A method for optimizing the location of rain gauges based on rainfall analysis

Abstract

This invention relates to the field of rainfall measurement technology, and particularly to a method for optimizing rain gauge location based on rainfall analysis. The invention divides all rain gauges into several rainfall sub-regions based on whether rainfall occurred at all stations during the same time period, and whether it was light or heavy rain. Rain gauges within each sub-region share strong rainfall commonalities. For rain gauges with the same rainfall commonalities, a neural network model is used to fit the rules between coordinate changes and rainfall changes. This neural network model can calculate rainfall in areas where no rain gauges are located. Then, using the arithmetic progression of rainfall from adjacent rain gauges as the optimization objective, a particle swarm optimization algorithm is used to adjust the location of rain gauges within each region. Because the adjustment is based on the existing conditions, it ensures reasonable rain gauge distribution and coverage while capturing as much varied rainfall as possible, significantly improving the accuracy of rainfall measurement within the region.

Description

Technical Field

[0001] This invention relates to the field of rainfall measurement technology, and in particular to a method for optimizing the location of rain gauges based on rainfall analysis. Background Technology

[0002] The data obtained from rain gauges are the most direct reflection of precipitation observations and can be used to support decision-making on practical issues such as weather forecasting and flood prevention, as well as for in-depth research on regional climate.

[0003] While densely packed rain gauges can accurately monitor actual rainfall, they consume significant manpower and resources, and the excessive amount of data increases information transmission and processing costs. However, overly sparse rain gauges can lead to errors in regional rainfall calculations. If the rainfall variation between adjacent rain gauges is too small, it results in an overly dense distribution of gauges; conversely, if the rainfall variation between adjacent rain gauges is too large, it leads to inaccurate regional rainfall estimates.

[0004] Currently, most of the rain gauges that have been deployed are poorly distributed, and there is no practical method for optimizing and adjusting the existing rain gauges in a region. Summary of the Invention

[0005] This invention discloses a method for optimizing the location of rain gauges based on rainfall analysis, which can optimize and adjust the location of existing rain gauges.

[0006] It is achieved through the following technical solution:

[0007] Get all rain gauges R in the area to be optimized i Location distribution information and historical rainfall information i represents the rain gauge station number, and j represents the number of rainfall events.

[0008] Based on the frequency of common rainfall types, all rain gauges R in the area to be optimized will be... i Divide into several rainfall sub-regions k;

[0009] For each rainfall sub-region k, the coordinates R of several rain gauge stations within the sub-region are used. k,x (x k,n y k,n ) and average rainfall are Training data was used to train a BP neural network model to fit the coordinate-rainfall distribution.

[0010] For existing rain gauge R k,n Based on the principle of full coverage of rainfall sub-regions, rain gauge stations R are designated for each rainfall sub-region. k,n Adjustable range;

[0011] Using the particle swarm optimization algorithm, the coordinates of all rain gauges within each rainfall sub-region k are adjusted within an adjustable range to generate new rain gauge coordinates R′. k,n (x′ k,n y′ k,n Its fitness function is From the new rain gauge coordinates R′ k,n (x′ k,n y′ k,n Input BP neural network model f k (x′ k,n y′ k,n The prediction is as follows: k,n is the current adjusted rainfall station in the k rainfall sub-region, and k,nm is the m rainfall stations in the same rainfall sub-region adjacent to the current adjusted rainfall station.

[0012] The advantage of this embodiment is that it divides all rain gauges into several rainfall sub-regions based on whether rain falls at the same time, whether it is light rain or heavy rain. The rain gauges in each rainfall sub-region have strong common rainfall characteristics. For rain gauges with the same rainfall characteristics, a neural network model is used to fit the rules of coordinate changes and rainfall changes. The neural network model can calculate the rainfall in areas where no rain gauges are deployed. Then, with the arithmetic variation of rainfall in adjacent rain gauges as the optimization objective, the particle swarm algorithm is used to adjust the position of rain gauges in each region. Since the adjustment is based on the original, it can ensure that the distribution and coverage of rain gauges are reasonable while obtaining as much varied rainfall as possible, which can significantly improve the accuracy of rainfall measurement in the region.

[0013] Furthermore, based on the frequency of common rainfall types, all rain gauge stations Pi in the area to be optimized are divided into several rainfall sub-regions k, as follows:

[0014] Sort all rainfall periods by time to generate a set of historical rainfall periods, and calculate the average rainfall for each rain gauge station during each rainfall period;

[0015] Through standardization, the average rainfall for each rainfall period at each rain gauge station is divided into three categories: less than average rainfall, equal to average rainfall, and exceeding average rainfall, thus generating a standardized bias set.

