CN110738355A - urban waterlogging prediction method based on neural network - Google Patents
urban waterlogging prediction method based on neural network Download PDFInfo
- Publication number
- CN110738355A CN110738355A CN201910885437.0A CN201910885437A CN110738355A CN 110738355 A CN110738355 A CN 110738355A CN 201910885437 A CN201910885437 A CN 201910885437A CN 110738355 A CN110738355 A CN 110738355A
- Authority
- CN
- China
- Prior art keywords
- waterlogging
- neural network
- data
- time
- time sequence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Abstract
The invention discloses an urban waterlogging prediction method based on a neural network, which comprises the steps of inputting time sequence characteristic data to be detected into a trained time sequence characteristic prediction model, outputting a time sequence characteristic prediction result by the time sequence characteristic prediction model, fusing the predicted time sequence characteristic data with space characteristic data to obtain time-space characteristic data to be detected, inputting the time-space characteristic data to be detected into the trained waterlogging prediction model, and outputting a grade prediction result of the waterlogging degree of a waterlogging point to be detected by the waterlogging prediction model.
Description
Technical Field
The invention relates to the technical field of data mining and prediction, in particular to urban waterlogging prediction methods based on a neural network.
Background
In recent years, with extreme rainfall in cities caused by global climate change and rapid urbanization process in China, waterlogging frequently occurs in various cities, particularly coastal cities, and various functions of the cities are nearly paralyzed due to a serious water logging problem, so that serious loss is brought to life safety and property of urban residents.
At present, most of urban floods are researched aiming at single disaster-causing factors such as rainstorm flood or urban water removal and the like, a prediction model is more traditional, data utilization is insufficient, and reliability is poor.
Accordingly, the prior art is yet to be improved and developed.
Disclosure of Invention
The present invention is based on the discovery and recognition by the inventors of the following facts and problems:
the inventor finds that in the key technology of urban flood control and disaster reduction by utilizing big data and machine learning technology to predict and simulate flood disasters, a drainage model is improved and jointly forms a ponding model in the prior art based on a rainfall model and a confluence model, the generalized problem of urban drainage system facilities in the traditional rainstorm waterlogging mathematical model is solved, the ponding depth is predicted, in addition, a digital combination mode is adopted to respectively calculate urban water removal and urban precipitation, software is utilized to convert and superpose the urban ponding effect graph, and the urban waterlogging occurrence can be predicted in advance by analyzing the urban ponding effect graph.
In view of the defects of the prior art, the invention aims to provide urban waterlogging prediction methods based on a neural network, and the urban waterlogging prediction method realizes prediction and forecast of urban waterlogging degree by integrating a plurality of urban waterlogging disaster-causing factors.
The technical scheme of the invention is as follows:
A neural network-based urban waterlogging prediction method, comprising the steps of:
inputting the time sequence feature data to be tested into a trained time sequence feature prediction model, and outputting a time sequence feature prediction result by the time sequence feature prediction model, wherein the time sequence feature prediction result is the predicted time sequence feature data at moment under the time sequence feature data to be tested;
fusing the predicted time sequence characteristic data with the spatial characteristic data to obtain space-time characteristic data to be measured;
and inputting the space-time characteristic data to be detected into a trained waterlogging prediction model, and outputting a grade prediction result of the waterlogging degree of the waterlogging point to be detected by the waterlogging prediction model.
The method further comprises step of inputting the time sequence characteristic data to be tested into the time sequence characteristic prediction model, further comprising:
constructing an inland inundation data set, wherein the inland inundation data set comprises time-space characteristic data and inland inundation point water immersion height grade data, and the time-space characteristic data comprises time sequence characteristic data and space characteristic data; and inputting the time sequence characteristic data into a preset GA-Elman neural network, and obtaining the time sequence characteristic prediction model after training.
According to the method, a step is provided, wherein before the time-space characteristic data to be detected are input into an inland inundation prediction model, the method further comprises the steps of inputting the time-space characteristic data and the inland inundation point water immersion height grade data into a preset GA-probabilistic neural network, and obtaining the inland inundation prediction model after training.
According to a further , the method for constructing the waterlogging data set includes:
extracting the relevant space-time characteristic data caused by waterlogging as characteristics;
extracting water logging height grade data of waterlogging points, judging the waterlogging degree grade according to the water logging height grade data of the waterlogging points, and taking the waterlogging degree grade as a label;
and constructing a data table according to the features and the labels to obtain the waterlogging data set.
The invention further comprises , wherein the preset GA-Elman neural network comprises an input layer, a hidden layer, a carrying layer and an output layer, and the training step of the time sequence characteristic prediction model comprises the following steps:
inputting the multi-dimensional sample time sequence characteristic data with the time length of T into a preset GA-Elman neural network, and outputting multi-dimensional sample time sequence characteristic prediction results of T1 at next moments by the preset GA-Elman neural network, wherein the time length of T is 3-5 hours;
calculating the average absolute relative deviation between the predicted value of the sample time series characteristic prediction result at the next moments T1 and the measured value of the sample time series characteristic prediction result at the next moments T1, and carrying out precision evaluation according to the predicted average absolute relative deviation;
and repeating the step of inputting the multi-dimensional and T-duration sample time sequence characteristic data into a preset GA-Elman neural network until the training of the preset GA-Elman neural network meets a preset condition, so as to obtain a time sequence characteristic prediction model.
