Multi-dimensional environment comprehensive evaluation method and system

Abstract

The invention relates to the technical field of environment evaluation, and discloses a multi-dimensional environment comprehensive evaluation method and system, and the method comprises the steps: carrying out the comprehensive image shooting of an environment through an unmanned plane carrying device, constructing an environment real scene model through combining laser scanning point cloud data and an algorithm, and providing a precise reference for the layout of a sensor. The method comprises the following steps: selecting proper sensors according to requirements and arranging the sensors; ensuring comprehensive coverage; setting an evaluation period; collecting environmental parameter values, constructing Bayesian network fusion data to obtain parameter evaluation values, calculating actual comprehensive evaluation values by combining subjective weights and objective weights, reflecting environmental conditions, and collecting historical data. And constructing an ARI MA model to predict an environmental parameter evaluation value in the next evaluation period, evaluating the prediction precision, if the precision is low, generating a prediction deviation signal, collecting data in real time in the evaluation period, constructing a complex time sequence, training a CNN-LSTM model for prediction, obtaining a prediction comprehensive evaluation value, and performing comprehensive evaluation on an environmental evaluation area.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental evaluation, and in particular to a multi-dimensional environmental comprehensive evaluation method and system. Background Art

[0002] In the field of environmental science, accurate and comprehensive assessment of the environment is of vital importance. With the acceleration of industrialization and urbanization, environmental problems are becoming increasingly complex and diverse. Traditional single-dimensional or simple environmental evaluation methods can no longer meet current needs. Researchers have gradually discovered a series of key issues that need to be addressed in practical exploration.

[0003] In the environmental data collection process, early technical means were limited, making it difficult to obtain comprehensive and high-precision environmental information. In the past, when collecting data for large-scale environmental assessment areas, they relied on ground monitoring stations, which had limited coverage and a large number of monitoring blind spots, and could not accurately reflect the overall environmental conditions of the region. In existing technologies, when constructing environmental real-scene models, due to the lack of effective data fusion methods, they only rely on a single data source, resulting in the constructed models being unable to accurately reflect the actual environmental conditions. For example, if modeling is based only on a small amount of ground measurement data, it is difficult to reflect the overall picture of the terrain and the spatial distribution relationship of environmental elements such as buildings. The integrity and accuracy of the model are poor, and it cannot provide reliable support for subsequent environmental assessments and decision-making.

[0004] Traditional methods have obvious flaws in environmental parameter assessment and comprehensive evaluation. When integrating multi-source environmental data, they are unable to effectively handle the complex relationships between data, resulting in inaccurate environmental parameter assessments. For example, when analyzing atmospheric environmental parameters, they fail to fully consider the mutual influence between meteorological factors and pollutant concentrations, causing the assessment results to deviate significantly from the actual situation. When determining the weights of environmental parameters, traditional methods rely heavily on subjective experience and lack objective data support, resulting in irrational weight distribution and an inability to truly reflect the importance of each environmental parameter in the comprehensive evaluation.

[0005] In response to the above problems, the present invention proposes a multi-dimensional environmental comprehensive evaluation method and system. Summary of the Invention

[0006] The object of the present invention is to provide a multi-dimensional environmental comprehensive evaluation method and system to solve at least one of the above-mentioned problems in the prior art.

[0007] In one aspect, the present invention proposes a multi-dimensional environmental comprehensive evaluation method, comprising the following steps:

[0008] The parameter evaluation values ​​of environmental parameters are obtained by fusing multi-location sensor data using Bayesian networks, and dynamic comprehensive weights are generated by combining subjective weights with objective weights. The actual comprehensive evaluation value is calculated through linear weighting.

[0009] Based on historical data, an ARIMA model is constructed to predict the environmental assessment area and obtain a predicted comprehensive evaluation value. The prediction accuracy of the ARIMA model is evaluated by comparing the comprehensive evaluation value with the actual value. If the error exceeds the limit, environmental parameter data is collected in real time and a CNN-LSTM prediction model is constructed. The DTW algorithm is used to determine the prediction accuracy of the CNN-LSTM prediction model and the model update strategy is adaptively adjusted.

[0010] Conduct a comprehensive evaluation of the environmental assessment area based on the predicted comprehensive evaluation value and the actual comprehensive evaluation value.

[0011] On the other hand, the present invention proposes a multi-dimensional environmental comprehensive evaluation system, which is used to implement the multi-dimensional environmental comprehensive evaluation method, including the following modules:

[0012] Environmental modeling module: establishes an environmental scene model of the environmental assessment area, sets up multiple sensors based on the environmental scene model, and arranges the multiple sensors;

[0013] Parameter quantification module: Uses Bayesian network to fuse multi-location sensor data to obtain parameter evaluation values ​​of environmental parameters, combines subjective weights with objective weights to generate dynamic comprehensive weights, and calculates the actual comprehensive evaluation value through linear weighting;

[0014] Model prediction module: Based on historical data, an ARIMA model is constructed to predict the environmental assessment area, and a predicted comprehensive evaluation value is obtained. The prediction accuracy of the ARIMA model is evaluated by comparing the comprehensive evaluation value with the actual value. If the error exceeds the limit, environmental parameter data is collected in real time and a CNN-LSTM prediction model is constructed. The DTW algorithm is used to determine the prediction accuracy of the CNN-LSTM prediction model, and the model update strategy is adaptively adjusted.

[0015] Comprehensive evaluation module: Comprehensively evaluate the environmental assessment area based on the predicted comprehensive evaluation value and the actual comprehensive evaluation value.

[0016] Beneficial effects of the present invention:

[0017] 1. The present invention uses drones equipped with high-resolution cameras and GPS to capture images of the environmental assessment area, which can overcome the limitations of traditional ground monitoring and obtain environmental images with wide coverage, rich details and accurate geographic location information. Edge feature points are extracted through advanced image processing technology and integrated with laser scanning point cloud data. The MVS algorithm is used to construct an extremely accurate environmental real-scene model. This process lays a solid foundation for subsequent environmental monitoring work, enables the sensor layout to be scientifically set according to the actual environmental conditions, comprehensively improves the accuracy and comprehensiveness of environmental data collection, provides high-quality data support for environmental assessment, and effectively improves the shortcomings of traditional methods in data acquisition and early model construction.

