Optimization method for layout of radar water level gauge on road

Abstract

The invention discloses an optimization method for the layout of a radar water level gauge on a road, and the method comprises the following steps: data preprocessing: selecting three dimensions of an input layer; determining the size of an output layer, wherein the output layer is composed of nine neurons in three rows and three columns; calculating an optimal neuron by using an SOM algorithm, updating the weight, and then repeatedly training, so as to determine the position layout of the water level gauge; the DBSCAN algorithm is used for optimizing the radar water level gauge layout again, and a final radar water level gauge layout scheme is obtained. According to the method, while the layout number of the radar water level gauges on the road is optimized, the defect that detection of partial areas in an existing method is blank is overcome, the topological structure between data can be kept, sample data points in other neurons can be fed back at a higher speed in the face of sudden weather, and therefore the method is faster and more accurate. The ponding coping scheme is accurately made, and the cost is further reduced while the accuracy of the scheme is ensured.

Description

Technical Field

[0001] The present invention relates to the field of urban hydrological monitoring, and particularly relates to an optimization method for the layout of radar water level gauges on roads. Background Art

[0002] Radar water level gauges have the advantages of non-contact measurement, high precision, high reliability, etc., and can accurately reflect the water accumulation situation in real time. However, it is difficult, costly and difficult to maintain in the later stage to install radar water level gauges on roads. Therefore, it is particularly important to optimize the layout of radar water level gauges. The existing layout methods of radar water level gauges on roads are roughly divided into two categories. One is the uniform layout method. For example, on a long straight urban road, a radar water level gauge is set every certain distance (such as 500 meters). Through the equidistant layout, it can ensure a relatively comprehensive monitoring of the road water accumulation situation to a certain extent. The other category is to plan the layout based on urban hydrological data. For example, radar water level gauges are set in places with low terrain, small density of drainage pipe outlets and large daily traffic flow. The present invention belongs to the second layout method in principle.

[0003] However, in the prior art, various influencing urban hydrological data are mostly analyzed subjectively by people, and then a layout plan for radar water level gauges is given. Inevitably, more refined areas will be missed, or comprehensive factors such as weather, traffic conditions, and dynamic terrain are not combined, resulting in a reduction in the rationality of the layout of radar water level gauges, causing waste of resources or detection blanks in some areas. Therefore, there is an urgent need for a new optimization method for the layout of radar water level gauges on roads to improve such problems. Summary of the Invention

[0004] Object of the Invention: To solve the problems mentioned in the background art, the present invention proposes an optimization method for the layout of radar water level gauges on roads, constructs a layout model of radar water level gauges on roads by using the SOM algorithm, combines the DBSCAN algorithm to perform secondary selection optimization on the layout of radar water level gauges on roads, and the two are combined to form an optimal layout plan.

[0005] Technical Solution:

[0006] The present invention discloses an optimization method for the layout of radar water level gauges on roads, and the method includes the following steps:

[0007] S1 Clean the obtained data related to the road network, remove noise points, and perform normalization processing on it to determine the dimension of the input layer of the model;

[0008] S2 Use the SOM algorithm to construct a layout model of radar water level gauges on roads;

[0009] S2.1 Determine the size of the output layer of the radar water level gauge layout model on the road, which consists of 9 neurons arranged in a three-row and three-column linear structure;

[0010] S2.2 Calculate the winning neuron according to the distance formula between the input vector and the output layer neurons and combined with the initial weights, and use PCA initialization to determine the initial weight values of the neurons;

[0011] S2.3 Update the weights and train repeatedly, and iterate and optimize the weights of the output layer neurons through the winning neuron and the weight update formula;

[0012] S3 Output the layout of the water level gauges, and use the DBSCAN algorithm to optimize the sub-selected sample data points to obtain the final radar water level gauge layout plan.

[0013] Furthermore, the input layer dimensions described in S1 include the daily traffic volume of the road, the density of drainage pipe outlets, and the slope; the normalization formula is:

[0014]

[0015] where date is the sample data value to be normalized; date min is the minimum value in the sample data values; date max is the maximum value in the sample data values; date new is the normalized data value.

