K-CNN-BiGRU-based ship outfield coating atmospheric pollution concentration prediction method
Through the K-CNN-BiGRU-based method, the problems of inadequate air pollution supervision and insufficient prediction methods in ship exterior coating operations are solved, and higher-precision air pollution concentration prediction is achieved, and scientific management guidance is provided.
Patent Information
- Application Number
- CN202411576351.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-06
- Publication Date
- 2025-02-18
AI Technical Summary
The supervision of air pollution in ship exterior coating operations is not in place, and the existing technology lacks effective prediction methods.
Using K-CNN-BiGRU-based method, we collect meteorological and pollutant data through environmental monitoring terminals, perform grid division and spatial interpolation, calculate spatial correlation using KNN tree algorithm, input CNN model to extract local spatiotemporal features, and in BiGRU Make global predictions to predict air pollution concentration.
It improves the accuracy of prediction of atmospheric pollution concentration in the outer field of the ship, provides scientific coating construction and safety and environmental protection management guidance, solves the problem of gradient disappearance and captures the two-way dependence relationship of sequence data.
Smart Images
Figure CN119475236A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of volatile organic matter monitoring, in particular to a method for predicting atmospheric pollution concentration of ship outdoor painting based on K-CNN-BiGRU. Background Art
[0002] The painting of the ship's main section, dock (platform) and pier, also known as outdoor painting, is the main source of pollution in shipbuilding and repair. At present, domestic and foreign shipyards are relatively mature in monitoring pollution at the factory boundary, but the outdoor painting in the factory area is in an open-air environment and it is difficult to collect spills, there are periodic moving objects such as lifting machinery, and the signal interference is strong. The operation area is large but the network wiring is imperfect and difficult to change flexibly. Therefore, related monitoring and prediction measures are not common.
[0003] Therefore, the present invention designs a method for predicting atmospheric pollution concentration of ship outdoor painting based on K-CNN-BiGRU to predict the trend of atmospheric pollution and provide auxiliary support for pollution control and prevention. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a method for predicting atmospheric pollution concentration of ship outdoor painting based on K-CNN-BiGRU, which can effectively solve the problems of inadequate supervision and insufficient prediction means of atmospheric pollution in ship outdoor painting operations in view of the shortcomings of the prior art.
[0005] The technical problem to be solved by the present invention is achieved through the following technical solutions. The present invention is a method for predicting atmospheric pollution concentration of ship outdoor painting based on K-CNN-BiGRU. The method is used to predict the atmospheric pollution concentration of the ship outdoor painting plant in the future period to guide employees to scientifically carry out painting construction and safety and environmental protection management. The method includes the following steps: (1) First, deploy environmental monitoring terminals to collect data on meteorology and pollutant concentrations, then divide the study area into grids, and then use IDW for spatial interpolation to calculate and organize the grid data set of atmospheric pollutant concentrations; (2) Use the KNN tree algorithm to calculate the spatial correlation and filter out the spatial correlation matrix, which is then arranged in time sequence and spliced into a spatiotemporal class diagram; (3) Input the spatiotemporal class graph into the CNN model to extract local spatiotemporal features; (4) The extracted spatiotemporal features are sent to BiGRU for global prediction.
[0006] The technical problem to be solved by the present invention can be further achieved by the following technical solutions. For the above-mentioned K-CNN-BiGRU-based ship outdoor painting atmospheric pollution concentration prediction method, in step (1), the collected data includes meteorological data, atmospheric pollutants in the ship outdoor painting area, ship painting operation data and time, wherein meteorological data and atmospheric pollutants are collected by cruise monitoring using a drone equipped with monitoring equipment, ship painting operation data are regularly reported by on-site construction personnel, and time values are obtained online by the system.