[0016] Calculate the information entropy of the standardized deviation set for each rain gauge station;

[0017] Based on the information entropy of the standardized deviation set of each rain gauge, rain gauges with information entropy less than the threshold are divided into Level 1 rainfall sub-regions;

[0018] Remove all rain gauges that have already been assigned to a rainfall sub-region from all rain gauges, and repeat the process of assigning the next level of rainfall sub-regions until the number of rainfall sub-regions reaches the preset target or all rain gauges have been assigned, resulting in k rainfall sub-regions.

[0019] The advantage of this embodiment is that, among all rain gauges, the rainfall characteristics of each rain gauge can be digitally expressed by the rainfall deviation set. Then, the predictability of the rainfall deviation set can be calculated by the information entropy, and a batch of rain gauges whose rainfall characteristics are closest to the average rainfall characteristics can be found. After finding the first batch of similar rain gauges, similar processing can be repeated for subsequent rain gauges to find several rain gauges with similar rainfall characteristics, thus completing the division of rainfall sub-regions.

[0020] Furthermore, the average rainfall at each rain gauge station during each rainfall period is calculated using the following method:

[0021] From all historical rainfall information In the middle, a set of historical rainfall periods (T) is generated by arranging them in chronological order. 1e -T 1s ,T 2e -T 2s ,...,T je -T js ); where j is the rainfall number, and all rain gauges R i During the same time period, any one or more instances of rainfall are recorded as one rainfall event, T js Let T be the start time of the j-th rainfall event. If two or more rain gauges register rainfall during the same time period, the start time of the rain gauge that registers the rainfall earliest is selected as the start time of the j-th rainfall event. je Let be the end time of the j-th rainfall. If there are two or more rain gauges with rainfall at the same time, the end time of the rain gauge with the latest end time is selected as the start time of the j-th rainfall.

[0022] Based on the historical rainfall period set (T 1e -T 1s ,T 2e -T 2s ,...,T je -T js Based on this, the average rainfall at each rain gauge station for the corresponding time period is recorded sequentially. The calculation formula is as follows:

[0023]

[0024] In the formula, Let rf be the average rainfall of the j-th rainfall at the i-th rain gauge station. i,j Let rf be the total rainfall at the i-th rain gauge during the j-th rainfall. If no rainfall occurs at the i-th rain gauge during the P-th rainfall period, then rf i,j =0.

[0025] The advantage of this embodiment is that the average rainfall is referenced to the common rainfall time of all rain gauges. Based on this, the parts of all rain gauges with the same rainfall characteristics can be found, and the positions of rain gauges with the same rainfall characteristics can be uniformly adjusted.

[0026] Furthermore, through standardization, the average rainfall for each rainfall period at each rain gauge station is categorized into below-average rainfall, equal to-average rainfall, and above-average rainfall. The specific method is as follows:

[0027] For each rain gauge, generate the average rainfall set for all rainfall periods.

[0028] The average rainfall per station for each rainfall event in the area to be optimized is calculated using the formula:

[0029]

[0030] For each rain gauge, the rainfall deviation set for all rainfall periods is calculated and generated according to the formula.

[0031] The set of rainfall deviations DR for all rainfall periods i Standardize the process;

[0032]

[0033] For each rain gauge, generate a set of standardized rainfall biases for all rainfall periods.

[0034] The advantage of this embodiment is that, by setting th1 and th2, each rain gauge station can be compared with the average rainfall during each rainfall period, and the comparison results of each rain gauge station can be digitized.

[0035] The information entropy of the standardized deviation set for each rain gauge is calculated using the following method:

[0036] Calculate the information entropy of each rain gauge using the formula:

[0037]

[0038] In the formula, n = -1, 0, 1, and p(n) is the set of standardized biases of rainfall at rain gauge station i. The probability of taking the value n.

[0039] The advantage of this embodiment is that information entropy can reflect whether a rain gauge is predictable, as it closely approximates the average rainfall characteristics, and rain gauges with the same predictability share common rainfall characteristics.

[0040] Furthermore, rain gauge stations with information entropy less than a threshold are classified as level 0 rainfall sub-regions, as follows:

[0041] On the map of the area to be optimized, the planar coordinates R of each rain gauge station are... i (x i ,y i The information entropy corresponding to the location is labeled to generate the original information entropy planar map;

[0042] Using the original information entropy planar map, the information entropy is labeled as... Several rain gauge stations are classified as level 0 rain gauge stations i0. This represents the minimum information entropy of all rain gauges.

[0043] If the level 0 rain gauge station i 0,l If the effective coverage area can be effectively connected, then the division of the level 0 rainfall sub-region is completed;

[0044] If the level 0 rain gauge station i 0,l If the effective coverage area cannot be effectively connected, adjust th2 and th1, increasing th2-th1, which can make the level 0 rain gauge i 0,l The number increases, completing the effective connection; th2-th1 decreases, allowing level 0 rain gauge i to... 0,l The number of stations decreased, and level 0 rain gauges that could not be connected were excluded.

[0045] The advantage of this embodiment is that by adjusting the difference between th1 and th2, rain gauges with the same rainfall characteristics can be interconnected, which facilitates the adjustment and optimization of all rain gauges in the same rainfall sub-region.

[0046] Furthermore, a BP neural network model for fitting the coordinate-rainfall distribution is trained, as follows:

[0047] For each rainfall sub-region k, construct a BP neural network model f. k (x k,n y k,n );

[0048] Obtain the coordinates R of all rain gauges within the rainfall sub-region. k,n (x k,n y k,n ) and the average rainfall at each rain gauge station is Construct real training data;

[0049] Select several sampling points in the rainfall sub-region that do not have rain gauges and obtain the coordinates of the sampling points. Based on the coordinates of adjacent rain gauge stations and the average rainfall, the average rainfall at the sampling points was calculated using interpolation. Construct sampling training data;

[0050] Combine real training data and sampled training data to generate all training data.

[0051] The first 80% of all training data is used as the training set, and the last 20% is used as the test set to train the BP neural network model f. k (x k,n y k,n ).

[0052] The advantage of this embodiment is that by sampling training data, the amount of training data can be increased, thus preventing underfitting.

[0053] Furthermore, the particle swarm optimization algorithm is described in the following specific steps:

[0054] The particle swarm is initialized using the number and coordinates of rain gauges within the rainfall sub-region, and the position x of each particle is initialized. i and velocity v i ;

[0055] Given that the effective range of the rain gauge can fully cover the area, set movement constraints for each particle;

[0056] Set the fitness function for each particle as follows: In the formula, Δrf represents the change in rainfall between adjacent rain gauge stations;

[0057] Update the particle fitness value. k,n With individual extreme values ​​pbest k,n (t) Comparison, if Fitness k,n pbest k,n Then use Fitness k,n Replace pbest k,n (t), otherwise keep pbest k,n (t);

[0058] Add up the fitness value of each particle and the global optimum gbest k Compared to (t), if the sum of the fitness values ​​∑Fitness k,n >gbest k Then replace gbest with the fitness value corresponding to the current particle state. k Otherwise, retain gbest. k ;

[0059] Iteratively update particle position x i and velocity v i ;

[0060] When the number of iterations reaches the preset value or the iteration target is reached, the particle swarm optimization algorithm stops and the current result is output.

[0061] By using the particle swarm optimization algorithm, with the goal of minimizing the difference in rainfall between adjacent rain gauges, the location of rain gauges can be made more reasonable, significantly improving the accuracy of regional rainfall statistics. Attached Figure Description

[0062] The accompanying drawings of this invention are described below.

[0063] Figure 1 This is a schematic diagram of the process of the present invention.

[0064] Figure 2 This is a training graph for a neural network model.

[0065] Figure 3 This is a schematic diagram of the particle swarm optimization algorithm. Detailed Implementation

[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0067] A method for optimizing rain gauge location based on rainfall analysis, such as Figure 1 As shown, the specific steps are as follows:

[0068] S1. Obtain all rainfall stations in the area to be optimized. i Location distribution information and historical rainfall information i represents the rain gauge station number, and j represents the number of rainfall events.

[0069] S2. Based on the frequency of common rainfall types, select all rain gauge stations R in the area to be optimized. i The region is divided into several rainfall sub-regions k, as follows:

[0070] S21. Sort all rainfall periods by time, generate a set of historical rainfall periods, and calculate the average rainfall for each rain gauge station during each rainfall period.