According to the invention, , the preset GA-probabilistic neural network comprises an input layer, a mode layer, a summation layer and an output layer, and the training of the waterlogging prediction model comprises the following steps:
inputting the sample time-space characteristic data and the sample waterlogging point water immersion height grade data into the preset GA-probabilistic neural network, and outputting a sample waterlogging degree prediction result by the preset GA-probabilistic neural network;
calculating the relative accuracy of the predicted value of the sample waterlogging degree prediction result and the measured value of the sample waterlogging degree prediction result, and carrying out precision evaluation according to the predicted relative accuracy;
and repeating the step of inputting the sample space-time characteristic data and the sample waterlogging point water immersion height grade data into the preset GA-probabilistic neural network until the training of the preset GA-probabilistic neural network meets a preset condition, and obtaining a waterlogging prediction model.
According to a further step, the step of fusing the predicted time sequence feature data with the spatial data to obtain spatio-temporal feature data to be measured includes:
inputting multi-dimensional time sequence feature data to be detected with T duration into the time sequence feature prediction model, and outputting time sequence feature prediction results of samples to be detected at moments T1 by the time sequence feature prediction model;
and (3) performing inverse regression on the predicted time sequence characteristic data to be detected at the next time T1, and fusing the time sequence characteristic data to be detected with the spatial characteristic data to form the time-space characteristic data to be detected and perform regression .
The invention further provides , wherein the step of performing the accuracy assessment based on the predicted mean absolute relative deviation further comprises:
initializing parameters of the preset GA-Elman neural network, optimizing weight thresholds from an input layer to a hidden layer, weight thresholds from the hidden layer to an output layer and weight thresholds of a carrying layer by utilizing a genetic algorithm, and training the preset GA-Elman neural network.
The invention further provides , wherein the step of performing the accuracy assessment based on the predicted relative accuracy further comprises:
initializing the parameters of the preset GA-probabilistic neural network, optimizing the smooth coefficient of the probabilistic neural network by using a genetic algorithm to obtain the optimal hyperparameter, and training the preset GA-probabilistic neural network.
The invention further provides , wherein the step of performing the accuracy assessment based on the predicted mean absolute relative deviation further comprises:
inputting characteristic data of a time window after multidimensional, T-duration and normalization processing into a preset GA-Elman neural network, wherein the preset GA-Elman neural network outputs a sample time sequence characteristic prediction result of the multidimensional and next moments T1;
the time window characteristic data with the time length of T slides backwards along with the time, and the sample characteristic data is updated;
adding hours later sample characteristic data, and removing corresponding hours old sample characteristic data;
and the time window continues to slide backwards along with time until the training of the time sequence characteristic prediction model is finished.
The invention provides urban waterlogging prediction methods based on neural networks, which comprise the steps of inputting time sequence characteristic data to be detected into a trained time sequence characteristic prediction model, outputting time sequence characteristic prediction results by the time sequence characteristic prediction model, fusing the predicted time sequence characteristic data with spatial characteristic data to obtain time-space characteristic data to be detected, inputting the time-space characteristic data to be detected into the trained waterlogging prediction model, and outputting a grade prediction result of the waterlogging degree of a waterlogging point to be detected by the waterlogging prediction model.
Drawings
Fig. 1 is a flow chart 1 of a neural network-based urban waterlogging prediction method in embodiments.
Fig. 2 is a flow chart 2 of a neural network-based urban waterlogging prediction method in embodiments.
Fig. 3 is a sample plot of the inland inundation data sets for a certain city along the sea in embodiments.
FIG. 4 is a comparison graph of predicted values and actual values at time T1 at below waterlogging point in a certain city along the sea in examples.
Fig. 5 is a block architecture diagram of an urban waterlogging prediction system based on a neural network in embodiments.
Detailed Description
The invention provides a urban waterlogging prediction method based on a neural network, which can meet the requirements of a specific scene, and the basic idea is to construct two neural network models, wherein network models are constructed on a training data set formed by time sequence characteristics, and network models are constructed on a training data set formed by time-space characteristics, which is equivalent to using the time sequence to predict network performance to expand the waterlogging prediction network performance.
In the embodiments and claims, unless the article "is specifically limited herein, the word" "and" the "can refer broadly to a single or a plurality.
Referring to fig. 1 to 4, the present invention provides preferred embodiments of urban waterlogging prediction methods based on neural networks.
