A data deduction method for a large-span assembled detection door

Abstract

The present invention relates to the technical field of water conservancy project detection, and specifically to a data deduction method for a large-span assembled detection door. The present invention non-linearly fits the wear degree data based on the Bayesian inference method and the genetic algorithm through the historical wear degree data, historical water quality data, and historical water pressure data of the large-span assembled detection door, and obtains the corresponding slope data and service life; trains a detection door influence parameter prediction model through these historical data, and simulates and predicts the service wear degree of the large-span assembled detection door based on the historical water pressure data and historical water quality data of the water conservancy detection target in the real environment, and corrects it in combination with the predicted slope data and service life data to generate a durability curve of the wear degree; through this method, the data deduction work of visualizing the wear degree can be carried out without use, so as to measure the service life of the large-span assembled detection door, so as to facilitate the subsequent generation of relevant use plans and the correction of the used plans.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy project detection, and specifically to a data deduction method for a large-span assembled detection gate. Background Technique

[0002] A large-span assembled detection gate is a special gate structure used for large-span river blocking gates, dam spillways and other large water conservancy projects. This detection gate is mainly used to control the water flow to facilitate daily maintenance and repair work. And due to its modular structure characteristics, it can be used across scales multiple times in different usage scenarios to save usage costs; at the same time, its modular structure is also convenient for transporting the detection gate and improving transportation efficiency.

[0003] Since the large-span assembled detection gate needs to control the water flow usage scenario, it requires good structural strength and water pressure resistance; during long-term use, the large-span assembled detection gate is easily affected by water corrosion and impact. Therefore, as the usage time increases, the service life of the large-span assembled detection gate will gradually decrease, and its durability and stability will also decrease, gradually affecting the degree of water flow control.

[0004] If a large-span assembled detection gate with low durability is used, it may lead to accidents due to the inability to stably control the water flow. Therefore, it is necessary to deduce the durability data of the large-span assembled detection gate, so as to more accurately measure the service life of the large-span assembled detection gate, and facilitate relevant technicians to more accurately formulate or revise the plan under visual data to ensure that the large-span assembled detection gate can be used more safely under the corresponding plan.

[0005] For this reason, a data deduction method for a large-span assembled detection gate is proposed. Summary of the Invention

[0006] The purpose of the present invention is to provide a data deduction method for a large-span assembled detection gate. Based on the historical wear degree data, historical water quality data and historical water pressure data of the large-span assembled detection gate, non-linear curve fitting is performed on the wear degree data by using the Bayesian inference method and the genetic algorithm, and the corresponding slope data and the first usage duration are obtained; a detection gate influence parameter prediction model is established and trained according to these historical data, and the historical water pressure data and historical water quality data of the corresponding position information of the water conservancy detection target in the real environment are combined to simulate the usage wear degree of the large-span assembled detection gate in the real usage environment for data prediction deduction, and the durability curve of the wear degree is generated by combining and correcting the predicted slope data and the first usage duration data; through this method, visual data deduction work on the wear degree can be carried out without use, so as to measure the service life of the large-span assembled detection gate, and facilitate the subsequent generation of relevant usage plans and the correction of existing usage plans.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A data deduction method for a large-span assembled detection door, comprising:

[0009] Obtain the material data of the large-span assembled detection door to be simulated, and obtain the historical wear degree data, historical water quality data, and historical water pressure data according to the large-span assembled detection door with the same material data; fit a historical change curve according to the historical wear degree data, and obtain the historical slope data and the first service life of the historical change curve; the historical wear degree data, the historical water quality data, and the historical water pressure data obtain corresponding data according to the fixed time recording point T, and the data collection point of the historical slope data is the same as the fixed time recording point T;

[0010] Further, taking the historical wear degree data as the vertical axis coordinate and time as the horizontal axis coordinate to generate a wear degree change coordinate, and performing non-linear fitting on the wear degree change coordinate to generate the historical change curve; the historical change curve performs non-linear fitting on the wear degree change coordinate through the Bayesian inference method and the genetic algorithm;

[0011] Establish a detection door influence parameter prediction model, and perform model training on the detection door influence parameter prediction model according to the first service life, the historical slope data, the historical water quality data, the historical water pressure data, and the historical wear degree data;

[0012] Further, the detection door influence parameter prediction model includes a detection door data encoding module and a detection door data decoding module, and the model training process is performed in the detection door data encoding module;

[0013] The detection door data encoding module includes a detection door influence data processing unit, a detection door influence data feature unit, and a detection door influence data training output unit;

[0014] The detection door influence data processing unit performs data preprocessing on the historical slope data, the first service life, the historical water quality data, the historical water pressure data, and the historical wear degree data to generate multiple pieces of detection door influence preprocessing data;

