A dam safety monitoring and early warning system and method based on multiple filling of missing data

Abstract

The present invention provides a dam safety monitoring and early warning system and method based on multiple filling of missing data. The system and method combine the ATTGAN filling method of missing values ​​of Euclidean space data and the ATTGAN filling method of missing values ​​of component data to fill missing values ​​for data sets with different missing values ​​and abnormal values, thereby compensating for the influence of too many missing values ​​on the dam early warning system and avoiding excessive error rate of missing value filling due to improper selection of missing value filling method. The data set after data processing is then used to predict data at future moments by using a memory-reduced online sequence extreme learning machine optimized by an exponential distribution optimizer, thereby improving the accuracy and completeness of the monitoring and early warning system. The data is sent to a monitoring platform to judge the safety status of the dam and issue an alarm in time, thereby meeting the requirements for real-time monitoring of dam safety and improving the accuracy of dam safety management.

Description

Technical Field

[0001] The present invention belongs to the technical field of dam safety prediction and early warning, and in particular relates to a dam safety monitoring and early warning system and method based on multiple filling of missing data. Background Art

[0002] Traditional reservoir dam safety monitoring methods mainly rely on manual inspections and local monitoring, and lack timely, accurate and comprehensive information feedback. They have problems such as limited monitoring scope, large influence of human factors, and slow response speed, which makes it difficult to meet the needs of modern dam safety monitoring. Summary of the Invention

[0003] The first object of the present invention is to provide a dam safety monitoring and early warning system based on multiple filling of missing data.

[0004] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0005] A dam safety monitoring and early warning system based on multiple filling of data gaps includes a data acquisition module, a data processing module, an EDO-MDOS-ELM prediction module, and a monitoring and early warning module;

[0006] The data acquisition module is used to collect data on dam displacement, dam settlement, water level and precipitation;

[0007] The data processing module is used to obtain the data collected by the data acquisition module, and select an appropriate filling method based on the attribute transformation generative adversarial network filling method for missing values ​​of Euclidean space data and the attribute transformation generative adversarial network filling method for missing values ​​of component data, by determining whether the missing value is greater than 5% of the entire data set, so as to achieve the purpose of multiple filling of missing data;

[0008] The EDO-MDOS-ELM prediction module includes an exponential distribution optimizer (EDO) and a memory-depleted online sequential extreme learning machine (MDOS-ELM). The EDO optimizer optimizes the output weights of the memory-depleted online sequential extreme learning machine, enabling the prediction model to produce faster and more accurate prediction results. The module then inputs the data output by the data processing module and outputs prediction results for dam displacement, dam settlement, water level, and precipitation.

[0009] The monitoring and early warning module includes a video monitoring point and an alarm module. The alarm module determines whether to issue an alarm based on the dam displacement and dam settlement prediction results output by the EDO-MDOS-ELM prediction module.

[0010] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0011] As a preferred technical solution of the present invention: the data acquisition module includes a multi-point sensor, an ultrasonic sensor and a weather station;

[0012] The multi-point sensor is arranged on the dam and is used to transmit signals to the Beidou satellite to obtain the real-time position coordinate information of the point, and observe the displacement and settlement of the dam body by recording the changes in the coordinates;

[0013] There are multiple ultrasonic sensors installed upstream of the dam, and the ultrasonic sensors are used to obtain real-time water level changes and output corresponding water level data;

[0014] The weather station is arranged near the dam and is used to obtain precipitation data.

[0015] As a preferred technical solution of the present invention: the data processing module is implemented as follows:

[0016] Step S1: Input the data collected by the data collection module, perform missing value statistics on the data, and calculate the missing value ratio;

[0017] Step S2: Create a data matrix ,in, , respectively represent the data of dam displacement, dam settlement, water level and precipitation; for a given variable , whose missing index is ;variable The observation value in is recorded as , missing values ​​are recorded as ;the remaining P -1 variable corresponding to the row in The data is recorded as , the corresponding row is The data is recorded as ;

[0018] Step S3: Determine whether the proportion of missing values ​​is greater than 5% of the entire data set. If so, proceed to step S4; otherwise, proceed to step S5.

[0019] Step S4: Execute the ATTGAN method to fill missing values ​​of Euclidean spatial data;

[0020] Step S41: Use mean filling method to Perform initial filling;

[0021] Step S42: The set of indices for the missing columns in Q And arrange the missing variables from small to large according to the missing rate;

[0022] Step S43: Determine whether the data meets the stopping criteria , when the stopping criterion is not met , then execute step S44; if the stopping criterion is met, the final filling matrix is ​​obtained; the above stopping criterion The loop stops if the difference between the new padding matrix and the previous padding matrix increases.

[0023] Step S44: Store the existing fill matrix, recorded as ;

[0024] for ,right and The attribute transformation generative adversarial network discriminator is used for classification. The discriminator is expressed as:

[0025] ;

[0026] According to the classification results, use predict , update the filling matrix using the obtained prediction value;

[0027] Step S5: Execute the ATTGAN method to fill in the missing values ​​of the component data;

[0028] Step S51: Select attribute conversion to generate adversarial network to obtain the initial filling matrix value of the filling matrix seat, and the initial matrix is ​​recorded as ;

[0029] Step S52: Through elementary transformation of the matrix, the missing value is adjusted to the first position of the data matrix. At this time, the matrix is ​​recorded as ;

[0030] Step S53: Use isometric logarithmic ratio transformation to swap the rows and columns of the data matrix From the simplex space to the Euclidean space, the transformed matrix is ​​recorded as ;

[0031] Step S54: Generate adversarial network based on the attribute transformation in step S44 to transform the data matrix Fill in and get the data matrix ;

[0032] Step S55: Using the inverse transformation of the isometric logarithmic ratio transformation, the data matrix Restore to component data and get the data matrix after inverse transformation ;

[0033] Step S56: Since the sum of each row of the original component data is different, it is necessary to fill the data matrix Make corrections; the final filling value is:

[0034]

[0035] Where, is the correction factor.

[0036] As a preferred technical solution of the present invention: the EDO-MDOS-ELM prediction module implementation process is as follows:

[0037] Initial stage: Determine the input of EDO-MDOS-ELM prediction module as the data of dam displacement, dam settlement, water level and precipitation filled by data processing module, and divide the input into time series z Sample blocks; input the first sample block Enter the MDOS-ELM initial model, which is represented as follows:

[0038] .

[0039] Where, is the input weight value after optimization by the Exponential Distribution Optimizer (EDO), is the weight vector, is the hidden layer output matrix; output weight vector ; In the online update phase, add training model samples After that, calculate the hidden layer output matrix , introducing the memory factor after the exponential distribution optimizer (EDO) optimization Then calculate the output weight vector , and update the MDOS-ELM model:

[0040]

[0041] On input z After 3 sample blocks, the MDOS-ELM model is obtained as follows:

[0042]

[0043] The output matrix is ​​calculated, which is the data of predicted dam displacement, dam settlement, water level and precipitation.

[0044] As a preferred technical solution of the present invention: the exponential distribution optimizer implementation process is as follows:

[0045] Step S1: Population initialization, its expression is as follows:

[0046]

[0047] Where, Indicates the input data after filling i The first dataj Dimensional variables, and Respectively represent j The lower and upper bounds of the dimensional variables, for Random numbers between ; the data are dam displacement, dam settlement, water level and precipitation;

[0048] Step S2: Development phase;

[0049] Step S21: Convert the guiding solution into the average of the first three solutions sorted from best to worst in terms of fitness value. The expression is as follows:

[0050]

[0051] Where, t is the current iteration number, It is t The best individual in the generation population, It is t The suboptimal individuals in the generation population, It is t The third best individual in the generation population;

[0052] Step S22: Define the original population matrix as the first memoryless position matrix, and then perform position update. The expression is as follows:

[0053]

[0054] Where, represents the first position in the memoryless position matrix i Individuals, is A random number uniformly generated in a 、 b is the weight factor, is the mean vector;

[0055] Step S3: exploration stage;

[0056] During the exploration phase, i The position update expression of each individual is as follows:

[0057]

[0058] Where, represents the average position of the population, c is the weight factor, and represents a random positive integer, and ;

[0059] Step S4: population update;

[0060] Output the new population after position update to form a new population matrix;

[0061] Step S5: Update the fitness value;

[0062] Update the fitness value, re-sort the population, and then store the optimal output weight and memory factor.

[0063] As a preferred technical solution of the present invention: the implementation process of the monitoring and early warning module is as follows:

[0064] The monitoring and early warning module includes video monitoring points and an alarm module. The dam displacement and dam settlement prediction results output by the model algorithm are input into the alarm module. The alarm module determines whether the distance between the coordinate points exceeds the threshold. If it exceeds the threshold, it means that there is a safety risk to the dam, and an alarm is issued. The water level and precipitation data tables, trend charts, and change trajectory charts displayed on the platform are then deeply analyzed, and video playback is watched in conjunction with the video monitoring points to judge the movement changes on the surface and inside of the dam. If risks arise, timely response and appropriate measures are taken.

[0065] Another object of the present invention is to provide a dam safety monitoring and early warning method based on multiple filling of missing data.

[0066] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0067] A dam safety monitoring and early warning method based on multiple filling of missing data includes the following steps:

[0068] First, the dam displacement, dam settlement, water level and precipitation data of a certain dam are collected at certain time intervals;

[0069] Secondly, based on the collected data, determine whether the missing value is greater than 5%, and select the ATTGAN filling method for missing values ​​of Euclidean spatial data or the ATTGAN filling method for missing values ​​of component data to complete the data processing;

[0070] Thirdly, during the model prediction process, the processed data set was divided into a training set and a test set in a ratio of 7:3 to reduce the risk of overfitting. At the same time, an exponential optimization distributor was used to optimize the output weights of the MDOS-ELM initial model, and finally the prediction results of various dam data were obtained.