As shown in fig. 1 and fig. 2, fig. 1 is a flowchart 1 of an urban waterlogging prediction method based on a neural network in embodiments, and fig. 2 is a flowchart 2 of an urban waterlogging prediction method based on a neural network in embodiments, and as shown in the figure, the method includes the steps of:
step S100, inputting the time sequence feature data to be tested into a trained time sequence feature prediction model, and outputting a time sequence feature prediction result by the time sequence feature prediction model, wherein the time sequence feature prediction result is the predicted time sequence feature data of moments under the time sequence feature data to be tested.
Before inputting the time sequence characteristic data to be tested into the time sequence characteristic prediction model, the method further comprises the following steps:
and step S10, constructing an inland inundation data set, wherein the inland inundation data set comprises space-time characteristic data and inland inundation point water immersion height grade data, and the space-time characteristic data comprises time sequence characteristic data and space characteristic data.
The construction of the waterlogging data set comprises the following steps:
step S11, extracting relevant space-time characteristic data to be tested of the flood caused by waterlogging, wherein the characteristic data comprises the following specific steps:
step S111, sorting factors related to the influence degree of waterlogging points, and extracting the time sequence data and the space data, wherein, referring to FIG. 3, FIG. 3 is an embodiment inland inundation data sample graph of a certain coastal city, the time sequence data comprises typhoon path data (typhoon path longitude, latitude, central air pressure, central wind speed, typhoon level at intervals of six hours), offshore meteorological data (mean station pressure at intervals of hours, mean sea level air pressure at intervals of time, mean air temperature at intervals of time, mean dew point temperature at intervals of time, mean relative humidity at times, total rainfall at times, prevailing wind direction at times, mean wind speed at times), tidal height data (tidal height at intervals of hours), urban meteorological data (total rainfall at intervals of hours, prevailing wind direction at times, mean wind speed at times), the space data comprises inland inundation point water state data (coefficient between 0 and 1), inland inundation point water drainage coefficient between dimension data (0 and 1), and the space data is more than 5, and the characteristic data is, such as longitude and latitude data, and inland inundation data.
Step S112, sorting the multidimensional time series data, and using the sorted multidimensional time series data system as sample data of a preset interval time, it should be noted that the preset interval time may be set according to an actual situation, which is set to 1 hour in this embodiment, that is, the sorted multidimensional time series data system is sample data of an interval of 1 hour.
And S113, supplementing the vacancy data by using a data interpolation method and a BP (back propagation) neural network according to the time-varying characteristic of the time sequence data to be detected, wherein the vacancy data is the weighted arithmetic mean of front and back adjacent sample data, and specifically, supplementing the vacancy data by using a data interpolation method and supplementing the typhoon grade vacancy data by using the BP neural network according to the time-varying characteristic of a typhoon path, meteorological data and tidal data, wherein the typhoon path is six-hour interval data, the BP neural network is constructed when the typhoon path is converted into -hour interval data, the longitude, the latitude, the central air pressure and the central air speed of the typhoon path are used as input, the typhoon grade is output in a fitting manner, the network prediction accuracy is over 98%, and the robustness is good.
Step S113, taking the time sequence data and the spatial data after the preprocessing as the characteristics; wherein the features include temporal features and spatial features.
Step S12, extracting water logging height grade data of the waterlogging point, judging the waterlogging degree grade according to the water logging height grade data of the waterlogging point, and using the waterlogging degree grade as a label;
the waterlogging degree grades can be divided into a blue grade, a yellow grade, an orange grade, a red grade and a black grade, and taking Australian in coastal cities as an example, the blue grade with the water level being below 0.5 meter above the road surface is defined as a mark 1, the yellow grade with the water level being 0.5-1.0 meter is defined as a mark 2, the orange grade with the water level being 1.0-1.5 meter is defined as a mark 3, the red grade with the water level being 1.5-2.5 meter is defined as a mark 4, and the black grade with the water level being above 2.5 meter is defined as a mark 5.
And step S13, constructing a data table according to the features and the labels to obtain the waterlogging data set.
As shown in FIG. 3, the waterlogging data set is a two-dimensional table of rows and columns, wherein the rows are horizontal and the columns are vertical, the sample characteristics of the waterlogging data set are related to the waterlogging disaster-causing factors, the sample labels are highly related to water logging, the characteristics and the labels can be expanded or adjusted, and the characterization can realize the waterlogging prediction by classifying the samples according to the probability.
Step S14, inputting the time sequence feature data into a preset GA-Elman neural network, and obtaining the time sequence feature prediction model after training; the preset GA-Elman neural network comprises the following components: an input layer, a hidden layer, a accepting layer and an output layer.