[0016] Furthermore, the specific calculation of the winning neuron described in S2.2 is as follows:

[0017] Calculate the best unit (winning neuron), calculate the first best unit according to the distance formula between the input vector and the output layer neurons and combined with the initial weights, and use PCA initialization to determine the initial weight values of the 9 neurons;

[0018] Standardize the data so that the mean of each feature is 0 and the variance is 1. The standardization formula is:

[0019]

[0020] where Y is the matrix, θ is the mean of each feature, and σ is the standard deviation of each feature;

[0021] Perform PCA on the standardized data Y standardized The covariance formula is:

[0022]

[0023] C = VΛV T

[0024] Among them, C is the covariance matrix, V is the eigenvector matrix, each column is a principal component direction, Λ is a diagonal matrix containing eigenvalues, and Y T standardized is the transpose of the standardized data matrix, and V T is the transpose of the eigenvector matrix V.

[0025] Calculate the initial weights of 9 neurons. The distance formula between the input vector and the neurons in the output layer, combined with the initial weights, is used to calculate the winning neuron. The initial weight formula is:

[0026] ω ij = θ + i * V 1 + j * V 2

[0027] Among them, ω ij is the weight value of the neuron in the i-th row and j-th column. θ is a constant term, usually used to adjust the initial value of the weight. i is the row number of the target neuron, j is the column number of the target neuron, and V 1 and V 2 are the first two principal component directions of the eigenvector matrix V, which are coefficients used to adjust the weight values.

[0028] Furthermore, the distance formula between the input vector and the neurons in the output layer, combined with the initial weights, is used to calculate the winning neuron, as follows:

[0029] The input vector X = (X 1 , X 2 , X 3 ), the position of the neuron in the output layer is (i, j), and the initial weight vector ω ij = (ω ij1 , ω ij2 , ω ij3 ). The distance formula from the input vector X to the neuron (i, j) in the output layer is:

[0030]

[0031] Among them, X = (X 1 , X 2 , X 3 ) is the eigenvector of the input data. X 1 is the value of the normalized daily traffic flow of the road. X 2 is the value of the normalized road slope. X 3 is the value of the normalized density of the road drainage pipe outlets. ω ij = (ω ij1 , ω ij2 , ω ij3 ) is the weight vector associated with the neuron in the output layer, and ω ij1 , ωij2 , ω ij3 are its three weight values, and d(X, ω ij ) is the Euclidean distance between the input vector and the weight vector. For the input vector X, the neuron in the output layer with the closest distance to it is the best neuron, that is, the winning neuron.

[0032] Furthermore, the weight update formula described in S2.3 is as follows:

[0033] ω ij (t + 1) = ω ij (t) + η(t) * h(t) * (X(t) - ω ij (t))

[0034] where t is the number of iterations, η(t) is the learning rate, and h(t) is the neighborhood function;

[0035]

[0036] where Τ 总 is the total number of cycles, d mi is the distance between neuron i and the best neuron, and σ(t) is the neighborhood radius that changes with time;

[0037] Repeat the operation of step S2.2 and continuously update the weights until all sample data is calculated, that is, through S2.2 and S2.3, the sample data is classified into different neurons according to their characteristics.

[0038] Furthermore, the specific operation of using the DBSCAN algorithm to optimize the layout of radar water level gauges described in S3 is as follows:

[0039] Convert the road network data into point data that is easy to process by the DBSCAN algorithm. Select the secondary sample data points from the radar water level gauge layout locations obtained from the radar water level gauge layout model on the road. By repeatedly defining the size of the minimum number of points MinPts and the parameter radius Eps and performing clustering analysis, obtain the minimum number of points and the parameter radius that are most suitable for the sample data. The center point of each cluster is the best location for the layout of the radar water level gauge, and obtain the final layout plan of the radar water level gauge.

[0040] Beneficial effects:

[0041] 1. The present invention uses the SOM algorithm to construct a radar water level gauge layout model on the road. Compared with the conventional K-means clustering method, the SOM algorithm can maintain the topological structure between data. In the face of sudden weather conditions, it can feedback the sample data points in other neurons at a faster speed, so as to make a water accumulation response plan more quickly and accurately.

[0042] 2. The present invention combines the DBSACN algorithm with the SOM algorithm to construct a layout scheme, and performs secondary optimization of sub-optimal points while obtaining the layout points of radar water level gauges in the road radar water level gauge layout model, so as to fully consider the water accumulation conditions of remote roads, such as suburbs and rural areas, making the layout scheme more comprehensive and further reducing the possibility of water accumulation hazards.