[0018] 2. The present invention integrates multi-source information by constructing a Bayesian network, rationally fuses sensor data, and combines scientifically determined dynamic comprehensive weights to calculate the actual comprehensive evaluation value that is more scientific and reliable. At the same time, time series data is constructed and predicted based on the ARIMA model, which can effectively grasp the changing trend of environmental parameters. When the prediction accuracy is insufficient, data is collected in real time and the CNN-LSTM prediction model is trained. The prediction accuracy of the model is evaluated by the DTW algorithm. This series of operations greatly improves the accuracy of environmental parameter prediction. Finally, an environmental comprehensive evaluation is performed based on the accurate predicted comprehensive evaluation value or the actual comprehensive evaluation value, which provides a more accurate and forward-looking basis for environmental management decision-making and significantly improves the overall efficiency of environmental assessment work. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 This is a flow chart of a multi-dimensional environmental comprehensive evaluation method provided in Example 1 of the present invention;

[0021] Figure 2 This is a flow chart of the steps for obtaining the actual comprehensive evaluation value in a multi-dimensional environmental comprehensive evaluation method provided in the first embodiment of the present invention;

[0022] Figure 3 This is a flowchart of the steps for evaluating the prediction accuracy of the ARIMA model in a multi-dimensional environmental comprehensive evaluation method provided in Example 1 of the present invention;

[0023] Figure 4 This is a flowchart of a multi-dimensional environmental comprehensive evaluation system provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0025] Example 1

[0026] like Figure 1As shown, the embodiment of the present invention provides a multi-dimensional environmental comprehensive evaluation method, which specifically includes the following steps:

[0027] S1. Use a drone equipped with a high-resolution camera and GPS to capture comprehensive images of the environmental assessment area. Use image processing technology to extract edge feature points. Combine laser scanning point cloud data with the MVS algorithm to construct a real-world model of the environment. Set up sensor layouts in the environmental assessment area based on the real-world model.

[0028] Use drones equipped with high-resolution cameras and high-precision GPS modules to capture images of the environmental assessment area;

[0029] Specifically, map information of the environmental assessment area is obtained, an optimal flight route is planned for the UAV based on the map information, and based on the optimal flight route, the UAV is controlled to fly around the environmental assessment area along the optimal flight route during an image capture period. The environmental assessment area is captured using a camera configured on the UAV at the capture time, and images captured during the image capture period are grouped into an environmental image group.

[0030] The image capture period refers to a period of time before environmental modeling is performed to capture images of the environmental assessment area. The capture time point is a time point within the image capture period. There are several capture time points within the image capture period. The intervals between adjacent capture time points are all the same and are preset by technical personnel to ensure that the captured environmental image group completely covers the environmental assessment area and has a high degree of overlap. During the capture process, the camera exposure parameters are adjusted in real time according to the lighting conditions to ensure the image quality of the environmental image group.

[0031] It should be noted that the high-resolution camera can capture environmental details, and the GPS module can accurately locate the shooting location. The optimal route is planned to ensure full coverage of the assessment area. Adjacent shooting points are set with the same interval length and high overlap to facilitate subsequent image stitching and analysis. Exposure parameters are adjusted according to light conditions to ensure clear images with high color reproduction, providing high-quality materials for subsequent environmental analysis and helping to fully understand the environmental conditions in the environmental assessment area.

[0032] Perform edge detection on the images in the obtained environment image group using the Canny algorithm to obtain edge feature points of terrain and buildings in the images in the environment image group. Generate a 128-dimensional descriptor for each feature point. The descriptor contains the gradient information of the pixels around the edge feature point. Calculate the Euclidean distance between the edge feature point descriptors in different images. If the Euclidean distance between the edge feature point descriptors is less than a distance threshold, the edge feature point is classified into a feature matching group.

[0033] Based on the obtained feature matching group, the plane image coordinates of the edge feature points in the feature matching group in the corresponding image are obtained based on the principle of multi-view geometry. The three-dimensional coordinates of the edge feature points are calculated. The point cloud data of the environmental assessment area is obtained through laser scanning. The MVS algorithm is used for three-dimensional reconstruction to construct an environmental real scene model of the environmental assessment area.

[0034] Select sensors for the environmental assessment area based on the environmental reality model and determine the layout of the sensors in the environmental assessment area. The specific location of the sensors needs to consider the mutual influence between sensors and the complementarity of data to ensure that the layout of sensors can cover the entire environmental assessment area.

[0035] For example, for atmospheric environment monitoring, high-precision gas sensors can be used, such as electrochemical sensors for detecting gas concentrations such as SO2 and NOx, and optical sensors for detecting particulate matter concentrations. For water environment monitoring, water quality sensors such as pH sensors, dissolved oxygen sensors, and COD sensors can be used. For noise monitoring, noise sensors can be used. For traffic flow monitoring, lidar sensors or geomagnetic sensors can be used.

[0036] For example, in a city, air quality sensors are required to be deployed in different blocks and at different building heights to comprehensively monitor the quality of the atmospheric environment. Air quality sensors and meteorological sensors are deployed together to simultaneously obtain information on atmospheric pollutant concentrations and meteorological conditions, and analyze the impact of meteorological factors on air quality.

[0037] It should be noted that the purpose of building an environmental reality model is to intuitively display the three-dimensional spatial structure of the assessment area, provide accurate spatial reference for sensor layout planning, and improve the scientificity and rationality of sensor layout;

[0038] S2. Set an environmental assessment cycle, collect environmental parameter values ​​at the end of the cycle, build a Bayesian network to integrate multi-source information, fuse sensor data to obtain parameter assessment values, combine subjective weights and objective weights to obtain dynamic comprehensive weights, and calculate the actual comprehensive evaluation value by combining the parameter assessment values ​​and dynamic comprehensive weights;

[0039] like Figure 2 As shown, the specific steps for obtaining the actual comprehensive evaluation value are as follows;

[0040] An environmental assessment cycle is set. At the end of the environmental assessment cycle, the set sensors are used to collect several environmental parameter values ​​within the environmental assessment area. The end of the environmental assessment cycle is marked as the parameter collection time. The parameter collection time of the current environmental assessment cycle is also the starting point of the next environmental assessment cycle.

[0041] It should be noted that the purpose of regularly collecting environmental parameters is to track the dynamic changes of the environment over time, clarify the time of parameter collection, and construct continuous time series data to provide a time dimension basis for subsequent environmental analysis, facilitating the study of environmental change laws and trends;

[0042] For example, the environmental parameters include temperature, humidity, gas concentrations such as SO2, particulate matter concentration, water pH value, water dissolved oxygen concentration, environmental noise parameters, etc.;

[0043] It should be noted that the reason why there are several environmental parameter values ​​in the environmental assessment area is that in order to ensure the comprehensiveness of the environmental assessment, it is necessary to set up the same type of sensors at different locations to collect the corresponding environmental parameter values ​​when conducting a comprehensive evaluation of the environment in the environmental assessment area;

[0044] Based on any environmental parameter, a Bayesian network structure is constructed. According to the relationship between sensors at different locations used to collect environmental parameters, a Bayesian network structure is constructed. The nodes in the network represent the environmental parameter values ​​collected by sensors at different locations, and the edges represent the dependencies between the nodes.

[0045] For example, if sensors at certain locations are greatly affected by environmental factors, while sensors at other locations are relatively stable, this dependency can be reflected in the Bayesian network;

[0046] Determine the prior probability distribution of each node and the conditional probability distribution between nodes, input the environmental parameter values ​​of each sensor collected at the current parameter collection time as evidence into the Bayesian network, and use the Bayesian inference algorithm to calculate the posterior probability distribution of the fused environmental parameters;

[0047] Extract the mean from the posterior probability distribution to obtain the parameter evaluation value of the environmental parameter;

[0048] It should be noted that the role of the Bayesian network is to effectively integrate multi-source information, reflect the dependency relationship between sensor data at different locations, and fuse the sensor data through prior and conditional probability distribution and inference algorithms to obtain more accurate environmental parameter assessment values, thereby improving the reliability and accuracy of environmental parameter assessment.

[0049] The judgment matrix is ​​preset according to the relative importance of each environmental parameter in the environmental assessment area. The element a of the judgment matrix is ij Indicates the importance of the i-th environmental parameter relative to the j-th environmental parameter. Perform a consistency test on the judgment matrix. If it passes the test, calculate the maximum eigenvalue and corresponding eigenvector of the judgment matrix. Normalize the eigenvector to obtain the subjective weight of the environmental parameter.

[0050] Among them, i and j represent the type number of environmental parameters;

[0051] The parameter evaluation values ​​of each environmental parameter at the time of parameter collection are standardized by the Z-score standardization method. Based on the standardized data, the entropy value H of the environmental parameter is calculated. i ;

[0052] It should be noted that the entropy value reflects the degree of discreteness of the data. The larger the entropy value, the greater the degree of discreteness of the data and the more information the environmental parameters provide.

[0053] Calculate the objective weight of environmental parameters based on entropy value The formula is:

[0054]

[0055] Where n is the total number of environmental parameter types;

[0056] The subjective weight and objective weight are combined by linear combination method to obtain the dynamic comprehensive weight w of environmental parameters. i ;

[0057] According to the calculated dynamic comprehensive weight, combined with the parameter evaluation value of each environmental parameter, the actual comprehensive evaluation value E of the environmental assessment area within the assessment period is calculated using linear weighted summation. re ;

[0058] It should be noted that the subjective weight reflects the expert's judgment on the relative importance of each environmental parameter, and the entropy weight reflects objective information based on the degree of data dispersion. The dynamic comprehensive weight obtained by combining the subjective and objective weights is more comprehensive and reasonable. The role of the linear weighted summation in calculating the actual comprehensive evaluation value is to comprehensively consider various environmental parameters and give a quantitative result of the overall environmental status of the assessment area within the assessment period.

[0059] S3. Construct time series data and build an ARIMA model based on the time series data, predict the environmental parameter evaluation value of the next evaluation period and calculate the predicted comprehensive evaluation value, compare the actual comprehensive evaluation value to evaluate the prediction accuracy of the ARIMA model, and generate a prediction deviation signal if the accuracy is low;

[0060] like Figure 3 As shown, the specific steps for evaluating the prediction accuracy of the ARIMA model are as follows;

[0061] Collect historical parameter data, including the parameter evaluation values ​​of each environmental parameter at several past parameter collection moments and the timestamps of the parameter collection moments;

[0062] The collected historical parameter data are cleaned, outliers are removed by using the Laida criterion, and missing values ​​are filled by linear interpolation;

[0063] It should be noted that outliers and missing values ​​may be caused by sensor failure, data transmission errors, etc. Data cleaning removes erroneous or outliers to ensure data quality. Linear interpolation is used to fill missing values ​​and maintain the integrity of the time series. This lays the foundation for building an accurate time series prediction model and ensures that the model can be trained and predicted based on complete and accurate historical data.

[0064] Based on any environmental parameter, the parameter evaluation values ​​of the environmental parameter at different collection time points are constructed according to the time series data X it , where i represents the type number of the environmental parameter, and t represents the time series number of the parameter collection moment corresponding to the parameter evaluation value in the historical parameter data;

[0065] According to the time series data X it Construct an ARIMA model, a time series forecasting model that combines autoregression (AR), difference (I), and moving average (MA):

[0066]

[0067] Among them, B is the lag operator, d is the difference order, is the autoregressive coefficient polynomial, θ(B) is the moving average coefficient polynomial, ε t is a white noise sequence;

[0068] Calculate the autocorrelation function (ACF) and partial autocorrelation function (PACF) of time series data, draw ACF and PACF graphs, and determine the order (p, d, q) of the ARIMA model, where p is the autoregressive order, d is the difference order, and q is the moving average order;

[0069] The least squares method is used to estimate the parameters of the ARIMA model and obtain the specific expression of the model;

[0070] Input the time series data into the ARIMA model with a determined order (p, d, q) and completed parameter estimation, predict the parameter evaluation values ​​of the environmental parameters in the next evaluation period, and output the predicted parameter evaluation values ​​of the environmental parameters;

[0071] Obtain the dynamic comprehensive weight calculated in the current assessment cycle, combine it with the evaluation value of the prediction parameters, and use linear weighted summation to calculate the predicted comprehensive evaluation value of the environmental assessment area in the next assessment cycle;

[0072] The predicted comprehensive evaluation value of the environmental assessment area obtained by the ARIMA model at the parameter collection time of the next assessment cycle is marked as E pr ;

[0073] At the next parameter collection moment, a linear weighted sum calculation is performed based on the dynamically adjusted dynamic comprehensive weight and the parameter evaluation value obtained by the collection calculation to obtain the actual comprehensive evaluation value E of the environmental assessment area at the parameter collection moment. re ;

[0074] According to the obtained prediction comprehensive evaluation value E pr Compared with the actual comprehensive evaluation value E re , through the formula

[0075]

[0076] Calculate the relative forecast error of the ARIMA model;

[0077] Taking into account factors such as environmental assessment accuracy requirements and historical prediction error distribution, a relative error threshold is preset and the calculated prediction relative error is compared with the relative error threshold;

[0078] If the prediction relative error is less than the relative error threshold, it means that the prediction accuracy of the ARIMA model is high, and the ARIMA model will continue to be used to predict the comprehensive evaluation value in the subsequent evaluation cycle;

[0079] If the forecast relative error is greater than or equal to the relative error threshold, it means that the forecast accuracy of the ARIMA model is low and a forecast deviation signal is generated;

[0080] It should be noted that the purpose of this step is to construct an ARIMA model to capture the temporal trends and periodic characteristics of environmental parameters, improve the model's prediction accuracy, and evaluate the model's effectiveness in predicting environmental parameters, thus providing a prediction reference for environmental assessment.

[0081] S4. After receiving the prediction deviation signal, collect environmental parameter data in real time, combine it with historical data to build a complex time series and train the CNN-LSTM prediction model. After the training is completed, use the model to make predictions and use the DTW algorithm to evaluate the similarity between the predicted data and the real-time data to evaluate the prediction accuracy of the CNN-LSTM prediction model.

[0082] If the generated prediction deviation signal is received, several collection time points are selected within the evaluation period, and each environmental parameter is collected in real time at the collection time points, wherein the intervals between adjacent collection time points are the same;

[0083] Calculate the parameter evaluation value of each environmental parameter in real time based on the collected data during the current evaluation period, obtain the collected historical parameter data, and combine the parameter evaluation values ​​within the evaluation period to form a complex time series data, where k represents the time sequence number of the parameter collection moment or collection time point corresponding to the parameter evaluation value in the complex time series data;

[0084] A CNN-LSTM prediction model is constructed using CNN as the time series feature extraction module and LSTM as the time series dependency modeling module. The CNN-LSTM prediction model is trained based on complex time series data, with the convolutional layer parameters fixed to retain common time series features.

[0085] When the parameter evaluation value at the current acquisition time point arrives, a sliding window mechanism is used to extract local time series segments, and online gradient descent is performed using the mean square error (MSE) as the loss function. The formula of the loss function L is as follows:

[0086]

[0087] Among them, T w It represents the window length of the sliding window, λ is the L2 regularization coefficient, which is used to prevent overfitting, and η is the parameter of the model. It represents the evaluation value of the model prediction parameter, y k It represents the actual parameter evaluation value obtained by collection and calculation;

[0088] Monitor the activation sparsity of each layer of the model, calculate the sparsity of each layer and compare it with the sparsity threshold. If the sparsity of any layer is continuously lower than the sparsity threshold for a period longer than the duration, remove redundant convolution kernels or LSTM units and perform complexity-aware pruning to reduce model complexity.

[0089] After training, a trained CNN-LSTM prediction model is obtained;

[0090] It should be noted that the above operations are used to collect and supplement the latest environmental information in real time, combine historical data to enrich time series features, use CNN to extract time series features, and use LSTM to model time series dependencies. The combination of the two improves the model's ability to capture changes in complex environmental parameters. The fixed convolutional layer retains common time series features, and the sliding window and mean square error training adapt the model to new data. Monitoring sparsity pruning reduces model complexity, prevents overfitting, and improves model training and prediction performance.

[0091] The trained CNN-LSTM prediction model is used to predict the parameter evaluation values ​​of each environmental parameter in the subsequent evaluation cycle. Based on the parameter evaluation values ​​and dynamic comprehensive weights, a linear weighted summation is used to calculate the predicted comprehensive evaluation value of the environmental assessment area.

[0092] The actual comprehensive evaluation value obtained through real-time collection and calculation is compared with the predicted comprehensive evaluation value, and the similarity of the change trends of the two is calculated. The dynamic time warping (DTW) algorithm is used to measure the similarity and construct two time series.

[0093] A={a1,a2,……,a m}

[0094] B={b1,b2,......,b m}

[0095] Where A represents the time series of the predicted comprehensive evaluation value, B represents the time series of the actual comprehensive evaluation value, and m represents the number of data in the time series.

[0096] D rs =d(a r , b s )

[0097] Among them, d is the distance measurement method such as Euclidean distance, and the cumulative distance matrix C is calculated by dynamic programming algorithm to calculate the final DTW distance C mm

[0098]

[0099] Compare the calculated DTW distance with the distance threshold;

[0100] If the DTW distance is less than the distance threshold, it indicates that the change trends of the predicted data of the CNN-LSTM prediction model and the real-time data are highly similar. The model is used for subsequent predictions, and real-time collection within the evaluation period is stopped. Environmental parameters are collected only at the parameter collection time of the evaluation period, and the DTW distance between the predicted comprehensive evaluation value and the actual comprehensive evaluation value at the parameter collection time is calculated.

[0101] If the DTW distance is greater than or equal to the distance threshold, it indicates that the similarity between the change trends of the CNN-LSTM prediction model's predicted data and the real-time data is low. Continue to collect environmental parameters in real time and train the CNN-LSTM prediction model within the evaluation period until the DTW distance is less than the distance threshold.

[0102] It should be noted that the purpose of the above operation is to evaluate the accuracy of model prediction by comparing the similarity between the prediction and the actual comprehensive evaluation value. The DTW algorithm can effectively measure the similarity of time series and decide whether to continue real-time collection and training based on the threshold, ensuring that the model prediction is closely aligned with real-time environmental changes. When the model prediction accuracy meets the requirements, it saves resources and improves the efficiency of environmental assessment.

[0103] S5. Conduct a comprehensive evaluation of the environmental assessment area based on the predicted comprehensive evaluation value or the actual comprehensive evaluation value;

[0104] Obtain the predicted comprehensive evaluation value obtained by the ARIMA model or CNN-LSTM prediction model and the actual comprehensive evaluation value obtained through collection and calculation;

[0105] Establish an evaluation model and input the obtained predicted comprehensive evaluation value or actual comprehensive evaluation value into the evaluation model. The evaluation model comprehensively evaluates the environmental assessment area by determining the range of deviation between the predicted comprehensive evaluation value or actual comprehensive evaluation value and the preset standard comprehensive evaluation value, and outputs the comprehensive evaluation result.

[0106] The comprehensive evaluation results include extremely poor environment, poor environment, normal environment, good environment and excellent environment;

[0107] If only the predicted comprehensive evaluation value exists at the current time point, the comprehensive evaluation result obtained based on the predicted comprehensive evaluation value is output as the comprehensive evaluation result of the environmental assessment area at the current time point;

[0108] If there is an actual comprehensive evaluation value at the current time point, the comprehensive evaluation result obtained according to the actual comprehensive evaluation value is output as the comprehensive evaluation result of the environmental assessment area at the current time point;

[0109] It should be noted that the purpose of this step is to quantify the environmental assessment area conditions into specific evaluation levels, intuitively present the environmental quality level, provide clear and specific environmental assessment conclusions for environmental management and decision-making, and help relevant departments formulate targeted environmental protection measures and plans;

[0110] The technical solution of the embodiment of the present invention is as follows: using a drone equipped with a high-resolution camera and GPS to comprehensively capture images of the environmental assessment area, extracting edge feature points through image processing technology, combining laser scanning point cloud data and the MVS algorithm to construct an environmental real scene model, setting a sensor layout in the environmental assessment area according to the environmental real scene model, setting an environmental assessment cycle, collecting environmental parameter values ​​at the end of the cycle, constructing a Bayesian network to integrate multi-source information, fusing sensor data to obtain parameter evaluation values, combining subjective weights and objective weights to obtain dynamic comprehensive weights, combining the parameter evaluation values ​​and the dynamic comprehensive weights to calculate actual comprehensive evaluation values, constructing time series data and constructing an ARIMA model based on the time series data, predicting environmental parameter evaluation values ​​for the next assessment cycle and calculating predicted comprehensive evaluation values, comparing the actual comprehensive evaluation values ​​to evaluate prediction accuracy, generating a prediction deviation signal if the accuracy is low, and upon receiving the prediction deviation signal, collecting environmental parameter data in real time, combining historical data to construct a complex time series and train a CNN-LSTM prediction model. After training is complete, using the model to make predictions, and evaluating the similarity between the predicted data and the real-time data through the DTW algorithm, evaluating the prediction accuracy of the CNN-LSTM prediction model, and comprehensively evaluating the environmental assessment area based on the obtained predicted comprehensive evaluation values ​​or the actual comprehensive evaluation values.

[0111] Example 2

[0112] like Figure 4As shown, the embodiment of the present invention provides a multi-dimensional environmental comprehensive evaluation system, which specifically includes the following modules:

[0113] Environmental modeling module: establishes an environmental scene model of the environmental assessment area, sets up multiple sensors based on the environmental scene model, and arranges the multiple sensors;

[0114] Use drones equipped with high-resolution cameras and high-precision GPS modules to capture images of the environmental assessment area;

[0115] Specifically, map information of the environmental assessment area is obtained, an optimal flight route is planned for the UAV based on the map information, and based on the optimal flight route, the UAV is controlled to fly around the environmental assessment area along the optimal flight route during an image capture period. The environmental assessment area is captured using a camera configured on the UAV at the capture time, and images captured during the image capture period are grouped into an environmental image group.

[0116] The image capture period refers to a period of time before environmental modeling is performed to capture images of the environmental assessment area. The capture time point is a time point within the image capture period. There are several capture time points within the image capture period. The intervals between adjacent capture time points are all the same and are preset by technical personnel to ensure that the captured environmental image group completely covers the environmental assessment area and has a high degree of overlap. During the capture process, the camera exposure parameters are adjusted in real time according to the lighting conditions to ensure the image quality of the environmental image group.

[0117] Perform edge detection on the images in the obtained environment image group using the Canny algorithm to obtain edge feature points of terrain and buildings in the images in the environment image group. Generate a 128-dimensional descriptor for each feature point. The descriptor contains the gradient information of the pixels around the edge feature point. Calculate the Euclidean distance between the edge feature point descriptors in different images. If the Euclidean distance between the edge feature point descriptors is less than a distance threshold, the edge feature point is classified into a feature matching group.