[0071] Specifically, step S21 involves analyzing all historical rainfall information. In the middle, a set of historical rainfall periods (T) is generated by arranging them in chronological order. 1e -T 1s ,T 2e -T 2s ,...,T je -T js ); where j is the rainfall number, and all rain gauges R i During the same time period, any one or more instances of rainfall are recorded as one rainfall event, T js Let T be the start time of the j-th rainfall event. If two or more rain gauges register rainfall during the same time period, the start time of the rain gauge that registers the rainfall earliest is selected as the start time of the j-th rainfall event. jeLet be the end time of the j-th rainfall. If there are two or more rain gauges with rainfall at the same time, the end time of the rain gauge with the latest end time is selected as the start time of the j-th rainfall.

[0072] Based on the historical rainfall period set (T 1e -T 1s ,T 2e -T 2s ,...,T je -T js Based on this, the average rainfall at each rain gauge station for the corresponding time period is recorded sequentially. The calculation formula is as follows:

[0073]

[0074] In the formula, Let rf be the average rainfall of the j-th rainfall at the i-th rain gauge station. i,j Let rf be the total rainfall at the i-th rain gauge during the j-th rainfall. If no rainfall occurs at the i-th rain gauge during the j-th rainfall period, then rf i,j =0.

[0075] S22. Through standardization, the average rainfall of each rainfall period at each rain gauge station is divided into less than average rainfall, equal to average rainfall, and exceeding average rainfall, generating a standardized deviation set.

[0076] Specifically, in step S22, for each rain gauge station, an average rainfall set is generated for all rainfall periods.

[0077] The average rainfall per station for each rainfall event in the area to be optimized is calculated using the formula:

[0078]

[0079] For each rain gauge, the rainfall deviation set for all rainfall periods is calculated and generated according to the formula.

[0080] The set of rainfall deviations DR for all rainfall periods i Standardize the process;

[0081]

[0082] For each rain gauge, generate a set of standardized rainfall biases for all rainfall periods.

[0083] S23. Calculate the information entropy of the standardized deviation set for each rain gauge station.

[0084] Specifically, in step S23, the information entropy of each rain gauge is calculated according to the formula:

[0085]

[0086] In the formula, n = -1, 0, 1, and p(n) is the set of standardized biases of rainfall at rain gauge station i. The probability of taking the value n.

[0087] S24. Based on the information entropy of the standardized deviation set of each rain gauge, the rain gauges with information entropy less than the threshold are divided into Level 1 rainfall sub-regions.

[0088] Specifically, in step S24, on the map of the area to be optimized, at the planar coordinates R of each rain gauge station... i (x i ),y i The information entropy corresponding to the location is labeled to generate the original information entropy planar map;

[0089] Using the original information entropy planar map, the information entropy is labeled as... Several rain gauge stations are classified as level 0 rain gauge stations i0. This represents the minimum information entropy of all rain gauges.

[0090] If the level 0 rain gauge station i 0,l If the effective coverage area can be effectively connected, then the division of the level 0 rainfall sub-region is completed;

[0091] If the level 0 rain gauge station i 0,l If the effective coverage area cannot be effectively connected, adjust th2 and th1, increasing th2-th1, which can make the level 0 rain gauge i 0,l The number increases, completing the effective connection; th2-th1 decreases, allowing level 0 rain gauge i to... 0,l The number of stations decreased, and level 0 rain gauges that could not be connected were excluded.

[0092] S25. Remove the rain gauges that have been assigned to the rainfall sub-region from all the rain gauges, and repeat the division of the next level of rainfall sub-regions until the number of rainfall sub-regions reaches the preset target or all rain gauges are divided, resulting in k rainfall sub-regions.

[0093] S3. For each rainfall sub-region k, use the coordinates R of several rain gauge stations within the sub-region. k,n (x k,n y k,n ) and average rainfall are Training data was used to train a BP neural network model to fit the coordinate-rainfall distribution.

[0094] The method for constructing the BP neural network model in step S3 is as follows: Figure 2 As shown:

[0095] S31. For each rainfall sub-region k, construct a BP neural network model f. k (x k,n y k,n );

[0096] S32. Obtain the coordinates R of all rain gauge stations within the rainfall sub-region. k,n (x k,n y k,n ) and the average rainfall at each rain gauge station is Construct real training data;

[0097] S33. Select several sampling points in the rainfall sub-region that do not have rain gauges and obtain the coordinates of the sampling points. Based on the coordinates of adjacent rain gauge stations and the average rainfall, the average rainfall at the sampling points was calculated using interpolation. Construct sampling training data;

[0098] S34. Combine real training data and sampled training data to generate all training data.