[0015] The detection door influence data feature unit extracts the features of each piece of detection door influence preprocessing data to generate detection door influence feature data; the detection door influence data training output unit predicts the slope data and the wear degree data according to the detection door influence feature data, and stores the detection door influence feature data;

[0016] Further, the detection gate impact data feature unit includes: a Transformer network, a residual convolutional network, an LSTM network, and a feature fusion network;

[0017] Input multiple pieces of the detection gate impact preprocessed data into the Transformer network for feature enhancement encoding;

[0018] The number of the residual convolutional networks is 8, and each residual convolutional network includes 3 1*1 convolutional networks and 1 1*3 convolutional network; the number of the LSTM networks is 12; the number of the feature fusion networks is 2;

[0019] 3 of the residual convolutional networks and 6 LSTM units are used to extract the first correlation features of the detection gate impact preprocessed data of the first usage duration, the historical slope data, the historical water quality data, the historical water pressure data, and the historical wear degree data after feature encoding;

[0020] 3 of the residual convolutional networks and 4 LSTM unit networks are used to extract the second correlation features of the detection gate impact preprocessed data corresponding to the first usage duration, the historical water quality data, the historical water pressure data, and the historical wear degree data after feature encoding;

[0021] 2 of the residual convolutional networks, 2 LSTM units, and 1 feature fusion network are used to extract the third correlation features of the detection gate impact preprocessed data corresponding to the first usage duration, the historical slope data, and the historical wear degree data;

[0022] The first correlation feature and the second correlation feature are fused through 1 of the feature fusion networks;

[0023] The first correlation feature and the third correlation feature are fused through 1 of the feature fusion networks;

[0024] Obtain the first usage duration, water quality data, and water pressure data corresponding to the position information of the water conservancy detection target;

[0025] Further, simulate the water conservancy detection target through a digital twin model, and the position information is all installable positions of the large-span assemblable detection gate;

[0026] And obtain the wear degree prediction data and slope prediction data according to the detection gate impact parameter prediction model, fit the first durability change curve according to the wear degree prediction data; correct the first durability change curve according to the slope prediction data;

[0027] Further, according to the detection gate data decoding module of the detection gate influence parameter prediction model, predict the wear degree prediction data and the slope prediction data; non-linearly fit the first durability change curve through Bayesian inference method and genetic algorithm;

[0028] The detection gate data decoding module includes a detection gate durability prediction processing unit, a detection gate durability prediction coding connection unit, and a detection gate durability prediction feature extraction and output unit;

[0029] The durability prediction processing unit performs data preprocessing on the first usage duration, the historical water quality data, and the historical water pressure data to generate durability prediction preprocessed data;

[0030] The detection gate durability prediction coding connection unit is connected to the detection gate data coding module; the detection gate durability prediction feature extraction and output unit combines the detection gate data coding module to extract the durability prediction features of the durability prediction preprocessed data; and predicts the wear degree prediction data and the slope prediction data based on the durability prediction features;

[0031] The detection gate durability prediction feature extraction and output unit includes: 1 Transformer network, 5 residual convolutional networks, and 6 LSTM networks; 1 Transformer network extracts the feature enhancement coding of the durability prediction preprocessed data and performs feature association with the features of the detection gate durability prediction coding connection unit; 5 residual convolutional networks and 6 LSTM networks extract the features of the durability prediction preprocessed data after feature association and perform prediction output;

[0032] Further, the detection gate data decoding module generates multiple prediction intervals based on the first usage duration, and each prediction interval is associated with the corresponding historical water quality data and historical water pressure data;

[0033] The nth prediction uses the wear degree prediction data of the (n - 1)th time as the prediction input data; the nth prediction input data includes the wear degree prediction data, the first usage duration, the historical water quality data, and the historical water pressure data, and the total number of prediction intervals is N, , ;

[0034] According to the wear degree data and the second usage duration of the large-span assembled detection gate with the same material data, fit and generate a second durability change curve; and secondarily correct the first durability change curve to generate a third durability change curve;

[0035] Further, with the second usage duration as the abscissa and the wear degree data as the ordinate, the second durability change curve is fitted by the Bayesian inference method and the genetic algorithm;

[0036] The first durability change curve is secondarily corrected to generate a third durability change curve as follows:

[0037] ;

[0038] wherein, is the wear degree of the third durability change curve at the first usage duration , is the wear degree of the first durability change curve at the first usage duration , is the wear degree corresponding to the second usage duration on the second durability change curve;

[0039] Based on the third durability change curve and the second durability change curve, the durability of the large-span assembled inspection door is simulated and deduced.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] 1. By collecting the historical wear degree data, historical water quality data, and historical water pressure data of the large-span assembled inspection door, and performing non-linear curve fitting on the wear degree data based on the Bayesian inference method and the genetic algorithm, the present invention can, through this method, understand the wear degree consumption data of each material of the large-span assembled inspection door under the influence of different water pressures and different water qualities based on the existing data, and obtain the corresponding slope data. On the one hand, these historical data can provide a large amount of training data for the subsequent simulation data deduction and prediction of the service life. At the same time, according to the curve slope fitted from the existing data, it is also possible to correct the data for the prediction result to prevent the error of the model deduction from being too large, resulting in a decrease in the accuracy of the result, and improving the reliability of the data deduction.