[0043] 3. The present invention constructs a radar water level gauge layout scheme through an algorithm model. While considering more comprehensive parameters, lower cost and higher precision of the layout scheme, it can add, subtract and update the model in real time according to needs, optimize the algorithm effect in real time, realize long-term update and iteration of the model, output the best possible prediction effect, and further maintain the quality of the maintenance scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is the logic flow chart of the present invention;

[0045] Figure 2 is the clustering result diagram of the output layer of the present invention;

[0046] Figure 3 is the comparison diagram of the SOM algorithm model and the K-means clustering results of the present invention;

[0047] Figure 4 is the comparison diagram of the original layout and the optimized results of the SOM+DBSCAN layout. DETAILED DESCRIPTION OF THE INVENTION

[0048] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] As Figure 1 - Figure 2 shown, the present invention proposes an optimization method for the layout of radar water level gauges on roads, and the relevant concepts in the method are defined as follows:

[0050] Definition 1: Determination of the input layer dimension; based on the data such as the slope of the target road, the density of drainage pipe outlets, the daily traffic volume, the degree of regional importance, and the past road water accumulation detection obtained. Considering that there is a high correlation between the slope and past water accumulation data, and between the daily traffic volume and the degree of regional importance, in order to improve the calculation efficiency and combine the correlation and data comprehensiveness among the five types of data, finally, the slope of the road, the density of drainage pipe outlets, and the daily traffic volume are selected as the three dimensions of the input layer.

[0051] Definition 2: Self-Organizing Map Neural Network Model (SOM) maps high-dimensional input data to a low-dimensional (two-dimensional) space while preserving the topological structure of the data. By combining the sample data we obtained and through experiments, it was found that when the size of the output layer is four rows and four columns, there will be no samples in two categories. To save resources and reduce waste, it was decided that the size of the output layer is three rows and three columns, consisting of 9 neurons and arranged in a linear structure. In this way, the arrangement order of the neurons can correspond to the road position order, facilitating understanding and subsequent layout planning.

[0052] Definition 3: Related definitions for the input vector and the neurons in the output layer; the directed input vector X = (X 1 , X 2 , X 3 ), the position of the neuron in the output layer is (i, j), and the initial weight vector ω ij = (ω ij1 , ω ij2 , ω ij3 ). The distance formula from the input vector X to the neuron (i, j) in the output layer is:

[0053]

[0054] For the input vector X, the neuron in the output layer with the closest distance to it is the best neuron (winning neuron).

[0055] Among them, X = (X 1 , X 2 , X 3 ) is the feature vector of the input data, X 1 is the value after normalizing the daily traffic flow of the road, X 2 is the value after normalizing the road slope, X 3 is the value after normalizing the density of the road drainage pipe outlets, ω ij = (ω ij1 , ω ij2 , ω ij3 ) is the weight vector associated with the neuron in the output layer, ω ij1 , ω ij2 , ω ij3 are its three weight values, and d(X, ω ij ) is the Euclidean distance between the input vector and the weight vector.

[0056] Definition 4: Weight optimization formula;

[0057] ω ij (t + 1) = ω ij (t) + η(t) * h(t) * (X(t) - ω ij (t))

[0058] where t is the number of iterations, η(t) is the learning rate, and h(t) is the neighborhood function.

[0059]

[0060] where Τ 总 is the total number of cycles, d mi is the distance between neuron i and the best neuron, and σ(t) is the neighborhood radius that changes with time.

[0061] Definition Five: Definitions related to the DBSCAN algorithm; first, convert the road network data into point data that is easy to process by the DBSCAN algorithm. By repeatedly defining the size of the minimum number of points (MinPts) and the parameter radius (Eps) and performing clustering analysis, the minimum number of points and the parameter radius that are most suitable for the sample data are obtained. The center point of each cluster is the best location for arranging the radar water level gauges.

[0062] The steps of the method of the present invention are as follows:

[0063] The first step: Data preprocessing. Clean the obtained road network-related data, remove noise points, and perform normalization processing on it. Then, select three types of data, namely the daily traffic volume of the road, the density of drainage pipe outlets, and the slope, as the three dimensions of the input layer;

[0064] The normalization formula is:

[0065]

[0066] where date is the sample data value to be normalized; date min is the minimum value in the sample data value; date max is the maximum value in the sample data value; date new is the normalized data value.