[0099] S35. Use the first 80% of all training data as the training set and the last 20% as the test set to train the BP neural network model f. k (x k,n y k,n ).

[0100] S4, regarding existing rain gauge stations R k,n Based on the principle of full coverage of rainfall sub-regions, rain gauge stations R are designated for each rainfall sub-region. k,n The adjustable range.

[0101] S5. Using the particle swarm optimization algorithm, adjust all rain gauges within each rainfall sub-region k within an adjustable range to generate new rain gauge coordinates R′. k,n (x′ k,n y′ k,n Its fitness function is From the new rain gauge coordinates R′ k,n (x′ k,n y′ k,n Input BP neural network model f k (x′ k,n y′ k,n The prediction is as follows: k,n is the current adjusted rainfall station in the k rainfall sub-region, and k,nm is the m rainfall stations in the same rainfall sub-region adjacent to the current adjusted rainfall station.

[0102] Specifically, the specific steps of the particle swarm optimization algorithm are as follows: Figure 3 As shown:

[0103] S51. Initialize the particle swarm using the number and coordinates of rain gauges within the rainfall sub-region, and initialize the position x of each particle. i and velocity v i ;

[0104] S52. Set movement constraints for each particle, assuming that the effective range of the rain gauge can fully cover the area.

[0105] S53. Set the fitness function for each particle as follows: In the formula, Δrf represents the change in rainfall between adjacent rain gauge stations;

[0106] S54. Update the particle fitness value. k,n With individual extreme values ​​pbest k,n (t) Comparison, if Fitness k,n pbest k,n Then use Fitness k,n Replace pbest k,n (t), otherwise keep pbest k,n (t);

[0107] S55. Add the fitness value of each particle together with the global optimum gbest. k Compared to (t), if the sum of the fitness values ​​∑Fitness k,n >gbest k Then replace gbest with the fitness value corresponding to the current particle state. k Otherwise, retain gbest. k ;

[0108] S56, Iteratively update particle position x i and velocity v i ;

[0109] S57. When the number of iterations reaches the preset value or the iteration target is reached, stop the particle swarm optimization algorithm and output the current result.

[0110] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for optimizing the location of rain gauges based on rainfall analysis, characterized in that, The specific method is as follows: Get all rain gauges in the area to be optimized Location distribution information and historical rainfall information , Number the rain gauge station. For the first This rainfall; Based on the frequency of common rainfall types, all rain gauge stations in the area to be optimized will be included. Divided into several rainfall sub-regions ; For each rainfall sub-region The coordinates of several rain gauge stations within the sub-region and average rainfall Training data was used to train a BP neural network model to fit the coordinate-rainfall distribution. ; For existing rain gauges Based on the principle of full coverage of rainfall sub-regions, rain gauge stations were designated for each rainfall sub-region. Adjustable range; The particle swarm optimization algorithm is used to adjust each rainfall sub-region within an adjustable range. New rain gauge coordinates are generated for all rain gauge stations. Its fitness function is , From the new rain gauge coordinates Input BP neural network model It was predicted; for Current adjustments to rain gauge stations within the rainfall sub-region For the current adjustment of the rain gauge station Rain gauge stations in the same sub-region of rainfall; Based on the frequency of common rainfall types, all rain gauge stations in the area to be optimized will be included. Divided into several rainfall sub-regions The specific method is as follows: Sort all rainfall periods by time to generate a set of historical rainfall periods, and calculate the average rainfall for each rain gauge station during each rainfall period; Through standardization, the average rainfall for each rainfall period at each rain gauge station is divided into three categories: less than average rainfall, equal to average rainfall, and exceeding average rainfall, thus generating a standardized bias set. Calculate the information entropy of the standardized deviation set for each rain gauge station; Based on the information entropy of the standardized deviation set of each rain gauge, rain gauges with information entropy less than the threshold are divided into Level 1 rainfall sub-regions; Remove all rain gauges that have already been assigned to a rainfall sub-region from all rain gauges, and repeat the process of assigning the next level of rainfall sub-regions until the number of rainfall sub-regions reaches the preset target or all rain gauges have been assigned. Each rainfall sub-region.

2. The method for optimizing rain gauge location based on rainfall analysis as described in claim 1, characterized in that, The method for calculating the average rainfall at each rain gauge station during each rainfall period is as follows: From all historical rainfall information In the middle, a set of historical rainfall periods is generated by arranging them in chronological order. ;in Number the number of rainfall events, all rain gauges If more than one rainfall event occurs within the same time period, it is recorded as one rainfall event. For the first If two or more rain gauges recorded rainfall at the same time, the start time of the rainfall at the first rain gauge is selected as the starting time of the next rainfall event. The start time of the next rainfall; For the first If two or more rain gauges recorded rainfall at the same time, the end time of the rain gauge that recorded the latest end time will be selected as the [number]th rainfall time. The end time of this rainfall; Based on historical rainfall periods Based on this, the average rainfall at each rain gauge station for the corresponding time period is recorded sequentially, and the calculation formula is as follows: In the formula, For the first The first rain gauge station Average rainfall of the rainfall event For the first The first rain gauge station The total rainfall of the second rainfall event, if the first The second rainfall period If no rainfall is recorded at any of the rain gauges, then... .

3. The method for optimizing rain gauge location based on rainfall analysis as described in claim 2, characterized in that, Through standardization, the average rainfall for each rainfall period at each rain gauge station is categorized into below-average rainfall, equal to-average rainfall, and above-average rainfall. The specific method is as follows: For each rain gauge, generate the average rainfall set for all rainfall periods. ; The average rainfall per station for each rainfall event in the area to be optimized is calculated using the formula: For each rain gauge, the rainfall deviation set for all rainfall periods is calculated and generated according to the formula. ; Set of rainfall deviations for all rainfall periods Standardization processing is required; For each rain gauge, generate a set of standardized rainfall biases for all rainfall periods. .

4. The method for optimizing rain gauge location based on rainfall analysis as described in claim 3, characterized in that, The information entropy of the standardized deviation set for each rain gauge is calculated using the following method: Calculate the information entropy of each rain gauge using the formula: In the formula, =-1, 0, 1, For rain gauge Rainfall Standardized Bias Set The value of the middle is The probability of.

5. The method for optimizing rain gauge location based on rainfall analysis as described in claim 1, characterized in that, Rain gauge stations with information entropy less than a threshold are classified as level 0 rainfall sub-regions. The specific method is as follows: On the map of the area to be optimized, the planar coordinates of each rain gauge station Label the corresponding information entropy and generate the original information entropy planar map; Using the original information entropy planar map, the information entropy is labeled as... Several rain gauge stations were classified as level 0 rain gauge stations. , This represents the minimum information entropy of all rain gauges. If a level 0 rain gauge If the effective coverage area can be effectively connected, then the division of the level 0 rainfall sub-region is completed; If the level 0 rain gauge station If the effective coverage area cannot be effectively connected, then adjust... and , Increasing the size can make a level 0 rain gauge station The number of connections increases, completing a valid connection; Reduction can reduce the level 0 rain gauge. The number of stations decreased, and level 0 rain gauges that could not be connected were excluded.

6. The method for optimizing rain gauge location based on rainfall analysis as described in claim 1, characterized in that, The specific method for training a BP neural network model to fit the coordinate-rainfall distribution is as follows: For each rainfall sub-region Construct a BP neural network model ; Obtain the coordinates of all rain gauges within the rainfall sub-region. The average rainfall at each rain gauge station was Construct real training data; Select several sampling points in the rainfall sub-region that do not have rain gauges and obtain the coordinates of the sampling points. Based on the coordinates of adjacent rain gauge stations and the average rainfall, the average rainfall at the sampling points was calculated using interpolation. Construct sampling training data; Combine real training data and sampled training data to generate all training data. The first 80% of all training data was used as the training set, and the last 20% was used as the test set to train the BP neural network model. .

7. The method for optimizing rain gauge location based on rainfall analysis as described in claim 1, characterized in that, The particle swarm optimization algorithm is described below: The particle swarm is initialized using the number and coordinates of rain gauges within the rainfall sub-region, and the position of each particle is initialized accordingly. and speed ; Given that the effective range of the rain gauge can fully cover the area, set movement constraints for each particle; Set the fitness function for each particle as follows: ; Updated particle fitness values With individual extreme values In comparison, if Then use Replace Otherwise, it will remain. ; Add the fitness value of each particle together with the global optimum. In comparison, if the fitness value after addition is... Then replace it with the fitness value corresponding to the current particle state. Conversely, retain ; Iteratively update particle position and speed ; When the number of iterations reaches the preset value or the iteration target is reached, the particle swarm optimization algorithm stops and the current result is output.

Images