The training step of the time sequence characteristic prediction model comprises the following steps:
s141, inputting multi-dimensional sample time sequence characteristic data with T duration into a preset GA-Elman neural network, and outputting multi-dimensional sample time sequence characteristic prediction results of T1 at next moments by the preset GA-Elman neural network, wherein the T duration is 3-5 hours;
step S142, calculating the average absolute relative deviation between the predicted value of the sample time series characteristic prediction result at the next moments T1 and the measured value of the sample time series characteristic prediction result at the next moments T1, and carrying out precision evaluation according to the predicted average absolute relative deviation;
the step of performing accuracy assessment according to the predicted mean absolute relative deviation further comprises:
step S143, inputting the characteristic data of the time window after multidimensional, T duration and normalization processing into a preset GA-Elman neural network, wherein the preset GA-Elman neural network outputs the multi-dimensional and next time T1 sample time sequence characteristic prediction results;
step S144, sliding the time window characteristic data with the time length of T backwards along with the time, and updating the sample characteristic data;
s145, adding characteristic data of a new moment after hours, and removing corresponding sample characteristic data of an old moment after hours;
and step S146, the time window continues to slide backwards along with time until the training of the time sequence characteristic prediction model is finished.
And S147, repeating the step of inputting the multi-dimensional and T-duration sample time sequence characteristic data into a preset GA-Elman neural network until the training of the preset GA-Elman neural network meets a preset condition, and obtaining a time sequence characteristic prediction model.
In specific implementation, a preset GA-Elman neural network is longitudinally constructed according to time sequence characteristics of an inland inundation data set, the preset GA-Elman neural network inputs 17 rows of time window characteristic data with T duration and subjected to quantization processing, outputs time sequence characteristic prediction results of T1 at the next 17 rows of time and times, takes 3-5 hours according to the weather characteristic T duration, carries out precision evaluation on the sample time sequence characteristic prediction result at the T1 at the time , when the precision is qualified, a time sequence characteristic prediction model is established, calculates the average absolute relative deviation between the prediction value at the next times T1 and the measured value at the next times T1, carries out precision according to the average absolute relative deviation of the prediction, and if the precision is unqualified, the steps S143 to S147 are repeated until the precision is qualified, and the time sequence characteristic prediction model is established.
The step of performing accuracy assessment according to the predicted mean absolute relative deviation further comprises:
and S148, initializing parameters of the preset GA-Elman neural network, optimizing the weight threshold values from the input layer to the hidden layer, the weight threshold values from the hidden layer to the output layer and the weight threshold value of the carrying layer by using a genetic algorithm, and training the preset GA-Elman neural network.
And S200, fusing the predicted time sequence characteristic data and the spatial characteristic data to obtain the space-time characteristic data to be measured.
The step of fusing the predicted time sequence characteristic data and the spatial data to obtain the space-time characteristic data to be measured comprises the following steps:
step S201, inputting multi-dimensional time sequence feature data to be detected with T duration into the time sequence feature prediction model, and outputting sample time sequence feature prediction results at T1 by the time sequence feature prediction model;
and S202, reversely reducing the predicted time sequence characteristic data to be detected at the time T1 of the next into , fusing the time sequence characteristic data to be detected with the spatial characteristic data to form the time-space characteristic data to be detected, and reducing the time-space characteristic data to be detected into .
In specific implementation, according to the time sequence characteristic prediction model, time sequence data of the waterlogging points to be detected with the same dimension and the same T duration are input, and a time sequence characteristic prediction result to be detected at the time T1 of is output.
In an actual prediction scene, 17 rows of time sequence data of a waterlogging point and time length T are extracted, the time sequence data are input into a time sequence feature prediction model Net1 after being preprocessed, Net1 outputs time sequence feature prediction results of T1 at the time of rows and below, the 17 rows of predicted time sequence features are reversely subjected to transformation and are fused with 5 rows of spatial feature data to form 22 new to-be-detected space-time features, and the 22 new to-be-detected space-time features are subjected to transformation.
And step S300, inputting the space-time characteristic data to be tested into a trained waterlogging prediction model, and outputting a grade prediction result of the waterlogging degree of the waterlogging point to be tested by the waterlogging prediction model.
Before inputting the space-time characteristic data to be detected into an inland inundation prediction model, the method further comprises the following steps:
step S30, inputting the space-time characteristic data and the water logging height grade data of the waterlogging point into a preset GA-probabilistic neural network, and obtaining the waterlogging prediction model after training; wherein the preset GA-probabilistic neural network comprises: an input layer, a mode layer, a summation layer, and an output layer.
For example, a preset GA-probabilistic neural network is transversely constructed according to the features and labels of the waterlogging data set, wherein the preset GA-probabilistic neural network inputs 22 columns of time-sequence features and spatial feature data after being subjected to quantization , and outputs the prediction level of the waterlogging degree, namely, the level 1-5.
The training of the waterlogging prediction model comprises the following steps:
step S301, inputting the sample space-time characteristic data and the sample waterlogging point water logging height grade data into the preset GA-probabilistic neural network, and outputting a sample waterlogging degree prediction result by the preset GA-probabilistic neural network;
step S302, calculating the relative accuracy of the predicted value of the sample waterlogging degree prediction result and the measured value of the sample waterlogging degree prediction result, and performing precision evaluation according to the predicted relative accuracy;
and S303, repeating the step of inputting the sample space-time characteristic data and the sample waterlogging point water logging height grade data into the preset GA-probabilistic neural network until the training of the preset GA-probabilistic neural network meets a preset condition, and obtaining a waterlogging prediction model.