[0007] The technical problem to be solved by the present invention can also be further achieved by the following technical solutions. For the above-mentioned K-CNN-BiGRU-based ship outdoor painting atmospheric pollution concentration prediction method, in step (1), the on-site spatial area is divided into grids, with the lower right corner as the starting point, and the Cartesian coordinates of each grid center point are stored. Then, the inverse distance weighted method is used for spatial interpolation to interpolate the atmospheric pollution concentration within the grid, and data preprocessing is completed. Then, the data set is divided into a training data set and a test data set according to a preset ratio.
[0008] The technical problem to be solved by the present invention can also be further achieved by the following technical solution. For the above-mentioned method for predicting atmospheric pollution concentration of ship outdoor painting based on K-CNN-BiGRU, in step (2), the characteristic variable data set at each time step of grid i is input into the KNN model to calculate the spatial correlation with the surrounding grids, and the first k grids are selected and arranged in ascending order of distance to generate a spatial correlation matrix. The spatial correlation matrix of each time step is spliced to obtain the spatiotemporal class diagram P of atmospheric pollutants.
[0009] The technical problem to be solved by the present invention can also be further achieved by the following technical solution. For the above-mentioned K-CNN-BiGRU-based ship outdoor painting atmospheric pollution concentration prediction method, in step (3), the spatiotemporal class graph is input into the CNN model, the local spatiotemporal features are extracted by the convolution layer, the spatiotemporal features are compressed by the pooling layer, and the spatiotemporal features are fused by the fully connected layer to obtain a one-dimensional local spatiotemporal correlation vector.
[0010] The technical problem to be solved by the present invention can also be further achieved by the following technical solution. For the above-mentioned method for predicting atmospheric pollution concentration for ship outdoor painting based on K-CNN-BiGRU, in step (4), the spatiotemporal correlation vector is input into the BiGRU model to predict the atmospheric pollution concentration prediction value of the dyeing grid i at time t.
[0011] The technical problem to be solved by the present invention can also be further achieved through the following technical solutions. For the above-mentioned K-CNN-BiGRU-based ship outdoor painting atmospheric pollution concentration prediction method, this method uses mean square error as the loss function of the model, uses Adam optimization algorithm as the model optimizer, and uses mean absolute percentage error and root mean square error to evaluate the prediction deviation of the model. After many experiments, a prediction model with lower loss is obtained for the prediction of atmospheric pollution of ship outdoor painting.
[0012] The technical problem to be solved by the present invention can also be further achieved through the following technical solutions. For the above-mentioned K-CNN-BiGRU-based ship outdoor painting atmospheric pollution concentration prediction method, in this method, the prediction result of the previous time step is added to the input of the next prediction, and multi-step prediction is performed to achieve medium- and long-term atmospheric pollution concentration prediction.
[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. Aiming at the problems of low accuracy of ship field coating environment monitoring and being easily affected by climate and environment, the present invention constructs a spatiotemporal correlation prediction model based on deep learning theory to improve the prediction accuracy of atmospheric pollution concentration in the correlation area; 2. The present invention adopts the K-CNN-BiGRU model to predict the atmospheric pollution concentration of ship outdoor painting, which solves the gradient vanishing problem and better captures the bidirectional dependency of sequence data, thereby improving the model's ability to understand sequence data and model performance. Compared with BiLSTM, the GRU unit has a simpler structure and fewer parameters, so the calculation speed may be faster, which has certain advantages in time series prediction tasks. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 The figure is a flow chart of a prediction method of the present invention. DETAILED DESCRIPTION
[0015] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0016] Reference Figure 1 , a K-CNN-BiGRU-based method for predicting atmospheric pollution concentration in ship outdoor painting plants, which is used to predict the atmospheric pollution concentration in the future period of the ship outdoor painting plant area and guide employees to carry out painting construction and safety and environmental protection management scientifically; The method includes the following 4 steps: Step 1: Deploy environmental monitoring terminals to collect data such as meteorological and pollutant concentrations, divide the study area into grids, use IDW for spatial interpolation, and calculate and organize the grid data set of atmospheric pollutant concentrations; Among them, the data to be collected are divided into meteorological data, atmospheric pollutants in the ship's external painting area, ship painting operation data and time. Meteorological data and atmospheric pollutants are collected by cruise monitoring with drones equipped with monitoring equipment. The ship painting operation data is reported regularly by on-site construction personnel, and the time value is obtained online by the system; the ship's external painting areas such as slipways and docks are mainly concave in shape, and the on-site three-dimensional space area is divided into grids, with the lower right corner as the starting point. The Cartesian coordinates are stored at the center point of each grid. Affected by the painting process and the occupation of physical objects such as the hull, it is impossible to deploy monitoring terminals in each grid to obtain pollution values. Therefore, the inverse distance weighted method is used for spatial interpolation to interpolate the atmospheric pollution concentration in the grid; after completing data preprocessing, the data set is then divided into a training data set and a test data set according to a preset ratio; Specifically, meteorological data mainly include temperature, humidity, wind speed and wind direction. Air pollutants from ship field painting mainly include non-methane total hydrocarbons, TSP, PM10 and PM2.5. Ship painting operation data mainly include paint type, painting area, paint usage and operation start and end time. Since ship field painting is an unorganized emission and is greatly affected by meteorological factors, it is appropriate to collect pollution data when the standard deviation of wind direction change is less than 29 degrees or the average wind speed is at [1,3] m / s or the atmospheric stability is at level F, E, or D. Under normal circumstances, the category of the worst suitable factor among the three meteorological factors of wind direction change, average wind speed, and atmospheric stability is used to estimate the overall suitability of meteorological conditions in this monitoring. Since the data units are different and the disparity is large, the grid data set must first be normalized so that the data is between [0,1] to eliminate the impact of the dimension. The min-max standardization is used for normalization. The dataset is divided into training and testing datasets in a ratio of 4:1 and decomposed into feature datasets of different categories.
[0017] Step 2: Use the KNN tree algorithm to calculate the spatial correlation and filter out the spatial correlation matrix, which is then arranged in time sequence and spliced into a space-time class diagram; Among them, the atmospheric pollution concentration value in the area changes over time under the influence of meteorological conditions and painting operations. At the same time, the concentration values in the associated grids will affect each other and cause the concentration values to change. Therefore, the prediction should take into account the spatiotemporal characteristics. The characteristic variable data set at each time step of grid i is input into the KNN model to calculate the spatial correlation with the surrounding grids. The first k grids are selected and arranged from small to large according to the distance to generate a spatial correlation matrix. The spatial correlation matrix of each time step is spliced, and the spatial correlation matrix data set is spliced using a sliding window to obtain the spatiotemporal class diagram P of atmospheric pollutants.
[0018] Step 3: Input the spatiotemporal class graph into the CNN model to extract local spatiotemporal features; Specifically, the spatiotemporal class graph is input into the CNN model, the convolutional layer extracts local spatiotemporal features, the pooling layer compresses the spatiotemporal features, and the fully connected layer fuses the spatiotemporal features to obtain a one-dimensional local spatiotemporal correlation vector.
[0019] Step 4: Send the extracted spatiotemporal features into BiGRU for global prediction.
[0020] Specifically, the spatiotemporal correlation vector is input into the BiGRU model to predict the atmospheric pollution concentration prediction value of the grid i at time t.
[0021] In practical application, the present invention adopts mean square error as the loss function of the model, uses Adam optimization algorithm as the model optimizer, and uses mean absolute percentage error and root mean square error to evaluate the prediction deviation of the model. After multiple experiments, a prediction model with lower loss is obtained for the prediction of atmospheric pollution during ship field painting. Among them, the prediction result of the previous time step can also be added to the input of the next prediction to achieve multi-step prediction and meet the needs of medium- and long-term atmospheric pollution concentration prediction.
[0022] The present invention is an outdoor coating environment monitoring and early warning system for the shipbuilding and repairing industry, which can realize the prediction and early warning of air pollution in the future period, and can be applied to slipways, docks and other places in various types of shipyards. It has broad application prospects and can assist management personnel to realize environmental and safety management and achieve cost reduction and efficiency improvement.
Claims
1. A method for predicting atmospheric pollution concentration of ship outdoor painting based on K-CNN-BiGRU, characterized by: This method is used to predict the atmospheric pollution concentration in the future period of the ship's external painting plant to guide employees to scientifically carry out painting construction and safety and environmental protection management. The method includes the following steps: (1) First, deploy environmental monitoring terminals to collect data on meteorology and pollutant concentrations, then divide the study area into grids, and then use IDW for spatial interpolation to calculate and organize the grid data set of atmospheric pollutant concentrations; (2) Use the KNN tree algorithm to calculate the spatial correlation and filter out the spatial correlation matrix, which is then arranged in time sequence and spliced into a spatiotemporal class diagram; (3) Input the spatiotemporal class graph into the CNN model to extract local spatiotemporal features; (4) The extracted spatiotemporal features are sent to BiGRU for global prediction.
2. The method for predicting atmospheric pollution concentration of ship outdoor painting based on K-CNN-BiGRU according to claim 1 is characterized in that: In step (1), the collected data include weather, atmospheric pollutants in the ship's off-site painting area, ship painting operation data and time. Among them, weather and atmospheric pollutants are collected by cruise monitoring using drones equipped with monitoring equipment. Ship painting operation data is reported regularly by on-site construction personnel, and time values are obtained online by the system.
3. The method for predicting atmospheric pollution concentration of ship outdoor painting based on K-CNN-BiGRU according to claim 1 or 2, characterized in that: In step (1), the on-site spatial area is divided into grids, with the lower right corner as the starting point. The Cartesian coordinates of each grid center point are stored, and then the inverse distance weighted method is used for spatial interpolation to interpolate the atmospheric pollution concentration within the grid to complete data preprocessing. Then, the data set is divided into a training data set and a test data set according to a preset ratio.
4. The method for predicting atmospheric pollution concentration during ship exterior painting based on K-CNN-BiGRU according to claim 1 is characterized in that: In step (2), the characteristic variable data set at each time step of grid i is input into the KNN model to calculate the spatial correlation with the surrounding grids. The first k grids are selected and arranged in ascending order of distance to generate a spatial correlation matrix. The spatial correlation matrices of each time step are spliced to obtain the spatiotemporal class diagram P of atmospheric pollutants.
5. The method for predicting atmospheric pollution concentration of ship outdoor painting based on K-CNN-BiGRU according to claim 1 is characterized in that: In step (3), the spatiotemporal class graph is input into the CNN model, the convolutional layer extracts the local spatiotemporal features, the pooling layer compresses the spatiotemporal features, and the fully connected layer fuses the spatiotemporal features to obtain a one-dimensional local spatiotemporal correlation vector.
6. The method for predicting atmospheric pollution concentration during ship exterior painting based on K-CNN-BiGRU according to claim 1 is characterized in that: In step (4), the spatiotemporal correlation vector is input into the BiGRU model to predict the atmospheric pollution concentration of the grid i at time t.
7. The method for predicting atmospheric pollution concentration during ship exterior painting based on K-CNN-BiGRU according to claim 1 is characterized in that: This method uses mean square error as the loss function of the model, Adam optimization algorithm as the model optimizer, mean absolute percentage error and root mean square error to evaluate the prediction deviation of the model. After many experiments, a prediction model with lower loss is obtained for the prediction of atmospheric pollution from ship outdoor painting.
8. The method for predicting atmospheric pollution concentration during ship exterior painting based on K-CNN-BiGRU according to claim 1 or 7, characterized in that: In this method, the prediction result of the previous time step is added to the input of the next prediction, and multi-step prediction is performed to achieve medium- and long-term prediction of atmospheric pollution concentration.