformation of the sluice body 1 ,Y 2 ,…,Y i ,…,Y n and the predicted data Y n+1 , calculate the acceleration rate α of the deformation of the sluice body. When α = 0, it means that the sluice body is in the uniform deformation stage and the current sluice is basically normal; when α < 1, it means that the sluice body is in the first stage of accelerated deformation and there are potential safety hazards in the current sluice; when α ≈ 1, it means that the sluice body is in the second stage of accelerated deformation and the current sluice is in an abnormal state; when α > 1, it means that the sluice body is in the third stage of accelerated deformation and the current sluice is in a dangerous situation.

[0091] In this embodiment, an electronic device includes a memory and a processor. The memory is used to store a program that supports the processor to execute the above-mentioned sluice safety status evaluation method, and the processor is configured to execute the program stored in the memory.

[0092] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium. When the computer program is run by a processor, it executes the steps of the above-mentioned sluice safety status evaluation method.

Claims

1. A method for evaluating the safety state of a sluice based on multi-source data fusion of mutual information, characterized in that it is carried out according to the following steps: Step 1: Obtain the time series set of various sluice deformation factors , where the time series of the th sluice deformation factor is denoted as ; represents the data at the i-th moment of the th sluice deformation factor, , , represents the time series length; Pair is standardized to obtain the data at the \(i\)-th moment of the \( \)-th sluice deformation factor after standardization , so as to obtain the time series \(X\) of the \( \)-th sluice deformation factor after standardization , thus obtaining the time series \(X\) of the \( \)-th sluice deformation factor after standardization ; that is, \(X j = [X 1j , X 2j , \cdots, X ij , \cdots, X nj ] T \); furthermore, the set of \( \)-th sluice deformation factors after standardization is obtained j =[X 1j ,X 2j ,…,X ij ,…,X nj T ; and then the set of \( \)-th sluice deformation factors after standardization is obtained ; ​​ Obtain the deformation data of the brake body at the i-th moment , so as to obtain the brake body deformation sequence ; Step 2: Screen influencing factors according to the principle of mutual information quantity, maximum correlation principle and minimum redundancy principle: Step 2.1: Calculate the time series of the th sluice deformation factor and the sluice body deformation sequence mutual information : (1) In formula (1), is the probability density function; is the deformation data of the brake body at the k-th moment; ; Step 2.2: Calculate successively according to Equation (1) the mutual information of kinds of sluice deformation factors and the sluice body deformation sequence and select the time series of the influencing factor with the largest mutual information and denote it as ; Define the set of influencing factors as , and initialize ; Step 2.3: After assigning to , according to the judgment condition shown in formula (2) , select from the influencing factor with the greatest correlation with the gate body deformation sequence and the least redundancy with ; ; (2) In formula (2), is the set of factors causing the deformation of the sluice gate The number of internal influencing factors, denotes the th time series of the factors causing the deformation of the sluice gate, is the th time series of the factors causing the deformation of the sluice gate and the th time series of the factors causing the deformation of the sluice gate The mutual information of; Step 2.4: After assigning to and assigning to , continuously update according to Equation (2), and screen out the time series of the influencing factors that have a greater impact on the deformation of the sluice, denoted as , where represents the time series of the th sluice deformation factor after screening, and X′ =[X′ l , X′ 1l , …, X′ 2l , …, X′ kl , …, X′ nl T , represents the data at the kth moment of the th sluice deformation factor after screening;​ Step 3: Construct a Stacking ensemble learning model, including: an RF network, an LSTM network and an SVM classification network in the first layer, and a GBDT model in the second layer; Step 4: Take the current moment data and the gate deformation data at the moment as inputs to the trained Stacking ensemble learning model, and output the predicted gate deformation data at the moment; Step 5: Combine the historical deformation data of the gate body and the predicted data , and calculate the deformation acceleration rate of the gate body . When , it indicates that the gate body is in the stage of uniform deformation, and the current sluice is basically normal; when , it indicates that the gate body is in the first stage of accelerated deformation, and there are potential safety hazards in the current sluice; when , it indicates that the gate body is in the second stage of accelerated deformation, and the current sluice is in an abnormal state; when , it indicates that the gate body is in the third stage of accelerated deformation, and the current sluice is in danger.