The step of assessing the accuracy based on the predicted relative accuracy may be preceded by the step of:
and S304, initializing parameters of the preset GA-probabilistic neural network, optimizing a smoothing coefficient of the probabilistic neural network by using a genetic algorithm to obtain an optimal hyperparameter, and training the preset GA-probabilistic neural network.
Specifically, parameters of the probabilistic neural network are initialized, operations such as encoding and decoding, fitness value calculation, selection, intersection, variation and the like are performed by utilizing a genetic algorithm, a smoothing coefficient of the GA-probabilistic neural network is optimized, an optimal hyper-parameter is obtained, and network training is performed.
The radial basis operation of the GA-probabilistic neural network is as follows:
wherein σ represents a smoothing coefficient, and is a value between 0 and 1, and the operation accuracy can be improved by adjusting, x is an input vector standardized to a unit length, and w is a value of a smoothing coefficientiFor weighting coefficients of class i levels that have been normalized to unit length, T is the matrix transpose and exp is an exponential function with a natural constant e as the base.
Specifically, 22 rows of space-time characteristic data to be tested obtained in a fusion mode in the step S202 are input into the waterlogging prediction model Net2 established in the step S30, and a waterlogging degree grade prediction result at the moment T1 below a certain waterlogging point is obtained, namely the T duration is 3 hours, and the 4 th-hour waterlogging degree of the certain waterlogging point is predicted to be 1-5 grades.
Referring to fig. 4, fig. 4 is a comparison graph of a predicted value and an actual value at a time T1 when a waterlogging point in a certain coastal city in embodiments, specifically, the waterlogging condition of the certain waterlogging point in the coastal city continues for 96 hours, the ordinate axis represents the waterlogging degree grade, the abscissa axis represents time, and the o point represents the actual value, and the o point represents the predicted value, from fig. 4, it can be known that the accuracy of the predicted relative accuracy can be evaluated through the difference between the predicted value and the actual value.
The urban waterlogging problem is analyzed and classified in a machine learning mode, and the network adopted for comparative evaluation is as follows:
the BP neural network is a 3-layer neural network, the input layer node is 4, the hidden layer node is 12, and the output layer node is 10;
the input layer node of the Elman neural network is 3, the hidden layer node is 10, the carrying layer node is 10, and the output layer node is 1; the input layer node of the probabilistic neural network is 22, the mode layer node is the number of training samples, the summation layer node is 6, and the output layer node is 1.
The invention also provides urban waterlogging prediction systems based on a neural network, as shown in fig. 5, fig. 5 is a module architecture diagram of the urban waterlogging prediction system based on the neural network in embodiments, the system includes a time series characteristic prediction result output module 100, a data fusion module 200 and a grade prediction result output module 300, wherein:
the time sequence characteristic prediction result output module 100 is used for inputting the time sequence characteristic data to be tested into a trained time sequence characteristic prediction model, and the time sequence characteristic prediction model outputs a time sequence characteristic prediction result, wherein the time sequence characteristic prediction result is the predicted time sequence characteristic data of moments under the time sequence characteristic data to be tested;
the data fusion module 200 is used for fusing the predicted time sequence characteristic data with the spatial characteristic data to obtain space-time characteristic data to be detected;
the level prediction result output module 300 is configured to input the to-be-detected time-space characteristic data into a trained waterlogging prediction model, where the waterlogging prediction model outputs a level prediction result of the waterlogging degree of the to-be-detected waterlogging point, which is specifically described in urban waterlogging prediction methods based on a neural network and is not described herein again.
In conclusion, the urban waterlogging prediction methods and systems based on the neural network, provided by the invention, have the advantages that a plurality of coastal urban waterlogging disaster-causing factors are integrated, a waterlogging data set comprising space-time characteristics and labels is constructed, the two networks are constructed for the data set to carry out fusion training, the prediction and prediction of the waterlogging degree at time below a waterlogging point are realized through the trained neural network, the accuracy is not lower than 92%, the precision is higher, the speed is high, the prediction is accurate, and the application value in the flood prevention and disaster reduction of coastal cities and the risk emergency management of smart cities is better.
It is to be understood that the invention is not limited to the examples described above, but that modifications and variations may be effected thereto by those of ordinary skill in the art in light of the foregoing description, and that all such modifications and variations are intended to be within the scope of the invention as defined by the appended claims.
Claims (10)
1, A neural network-based urban waterlogging prediction method, characterized in that the method comprises the steps of:
inputting the time sequence feature data to be tested into a trained time sequence feature prediction model, and outputting a time sequence feature prediction result by the time sequence feature prediction model, wherein the time sequence feature prediction result is the predicted time sequence feature data at moment under the time sequence feature data to be tested;
fusing the predicted time sequence characteristic data with the spatial characteristic data to obtain space-time characteristic data to be measured;
and inputting the space-time characteristic data to be detected into a trained waterlogging prediction model, and outputting a grade prediction result of the waterlogging degree of the waterlogging point to be detected by the waterlogging prediction model.