[0067] Import the sample data after preprocessing and normalization into the MATLAB software in the.mat file format to complete the data processing and import work;

[0068] The second step: Use the SOM algorithm to construct a layout model of radar water level gauges on the road, determine the size of the output layer in the SOM algorithm, and based on the number of sample data obtained in the previous step and through repeated testing by the software, it is found that when the size of the output layer is four rows and four columns, there will be a situation where there are no samples in two categories. In order to save resources and reduce waste, it is therefore determined that the output layer consists of nine neurons arranged in a three-row and three-column linear structure;

[0069] Step 3: Calculate the best unit (winning neuron). Calculate the first best unit according to the distance formula between the input vector and the neurons in the output layer and combined with the initial weights, that is, classify the first sample data into a category through the distance formula and the weight formula. Since the results of randomly initializing the weights are unstable, in order to improve the operation accuracy and facilitate the rapid convergence of the sample data, here we use PCA (Principal Component Analysis) to initialize and determine the initial weight values of 9 neurons.

[0070] First, standardize the data so that the mean of each feature is 0 and the variance is 1. The standardization formula is:

[0071]

[0072] where Y is a 64-row and 3-column matrix, θ is the mean of each feature, and σ is the standard deviation of each feature. According to the sample data processed in the first step, we have: θ = [0.266 0.424 0.448], σ = [0.311 0.221 0.239],

[0073] Then, perform PCA on the standardized data Y standardized The covariance formula is:

[0074]

[0075] CVΛV T

[0076] where C is the covariance matrix, n = 64, V is the eigenvector matrix, Y T standardized is the transpose of the standardized data matrix, V T is the transpose of the eigenvector matrix V, each column is a principal component direction, and Λ is a diagonal matrix containing the eigenvalues. Calculate the eigenvalues and eigenvectors: λ = [1.512 0.876 0.612], V = The first two principal component directions are:

[0077] Then, calculate the initial weights of 9 neurons. The initial weight formula is:

[0078] ωij = θ + i * V 1 + j * V 2

[0079] where ω ij is the weight value of the neuron in the i-th row and j-th column, θ is a constant term, usually used to adjust the initial value of the weight, i is the row number of the target neuron, j is the column number of the target neuron, V 1 and V 2are the first two principal component directions of the eigenvector matrix V and are coefficients used to adjust the weight values.

[0080] Calculated: ω 0 0 = [0.266 0.424 0.448], ω 0 0.5 = [0.266 0.832 0.24], ω 0 1 = [0.266 1.24 0.032], ω 0.5 0 = [0.62 0.424 0.096], ω 0.5 0.5 = [0.62 0.832 -0.112], ω 0.5 1 = [0.62 1.24 -0.32], ω 1 0 = [0.973 0.424 -0.256], ω 1 0.5 = [0.973 0.832 -0.464], ω 1 1 = [0.973 1.24 -0.672].

[0081] In this embodiment, by defining three, the winning neuron of the first sample ([0.0667, 0.4, 0.5455]) is ω 0 0 .

[0082] Fourth step: Update the weights and train repeatedly. Optimize the weights of the output layer neurons through the best unit and the weight update formula (Definition Four) to obtain the updated neuron weights as: ω 0 0 (t + 1) = [0.266, 0.424, 0.448], ω 0 0.5 (t + 1) = [0.266, 0.832, 0.24], ω 0 1 (t + 1) = [0.266, 1.24, 0.032], ω 0.5 0 (t + 1) = [0.62, 0.424, 0.096], ω 0.5 0.5 (t + 1) = [0.62, 0.832, -0.112], ω 0.5 1 (t + 1) = [0.62, 1.24, -0.32], ω 1 0 (t + 1) = [0.973, 0.424, -0.256], ω 1 0.5 (t + 1) = [0.973, 0.832, -0.464], ω 1 1 (t + 1) = [0.973, 1.24, -0.672].

[0083] Next, repeat the operation of the third step and continuously update the weights until all sample data is calculated, that is, divide the sample data into different neurons according to their characteristics through the third and fourth steps.

[0084] Step 5: Determine the layout of the radar water level gauges. To ensure the feasibility of the SOM algorithm, the sample data is also subjected to cluster analysis using the K-means algorithm, with K = 9 to ensure that the number of categories in the two algorithms is the same. By comparing the three-dimensional spatial diagrams of the two algorithms, it can be seen that their clustering effects are almost the same, but the SOM algorithm can maintain the topological structure between data, which is one of the important reasons why the SOM algorithm is superior to other algorithms such as K-means. The comparison results of the two are as Figure 3 shown.