2. A method for evaluating the safety state of a sluice based on multi-source data fusion of mutual information according to claim 1, characterized in that said Step 3 is carried out according to the following steps: Step 3.1: From the data of the group of influencing factors at the k-th moment and the gate deformation data at the k-th moment constitute the k-th group of relationship data, so as to obtain the relationship matrix with the dimension of by using Equation (3), and split it into the training set with the dimension of and the test set with the dimension of , where ; ​ Select the relationship matrix using Equation (4). The gate deformation data at the next moment in is used as the prediction set and is correspondingly divided into a training prediction set with a dimension of and a test prediction set with a dimension of . ;​ (3) O = [Y 2 , …, Y k , …, Y n T (4)​ Step 3.2: Divide the training set and the training prediction set into A equal parts respectively to obtain A equal parts of training subsets and A equal parts of training prediction subsets; Step 3.3: Respectively take A - 1 equal parts of the training subsets as the inputs of the RF network, LSTM network and SVM classification in the first layer, and take the corresponding A - 1 equal parts of the training prediction subsets as the outputs of the RF network, LSTM network and SVM classification network in the first layer; thus, train the RF network, LSTM network and SVM classification in the first layer respectively to obtain the trained RF model, LSTM network and SVM classification network; Take another equal part of the training subset as the validation set and input it into the trained RF model, LSTM network and SVM classification network respectively, so as to obtain a prediction result of the RF model validation set, a prediction result of the LSTM network validation set, and a prediction result of the SVM classification network validation set; Step 3.4, input the test set into the trained RF model, LSTM network, and SVM classification network respectively, so as to obtain a prediction result of the RF model test set, a prediction result of the LSTM network test set, and a prediction result of the SVM classification network test set accordingly; Step 3.5: According to the process of Step 3.3 - Step 3.4, use the cross - validation method to obtain A prediction results of the RF model validation set, A prediction results of the LSTM network validation set, A prediction results of the SVM classification network validation set, as well as A prediction results of the RF model test set, A prediction results of the LSTM network test set, and A prediction results of the SVM classification network test set; Step 3.6: Combine the prediction results of the A RF model validation sets to obtain a new RF model training set; Combine the prediction results of the A LSTM network validation sets to obtain a new LSTM network training set; Combine the prediction results of the A SVM classification network validation sets to obtain a new SVM classification network training set; Merge the new RF model training set, the new LSTM network training set, and the new LSTM network training set into a new training set ; Average the prediction results of the A RF model test sets to obtain a new RF model test set; Average the prediction results of the A LSTM network test sets to obtain a new LSTM network test set; Average the prediction results of the A SVM classification network test sets to obtain a new SVM classification network test set; Merge the new RF model test set, the new LSTM network test set, and the new LSTM network test set into a new test set ; Step 3.7: Use the new training set and the new test set as the input of the GBDT model in the second layer, and use the training prediction set , the test prediction set as the output of the GBDT model, so as to train the GBDT model and obtain the optimal GBDT model; The trained Stacking ensemble learning model is composed of the trained RF model, LSTM network, SVM classification network and the optimal GBDT model.

3. An electronic device, including a memory and a processor, characterized in that the memory is used to store a program that supports the processor to execute the method for evaluating the safety state of the sluice according to claim 1 or 2, and the processor is configured to execute the program stored in the memory.

4. A computer - readable storage medium, on which a computer program is stored, characterized in that When the computer program is run by a processor, it executes the steps of the lock safety status evaluation method according to claim 1 or 2.

Images