2. The urban waterlogging prediction method based on the neural network as claimed in claim 1, wherein before inputting the time series characteristic data to be tested into the time series characteristic prediction model, the method further comprises the steps of:
constructing an inland inundation data set, wherein the inland inundation data set comprises time-space characteristic data and inland inundation point water immersion height grade data, and the time-space characteristic data comprises time sequence characteristic data and space characteristic data;
and inputting the time sequence characteristic data into a preset GA-Elman neural network, and obtaining the time sequence characteristic prediction model after training.
3. The urban waterlogging prediction method based on the neural network as claimed in claim 2, wherein before inputting the spatiotemporal feature data to be tested into the waterlogging prediction model, the method further comprises the following steps:
and inputting the space-time characteristic data and the water logging height grade data of the waterlogging point into a preset GA-probabilistic neural network, and training to obtain the waterlogging prediction model.
4. The neural network-based urban waterlogging prediction method of claim 2, wherein said step of constructing a waterlogging dataset comprises:
extracting the relevant space-time characteristic data caused by waterlogging as characteristics;
extracting water logging height grade data of waterlogging points, judging the waterlogging degree grade according to the water logging height grade data of the waterlogging points, and taking the waterlogging degree grade as a label;
and constructing a sample data table according to the features and the tags to obtain the waterlogging data set.
5. The method for urban waterlogging prediction based on neural network as claimed in claim 2, wherein the preset GA-Elman neural network comprises: the device comprises an input layer, a hidden layer, a carrying layer and an output layer; the training step of the time sequence characteristic prediction model comprises the following steps:
inputting the multi-dimensional sample time sequence characteristic data with the time length of T into a preset GA-Elman neural network, and outputting multi-dimensional sample time sequence characteristic prediction results of T1 at next moments by the preset GA-Elman neural network, wherein the time length of T is 3-5 hours;
calculating the average absolute relative deviation between the predicted value of the sample time series characteristic prediction result at the next moments T1 and the measured value of the sample time series characteristic prediction result at the next moments T1, and carrying out precision evaluation according to the predicted average absolute relative deviation;
and repeating the step of inputting the multi-dimensional and T-duration sample time sequence characteristic data into a preset GA-Elman neural network until the training of the preset GA-Elman neural network meets a preset condition, and obtaining the time sequence characteristic prediction model.
6. The method of claim 3, wherein the pre-set GA-probabilistic neural network comprises: an input layer, a mode layer, a summation layer, and an output layer; the training of the waterlogging prediction model comprises the following steps:
inputting the sample time-space characteristic data and the sample waterlogging point water immersion height grade data into the preset GA-probabilistic neural network, and outputting a sample waterlogging degree prediction result by the preset GA-probabilistic neural network;
calculating the relative accuracy of the predicted value of the sample waterlogging degree prediction result and the measured value of the sample waterlogging degree prediction result, and carrying out precision evaluation according to the predicted relative accuracy;
and repeating the step of inputting the sample space-time characteristic data and the sample waterlogging point water immersion height grade data into the preset GA-probabilistic neural network until the training of the preset GA-probabilistic neural network meets a preset condition, and obtaining the waterlogging prediction model.
7. The urban waterlogging prediction method based on the neural network as claimed in claim 1, wherein the step of fusing the prediction time series feature data with the spatial data to obtain the spatiotemporal feature data to be measured comprises:
inputting multi-dimensional time sequence feature data to be detected with T duration into the time sequence feature prediction model, and outputting time sequence feature prediction results to be detected at moments T1 by the time sequence feature prediction model;
and (3) performing inverse regression on the predicted time sequence characteristic data to be detected at the next time T1, and fusing the time sequence characteristic data to be detected with the spatial characteristic data to form the time-space characteristic data to be detected and perform regression .
8. The neural network-based urban waterlogging prediction method of claim 5, wherein said step of assessing accuracy based on predicted mean absolute relative deviation is preceded by the steps of:
initializing parameters of the preset GA-Elman neural network, optimizing weight thresholds from an input layer to a hidden layer, weight thresholds from the hidden layer to an output layer and weight thresholds of a carrying layer by utilizing a genetic algorithm, and training the preset GA-Elman neural network.
9. The neural network-based urban waterlogging prediction method of claim 6, wherein said step of assessing the accuracy based on the relative accuracy of the prediction further comprises:
initializing the parameters of the preset GA-probabilistic neural network, optimizing the smooth coefficient of the probabilistic neural network by using a genetic algorithm to obtain the optimal hyperparameter, and training the preset GA-probabilistic neural network.