[0042] 2. Through the encoder and decoder structures, the present invention conducts feature training on the predicted data in the encoder. To improve the prediction accuracy, a combined model of LSTM and residual convolution is adopted based on time-series data for comprehensive feature learning from both the time-series perspective and the data form perspective, and the trained features are stored. When actual prediction is required, feature extraction is performed through the decoder unit, and at the same time, the stored features of the encoder unit are called for comprehensive learning to improve the prediction accuracy. Through this method, modular splitting can be carried out during the training and prediction processes, and the training model and prediction model can be independently optimized to adapt to the data deduction and prediction work under different actual scenarios, so as to improve the accuracy of the simulation and deduction prediction of the service life. At the same time, it is also convenient for subsequent independent optimization of the model and parameters for training and prediction work to further improve the accuracy of the simulation and deduction prediction of the service life.

[0043] 3. The present invention obtains the prediction result through the simulation and deduction of the service life of the prediction model, combines the Bayesian inference method and the genetic algorithm to fit the data to generate a curve, corrects the data of the curve in combination with the predicted slope data, and corrects the data of the quadratic curve in combination with the data relationship between the usage duration and the wear degree. Through this method, the data can be visualized based on the prediction and deduction results, which is convenient for assisting relevant technical personnel to generate corresponding solutions or correct the sampled solutions based on the visualization results. At the same time, the corresponding curves of other life-influencing data are also generated for visualization operations to further facilitate the auxiliary work of relevant technical personnel for data analysis. At the same time, the independent visualization of each data is also convenient for the subsequent optimization of the entire data deduction and prediction work to improve the accuracy of the simulation and deduction prediction of the service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a schematic flow chart of the method of the present invention;

[0045] Figure 2 is a schematic diagram of the digital twin of the hydropower station of the present invention;

[0046] Figure 3 is a schematic flow chart of the detection door data encoding module of the present invention;

[0047] Figure 4 is a schematic flow chart of the detection door data decoding module of the present invention;

[0048] Figure 5 is a schematic model diagram of the detection door data encoding module and the detection door data decoding module of the present invention;

[0049] Figure 6 is a schematic structure diagram of the residual convolution group of the present invention;

[0050] Figure 7 is a schematic flow chart of the model of the comparative model M03 of the present invention;

[0051] Figure 8 This is the first durability change curve diagram of the large-span assembled detection door of the present invention;

[0052] Figure 9 This is the third durability change curve diagram of the large-span assembled detection door of the present invention;

[0053] Figure 10 This is the durability curve diagram of the large-span assembled detection door of the present invention. Detailed implementation manners

[0054] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0055] Due to the combined structural characteristics of the large-span assembled detection door, this detection door has the characteristics of being easy to transport and applicable to various scenarios, can be used in a large span, is convenient for water flow control in multiple areas, and realizes the maintenance and transportation of relevant personnel; but also because of its combined structural characteristics, the large-span assembled detection door has relatively high stability requirements. Due to the long-term influence of water pressure and water corrosion, its stability will also decrease with the increase of service time. Therefore, it is necessary to perform data deduction on the durability of the large-span assembled detection door in order to accurately measure the service life of the large-span assembled detection door. For this reason, the present invention provides a data deduction method for a large-span assembled detection door. Referring to Figure 1 as shown, the technical solution is as follows:

[0056] Obtain the material data of the large-span assembled detection door to be simulated, and obtain the historical wear degree data, historical water quality data, and historical water pressure data based on the large-span assembled detection door with the same material data; fit the historical wear degree data to generate a historical change curve, and obtain the historical slope data and the first service duration of the historical change curve;

[0057] Establish a detection door influence parameter prediction model, and perform model training on the detection door influence parameter prediction model based on the first service duration, the historical slope data, historical water quality data, the historical water pressure data, and historical wear degree data;

[0058] Obtain the first usage duration, water quality data, and water pressure data corresponding to the position information of the water conservancy detection target, and obtain the wear degree prediction data and slope prediction data according to the detection door influence parameter prediction model. Fit the first durability change curve according to the wear degree prediction data; correct the first durability change curve according to the slope prediction data;

[0059] Based on the wear degree data and the second usage duration of the large-span assembled detection door with the same material data, fit and generate a second durability change curve; and correct the first durability change curve twice to generate a third durability change curve;

[0060] Simulate and deduce the durability of the large-span assembled detection door based on the third durability change curve and the second durability change curve.

[0061] In order to be able to perform data deduction on the large-span assembled detection door, through the material data of the large-span assembled detection door to be simulated, obtain the historical wear degree data of the large-span assembled detection door with the same used material, and obtain the historical water quality data and historical water pressure data in its corresponding usage environment. Establish a change curve of the historical wear degree data through these data. Through this change curve, the change situation of the wear degree in the actual usage process can be understood. Therefore, collect the slope data of each point through the curve as the actual change data, and then obtain the first usage duration data through the curve as one of the parameters of the wear degree change at different usage times, so as to correct the data during the data deduction process to achieve a more accurate data deduction result;

[0062] At the same time, based on the collected data, through the detection door parameter prediction model, in the training stage of data prediction and deduction, using the duration data, wear data, water quality data, water pressure data, and slope data as time series data input, the model can be trained and learn the direct hidden data features and the change features of the data itself, and store the data features learned by the model; combined with the required deduction scenario, the large-span assembled detection door at each position of the usage target can be based on the water pressure data and water quality data collected in the real environment, combined with the set first usage duration data for data prediction to reach the result of simulation and deduction. Calculate the change relationship of its wear degree according to multiple prediction results to generate the first durability curve, so as to intuitively understand the wear data of the large-span assembled detection door, facilitate the generation of subsequent related solutions and the correction of the generated solutions, and provide more intuitive visual data;

[0063] To further improve the accuracy, a second durability change curve is established based on the relationship between the second usage duration and the wear degree. Further correction is performed using this data. After excluding the influence of the second usage duration on the wear degree, the durability change curve in the actual process is displayed. Through this method, the usage data in different situations can be visually displayed to further assist the work of relevant technical personnel. At the same time, it can also intuitively display the service life situation and facilitate the debugging of existing models, improving the accuracy and precision of the data prediction and deduction results.

[0064] Embodiment 1

[0065] To elaborate on the model training and prediction process in detail, it is described in combination with the following embodiments:

[0066] Obtain the material data of the large-span assembled inspection door to be simulated (such as concrete, metal, etc.). Based on the large-span assembled inspection door with the same material data, obtain the historical wear degree data, historical water quality data, and historical water pressure data; fit a historical change curve based on the historical wear degree data, and obtain the historical slope data and the first usage duration of the historical change curve; the first usage duration is the usage duration of the large-span assembled inspection door during the maintenance process; the second usage duration is the usage duration that generates wear data during assembly and transportation;

[0067] The historical wear degree data obtains the wear degree data by identifying the wear degree characteristics on the large-span assembled inspection door caused by water pressure or water quality corrosion through the yolov5 model. The specific calculation is as follows:

[0068] ;

[0069] Among them, is the wear degree data, M is the total number of images, is the number of pixel points of the wear degree characteristics of water pressure or water quality corrosion in the i-th image identified by yolov5, is the total number of pixel points in the i-th image, is the wear degree data of water pressure or water quality corrosion identified based on a large amount of data combined with expert experience, and are the wear degree weights, defaulting to 0.5 and 0.5, and can be specifically modified according to the specific implementation situation; the minimum wear data is 0 representing no wear, and the maximum wear data is 1 representing complete wear;

[0070] Establish a prediction model for the inspection door influence parameters, and train the prediction model for the inspection door influence parameters based on the first usage duration, the historical slope data, historical water quality data, the historical water pressure data, and historical wear degree data;

[0071] Furthermore, the detection gate influence parameter prediction model includes a detection gate data encoding module and a detection gate data decoding module, and the model training process is carried out in the detection gate data encoding module;

[0072] The detection gate data encoding module includes a detection gate influence data processing unit, a detection gate influence data feature unit, and a detection gate influence data training output unit, as shown in Figure 3 the figure; the detection gate influence data processing unit preprocesses the historical slope data, the first usage duration, the historical water quality data, the historical water pressure data, and the historical wear degree data to generate multiple pieces of detection gate influence preprocessed data;

[0073] The detection gate influence data feature unit extracts the features of each piece of the detection gate influence preprocessed data to generate detection gate influence feature data; the detection gate influence data training output unit predicts the slope data and the wear degree data based on the detection gate influence feature data, and stores the detection gate influence feature data;

[0074] For the convenience of the model training and prediction processes, the model is split into an encoding module part and a decoding module part. The encoding module can extract features based on the existing historical data for training, and the decoding part conducts real training. Considering that in the prediction simulation deduction of actual data, only the water pressure data and water quality data simulating the real environment are input, the training and prediction processes are split. During the training process, the first usage duration data, historical water pressure data, historical water quality data, historical slope data, and historical wear degree are used to extract features through the model training process to obtain the hidden feature connections between various data and the change trends of each data, so as to deeply mine the features of each data to improve the accuracy of training prediction and the depth of storing data features, in order to provide a solid model feature basis for the subsequent simulation deduction data process and improve the accuracy in the subsequent simulation deduction process;

[0075] Furthermore, the detection gate influence data feature unit includes: a Transformer network, a residual convolutional network, an LSTM network, and a feature fusion network, as shown in Figure 5 the detection gate influence data encoding module part;

[0076] Input multiple pieces of the detection gate influence preprocessed data into the Transformer network for feature enhancement encoding;

[0077] The number of the residual convolutional networks is 8, and each residual convolutional network includes 3 1*1 convolutional networks and 1 1*3 convolutional network; the residual convolutional network is shown in Figure 6 the figure; the number of the LSTM networks is 12; the number of the feature fusion networks is 2;

[0078] Three of the residual convolutional networks and six LSTM units are used to extract the first correlation features of the preprocessed data affected by the detection gates of the first usage duration, the historical slope data, the historical water quality data, the historical water pressure data, and the historical wear degree data after feature encoding;

[0079] Three of the residual convolutional networks and four LSTM unit networks are used to extract the second correlation features of the preprocessed data affected by the detection gates corresponding to the first usage duration, the historical water quality data, the historical water pressure data, and the historical wear degree data after feature encoding;

[0080] Two of the residual convolutional networks, two LSTM units, and one feature fusion network are used to extract the third correlation features of the preprocessed data affected by the detection gates corresponding to the first usage duration, the historical slope data, and the historical wear degree data;

[0081] The first correlation feature and the second correlation feature are fused through one of the feature fusion networks;

[0082] The first correlation feature and the third correlation feature are fused through one of the feature fusion networks;

[0083] In this embodiment, we obtained the historical wear degree data, the historical water quality data, and the historical water pressure data through relevant technical personnel. After obtaining the slope data through the above method, we carried out the training process of the model detection gate data encoding module, and used multiple groups of data for training. Each group of data would generate multiple results. Therefore, the average value of the results of each group of data was used as the accuracy rate of that group of data, and the model parameters with the highest accuracy rate during the training process were stored. Due to too much data, only partial results are shown in this table. The accuracy rate results of the training are shown in Table 1:

[0084] Table 1 Simulation Deduction Training Accuracy Rate of the Detection Gate Data Encoding Module

[0085] ;

[0086] As can be seen from Table 1, except for the accuracy rates of individual data being lower than 90%, the training accuracy rates of the overall data are all higher than 90%, and the training results are good;

[0087] During the training process of the model, first, obtain the global hidden correlation features in terms of wear influence correlation and wear change trend through the first usage duration, the historical slope data, the historical water quality data, the historical water pressure data, and the historical wear degree data; then, obtain the local influence correlation weight features of these data on the wear degree through the first usage duration, the historical water pressure data, the historical water quality data, and the historical wear degree. At the same time, during the training process, learn the local influence change trend features of the wear degree data changing with time between the first usage duration, the wear degree data, and the slope. Through this training process, the data can be trained from the dimensions of global data and local data to achieve the purpose of more accurately grasping the data features, not only from the feature correlation of the data before, but also from the change dimension of the data itself for feature training to improve the accuracy in the subsequent simulation and deduction process;

[0088] Obtain the first usage duration, water quality data, and water pressure data corresponding to the position information of the water conservancy detection target; and obtain the wear degree prediction data and slope prediction data according to the detection gate influence parameter prediction model, and fit the first durability change curve according to the wear degree prediction data; correct the first durability change curve according to the slope prediction data;

[0089] Further, predict the wear degree prediction data and the slope prediction data according to the detection gate data decoding module of the detection gate influence parameter prediction model; non-linearly fit the first durability change curve by the Bayesian inference method and the genetic algorithm;

[0090] The detection gate data decoding module includes a detection gate durability prediction processing unit, a detection gate durability prediction coding connection unit, and a detection gate durability prediction feature extraction and output unit, as shown in Figure 4 shown;

[0091] The durability prediction processing unit performs data preprocessing on the first usage duration, the historical water quality data, and the historical water pressure data to generate durability prediction preprocessed data;

[0092] The detection gate durability prediction coding connection unit is connected to the detection gate data coding module; the detection gate durability prediction feature extraction and output unit combines the detection gate data coding module to extract the durability prediction features of the durability prediction preprocessed data; and predicts the wear degree prediction data and the slope prediction data based on the durability prediction features;

[0093] In the decoding unit, since the data for prediction deduction is only water quality data and water pressure data, in addition to extracting the environmental data for simulation deduction, the data features stored during the training process are also added. Based on the environmental data for simulation deduction, the associated feature data of wear degree training is added to perform data prediction work, so as to predict the corresponding wear degree data through the features of the environmental data, achieving the purpose of predicting the wear degree data and meeting the requirements of life data deduction for the large-span assembled detection door;

[0094] The detection door durability prediction feature extraction and output unit includes: 1 Transformer network, 5 residual convolutional networks, and 6 LSTM networks; 1 of the Transformer networks extracts the feature enhancement encoding of the durability prediction preprocessed data and performs feature association with the features of the detection door durability prediction encoding connection unit; 5 of the residual convolutional networks and 6 of the LSTM networks extract the features of the durability prediction preprocessed data after feature association and perform prediction output;

[0095] In this embodiment, some data that did not participate in the training are selected to simulate real environmental data for deduction prediction. By comparing with the real data, the prediction accuracy is obtained. Combining with expert experience, if the error is within 2%, the prediction accuracy is considered 100%. And so on, the accuracy calculation formula is as follows:

[0096] ;

[0097] Among them, is the accuracy, is the predicted value, is the real value;

[0098] In this embodiment, 5 groups of data that did not participate in the training are selected as the data for simulating the real environment for data deduction. The prediction accuracy results of the detection door data decoding module are shown in Table 2:

[0099] Table 2 Prediction accuracy results of the detection door data decoding module

[0100] ;

[0101] This embodiment also compares the present model M01, the model that only uses the detection door data encoding module M02, and the overall model M03 that combines the detection door data encoding module and the detection door data decoding module. M01 refers to Figure 5 as shown, M02 refers to Figure 5 shown in the detection door impact data encoding module, and M03 refers to Figure 7 as shown. The comparison of prediction data is carried out, and the data used is A01 and A02. The specific results are shown in Table 3:

[0102] Table 3 Comparison Results of Model Data Prediction Deduction

[0103]

[0104] As can be seen from Table 3, the model M01 in this paper has the highest accuracy in wear degree prediction and slope prediction, and can stably stay around 90%, with good prediction effect;

[0105] Furthermore, the detection gate data decoding module generates multiple prediction intervals based on the first usage duration, and each prediction interval is associated with the corresponding historical water quality data and historical water pressure data;

[0106] The nth prediction will use the wear degree prediction data of the (n - 1)th time as the prediction input data; the prediction input data of the nth time includes the wear degree prediction data, the first usage duration, the historical water quality data and the historical water pressure data, and the total number of prediction intervals is N, , ;

[0107] In the actual prediction process, multiple first usage duration parameters can be set in combination with expert experience, and these first usage duration parameters are formed into a new time series and input together with water quality data and water pressure data. In the actual operation of the model, the first usage duration is used as the prediction interval, and predictions are made according to the water quality data and water pressure data under each first usage duration; during the prediction process of the first usage duration at the second node, the predicted wear degree data of the previous node is substituted for input into the next prediction interval. Through this method, multiple result predictions can be obtained to get multiple data for subsequent curve generation and visualization operations. At the same time, by inputting the first usage duration to generate the number of predictions, it is also convenient for relevant technical personnel to perform quantitative operations, expanding the operation space of model prediction, and making it more convenient for relevant technical personnel to conduct durability data deduction work on large-span assembled detection gates, enabling more accurate measurement of the deduction of their service life and facilitating the modification or generation of corresponding technical solutions;

[0108] Based on the wear degree data and the second usage duration of the large-span assembled detection gate with the same material data, a second durability change curve is fitted; and the first durability change curve is corrected twice to generate a third durability change curve; the durability of the large-span assembled detection gate is simulated and deduced based on the third durability change curve and the second durability change curve.

[0109] Example Two

[0110] The curve fitting process is described through the following examples:

[0111] Obtain the material data of the large-span assembled inspection door to be simulated, and obtain the historical wear degree data, historical water quality data, and historical water pressure data based on the large-span assembled inspection door with the same material data; fit the historical wear degree data to generate a historical change curve, and obtain the historical slope data and the first usage duration of the historical change curve;

[0112] The historical wear degree data, the historical water quality data, and the historical water pressure data obtain corresponding data according to the fixed time recording point T, and the data collection point of the historical slope data is the same as the fixed time recording point T; by default, the first usage duration is the time interval between each time recording point;

[0113] In order to obtain the change situation of the wear data, so that the wear data of the large-span assembled inspection door can be visually displayed during the data deduction process. Therefore, when collecting data, the water quality data, wear data, and water pressure data are formed into time series data through multiple recording points, so as to be displayed on the coordinate. At the same time, based on the data obtained by curve fitting, the corresponding slope data and the first usage duration data are obtained as additional parameters for later data deduction; through this method, the data can be quantified, which is convenient for analysis and prediction, provides a data basis for subsequent data deduction calculations, and can also obtain more influencing parameters from the existing data, improving the analysis dimension of data deduction and the accuracy of subsequent deduction;

[0114] Furthermore, taking the historical wear degree data as the vertical axis coordinate and time as the horizontal axis coordinate, generate a wear degree change coordinate, and perform non-linear fitting on the wear degree change coordinate to generate the historical change curve; through the fitted curve, the wear degree data can be obtained during the actual use process, along with the influence of different water pressure data and water quality data on the wear degree. At the same time, based on this data, the slope data at different time points can be extracted, and the hidden data relationship between the wear data and its influencing data can be mined to increase the dimension of the analyzable data and improve the accuracy of subsequent deduction;

[0115] The historical change curve performs non-linear fitting on the wear degree change coordinate through the Bayesian inference method and the genetic algorithm;

[0116] Through the Bayesian inference method, non-linear fitting can be performed based on the existing data. The genetic algorithm can optimize the fitting of non-linear curves, making the fitted curve more accurate. At the same time, the curve optimized by the Bayesian inference method and the genetic algorithm can obtain a more accurate slope change relationship, which is convenient for providing more accurate slope data, so as to perform later feature training and data correction to improve the data accuracy of later deduction;

[0117] Establish a prediction model for the influencing parameters of the inspection gate, and train the prediction model for the influencing parameters of the inspection gate based on the first usage duration, the historical slope data, the historical water quality data, the historical water pressure data, and the historical wear degree data;

[0118] Obtain the first usage duration, water quality data, and water pressure data corresponding to the location information of the water conservancy inspection target;

[0119] Furthermore, simulate the water conservancy inspection target (such as a dam, a hydropower station, etc., referring to the digital twin of the hydropower station shown Figure 2 ), and the location information is all installable positions of the large-span assembled inspection gate; simulate the water conservancy inspection target through the digital twin model, so as to establish a data model, perform data operations on all actual data based on the simulated virtual data, facilitate the subsequent deduction work, and at the same time, based on the location annotation of the digital twin model, it is also convenient for relevant technical personnel to simulate the data deduction process of the large-span assembled inspection gate, realizing visual and digital display;

[0120] And obtain the wear degree prediction data and slope prediction data according to the prediction model for the influencing parameters of the inspection gate, fit the first durability change curve according to the wear degree prediction data; correct the first durability change curve according to the slope prediction data;

[0121] Furthermore, non-linearly fit the first durability change curve by the Bayesian inference method and the genetic algorithm;

[0122] Fit and generate a second durability change curve based on the wear degree data and the second usage duration of the large-span assembled inspection gate with the same material data; and perform a secondary correction on the first durability change curve to generate a third durability change curve;

[0123] Furthermore, with the second usage duration as the abscissa and the wear degree data as the ordinate, fit the second durability change curve by the Bayesian inference method and the genetic algorithm;

[0124] Classify according to the historical assembly time and historical transportation time, perform data fitting based on the historical assembly time and wear degree data to generate a second durability change curve of the assembly time, perform data fitting based on the historical transportation time and wear degree data to generate a second durability change curve of the transportation time, combine the mean values of multiple historical data and expert experience to judge the organization time and transportation time in different prediction intervals, and obtain the corresponding wear degree data according to the corresponding second durability change curve;

[0125] The wear degree data generated during assembly and transportation is calculated in the same way as the wear degree data caused by water pressure or water quality corrosion; that is, the corresponding wear degree data generated during assembly and transportation is obtained through comprehensive weighting of the yolov5 recognition features and the expert experience of a large amount of data;

[0126] The first durability change curve is secondarily corrected to generate a third durability change curve as follows:

[0127] ;

[0128] Among them, is the wear degree of the third durability change curve at the first usage duration , is the wear degree of the first durability change curve at the first usage duration , is the wear degree corresponding to the second usage duration on the second durability change curve;

[0129] In this embodiment, relevant technicians adopted some un-trained data for data deduction simulation. A total of three first usage durations were set. The first first usage duration was 185 minutes, the second first usage duration was 95 minutes, and the third first usage duration was 90 minutes to simulate the deduction simulation of the same large-span assembled inspection door under continuous use. The generated first durability change curve is as Figure 8 shown, and the third durability change curve generated after being corrected by the second durability change curve is as Figure 9 shown;

[0130] Based on the second usage duration and the wear degree data, the second durability change curve is used to correct the first durability change curve. Based on this scheme, the wear of the large-span assembled inspection door data caused by non-water pressure and water corrosion can be excluded, such as the wear during transportation or the wear during the organization process. Through this method, the wear of the large-span assembled inspection door in different situations can be further understood for estimation, so as to evaluate the service life of the large-span assembled inspection door after the current first usage duration. At the same time, the change of the second usage duration such as transportation can also be visualized, which is convenient for relevant technicians to formulate and correct the scheme, and improves the accuracy of data deduction;

[0131] Simulate and deduce the durability of the large-span assembled inspection door according to the third durability change curve and the second durability change curve; the calculation of the durability is: ; In this embodiment, 1 represents the initial durability in the un-worn state; the wear degree corresponds to the consumed durability;

[0132] In this embodiment, according toFigure 8 and Figure 9 The results generate a visualization curve of the durability of the large-span assembled inspection door, as shown in reference to Figure 10 shown

[0133] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for deducing data of a large-span assembleable detection door, characterized in that: include: Obtain material data of the large-span assemblable inspection door to be simulated, and obtain historical wear data, historical water quality data, and historical water pressure data based on the large-span assemblable inspection door with the same material data; generate a historical change curve based on the historical wear data, and obtain historical slope data and a first usage duration of the historical change curve; Establishing a detection door influencing parameter prediction model, and performing model training on the detection door influencing parameter prediction model according to the first usage duration, the historical slope data, the historical water quality data, the historical water pressure data, and the historical wear data; The detection gate influence parameter prediction model includes a detection gate data encoding module and a detection gate data decoding module, and the model training process is trained in the detection gate data encoding module; the detection gate data encoding module includes a detection gate influence data processing unit, a detection gate influence data feature unit and a detection gate influence data training output unit; the detection gate influence data feature unit includes: a Transformer network, a residual convolution network, an LSTM network and a feature fusion network; Acquire the first usage time, water quality data and water pressure data corresponding to the location information of the water conservancy detection target, and acquire wear prediction data and slope prediction data according to the detection door influencing parameter prediction model, and fit the first durability change curve according to the wear prediction data; and correct the first durability change curve according to the slope prediction data; According to the detection gate data decoding module of the detection gate influencing parameter prediction model, the wear degree prediction data and the slope prediction data are predicted; the first durability change curve is nonlinearly fitted by Bayesian inference method and genetic algorithm; the detection gate data decoding module includes a detection gate durability prediction processing unit, a detection gate durability prediction coding connection unit and a detection gate durability prediction feature extraction and output unit; the detection gate durability prediction feature extraction and output unit includes: 1 Transformer network, 5 residual convolution networks and 6 LSTM networks; 1 Transformer network extracts the feature enhancement coding of the durability prediction preprocessing data, and performs feature association with the features of the detection gate durability prediction coding connection unit; 5 residual convolution networks and 6 LSTM networks extract the features of the durability prediction preprocessing data after feature association, and perform prediction output; According to the wear data of the large-span assemblable detection door with the same material data and the second use time, a second durability change curve is generated by fitting; and the first durability change curve is corrected twice to generate a third durability change curve; The durability of the large-span assemblable inspection door is simulated and deduced based on the third durability change curve and the second durability change curve.

2. A method for deducing data of a large-span assembleable detection door according to claim 1, characterized in that: The historical wear data, the historical water quality data and the historical water pressure data obtain corresponding data based on a fixed time recording point T, and the data collection point of the historical slope data is the same as the fixed time recording point T.

3. A method for deducing data of a large-span assembleable detection door according to claim 1, characterized in that: Generating a historical change curve by fitting the historical wear data includes: generating wear change coordinates with the historical wear data as the vertical axis coordinate and time as the horizontal axis coordinate, and performing nonlinear fitting on the wear change coordinates to generate the historical change curve.

4. A method for deducing data of a large-span assembleable detection door according to claim 3, characterized in that: The wear degree variation coordinates are nonlinearly fitted by using Bayesian inference method and genetic algorithm.

5. The method for deducing data of a large-span assembleable detection door according to claim 1 is characterized in that: The water conservancy detection target is simulated by a digital twin model, and the position information is all installable positions of the large-span assembled detection door.

6. A method for deducing data of a large-span assembleable detection door according to claim 1, characterized in that: The prediction output of the detection gate data decoding module also includes: The detection gate data decoding module generates a plurality of prediction intervals based on the first usage duration, each prediction interval being associated with the corresponding historical water quality data and the historical water pressure data; The nth prediction will use the wear prediction data of the n-1th time as prediction input data; the nth prediction input data includes the wear prediction data, the first usage time, the historical water quality data and the historical water pressure data, and the total number of the prediction intervals is N. , .

7. A method for deducing data of a large-span assembleable detection door according to claim 1, characterized in that: According to the wear data of the large-span assemblable detection door with the same material data and the second use time, a second durability change curve is generated by fitting; And secondarily correcting the first durability change curve to generate a third durability change curve comprises: With the second usage time as the horizontal axis and the wear data as the vertical axis, the second durability change curve is fitted by Bayesian inference method and genetic algorithm; The first durability change curve is corrected twice to generate a third durability change curve as follows: ； in, is the wear data, The third durability change curve is the first use time The degree of wear, The first durability change curve is the first usage time The degree of wear, The second usage time on the second durability change curve The corresponding wear degree.

Images