[0118] Based on the obtained feature matching group, the plane image coordinates of the edge feature points in the feature matching group in the corresponding image are obtained based on the principle of multi-view geometry. The three-dimensional coordinates of the edge feature points are calculated. The point cloud data of the environmental assessment area is obtained through laser scanning. The MVS algorithm is used for three-dimensional reconstruction to construct an environmental real scene model of the environmental assessment area.

[0119] Select sensors for the environmental assessment area based on the environmental reality model and determine the layout of the sensors in the environmental assessment area. The specific location of the sensors needs to consider the mutual influence between sensors and the complementarity of data to ensure that the layout of sensors can cover the entire environmental assessment area.

[0120] Parameter quantification module: Uses Bayesian network to fuse multi-location sensor data to obtain parameter evaluation values ​​of environmental parameters, combines subjective weights with objective weights to generate dynamic comprehensive weights, and calculates the actual comprehensive evaluation value through linear weighting;

[0121] An environmental assessment cycle is set. At the end of the environmental assessment cycle, the set sensors are used to collect several environmental parameter values ​​within the environmental assessment area. The end of the environmental assessment cycle is marked as the parameter collection time. The parameter collection time of the current environmental assessment cycle is also the starting point of the next environmental assessment cycle.

[0122] Based on any environmental parameter, a Bayesian network structure is constructed. According to the relationship between sensors at different locations used to collect environmental parameters, a Bayesian network structure is constructed. The nodes in the network represent the environmental parameter values ​​collected by sensors at different locations, and the edges represent the dependencies between the nodes.

[0123] Determine the prior probability distribution of each node and the conditional probability distribution between nodes, input the environmental parameter values ​​of each sensor collected at the current parameter collection time as evidence into the Bayesian network, and use the Bayesian inference algorithm to calculate the posterior probability distribution of the fused environmental parameters;

[0124] Extract the mean from the posterior probability distribution to obtain the parameter evaluation value of the environmental parameter;

[0125] The judgment matrix is ​​preset according to the relative importance of each environmental parameter in the environmental assessment area. The element a of the judgment matrix is ij Indicates the importance of the i-th environmental parameter relative to the j-th environmental parameter. Perform a consistency test on the judgment matrix. If it passes the test, calculate the maximum eigenvalue and corresponding eigenvector of the judgment matrix. Normalize the eigenvector to obtain the subjective weight of the environmental parameter.

[0126] Among them, i and j represent the type number of environmental parameters;

[0127] The parameter evaluation values ​​of each environmental parameter at the time of parameter collection are standardized by the Z-score standardization method. Based on the standardized data, the entropy value H of the environmental parameter is calculated. i ;

[0128] Calculate the objective weight of environmental parameters based on entropy value The formula is:

[0129]

[0130] Where n is the total number of environmental parameter types;

[0131] The subjective weight and objective weight are combined by linear combination method to obtain the dynamic comprehensive weight w of environmental parameters. i ;

[0132] According to the calculated dynamic comprehensive weight, combined with the parameter evaluation value of each environmental parameter, the actual comprehensive evaluation value E of the environmental assessment area within the assessment period is calculated using linear weighted summation. re ;

[0133] Model prediction module: Based on historical data, an ARIMA model is constructed to predict the environmental assessment area, and a predicted comprehensive evaluation value is obtained. The prediction accuracy of the ARIMA model is evaluated by comparing the comprehensive evaluation value with the actual value. If the error exceeds the limit, environmental parameter data is collected in real time and a CNN-LSTM prediction model is constructed. The DTW algorithm is used to determine the prediction accuracy of the CNN-LSTM prediction model, and the model update strategy is adaptively adjusted.

[0134] Collect historical parameter data, including the parameter evaluation values ​​of each environmental parameter at several past parameter collection moments and the timestamps of the parameter collection moments;

[0135] The collected historical parameter data are cleaned, outliers are removed by using the Laida criterion, and missing values ​​are filled by linear interpolation;

[0136] Based on any environmental parameter, the parameter evaluation values ​​of the environmental parameter at different collection time points are constructed according to the time series data X it , where i represents the type number of the environmental parameter, and t represents the time series number of the parameter collection moment corresponding to the parameter evaluation value in the historical parameter data;

[0137] According to the time series data X it Construct an ARIMA model, a time series forecasting model that combines autoregression (AR), difference (I), and moving average (MA):

[0138]

[0139] Among them, B is the lag operator, d is the difference order, is the autoregressive coefficient polynomial, θ(B) is the moving average coefficient polynomial, ε t is a white noise sequence;

[0140] Calculate the autocorrelation function (ACF) and partial autocorrelation function (PACF) of time series data, draw ACF and PACF graphs, and determine the order (p, d, q) of the ARIMA model, where p is the autoregressive order, d is the difference order, and q is the moving average order;

[0141] The least squares method is used to estimate the parameters of the ARIMA model and obtain the specific expression of the model;

[0142] Input the time series data into the ARIMA model with a determined order (p, d, q) and completed parameter estimation, predict the parameter evaluation values ​​of the environmental parameters in the next evaluation period, and output the predicted parameter evaluation values ​​of the environmental parameters;

[0143] Obtain the dynamic comprehensive weight calculated in the current assessment cycle, combine it with the evaluation value of the prediction parameters, and use linear weighted summation to calculate the predicted comprehensive evaluation value of the environmental assessment area in the next assessment cycle;

[0144] The predicted comprehensive evaluation value of the environmental assessment area obtained by the ARIMA model at the parameter collection time of the next assessment cycle is marked as E pr ;

[0145] At the next parameter collection moment, a linear weighted sum calculation is performed based on the dynamically adjusted dynamic comprehensive weight and the parameter evaluation value obtained by the collection calculation to obtain the actual comprehensive evaluation value E of the environmental assessment area at the parameter collection moment. re ;

[0146] According to the obtained prediction comprehensive evaluation value E pr Compared with the actual comprehensive evaluation value E re , through the formula

[0147]

[0148] Calculate the relative forecast error of the ARIMA model;

[0149] Taking into account factors such as environmental assessment accuracy requirements and historical prediction error distribution, a relative error threshold is preset and the calculated prediction relative error is compared with the relative error threshold;

[0150] If the prediction relative error is less than the relative error threshold, it means that the prediction accuracy of the ARIMA model is high, and the ARIMA model will continue to be used to predict the comprehensive evaluation value in the subsequent evaluation cycle;

[0151] If the forecast relative error is greater than or equal to the relative error threshold, it means that the forecast accuracy of the ARIMA model is low and a forecast deviation signal is generated;

[0152] If the generated prediction deviation signal is received, several collection time points are selected within the evaluation period, and each environmental parameter is collected in real time at the collection time points, wherein the intervals between adjacent collection time points are the same;

[0153] Calculate the parameter evaluation value of each environmental parameter in real time based on the collected data during the current evaluation period, obtain the collected historical parameter data, and combine the parameter evaluation values ​​within the evaluation period to form a complex time series data, where k represents the time sequence number of the parameter collection moment or collection time point corresponding to the parameter evaluation value in the complex time series data;

[0154] A CNN-LSTM prediction model is constructed using CNN as the time series feature extraction module and LSTM as the time series dependency modeling module. The CNN-LSTM prediction model is trained based on complex time series data, with the convolutional layer parameters fixed to retain common time series features.

[0155] When the parameter evaluation value at the current acquisition time point arrives, a sliding window mechanism is used to extract local time series segments, and online gradient descent is performed using the mean square error (MSE) as the loss function. The formula of the loss function L is as follows:

[0156]

[0157] Among them, T w It represents the window length of the sliding window, λ is the L2 regularization coefficient, which is used to prevent overfitting, and η is the parameter of the model. It represents the evaluation value of the model prediction parameter, y k It represents the actual parameter evaluation value obtained by collection and calculation;

[0158] Monitor the activation sparsity of each layer of the model, calculate the sparsity of each layer and compare it with the sparsity threshold. If the sparsity of any layer is continuously lower than the sparsity threshold for a period longer than the duration, remove redundant convolution kernels or LSTM units and perform complexity-aware pruning to reduce model complexity.

[0159] After training, a trained CNN-LSTM prediction model is obtained;

[0160] The trained CNN-LSTM prediction model is used to predict the parameter evaluation values ​​of each environmental parameter in the subsequent evaluation cycle. Based on the parameter evaluation values ​​and dynamic comprehensive weights, a linear weighted summation is used to calculate the predicted comprehensive evaluation value of the environmental assessment area.

[0161] The actual comprehensive evaluation value obtained through real-time collection and calculation is compared with the predicted comprehensive evaluation value, and the similarity of the change trends of the two is calculated. The dynamic time warping (DTW) algorithm is used to measure the similarity and construct two time series.

[0162] A={a1,a2,......,a m}

[0163] B={b1,b2,......,bm}

[0164] Where A represents the time series of the predicted comprehensive evaluation value, B represents the time series of the actual comprehensive evaluation value, and m represents the number of data in the time series.

[0165] D rs =d(a r , b s )

[0166] Among them, d is the distance measurement method such as Euclidean distance, and the cumulative distance matrix C is calculated by dynamic programming algorithm to calculate the final DTW distance C mm

[0167]

[0168] Compare the calculated DTW distance with the distance threshold;

[0169] If the DTW distance is less than the distance threshold, it indicates that the change trends of the predicted data of the CNN-LSTM prediction model and the real-time data are highly similar. The model is used for subsequent predictions, and real-time collection within the evaluation period is stopped. Environmental parameters are collected only at the parameter collection time of the evaluation period, and the DTW distance between the predicted comprehensive evaluation value and the actual comprehensive evaluation value at the parameter collection time is calculated.

[0170] If the DTW distance is greater than or equal to the distance threshold, it indicates that the similarity between the change trends of the CNN-LSTM prediction model's predicted data and the real-time data is low. Continue to collect environmental parameters in real time and train the CNN-LSTM prediction model within the evaluation period until the DTW distance is less than the distance threshold.

[0171] Comprehensive evaluation module: conducts comprehensive evaluation of the environmental assessment area based on the predicted comprehensive evaluation value and the actual comprehensive evaluation value;

[0172] Obtain the predicted comprehensive evaluation value obtained by the ARIMA model or CNN-LSTM prediction model and the actual comprehensive evaluation value obtained through collection and calculation;

[0173] Establish an evaluation model and input the obtained predicted comprehensive evaluation value or actual comprehensive evaluation value into the evaluation model. The evaluation model comprehensively evaluates the environmental assessment area by determining the range of deviation between the predicted comprehensive evaluation value or actual comprehensive evaluation value and the preset standard comprehensive evaluation value, and outputs the comprehensive evaluation result.

[0174] The comprehensive evaluation results include extremely poor environment, poor environment, normal environment, good environment and excellent environment;

[0175] If only the predicted comprehensive evaluation value exists at the current time point, the comprehensive evaluation result obtained based on the predicted comprehensive evaluation value is output as the comprehensive evaluation result of the environmental assessment area at the current time point;

[0176] If there is an actual comprehensive evaluation value at the current time point, the comprehensive evaluation result obtained according to the actual comprehensive evaluation value is output as the comprehensive evaluation result of the environmental assessment area at the current time point.

[0177] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A multi-dimensional environmental comprehensive evaluation method, characterized in that: The following steps are involved: The parameter evaluation values ​​of environmental parameters are obtained by fusing multi-location sensor data using Bayesian networks, and dynamic comprehensive weights are generated by combining subjective weights with objective weights. The actual comprehensive evaluation value is calculated through linear weighting. Based on historical environmental parameter data, an ARIMA model is constructed to predict the environmental assessment area and obtain a predicted comprehensive evaluation value. The prediction accuracy of the ARIMA model is evaluated by comparing the comprehensive evaluation value with the actual value. If the error exceeds the limit, the environmental parameter data is collected in real time and a CNN-LSTM prediction model is constructed. The DTW algorithm is used to determine the prediction accuracy of the CNN-LSTM prediction model and the model update strategy is adaptively adjusted. Conduct a comprehensive evaluation of the environmental assessment area based on the predicted comprehensive evaluation value and the actual comprehensive evaluation value.

2. A multi-dimensional environmental comprehensive evaluation method according to claim 1, characterized in that: The actual comprehensive evaluation value is obtained as follows: A judgment matrix is ​​preset based on the relative importance of each environmental parameter. The judgment matrix is ​​processed to obtain the subjective weight of the environmental parameters. The parameter evaluation values ​​of each environmental parameter at the time of parameter collection are standardized and entropy value calculated using the Z-score standardization method to obtain the objective weight of the environmental parameters. The subjective weight and objective weight are fused to obtain the dynamic comprehensive weight of environmental parameters; The parameter evaluation value of each environmental parameter is combined with the dynamic comprehensive weight to perform weighted calculation to obtain the actual comprehensive evaluation value of the environmental assessment area during the assessment period.

3. A multi-dimensional environmental comprehensive evaluation method according to claim 2, characterized in that: The parameter evaluation value is obtained as follows: Based on the relationship between sensors at different locations used to collect environmental parameters, a Bayesian network structure is constructed to determine the prior probability distribution of each node and the conditional probability distribution between nodes. The environmental parameter values ​​of each sensor are input into the Bayesian network as evidence, and the Bayesian inference algorithm is used to calculate the posterior probability distribution of the fused environmental parameters. The mean is extracted from the posterior probability distribution to obtain the parameter evaluation value of the environmental parameter.

4. A multi-dimensional environmental comprehensive evaluation method according to claim 1, characterized in that: The method for evaluating the prediction accuracy of the ARIMA model is: Obtain the forecast relative error of the ARIMA model. If the forecast relative error is greater than or equal to the relative error threshold, the forecast accuracy of the ARIMA model is judged to be low, and a forecast deviation signal is generated.

5. A multi-dimensional environmental comprehensive evaluation method according to claim 4, characterized in that: The method for obtaining the prediction relative error is: Obtain the dynamic comprehensive weight calculated in the current evaluation cycle, combine it with the predicted parameter evaluation value at the parameter collection time of the next evaluation cycle predicted by the ARIMA model, and perform data processing to obtain the predicted comprehensive evaluation value; The actual comprehensive evaluation value at the parameter collection moment of the next evaluation cycle is obtained, and the data is processed in combination with the obtained predicted comprehensive evaluation value to obtain the prediction relative error of the ARIMA model.

6. A multi-dimensional environmental comprehensive evaluation method according to claim 5, characterized in that: The method for obtaining the prediction parameter evaluation value is as follows: The parameter evaluation values ​​of environmental parameters at different collection time points are constructed into time series data according to the time series, and the ARIMA model is constructed according to the time series data; According to the ARIMA model, the parameter evaluation values ​​of the environmental parameters in the next evaluation period are predicted, and the predicted parameter evaluation values ​​of the environmental parameters are output.

7. A multi-dimensional environmental comprehensive evaluation method according to claim 1, characterized in that: The CNN-LSTM prediction model is constructed as follows: If a prediction deviation signal is received, each environmental parameter is collected in real time during the evaluation period, and historical parameter data are combined to obtain complex time series data; A CNN-LSTM prediction model is constructed using CNN as the time series feature extraction module and LSTM as the time series dependency modeling module. The CNN-LSTM prediction model is trained based on complex time series data to obtain a trained CNN-LSTM prediction model.

8. A multi-dimensional environmental comprehensive evaluation method according to claim 7, characterized in that: The method for judging the prediction accuracy of the CNN-LSTM prediction model is: The trained CNN-LSTM prediction model is used to predict the parameter evaluation values ​​of each environmental parameter in the subsequent evaluation cycle to obtain the predicted comprehensive evaluation value; Construct the time series of predicted comprehensive evaluation values ​​and the time series of actual comprehensive evaluation values, use the DTW algorithm to calculate the DTW distance between the two time series and compare it with the distance threshold; If the DTW distance is less than the distance threshold, the prediction accuracy of the CNN-LSTM prediction model is judged to be high; otherwise, the prediction accuracy of the CNN-LSTM prediction model is judged to be low.

9. A multi-dimensional environmental comprehensive evaluation method according to claim 1, characterized in that: The method for conducting comprehensive evaluation of the environmental assessment area is as follows: An evaluation model is established, and the obtained predicted comprehensive evaluation value or actual comprehensive evaluation value is input into the evaluation model. The evaluation model conducts a comprehensive evaluation of the environmental assessment area by judging the range of deviation between the input value and the preset standard comprehensive evaluation value, and outputs the comprehensive evaluation result.

10. A multi-dimensional environmental comprehensive evaluation system, the system being used to implement the evaluation method according to any one of claims 1 to 9, characterized in that: include: Environmental modeling module: establishes an environmental scene model of the environmental assessment area, sets up multiple sensors based on the environmental scene model, and arranges the multiple sensors; Parameter quantification module: Uses Bayesian network to fuse multi-location sensor data to obtain parameter evaluation values ​​of environmental parameters, combines subjective weights with objective weights to generate dynamic comprehensive weights, and calculates the actual comprehensive evaluation value through linear weighting; Model prediction module: Based on historical data, the ARIMA model is constructed to predict the environmental assessment area, and the predicted comprehensive evaluation value is obtained. The prediction accuracy of the ARIMA model is evaluated by comparing the comprehensive evaluation value with the actual value. If the error exceeds the limit, the environmental parameter data is collected in real time and a CNN-LSTM prediction model is constructed. The DTW algorithm is used to judge the prediction accuracy of the CNN-LSTM prediction model, and the model update strategy is adaptively adjusted. Comprehensive evaluation module: Comprehensively evaluate the environmental assessment area based on the predicted comprehensive evaluation value and the actual comprehensive evaluation value.

Images