10. The neural network-based urban waterlogging prediction method of claim 5, wherein said step of assessing accuracy based on predicted mean absolute relative deviation is preceded by the steps of:
inputting characteristic data of a time window after multidimensional, T-duration and normalization processing into the preset GA-Elman neural network, wherein the preset GA-Elman neural network outputs a sample time sequence characteristic prediction result of the multidimensional and next moments T1;
the time window characteristic data with the time length of T slides backwards along with the time, and the sample characteristic data is updated;
adding hours later sample characteristic data, and removing corresponding hours old sample characteristic data;
and the time window continues to slide backwards along with time until the training of the time sequence characteristic prediction model is finished.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910885437.0A CN110738355B (en) | 2019-09-19 | 2019-09-19 | Urban waterlogging prediction method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910885437.0A CN110738355B (en) | 2019-09-19 | 2019-09-19 | Urban waterlogging prediction method based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110738355A true CN110738355A (en) | 2020-01-31 |
| CN110738355B CN110738355B (en) | 2023-07-04 |
Family
ID=69269344
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910885437.0A Active CN110738355B (en) | 2019-09-19 | 2019-09-19 | Urban waterlogging prediction method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110738355B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111368397A (en) * | 2020-02-04 | 2020-07-03 | 中国水利水电科学研究院 | Method and device for predicting waterlogging risk |
| CN111370122A (en) * | 2020-02-27 | 2020-07-03 | 西安交通大学 | Knowledge guidance-based time sequence data risk prediction method and system and application thereof |
| CN111582541A (en) * | 2020-03-27 | 2020-08-25 | 合肥学院 | Firefly algorithm-based inland inundation model prediction method |
| CN111882830A (en) * | 2020-07-31 | 2020-11-03 | 珠江水利委员会珠江水利科学研究院 | Urban waterlogging monitoring, forecasting and early warning method, device and system and storage medium |
| CN113239614A (en) * | 2021-04-22 | 2021-08-10 | 西北工业大学 | Atmospheric turbulence phase space-time prediction algorithm |
| CN113344305A (en) * | 2021-08-05 | 2021-09-03 | 中国水利水电科学研究院 | Rapid prediction method for rainstorm waterlogging event |
| CN114970315A (en) * | 2022-04-19 | 2022-08-30 | 河海大学 | Urban accumulated water simulation and rapid prediction method based on spatial dynamic characteristic deep learning |
| CN115471078A (en) * | 2022-09-15 | 2022-12-13 | 镇江市勘察测绘研究院有限公司 | Flood risk point assessment method and device based on urban water affair system |
| CN117688790A (en) * | 2024-02-04 | 2024-03-12 | 中科智远科技江苏有限公司 | Waterlogging flooding point pre-judging method based on multidimensional real-time monitoring and big data analysis |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103279671A (en) * | 2013-06-03 | 2013-09-04 | 南京大学 | Urban water disaster risk prediction method based on RBF (radial basis function) neural network-cloud model |
| CN104239489A (en) * | 2014-09-05 | 2014-12-24 | 河海大学 | Method for predicting water level by similarity search and improved BP neural network |
| CN106373070A (en) * | 2016-08-29 | 2017-02-01 | 西安理工大学 | Four-prevention method for responding to city rainstorm waterlogging |
| CN106650767A (en) * | 2016-09-20 | 2017-05-10 | 河海大学 | Flood forecasting method based on cluster analysis and real time correction |
| CN106970986A (en) * | 2017-03-29 | 2017-07-21 | 武汉大学 | Urban waterlogging influence degree method for digging and system based on deep learning |
| CN108280553A (en) * | 2018-02-24 | 2018-07-13 | 中山大学 | Regional Torrent Risk Zonation based on GIS- Artificial neural network ensembles and prediction technique |
| CN108288109A (en) * | 2018-01-11 | 2018-07-17 | 安徽优思天成智能科技有限公司 | Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks |
-
2019
- 2019-09-19 CN CN201910885437.0A patent/CN110738355B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103279671A (en) * | 2013-06-03 | 2013-09-04 | 南京大学 | Urban water disaster risk prediction method based on RBF (radial basis function) neural network-cloud model |
| CN104239489A (en) * | 2014-09-05 | 2014-12-24 | 河海大学 | Method for predicting water level by similarity search and improved BP neural network |
| CN106373070A (en) * | 2016-08-29 | 2017-02-01 | 西安理工大学 | Four-prevention method for responding to city rainstorm waterlogging |
| CN106650767A (en) * | 2016-09-20 | 2017-05-10 | 河海大学 | Flood forecasting method based on cluster analysis and real time correction |
| CN106970986A (en) * | 2017-03-29 | 2017-07-21 | 武汉大学 | Urban waterlogging influence degree method for digging and system based on deep learning |
| CN108288109A (en) * | 2018-01-11 | 2018-07-17 | 安徽优思天成智能科技有限公司 | Motor-vehicle tail-gas concentration prediction method based on LSTM depth space-time residual error networks |
| CN108280553A (en) * | 2018-02-24 | 2018-07-13 | 中山大学 | Regional Torrent Risk Zonation based on GIS- Artificial neural network ensembles and prediction technique |