[0085] The topological structure means that adjacent two neurons have high similarity. Taking the first neuron as an example, the characteristics of the sample data it contains are large daily traffic flow, small slope, and small density of drainage pipe outlets. Then the sample data of the adjacent second neuron has the characteristics of large daily traffic flow, small slope, and large density of drainage pipe outlets; similarly, the similarity between the first neuron and the ninth neuron, which is the farthest from it, is the lowest. This characteristic of the SOM algorithm can bring a lot of convenience to the layout of radar water level gauges on roads. For example, in heavy rain weather in summer, due to the huge rainfall, even roads with a large density of drainage pipe outlets will experience waterlogging. At this time, radar water level gauges can be installed at the positions of the sample data points in the second neuron, so as to detect the waterlogging situation of relevant sections in a timely manner and reduce the probability of sudden situations such as road waterlogging. This is an advantage that the K-means algorithm does not have.

[0086] There is a high degree of relevance between the SOM algorithm and the layout optimization of radar water level gauges on roads, which is its second advantage. The layout of infrastructure such as street lights and trash cans on roads focuses on the uniformity of distribution; while the layout of radar water level gauges on roads needs to combine multi-dimensional data such as rainfall, traffic volume, slope, and drainage pipe density. SOM can effectively process these multi-dimensional data and find the internal structure of the data through dimensionality reduction and clustering. Therefore, compared with the layout of infrastructure such as street lights, there is a higher correlation between the SOM algorithm and the layout of radar water level gauges.

[0087] According to the analysis results of the neurons and considering the actual situation, the data sample points in the neuron with large daily traffic flow, small slope, and small density of drainage pipe outlets will be used as the installation locations of the radar water level gauges.

[0088] Step 6: Use the DBSCAN algorithm to optimize the layout of the radar water level gauges again. First, quantify the road network data into point data for subsequent analysis. Relying on the installation locations of the radar water level gauges obtained by the SOM algorithm, optimize and analyze the remaining sample data points, define their minimum number of points (MinPts) and parameter radius (Eps) respectively, and perform cluster analysis to obtain the final layout plan of the radar water level gauges.

[0089] Since the number of sampling sample points in this embodiment is 64 and all sample points are roughly distributed in a two-dimensional space of 30*40 size, the sample size is small and the distribution is small. Therefore, 5 is taken as the reference number. First, ensure that Eps is 5 and continuously adjust the value of MinPts. Experiments show that when MinPts is 6, the points in the suburbs around the road network are all identified as noise points. Therefore, only the case where MinPts is less than or equal to 5 needs to be considered. Then, by controlling MinPts to be 1, 2, 3, 4, or 5, continuously adjust the value of Eps. When MinPts is 2, 3, or 4, if Eps = 6, the number of clustering categories is too small, only 1 or 2 categories; if Eps = 4, there are too many noise points. When MinPts is 1, if Eps = 6, the number of clustering categories is too small, only 1 or 2 categories; if Eps = 4, the number of clustering categories is too large, up to 33 categories. Considering all factors, when MinPts is 5, the best clustering optimization effect can be achieved. By comparing when Eps is from 1 to 5, it is found that when it is 1, the number of noise points in the entire network is the least and the number of categories is moderate, which is the best solution.

[0090] As Figure 4 shown, compared with the uniform distribution, the clustering centers obtained by the present invention and the number of radar water level gauge layout points solved by the SOM algorithm have an optimization effect. The purpose of the secondary optimization of the DBSCAN algorithm is to fully consider the water accumulation situation on remote roads to solve the problems existing in the current background.

[0091] The above description of the embodiments enables those skilled in the art to implement or use the present invention. Various modifications to the embodiments will be obvious to those skilled in the art. The general principles of the present invention can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention should not be limited to the embodiments shown herein, but should cover the widest range consistent with the principles and novel features disclosed in the present invention.

Claims

1. A method for optimizing the layout of radar water level gauges on roads, characterized in that: The method comprises the following steps: S1 cleans the acquired road network related data, removes noise points, and normalizes it to determine the model input layer dimension; S2 uses the SOM algorithm to build a layout model of radar water level gauges on the road; S2.1 determines the output layer size of the road radar water level gauge layout model, which consists of three rows and three columns with a total of 9 neurons and is arranged in a linear structure; S2.2 calculates the winning neuron based on the distance formula between the input vector and the output layer neuron and the initial weight, and uses PCA initialization to determine the initial weight value of the neuron; S2.3 Update the weights and train repeatedly, iterating and optimizing the weights of the output layer neurons through the winning neurons and the weight update formula; S3 outputs the water level gauge location layout, uses the DBSCAN algorithm to optimize the secondary sample data points, and obtains the final radar water level gauge layout plan.

2. The optimization method for the layout of radar water level gauges on roads according to claim 1, characterized in that: The input layer dimensions of S1 include the daily traffic volume of the road, the density of drainage pipe outlets, and the slope; the normalization formula is: Among them, date is the sample data value to be normalized; date min Is the minimum date in the sample data value max is the maximum value among the sample data values; date new is the normalized data value.

3. The optimization method for the layout of radar water level gauges on roads according to claim 1, characterized in that: The calculation of the winning neuron described in S2.2 is as follows: Calculate the best unit (winning neuron), calculate the first best unit based on the distance formula between the input vector and the output layer neuron and the initial weight, and use PCA initialization to determine the initial weight values ​​of the 9 neurons; The data is standardized so that the mean of each feature is 0 and the variance is 1. The standardization formula is: Where Y is a matrix, θ is the mean of each feature, and σ is the standard deviation of each feature; For the standardized data Y standardized Perform PCA, and the covariance formula is: C=VΛV T Where C is the covariance matrix, V is the eigenvector matrix, each column is a principal component direction, Λ is a diagonal matrix containing eigenvalues, Y T standardized is the transpose of the normalized data matrix, V T is the transpose of the eigenvector matrix V; Calculate the initial weights of the 9 neurons, the distance formula between the input vector and the output layer neuron and combine it with the initial weight to calculate the winning neuron. The initial weight formula is: oh ij =θ+i*V1+j*V2 Among them, ω ij is the weight value of the neuron in the i-th row and j-th column, θ is a constant term, which is usually used to adjust the initial value of the weight, i is the number of rows of the target neuron, j is the number of columns of the target neuron, V1 and V2 are the first two principal component directions of the eigenvector matrix V, which are coefficients used to adjust the weight value.

4. The optimization method for the layout of radar water level gauges on roads according to claim 3, characterized in that: The distance formula between the input vector and the output layer neuron is combined with the initial weight to calculate the winning neuron, as follows: Input vector X = (X1, X2, X3), the output layer neuron position is (i, j), the output layer initial weight vector ω ij =(ω ij1 ,ω ij2 ,ω ij3 ), the distance formula from the input vector X to the output layer neuron (i, j) is: Among them, X = (X1, X2, X3) is the characteristic vector of the input data, X1 is the normalized value of the road daily traffic flow, X2 is the normalized value of the road slope, X3 is the normalized value of the road drainage pipe outlet density, ω ij =(ω ij1 ,ω ij2 ,ω ij3 ) is the weight vector associated with the output layer neurons, ω ij1 ,ω ij2 ,ω ij3 are its three weight values, d(X,ω ij ) is the Euclidean distance between the input vector and the weight vector. For the input vector X, the neuron in the output layer that is closest to it is the best neuron, that is, the winning neuron.

5. The optimization method for the layout of radar water level gauges on roads according to claim 1, characterized in that: The weight update formula described in S2.3 is as follows: oh ij (t+1)=ω ij (t)+η(t)*h(t)*(X(t)-ω ij (t)) Where t is the number of iterations, η(t) is the learning rate, and h(t) is the neighborhood function; where Τ 总 is the total number of cycles, d mi is the distance between neuron i and the best neuron, σ(t) is the neighborhood radius that changes over time; Repeat step S2.2 and continuously update the weights until all sample data are calculated, that is, the sample data are divided into different neurons according to their characteristics through S2.2 and S2.

3.

6. The optimization method for the layout of radar water level gauges on roads according to claim 1, characterized in that: S3 describes the use of the DBSCAN algorithm to optimize the radar water level meter layout again as follows: The road network data is converted into point data that is easy to process by the DBSCAN algorithm. The radar water level meter layout model on the road obtains the radar water level meter layout locations and selects the second-selected sample data points. By repeatedly defining the minimum number of points MinPts and the size of the parameter radius Eps and performing cluster analysis, the minimum number of points and parameter radius that best suit the sample data are obtained. Each cluster center point is the optimal location for the radar water level meter layout, and the final radar water level meter layout plan is obtained.

Images