[0071] Finally, based on the prediction results of dam displacement and dam settlement, it is determined whether to issue an alarm.

[0072] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0073] As a preferred technical solution of the present invention: the data processing process is as follows:

[0074] Step S1: Perform missing value statistics on the collected data and calculate the missing value ratio;

[0075] Step S2: Create a data matrix ,in, , respectively represent the data of dam displacement, dam settlement, water level and precipitation; for a given variable , whose missing index is ;variable The observation value in is recorded as , missing values ​​are recorded as ;the remaining P -1 variable corresponding to the row in The data is recorded as , the corresponding row is The data is recorded as ;

[0076] Step S3: Determine whether the proportion of missing values ​​is greater than 5% of the entire data set. If so, proceed to step S4; otherwise, proceed to step S5.

[0077] Step S4: Execute the ATTGAN method to fill missing values ​​of Euclidean spatial data;

[0078] Step S41: Use mean filling method to Perform initial filling;

[0079] Step S42: The set of indices for the missing columns in Q And arrange the missing variables from small to large according to the missing rate;

[0080] Step S43: Determine whether the data meets the stopping criteria , when the stopping criterion is not met , then execute step S44; if the stopping criterion is met, the final filling matrix is ​​obtained; the above stopping criterion The loop stops if the difference between the new padding matrix and the previous padding matrix increases.

[0081] Step S44: Store the existing fill matrix, recorded as ;

[0082] for ,right and The attribute transformation generative adversarial network discriminator is used for classification. The discriminator is expressed as:

[0083] ;

[0084] According to the classification results, use predict , update the filling matrix using the obtained prediction value;

[0085] Step S5: Execute the ATTGAN method to fill in the missing values ​​of the component data;

[0086] Step S51: Select attribute conversion to generate adversarial network to obtain the initial filling matrix value of the filling matrix seat, and the initial matrix is ​​recorded as ;

[0087] Step S52: Through elementary transformation of the matrix, the missing value is adjusted to the first position of the data matrix. At this time, the matrix is ​​recorded as ;

[0088] Step S53: Use isometric logarithmic ratio transformation to swap the rows and columns of the data matrix From the simplex space to the Euclidean space, the transformed matrix is ​​recorded as ;

[0089] Step S54: Generate adversarial network based on the attribute transformation in step S44 to transform the data matrix Fill in and get the data matrix ;

[0090] Step S55: Using the inverse transformation of the isometric logarithmic ratio transformation, the data matrix Restore to component data and get the data matrix after inverse transformation ;

[0091] Step S56: Since the sum of each row of the original component data is different, it is necessary to fill the data matrix Make corrections; the final filling value is:

[0092]

[0093] Where, is the correction factor.

[0094] As a preferred technical solution of the present invention, the initialization process of the MDOS-ELM initial model is as follows: the data of dam displacement, dam settlement, water level and precipitation filled by data processing are divided into z Sample blocks; input the first sample block Enter the MDOS-ELM initial model, which is represented as follows:

[0095] .

[0096] Where, is the input weight value after optimization by the Exponential Distribution Optimizer (EDO), is the weight vector, is the hidden layer output matrix; output weight vector ; In the online update phase, add training model samples After that, calculate the hidden layer output matrix , introducing the memory factor after the exponential distribution optimizer (EDO) optimization Then calculate the output weight vector , and update the MDOS-ELM model:

[0097]

[0098] On input z After 3 sample blocks, the MDOS-ELM model is obtained as follows:

[0099]

[0100] The output matrix is ​​calculated, which is the data of predicted dam displacement, dam settlement, water level and precipitation.

[0101] The implementation process of the exponential distribution optimizer is as follows:

[0102] Step S1: Population initialization, its expression is as follows:

[0103]

[0104] Where, Indicates the input data after filling i The first data j Dimensional variables, and Respectively represent j The lower and upper bounds of the dimensional variables, for Random numbers between ; the data are dam displacement, dam settlement, water level and precipitation;

[0105] Step S2: Development phase;

[0106] Step S21: Convert the guiding solution into the average of the first three solutions sorted from best to worst in terms of fitness value. The expression is as follows:

[0107]

[0108] Where, t is the current iteration number, It is t The best individual in the generation population, It is t The suboptimal individuals in the generation population, It is t The third best individual in the generation population;

[0109] Step S22: Define the original population matrix as the first memoryless position matrix, and then perform position update. The expression is as follows:

[0110]

[0111] Where, represents the first position in the memoryless position matrix i Individuals, is A random number uniformly generated in a 、 b is the weight factor, is the mean vector;

[0112] Step S3: exploration stage;

[0113] During the exploration phase, i The position update expression of each individual is as follows:

[0114]

[0115] Where, represents the average position of the population, c is the weight factor, and represents a random positive integer, and ;

[0116] Step S4: population update;

[0117] Output the new population after position update to form a new population matrix;

[0118] Step S5: Update the fitness value;

[0119] Update the fitness value, re-sort the population, and then store the optimal output weight and memory factor.

[0120] As a preferred technical solution of the present invention: the alarm issuance process is as follows: determine whether the distance between the coordinate points exceeds the threshold. If it exceeds the threshold, it indicates that there is a risk, and an alarm is issued; in addition, the water level and precipitation data tables, trend charts, and change trajectory charts displayed on the platform are deeply analyzed, and video playback is viewed in combination with video monitoring points to determine the movement changes on the surface and inside of the dam body. If a risk occurs, timely response and appropriate measures are taken.

[0121] The present invention provides a dam safety monitoring and early warning system and method based on multiple data missing fills. The system utilizes high-precision sensors and Beidou satellites to detect parameter data such as dam displacement, dam settlement, water level, and precipitation in real time. Taking into account the influence of weather and geological movement, some sensors may fail or be damaged, or be interfered with by signal noise, resulting in errors in data collection and affecting the system's monitoring of dam safety. Therefore, the present invention combines the ATTGAN method for filling missing values ​​in Euclidean spatial data with the ATTGAN method for filling missing values ​​in component data to fill missing values ​​in data sets with different missing values ​​and abnormal values. This method not only compensates for the impact of excessive missing values ​​on the dam early warning system, but also avoids excessive missing value filling errors caused by improper selection of missing value filling methods. The system then uses the processed data set to predict future data using a memory-loss online sequence extreme learning machine optimized by an exponential distribution optimizer, thereby improving the accuracy and integrity of the monitoring and early warning system. The data is then sent to a monitoring platform to determine the safety status of the dam and issue a timely alarm. This method effectively meets the requirements for real-time dam safety monitoring and improves the accuracy of dam safety management.

[0122] Specifically, compared with the prior art, the present invention has the following beneficial effects:

[0123] 1) By collecting six different parameters, including dam displacement, dam settlement, water level, and precipitation, the present invention can accurately monitor and analyze key information such as dam structure and environmental parameters. This will significantly improve the safety performance of the dam and prevent and reduce the risk of accidents caused by structural damage, natural disasters, and other reasons.

[0124] 2) The present invention selects the ATTGAN filling method for missing values ​​of Euclidean spatial data or the ATTGAN filling method for missing values ​​of component data by determining whether the proportion of missing values ​​in the entire data is greater than 5%, thereby avoiding excessive error rate in missing value filling due to improper selection of missing value filling method.

[0125] 3) The present invention is based on multi-source data prediction and early warning based on data filling, and uses a memory-loss online sequence extreme learning machine optimized by an exponential distribution optimizer to predict data at future moments, thereby improving the accuracy and completeness of the monitoring and early warning system.

[0126] 4) Traditional dam inspections require significant manpower and time, and are significantly affected by human factors. This invention, however, can achieve automated monitoring and early warning functions, significantly reducing the workload of manual inspections and improving work efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0127] Figure 1 It is the overall flow chart of the present invention.

[0128] Figure 2 This is the flow chart of the exponential distribution optimizer proposed in the present invention.

[0129] Figure 3 A flow chart of a data processing module is proposed for the present invention.

[0130] Figure 4 The following is a histogram of the RMSE indicators of four data under different model predictions.

[0131] Figure 5 This is a comparison chart of the actual value and predicted value of the EDO-MDOS-ELM model provided by the present invention in water level data prediction.

[0132] Figure 6 This is a comparison chart of the actual value and predicted value of precipitation data predicted by the EDO-MDOS-ELM model provided by the present invention.

[0133] Figure 7 This is a comparison chart of the actual value and predicted value of the EDO-MDOS-ELM model provided by the present invention in the dam displacement data prediction.

[0134] Figure 8 This is a comparison chart of the actual value and predicted value of the EDO-MDOS-ELM model provided by the present invention in predicting dam settlement data. DETAILED DESCRIPTION

[0135] The present invention will be further described in detail below with reference to the accompanying drawings.

[0136] like Figure 1 As shown, a dam safety monitoring and early warning system based on multiple filling of data gaps includes a data acquisition module; a data processing module; an EDO-MDOS-ELM prediction module and a monitoring and early warning module.

[0137] The data acquisition module is used to collect data on dam displacement, dam settlement, water level and precipitation.

[0138] The data processing module is divided into the attribute transformation generative adversarial network method for filling missing values ​​of Euclidean space data and the attribute transformation generative adversarial network method for filling missing values ​​of component data by determining the amount of data missing and anomalies.

[0139] The EDO-MDOS-ELM prediction module optimizes the output weights of the MDOS-ELM model through an exponential distribution optimizer, and then predicts dam displacement, dam settlement, water level and precipitation through the optimized model.

[0140] The monitoring and early warning module consists of video monitoring points and alarm modules.

[0141] like Figure 2As shown in Figure 2, the optimization process of the exponential distribution optimizer in the EDO-MDOS-ELM prediction module is as follows:

[0142] Step S1: Population initialization. Its expression is as follows:

[0143]

[0144] Where, Indicates the input data after filling i The first data (dam displacement, dam settlement, water level and precipitation) j Dimensional variables, and Respectively represent j The lower and upper bounds of the dimensional variables, for A random number between .

[0145] Step S2: Development phase.

[0146] Step S21: Convert the guiding solution into the average of the first three solutions sorted from best to worst in terms of fitness value. The expression is as follows:

[0147]

[0148] Where, t is the current iteration number, It is t The best individual in the generation population, It is t The suboptimal individuals in the generation population, It is t The third best individual in the population.

[0149] Step S22: Define the original population matrix as the first memoryless position matrix, and then perform position update. The expression is as follows:

[0150]

[0151] Where, represents the first position in the memoryless position matrix i Individuals, is A random number uniformly generated in a 、 b is the weight factor, is the mean vector;

[0152] Step S3: Exploration stage.

[0153] During the exploration phase, i The position update expression of each individual is as follows:

[0154]

[0155] Where, represents the average position of the population, c is the weight factor, and represents a random positive integer, and ;

[0156] Step S4: Population update.

[0157] The new population after the output position is updated forms a new population matrix.

[0158] Step S5: Update the fitness value.

[0159] Update the fitness value, re-sort the population, and then store the optimal output weight and memory factor.

[0160] like Figure 3 As shown in the figure, the data processing module implementation process is as follows:

[0161] Step S1: Input the data collected by the data collection module, perform missing value statistics on the data, and calculate the missing value ratio;

[0162] Step S2: Create a data matrix ,in , respectively represent the data of dam displacement, dam settlement, water level and precipitation; for a given variable , whose missing index is .variable The observation value in is recorded as , missing values ​​are recorded as ;the remaining P -1 variable corresponding to the row in The data is recorded as , the corresponding row is The data is recorded as ;

[0163] Step S3: Determine whether the proportion of missing values ​​is greater than 5% of the entire data set. If so, proceed to step S4; otherwise, proceed to step S5.

[0164] Step S4: Execute the ATTGAN method to fill missing values ​​of Euclidean spatial data;

[0165] Step S41: Use mean filling method to Perform initial filling;

[0166] Step S42: The indicator set of missing columns in is denoted as Q, and the missing variables are arranged from small to large according to the missing rate;

[0167] Step S43: Determine whether the data meets the stopping criteria , when the stopping criterion is not met , then execute step S44; if the stopping criterion is met, the final filling matrix is ​​obtained.

[0168] The above stopping criterion is: if the difference between the new padding matrix and the previous padding matrix increases, then the loop stops.

[0169] Step S44: Store the existing fill matrix, recorded as ;

[0170] for ,right and The attribute transformation generative adversarial network discriminator is used for classification. The discriminator is expressed as:

[0171] ;

[0172] According to the classification results, use predict , update the filling matrix using the obtained prediction value;

[0173] Step S5: Execute the ATTGAN method to fill in the missing values ​​of the component data;

[0174] Step S51: Select attribute conversion to generate adversarial network to obtain the initial filling matrix value of the filling matrix seat, and the initial matrix is ​​recorded as ;

[0175] Step S52: Through elementary transformation of the matrix, the missing value is adjusted to the first position of the data matrix. At this time, the matrix is ​​recorded as ;

[0176] Step S53: Use isometric logarithmic ratio transformation to swap the rows and columns of the data matrix From the simplex space to the Euclidean space, the transformed matrix is ​​recorded as ;

[0177] Step S54: Generate adversarial network based on the attribute transformation in step S44 to transform the data matrix Fill in and get the data matrix ;

[0178] Step S55: Using the inverse transformation of the isometric logarithmic ratio transformation, the data matrix Restore to component data and get the data matrix after inverse transformation ;

[0179] Step S56: Since the sum of each row of the original component data is different, it is necessary to fill the data matrix Make corrections; get the final filling value

[0180]

[0181] Where, is the correction factor.

[0182] To verify the accuracy of the hybrid model (EDO-MDOS-ELM) proposed in this paper in predicting dam displacement, dam settlement, water level and precipitation, first, the dam displacement, dam settlement, water level and precipitation data of a domestic dam were collected at a sampling interval of once a day. Based on the collected data, it was determined whether the missing values ​​were greater than 5%, and an appropriate filling method was selected to complete the data processing. During the model prediction process, the processed dataset was divided into training and test sets in a ratio of 7:3 to reduce the risk of overfitting. At the same time, an exponential optimization distributor was used to optimize the model's output weights, and finally the prediction results of various dam data were obtained.

[0183] The memory-loss online sequential extreme learning machine model proposed in the present invention is used to predict dam safety data. In order to verify the effectiveness of the present invention, an algorithm program is written in MATLAB language, and the prediction model of the present invention and three groups of control prediction models are constructed respectively.

[0184] The hybrid model is compared with other models based on the four data of dam water level, precipitation, dam displacement, and dam settlement. Table 1 shows the RMSE evaluation index of each model.

[0185] Table 1 Statistical table of performance indicators of the present invention model and the control group model

[0186] Model RMSE RMSE RMSE RMSE ELM 12.56422 1.988977 0.530962 2.155718 FILL-ELM 6.515984 1.60886 0.419604 1.661875 MDOS-ELM 8.543714 0.788173 0.357702 1.242328 EDO-MDOS-ELM 4.643145 0.71954 0.27308 0.781586

[0187] In order to show the error index value more intuitively, Figure 4 The RMSE values ​​of the four models corresponding to different data are shown. Figure 4 It can be concluded that the EDO-MDOS-ELM model performs best compared to other models and has superior nonlinear fitting capabilities. The comparison between the models shows that the algorithm can effectively optimize the parameter selection of the ELM model, thereby improving the prediction accuracy.

[0188] In order to more intuitively demonstrate the prediction results of the prediction model of the present invention and further verify the effectiveness of the proposed model, Figure 5 、 Figure 6 、 Figure 7 、 Figure 8The line comparison charts of the prediction results of different data of the EDO-MDOS-ELM model and the true values ​​are given respectively. It can be clearly seen from the figures that the predicted values ​​are basically consistent with the actual values, which fully proves the effectiveness of the model proposed in this invention.

[0189] The above embodiments are intended only to illustrate the technical concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. They are not intended to limit the scope of protection of the present invention. Any equivalent changes or modifications made in accordance with the spirit of the present invention are intended to be covered by the scope of protection of the present invention.

Claims

1. A dam safety monitoring and early warning system based on multiple filling of missing data, characterized by: It includes data acquisition module, data processing module, EDO-MDOS-ELM prediction module and monitoring and early warning module; The data acquisition module is used to collect data on dam displacement, dam settlement, water level and precipitation; The data processing module is used to obtain the data collected by the data collection module, and select an appropriate filling method according to the two methods of attribute conversion generative adversarial network filling method for missing values ​​of Euclidean space data and attribute conversion generative adversarial network filling method for missing values ​​of component data, by determining whether the missing value is greater than 5% of the entire data set, so as to achieve the purpose of multiple filling of missing data; The EDO-MDOS-ELM prediction module includes an exponential distribution optimizer and a memory-reduced online sequence extreme learning machine. The exponential distribution optimizer is used to optimize the output weight of the memory-reduced online sequence extreme learning machine, so that the prediction model can have faster and more accurate prediction results. Then, the data output by the data processing module is input to output the prediction results of dam displacement, dam settlement, water level and precipitation. The monitoring and early warning module includes a video monitoring point and an alarm module. The alarm module determines whether to issue an alarm based on the dam displacement and dam settlement prediction results output by the EDO-MDOS-ELM prediction module; The data processing module implementation process is as follows: Step S1: input the data collected by the data collection module, perform missing value statistics on the data, and calculate the missing value ratio; Step S2: Create a data matrix Among them, α=1,2,3,4, which represent the data of dam displacement, dam settlement, water level and precipitation respectively; for a given variable Its missing index is variable The observed values ​​in are recorded as Missing values ​​are recorded as The rows corresponding to the remaining P-1 variables are in The data is recorded as The corresponding line is The data is recorded as Step S3: Determine whether the proportion of missing values ​​is greater than 5% of the entire data set. If so, execute step S4; otherwise, execute step S5; Step S4: Execute the ATTGAN filling method for missing values ​​of Euclidean spatial data; Step S41: Use mean filling method to fill X α Perform initial filling; Step S42: X α The indicator set of the missing columns in is denoted as Q, and the missing variables are arranged from small to large according to the missing rate; Step S43: Determine whether the data satisfies the stopping criterion β. If the stopping criterion β is not satisfied, execute step S44. If the stopping criterion is satisfied, the final padding matrix is ​​obtained. The above-mentioned stopping criterion β is: if the difference between the new padding matrix and the previous padding matrix increases, then the loop stops. Step S44: Store the existing padding matrix, denoted as For ρ∈Q, and The attribute conversion generative adversarial network discriminator is used for classification, and the discriminator is expressed as: According to the classification results, use predict Update the filling matrix using the obtained prediction values; Step S5: Execute the ATTGAN filling method for missing values ​​of component data; Step S51: Select attribute conversion to generate adversarial network to obtain the initial padding matrix value of the padding matrix seat, and the initial matrix is ​​recorded as Y0; Step S52: By performing elementary transformation of the matrix, the missing value is adjusted to the first position of the data matrix. The matrix is ​​then recorded as Y0'; Step S53: transform the data matrix Y0' after the row and column exchange from the simplex space to the Euclidean space by using the isometric logarithmic ratio transformation, and the transformed matrix is ​​recorded as Y1'; Step S54: fill the data matrix Y1' using the attribute conversion generative adversarial network method in step S44 to obtain the data matrix Y2'; Step S55: using the inverse transformation of the isometric logarithmic ratio transformation, the data matrix Y2' is restored to component data to obtain the inverse transformed data matrix Y3'; Step S56: Since the sum of each row of the original component data is different, the padded data matrix Y3' needs to be corrected; the final padded value is: Y=λ·Y3', In the formula, λ is the correction factor.

2. The system according to claim 1, characterized in that: The data acquisition module includes a multi-point sensor, an ultrasonic sensor and a weather station; The multi-point sensor is arranged on the dam, and is used to transmit signals to the Beidou satellite to obtain the real-time position coordinate information of the point, and observe the displacement and settlement of the dam body by recording the change of the coordinates; There are multiple ultrasonic sensors installed upstream of the dam, and the ultrasonic sensors are used to obtain real-time water level changes and output corresponding water level data; The weather station is arranged near the dam and is used to obtain precipitation data.

3. The system according to claim 1, characterized in that: The EDO-MDOS-ELM prediction module implementation process is as follows: In the initial stage, the input of the EDO-MDOS-ELM prediction module is determined to be the data of dam displacement, dam settlement, water level and precipitation filled by the data processing module, and the input is divided into z sample blocks according to the time series; the first sample block S1 is input into the MDOS-ELM initial model, and the initial model is expressed as follows: Where λ1 is the input weight value after optimization by the exponential distribution optimizer (EDO), is the weight vector, H0 is the hidden layer output matrix; output weight vector In the online update phase, after adding the training model sample S2, the hidden layer output matrix H1 is calculated, and the output weight vector is calculated after introducing the memory factor η1 optimized by the exponential distribution optimizer (EDO) And update the MDOS-ELM model: After inputting z sample blocks, the MDOS-ELM model is obtained as: The output matrix is ​​calculated, i.e. the data of predicted dam displacement, dam settlement, water level and precipitation.

4. The system according to claim 1 or 3, characterized in that: The exponential distribution optimizer implementation process is as follows: Step S1: Population initialization, its expression is as follows: m ij =lb j +ζ·(ub j -lb j ), In the formula, m ij represents the variable of the jth dimension of the i-th data after the input data is padded, lb j andub j They represent the lower and upper limits of the j-th dimension variable, and ζ is a random number between [0,1]; The data are dam displacement, dam settlement, water level and precipitation; Step S2: Development phase; Step S21: Convert the guidance solution into the average of the first three solutions sorted from best to worst in terms of fitness value, as shown in the following expression: Where t is the current iteration number, is the best individual in the t-th generation population, is the second best individual in the t-th generation population, It is the third best individual in the t-th generation population; Step S22: define the original population matrix as the first memoryless position matrix, and then perform position update, the expression is as follows: In the formula, represents the i-th individual in the memoryless position matrix, φ is a random number uniformly generated in [0,1], a and b are weight factors, and σ is the mean vector; Step S3: exploration stage; In the exploration phase, the position update expression of the i-th individual is as follows: In the formula, represents the average position of the population, c is the weight factor, r1 and r2 represent random positive integers, and r1≠r2; Step S4: population update; Output the new population after the position is updated to form a new population matrix; Step S5: Update the fitness value; Update the fitness value, re-sort the population, and then store the optimal output weight and memory factor.

5. The system according to claim 1, characterized in that: The monitoring and early warning module is implemented as follows: The monitoring and early warning module includes video monitoring points and an alarm module. The dam displacement and dam settlement prediction results output by the model algorithm are input into the alarm module, and the alarm module determines whether the distance between the coordinate points exceeds the threshold. If it exceeds the threshold, it means that there is a safety risk to the dam, and an alarm is issued. The water level and precipitation data tables, trend charts, and change trajectory charts displayed on the platform are then analyzed in depth, and video playback is watched in conjunction with the video monitoring points to determine the movement changes on the surface and inside of the dam. If risks arise, timely responses and appropriate measures are taken.

6. A dam safety monitoring and early warning method based on multiple filling of missing data, characterized by: The method comprises the following steps: First, the dam displacement, dam settlement, water level and precipitation data of a certain dam are collected at certain time intervals; Secondly, according to the collected data, determine whether the missing value is greater than 5%, and select the ATTGAN filling method for missing values ​​of Euclidean spatial data or the ATTGAN filling method for missing values ​​of component data to complete the data processing; Thirdly, in the process of model prediction, the processed data set is divided into training set and test set in a ratio of 7:3 to reduce the risk of overfitting prediction. At the same time, the exponential optimization distributor is used to optimize the output weights of the MDOS-ELM initial model, and finally the prediction results of various dam data are obtained; Finally, based on the prediction results of dam displacement and dam settlement, it is determined whether to issue an alarm.

7. The method according to claim 6, characterized in that: The data processing process is as follows: Step S1: Perform missing value statistics on the collected data and calculate the missing value ratio; Step S2: Create a data matrix Among them, α=1,2,3,4, which represent the data of dam displacement, dam settlement, water level and precipitation respectively; for a given variable Its missing index is variable The observed values ​​in are recorded as Missing values ​​are recorded as The rows corresponding to the remaining P-1 variables are in The data is recorded as The corresponding line is The data is recorded as Step S3: Determine whether the proportion of missing values ​​is greater than 5% of the entire data set. If so, execute step S4; otherwise, execute step S5; Step S4: Execute the ATTGAN filling method for missing values ​​of Euclidean spatial data; Step S41: Use mean filling method to fill X α Perform initial filling; Step S42: X α The indicator set of the missing columns in is denoted as Q, and the missing variables are arranged from small to large according to the missing rate; Step S43: Determine whether the data satisfies the stopping criterion β. If the stopping criterion β is not satisfied, execute step S44. If the stopping criterion is satisfied, the final padding matrix is ​​obtained. The above-mentioned stopping criterion β is: if the difference between the new padding matrix and the previous padding matrix increases, then the loop stops. Step S44: Store the existing padding matrix, denoted as For ρ∈Q, and The attribute conversion generative adversarial network discriminator is used for classification, and the discriminator is expressed as: According to the classification results, use predict Update the filling matrix using the obtained prediction values; Step S5: Execute the ATTGAN filling method for missing values ​​of component data; Step S51: Select attribute conversion to generate adversarial network to obtain the initial padding matrix value of the padding matrix seat, and the initial matrix is ​​recorded as Y0; Step S52: By performing elementary transformation of the matrix, the missing value is adjusted to the first position of the data matrix. The matrix is ​​then recorded as Y0'; Step S53: transform the data matrix Y0' after the row and column exchange from the simplex space to the Euclidean space by using the isometric logarithmic ratio transformation, and the transformed matrix is ​​recorded as Y1'; Step S54: fill the data matrix Y1' using the attribute conversion generative adversarial network method in step S44 to obtain the data matrix Y2'; Step S55: using the inverse transformation of the isometric logarithmic ratio transformation, the data matrix Y2' is restored to component data to obtain the inverse transformed data matrix Y3'; Step S56: Since the sum of each row of the original component data is different, the padded data matrix Y3' needs to be corrected; the final padded value is: Y=λ·Y3', In the formula, λ is the correction factor.

8. The method according to claim 6, characterized in that: The initialization process of the MDOS-ELM initial model is as follows: the data of dam displacement, dam settlement, water level and precipitation filled by data processing are divided into z sample blocks according to the time series; the first sample block S1 is input into the MDOS-ELM initial model, and the initial model is expressed as follows: Where λ1 is the input weight value after optimization by the exponential distribution optimizer (EDO), is the weight vector, H0 is the hidden layer output matrix; output weight vector In the online update phase, after adding the training model sample S2, the hidden layer output matrix H1 is calculated, and the output weight vector is calculated after introducing the memory factor η1 optimized by the exponential distribution optimizer (EDO) And update the MDOS-ELM model: After inputting z sample blocks, the MDOS-ELM model is obtained as: Calculate the output matrix, i.e. the predicted dam displacement, dam settlement, water level and precipitation data; The implementation process of the exponential distribution optimizer is as follows: Step S1: Population initialization, its expression is as follows: m ij =lb j +ζ·(ub j -lb j ), In the formula, m ij represents the variable of the jth dimension of the i-th data after the input data is padded, lb j andub j They represent the lower and upper limits of the j-th dimension variable, and ζ is a random number between [0,1]; The data are dam displacement, dam settlement, water level and precipitation; Step S2: Development phase; Step S21: Convert the guidance solution into the average of the first three solutions sorted from best to worst in terms of fitness value, as shown in the following expression: Where t is the current iteration number, is the best individual in the t-th generation population, is the second best individual in the t-th generation population, It is the third best individual in the t-th generation population; Step S22: define the original population matrix as the first memoryless position matrix, and then perform position update, the expression is as follows: In the formula, represents the i-th individual in the memoryless position matrix, φ is a random number uniformly generated in [0,1], a and b are weight factors, and σ is the mean vector; Step S3: exploration stage; In the exploration phase, the position update expression of the i-th individual is as follows: In the formula, represents the average position of the population, c is the weight factor, r1 and r2 represent random positive integers, and r1≠r2; Step S4: population update; Output the new population after the position is updated to form a new population matrix; Step S5: Update the fitness value; Update the fitness value, re-sort the population, and then store the optimal output weight and memory factor.

9. The method according to claim 6, characterized in that: The alarm issuance process is as follows: determine whether the distance between the coordinate points exceeds the threshold. If it exceeds the threshold, it means there is a risk and an alarm is issued. In addition, conduct in-depth analysis of the water level and precipitation data tables, trend charts, and change trajectory charts displayed on the platform, and watch video playback in combination with video monitoring points to determine the movement changes on the surface and inside of the dam. If a risk occurs, respond in a timely manner and take appropriate measures.

Images