Non-Patent Citations (4)
| Title |
|---|
| APOORV GUPTA ET AL.: "\"Urban Waterlogging Detection and Severity Prediction using Artificial Neural Networks\"", pages 42 - 49 * |
| 张梦 等: ""PSO-ESN在城市内涝点降雨积水预测中的应用"", no. 6, pages 56 - 59 * |
| 潘银 等: ""利用深度学习模型进行城市内涝影响分析"", vol. 44, no. 44, pages 132 - 138 * |
| 饶群华;夏洪;: "基于神经网络的洪水水位预测算法", 信息技术, no. 02, pages 42 - 44 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111368397B (en) * | 2020-02-04 | 2021-02-12 | 中国水利水电科学研究院 | Method and device for predicting waterlogging risk |
| CN111368397A (en) * | 2020-02-04 | 2020-07-03 | 中国水利水电科学研究院 | Method and device for predicting waterlogging risk |
| CN111370122B (en) * | 2020-02-27 | 2023-12-19 | 西安交通大学 | Time sequence data risk prediction method and system based on knowledge guidance and application thereof |
| CN111370122A (en) * | 2020-02-27 | 2020-07-03 | 西安交通大学 | Knowledge guidance-based time sequence data risk prediction method and system and application thereof |
| CN111582541A (en) * | 2020-03-27 | 2020-08-25 | 合肥学院 | Firefly algorithm-based inland inundation model prediction method |
| CN111882830A (en) * | 2020-07-31 | 2020-11-03 | 珠江水利委员会珠江水利科学研究院 | Urban waterlogging monitoring, forecasting and early warning method, device and system and storage medium |
| CN113239614A (en) * | 2021-04-22 | 2021-08-10 | 西北工业大学 | Atmospheric turbulence phase space-time prediction algorithm |
| CN113344305A (en) * | 2021-08-05 | 2021-09-03 | 中国水利水电科学研究院 | Rapid prediction method for rainstorm waterlogging event |
| CN113344305B (en) * | 2021-08-05 | 2021-10-29 | 中国水利水电科学研究院 | Rapid prediction method for rainstorm waterlogging event |
| CN114970315A (en) * | 2022-04-19 | 2022-08-30 | 河海大学 | Urban accumulated water simulation and rapid prediction method based on spatial dynamic characteristic deep learning |
| CN114970315B (en) * | 2022-04-19 | 2023-08-04 | 河海大学 | Urban ponding simulation and rapid prediction method based on space dynamic characteristic deep learning |
| CN115471078A (en) * | 2022-09-15 | 2022-12-13 | 镇江市勘察测绘研究院有限公司 | Flood risk point assessment method and device based on urban water affair system |
| CN115471078B (en) * | 2022-09-15 | 2024-02-02 | 镇江市勘察测绘研究院有限公司 | Flood risk point assessment method and device based on urban water affair system |
| CN117688790A (en) * | 2024-02-04 | 2024-03-12 | 中科智远科技江苏有限公司 | Waterlogging flooding point pre-judging method based on multidimensional real-time monitoring and big data analysis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110738355B (en) | 2023-07-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110738355A (en) | urban waterlogging prediction method based on neural network | |
| CN106650767B (en) | Flood forecasting method based on cluster analysis and real-time correction | |
| CN109658695B (en) | Multi-factor short-term traffic flow prediction method | |
| CN111310968B (en) | LSTM neural network circulating hydrologic forecasting method based on mutual information | |
| CN108280553B (en) | Mountain torrent disaster risk zoning and prediction method based on GIS-neural network integration | |
| CN111832814A (en) | Air pollutant concentration prediction method based on graph attention machine mechanism | |
| CN110047291B (en) | Short-term traffic flow prediction method considering diffusion process | |
| CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
| CN112506990B (en) | Hydrological data anomaly detection method based on spatiotemporal information | |
| CN109840587B (en) | Reservoir warehousing flow prediction method based on deep learning | |
| CN109086540B (en) | Method and device for constructing tropical cyclone path forecasting model | |
| CN111008337B (en) | Deep attention rumor identification method and device based on ternary characteristics | |
| CN108133225A (en) | A kind of icing flashover fault early warning method based on support vector machines | |
| CN112149887A (en) | PM2.5 concentration prediction method based on data space-time characteristics | |
| CN114565124A (en) | Ship traffic flow prediction method based on improved graph convolution neural network | |
| CN110276477B (en) | Flood forecasting method based on hierarchical Bayesian network and incremental learning | |
| CN110781266A (en) | Urban perception data processing method based on time-space causal relationship | |
| CN115267945A (en) | Thunder and lightning early warning method and system based on graph neural network | |
| Wegayehu et al. | Multivariate streamflow simulation using hybrid deep learning models | |
| CN113516304A (en) | Space-time joint prediction method and device for regional pollutants based on space-time graph network | |
| CN113985496B (en) | Storm surge intelligent forecasting method based on LSTM-GM neural network model | |
| CN115310532A (en) | Basin multipoint prediction early warning method based on space-time association mixed deep learning model | |
| CN114594532A (en) | Method and device for predicting cold weather, electronic equipment and computer readable medium | |
| CN116434479A (en) | Regional flood disaster grade prediction and early warning method and system | |
| CN115829150A (en) | Accumulated water prediction system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |