A water level prediction method and device for hydraulic engineering

Abstract

The application discloses a water level prediction method and device for water conservancy projects. The method comprises the following steps: reading water level time series data of a reservoir at a preset interval; obtaining external influence data of the water level of the reservoir and performing feature extraction to obtain external influence features and store the external influence features to a block chain; performing correlation analysis on the external influence features and the water level of the reservoir, and assigning weights to the external influence features according to the analysis result; inputting the water level time series data and the external influence features carrying the weights into a pre-trained water level prediction model to obtain an initial predicted water level; calculating first internal influence data of the water level of the reservoir through the inflow of the reservoir according to water quality information and geological information of inflow rivers of the reservoir; measuring underground water level data and underground geological data of the reservoir to determine second internal influence data of the water level of the reservoir; and correcting the initial predicted water level through a water level correction model based on the first internal influence data and the second internal influence data to obtain a final predicted water level.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a water level prediction method and device for water conservancy projects. BACKGROUND

[0002] Water conservancy projects have positive significance for rational utilization of water resources, protection of people's water for living, water for industry and agriculture, and reduction of the impact of natural disasters such as floods and droughts on people's production and life.

[0003] As an important part of water conservancy projects, reservoirs have multiple functions such as water storage, flow regulation, power generation, and flood control. Accurate prediction of reservoir water level fluctuations is of great significance for the scientific and rational utilization of water resources and the reduction of the impact of natural disasters such as floods and droughts.

[0004] However, in existing reservoir water level prediction schemes, most of them are realized by neural network models, and historical water level data and influencing factors of the reservoir are input into a pre-trained model to output the predicted water level result. However, in this prediction process, on the one hand, in order to improve the prediction accuracy, most of them are improved at the model level, and better prediction performance models or combined models are selected, while the selection of model input influencing factors is basically the same, and most of them only consider single factors such as rainfall, flood discharge, and evaporation. Other factors that may affect the water level of the reservoir, such as the underground water level of the reservoir and the geographical environmental factors around the reservoir or the geographical environmental factors around the river flowing into the reservoir, are not considered or directly excluded. Although these other factors do not have as high an impact on the water level of the reservoir as rainfall, flood discharge, and evaporation, the water level of the reservoir will still change to some extent under the influence of these other factors. Therefore, including these factors in the consideration of influencing factors has certain significance for improving the accuracy of water level prediction. On the other hand, existing schemes for predicting water level by model generally extract features of influencing factors within the model, and use the performance of the model to extract features that affect the water level. In the case of a certain model performance, the feature extraction process is not targeted. At the same time, after the feature extraction is completed, the model generally directly uses the extracted influencing features to predict the water level, only considering that these features will affect the water level, but not considering the degree of influence of each feature on the water level, which will inevitably affect the accuracy of the prediction model. SUMMARY

[0005] The present application provides a water level prediction method and device for water conservancy projects to solve at least one of the above technical problems.

[0006] The present application adopts the following technical solutions:

[0007] A water level prediction method for a water conservancy project, the water conservancy project comprising at least a reservoir, the method comprising: reading water level time series data of the reservoir at preset intervals; obtaining external influence data of the reservoir water level, and performing feature extraction on the external influence data to obtain external influence features and store them to a blockchain, then performing correlation analysis on the external influence features and the reservoir water level, and assigning weights to the external influence features according to the analysis results; inputting the water level time series data and the external influence features carrying the weights into a pre-trained water level prediction model to obtain an initial predicted water level; calculating first internal influence data of the reservoir water level according to water quality information and geological information of a river flowing into the reservoir, through the inflow of the reservoir; measuring underground water level data and underground geological data of the reservoir, and determining second internal influence data of the reservoir water level through the measurement results; correcting the initial predicted water level through a water level correction model based on the first internal influence data and the second internal influence data to obtain a final predicted water level.

[0008] In a possible implementation manner of the present application, the method further comprises: reading geographical position data of the reservoir, and obtaining meteorological data of the geographical position data between a current time node and a prediction time node; extracting rain condition data and temperature and humidity data from the meteorological data to perform feature extraction on the rain condition data to obtain rainfall features of the reservoir corresponding to the prediction time node, and perform feature extraction on the temperature and humidity data to obtain evaporation features of the reservoir corresponding to the prediction time node in combination with measured water temperature data; determining river information associated with the reservoir, and determining inflow features of the reservoir based on the river information and meteorological information corresponding to the river information between the current time node and the prediction time node; obtaining water scheduling schemes of the reservoir between the current time node and the prediction time node to extract outflow features of the reservoir according to the water scheduling schemes; and constructing external influence features of the reservoir water level based on the rainfall features, the evaporation features, the inflow features and the outflow features.

[0009] In a possible implementation of the present application, the extraction / determination process of the external influence feature includes: when the geographic location data is administrative division data, obtaining a rainfall dynamic cloud chart of the administrative division data between a current time node and a prediction time node, and dynamically adding the temperature and humidity data into the rainfall dynamic cloud chart; determining a reservoir area in the administrative division data, and intercepting the corresponding reservoir area in the rainfall dynamic cloud chart to obtain a rainfall dynamic cloud chart of the reservoir area; setting a plurality of sampling areas in the rainfall dynamic cloud chart corresponding to the reservoir area according to spatial positions and / or rainfall distributions, and generating a corresponding two-dimensional array for each sampling area, wherein a first-dimensional element in the two-dimensional array is used to indicate rainfall time series data corresponding to the current sampling area, and a second-dimensional element in the two-dimensional array is used to indicate temperature and humidity time series data corresponding to the current sampling area; extracting the first-dimensional elements in all two-dimensional arrays, superimposing rainfall data corresponding to the same time, and outputting the rainfall feature; extracting the second-dimensional elements in all two-dimensional arrays, and outputting temperature and humidity features after mean value processing of temperature and humidity data corresponding to the same time; obtaining current water level data of the reservoir in the water level time series data, and updating the current water level data through the rainfall feature; determining a temperature influence coefficient α and a humidity influence coefficient β according to the water temperature data and the temperature and humidity features, and calculating an evaporation amount feature of the reservoir through the updated current water level data and a horizontal area of the reservoir by the following formula:

[0010]

[0011] wherein E i is the evaporation amount feature, α is the temperature influence coefficient, β is the humidity influence coefficient, f(W i is a wind speed function obtained according to meteorological data, S is the horizontal area of the reservoir, Q0 is the current water level data, P0 is the current water vapor pressure, T0 is the water temperature data, T i is the temperature feature, U i is the humidity feature, C0 is an adjustment parameter, i is any time node from the current time node to the prediction time node, and n is the number of all time nodes from the current time node to the prediction time node.

[0012] In a possible implementation manner of the present application, the extraction / determination process of the external influence feature further includes: obtaining historical runoff data and historical meteorological data of the river information in a time period corresponding to the current time node and the predicted time node, the historical runoff data being a mean value of the runoff data in a historical time period corresponding to the previous N2 years of the time period, and the historical meteorological data being meteorological data in a historical time period corresponding to the previous N1 years of the time period, where N1≥N2, and the initial value of N2 is 1; performing similarity calculation on the historical meteorological data and the meteorological information, the similarity calculation including at least rainfall amount similarity calculation and temperature and humidity similarity calculation; when a result of the similarity calculation is greater than or equal to a preset similarity threshold, determining the historical flow data as the runoff flow feature, and when the result of the similarity calculation is less than the preset similarity threshold, then performing N2=N2+1, and performing similarity calculation again until the result of the similarity calculation is greater than or equal to the preset similarity threshold and / or the value of N2 is greater than 5; identifying, in the water regulation scheme, the outflow, the outflow including at least any one or more of irrigation water, power generation water, navigation water, urban water supply, discharge flow and leakage loss of the reservoir in the time period corresponding to the current time node and the predicted time node; and obtaining the outflow feature after time series processing of the outflow.

[0013] In a possible implementation manner of the present application, the correlation analysis on the external influence feature and the reservoir water level includes: extracting, in the water level time series data, historical water level data Q[q1, q2,..., qn] corresponding to a plurality of time nodes between the current time node and the predicted time node, and obtaining historical rainfall amount features R[r1, r2,..., rn], historical evaporation amount features E[e1, e2,..., ei], historical runoff flow features D[d1, d2,..., dn] and historical outflow features F[f1, f2,..., fn] corresponding to the historical water level data respectively; calculating mean values and variances corresponding to Q, R, E, D and F respectively, calculating covariance between Q and R, Q and E, Q and D and Q and F based on the mean values corresponding to Q, R, E, D and F respectively, and calculating standard deviations between Q and R, Q and E, Q and D and Q and F based on the variances corresponding to Q, R, E, D and F respectively; determining a ratio of the covariance and the standard deviation as a correlation coefficient between Q and R, Q and E, Q and D and Q and F, and generating a correlation coefficient matrix; and after normalizing elements in the correlation coefficient matrix, corresponding values of the elements are assigned as weights of the external influence feature.

[0014] In a possible implementation manner of the present application, the water quality information at least includes physical entity content, and the geological information at least includes river sediment content; the calculation process of the first internal influence data includes: obtaining a historical mean inflow of the incoming river in a non-rainy season, and calculating a predicted inflow of the incoming river in a time period corresponding to a preset time node from a current time node according to the historical mean inflow; extracting rainfall data of a region corresponding to the incoming river in a dynamic rainfall map, and accumulating the predicted inflow according to the rainfall data and the predicted inflow to obtain an actual inflow; obtaining vegetation coverage in a preset range on both sides of the incoming river; when the vegetation coverage is higher than a preset coverage threshold, predicting the physical entity content of the incoming river under the influence of the rainfall data through the vegetation coverage, the physical entity at least including any one or more of leaves, branches and trunks, and when the vegetation coverage is lower than the preset coverage threshold, taking the physical entity content as zero; determining non-water mass quality brought into the reservoir by the incoming river according to a first product between the river sediment content and the actual inflow, and a second product between the physical entity content and the actual inflow; and calculating first influence data of the incoming river on the water level of the reservoir through the non-water mass quality and a reservoir capacity of the reservoir.

[0015] In a possible implementation manner of the present application, the underground geological data at least includes a water absorption index, and the second internal influence data is determined by the following formula: wherein, λ is the second internal influence data, k is the water absorption index, Q0 is an initial predicted water level, Q u is underground water level data, P c is a reservoir bottom edge pressure, P u is an underground water level pressure, and R is a reservoir bottom shape parameter.

[0016] In a possible implementation manner of the present application, the water level prediction model adopts a time series model and / or a neural network model; and the correcting of the initial predicted water level includes: performing hyperparameter optimization on the water level prediction model by using a particle swarm algorithm; updating the water level prediction model based on an optimization result, and performing secondary training on the updated water level prediction model to obtain a water level correction model; inputting the initial predicted water level, the first internal influence data and the second internal influence data into the water level correction model to perform secondary prediction on the initial predicted water level, to obtain a final predicted water level.

[0017] In a possible implementation of the present application, the correcting the initial predicted water level further includes: determining a positive influence parameter corresponding to the first internal influence data and a negative influence parameter corresponding to the second internal influence data based on the initial predicted water level, wherein the influence parameter is used to indicate the influence degree of the first internal influence data and / or the second internal influence data on the upward and downward floating of the initial predicted water level; calculating the product of the initial predicted water level and the positive influence parameter and the negative influence parameter, and determining the final predicted water level according to the product.

[0018] In another aspect, the present application also provides a water level prediction device for water conservancy projects, which comprises: at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, so that the at least one processor can execute a water level prediction method for water conservancy projects as described in any one of the above embodiments.

[0019] The water level prediction method and device for water conservancy projects provided by the present application have high accuracy or prediction precision, which is embodied in the following two aspects:

[0020] On the basis of considering the factors such as rainfall, evaporation, inflow and outflow of reservoirs, which directly affect or have a high influence degree on the water level of reservoirs, the present application further considers other factors such as river weather information, river information, underground water level information and underground geological information, which may affect the water level of reservoirs, so that the selection or consideration of influence factors is more comprehensive. At the same time, in order not to affect the prediction precision of the model, the present application selects to input the factors such as rainfall, evaporation, inflow and outflow of reservoirs, which have a high influence degree, for prediction in the initial prediction process, and then uses the above other possible influence factors to correct or adjust the initial predicted water level after obtaining the initial predicted water level. In this way, the water level prediction model can be ensured not to be affected by too many input factors, the initial water level with high accuracy can be predicted, and at the same time, the other possible influence factors can be considered, so that they can correct the predicted water level, and further improve the precision of the water level prediction result.

[0021] In another aspect, the feature extraction process of the influencing factors is pre-processed, so that different influencing factors can be extracted more specifically. For example, for rainfall, the traditional prediction process is to input the rainfall of the entire region where the reservoir is located into the model as input data. The rainfall data input into the model will inevitably increase relative to the corresponding rainfall. By introducing the dynamic rainfall cloud map, the actual rainfall of the reservoir area can be directly obtained, so that only this part of the rainfall feature is input into the model for prediction, making the input of the model more accurate. In this way, the correlation between the data input into the model and the water level is higher, so that the prediction result of the model is more accurate. At the same time, after the feature extraction is completed, the correlation analysis between the features and the water level is performed, so that the features input into the model carry the weight determined according to the correlation analysis result, and the influence degree of each factor on the water level of the reservoir is characterized and reflected, so that the model can pay attention to the influence degree of each factor during prediction, and the prediction accuracy of the model can be further improved. The above two aspects can obtain a prediction result with higher accuracy, ensure the prediction accuracy of the water level of the reservoir, and provide a more valuable decision basis for the scheduling, early warning and the like of the reservoir. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor. In the drawings:

[0023] Figure 1 A flow chart of a water level prediction method for water conservancy projects provided by the present application;

[0024] Figure 2 A structural schematic diagram of a water level prediction device for water conservancy projects provided by the present application. DETAILED DESCRIPTION

[0025] In order to make the person skilled in the art better understand the technical solutions in the present application, the technical solutions in the present application will be described clearly and completely in the following with reference to the drawings in the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0026] The application provides a water level prediction method and device for water conservancy projects, which realizes accurate prediction of reservoir water level in water conservancy projects by prepositioning the feature extraction process of external influence data and correcting the prediction results and / or prediction model, thereby providing decision basis for flood warning downstream of the reservoir.

[0027] The method in the application will be described in detail below with reference to the drawings.

[0028] Figure 1 As shown in the flowchart of the water level prediction method for water conservancy projects provided by the application, Figure 1 the water level prediction method in the application at least includes the following execution steps:

[0029] Step 101: reading the water level time series data of the reservoir at preset intervals.

[0030] The reservoir water level prediction scheme in the application is realized based on a time series model or a neural network model, for example, an initial prediction model such as an LSTM model. Since the reservoir water level has a certain correlation in time, the application first reads the water level time series data of the reservoir when predicting.

[0031] In one example, the water level time series data is obtained by extracting the water level data of each historical time node forwardly at preset intervals from the current time node as the starting point, and then sorting in time sequence in ascending order. For example, the current time node is t0 days, and the preset time interval is 1 day, so the water level time series data read is: [...,t 0-2 ,t 0-1 ,t0] and the data before t 0-2 is determined by the length of the read water level time series data. Of course, according to the length requirement of the water level time series data, it can also be selected to read from which time node, for example, it is required to read the water level data within 10 days to constitute the water level time series data, and the preset time interval is 1 day, so the finally obtained water level time series data is: [t 0-9 ,...,t 0-2 ,t 0-1 ,t0].

[0032] It should be noted that if the water level time series data is in units of days, then preferably, each water level data corresponds to the water level data at the same time point in a day.

[0033] Step 102: obtaining external influence data of the reservoir water level, and performing feature extraction on the external influence data to obtain external influence features and store them to the blockchain.

[0034] In a possible implementation of the present application, the external influence data of the reservoir water level refers to rainfall data, evaporation data, inflow data and outflow data which have a greater influence on the water level of the reservoir, wherein each data corresponds to rainfall features, evaporation features, inflow features and outflow features after feature extraction. The present application pre-positions the feature extraction process of the model through this step, or this step can also be considered as processing the obtained original data to obtain input data more suitable for the model prediction process, and the feature extraction process or processing process of each data is targeted, so that the processed data or extracted features can better represent the influence on the water level of the reservoir.

[0035] In addition, the features proposed in the present application can also be stored in the blockchain, and the characteristics of the blockchain make the data tamper-proof, which can support the tracing of the influence data of the water level of the reservoir.

[0036] Specifically, the acquisition process and feature extraction process of each influence data are as follows:

[0037] 1) For rainfall data:

[0038] The present application first acquires the geographical position data of the reservoir, which can be latitude and longitude data or administrative division data, and then acquires the meteorological data of the area where the reservoir is located between the current time node and the prediction time node according to the acquired geographical position, for example, the current time node is the current day, and the prediction time node is 3 days later, so the meteorological data from the current day to the future 3 days is acquired. The way to acquire it can be through the meteorological monitoring station around the reservoir, or the meteorological data on the mobile terminal. The rain data and temperature and humidity data are extracted from the acquired meteorological data, the rain data corresponds to the rainfall data, and the temperature and humidity data correspond to the evaporation data, so the rainfall features are obtained by feature extraction on the rain data.

[0039] Specifically, when the geographic position data corresponding to the reservoir is administrative division data, the dynamic rainfall cloud chart of the administrative division data between the current time node and the predicted time node is obtained, and then the corresponding reservoir area is intercepted in the dynamic rainfall cloud chart based on the corresponding area position of the reservoir in the administrative division data, to obtain the dynamic rainfall cloud chart of the reservoir area. For example, the reservoir is located in X city, and the rainfall data generally obtained is the rainfall cloud chart displayed on the X city map. In order to ensure the influence accuracy of rainfall on the water level of the reservoir, the application only counts the rainfall corresponding to the reservoir area, so the reservoir area is intercepted on the X city map, and the dynamic rainfall cloud chart of the reservoir is obtained. It should be noted that the dynamic here means that for a certain position point of the reservoir, the cloud chart displays all the rainfall information of the position point in the time period between the current time node and the predicted time node. Then, a plurality of sampling areas are set in the dynamic rainfall cloud chart corresponding to the reservoir area according to the spatial position and / or rainfall distribution, and a corresponding two-dimensional array is generated for each sampling area. The first dimension element in the two-dimensional array is used to indicate the rainfall time series data corresponding to the current sampling area. Generally speaking, the rainfall degree displayed on the dynamic rainfall cloud chart corresponding to the reservoir area is similar and has little difference, so the sampling area is preferably set according to the spatial position. Assuming that there are four sampling points or sampling areas in the reservoir area, then according to the dynamic rainfall cloud chart of the reservoir, four two-dimensional data corresponding to the sampling areas can be extracted, and then the first dimension elements in the four two-dimensional data are extracted. Each group of first dimension elements contains rainfall data of the sampling area from the current time node to the predicted time node, so the rainfall feature is obtained by superimposing the rainfall data corresponding to the same time in the extracted first dimension elements.

[0040] The visualization of the above process is to display the dynamic rainfall cloud chart corresponding to the reservoir on the upper layer of the visualization plane map of the reservoir, display the two-dimensional array analyzed from the dynamic rainfall cloud chart on the upper layer, extract the first dimension elements in the two-dimensional array to form a new array, superimpose the elements of the same time position in each new data in the upper layer, display the dynamic superimposition process, and display the rainfall feature obtained by superimposition on the uppermost layer.

[0041] It should be noted that even if the rainfall of the reservoir area has little difference, in order to improve the prediction accuracy of the water level of the reservoir, the data accuracy needs to be ensured from the aspect of features or input data, so the application superimposes the rainfall of each area by setting sampling areas to replace the rainfall of the whole area, which is equivalent to dividing the reservoir into finer dimensions, so that the rainfall data is more accurate.

[0042] 2) For evaporation data:

[0043] The evaporation amount of the reservoir is mainly related to the temperature and humidity data corresponding to the meteorological information of the reservoir. Therefore, the temperature and humidity data are also added to the dynamic rainfall cloud chart of the reservoir area, and the temperature and humidity data corresponding to the same time are added to the rainfall data at the same time during the addition. In this way, the second dimension elements of the two-dimensional array can be used to represent the temperature and humidity data. In an example, the temperature and humidity data here can be a temperature range and / or a humidity range, and in order to make the final temperature and humidity characteristics represent the temperature and humidity of the entire reservoir area, not the temperature and humidity of a sampling area in the reservoir, the application will average the temperature and humidity data corresponding to the same time in the extracted second dimension elements, and obtain the temperature and humidity characteristics after the processing.

[0044] Further, the current water level data of the current time node is extracted from the water level time series data of the reservoir, and the current water level data is updated by the aforementioned rainfall characteristics. The purpose of this step is to consider that the rainfall falls entirely into the reservoir area, that is, the evaporation amount of the rainwater during the falling process is not considered, but the evaporation amount after the water level change caused by the rainfall is considered after the rainfall falls entirely. At the same time, the evaporation amount is also related to the water temperature data, so when determining the evaporation amount characteristics, the application first calculates the temperature influence coefficient and the humidity influence coefficient according to the water temperature data and the temperature and humidity characteristics. It should be noted that the temperature influence coefficient has a positive effect on the evaporation amount characteristics of the reservoir, and the humidity influence coefficient has a negative effect on the evaporation amount characteristics of the reservoir; then, the evaporation amount characteristics are calculated based on the following formula:

[0045]

[0046] wherein E i is the evaporation amount characteristics, a is the temperature influence coefficient, and β is the humidity influence coefficient; f(W i ) is a wind speed function obtained according to meteorological data, which can be obtained according to existing algorithms; S is the horizontal area of the reservoir, Q0 is the current water level data, P0 is the current water vapor pressure, and T0 is the water temperature data, which are all known quantities; T i is the temperature characteristics, U i is the humidity characteristics, and C0 is an adjustment parameter, which is a constant, i is any time node from the current time node to the prediction time node, and n is the number of all time nodes from the current time node to the prediction time node.

[0047] It should be noted that the calculated evaporation amount characteristics are also time series characteristics, that is, assuming that the evaporation amount characteristics are represented by an array, there is a time sequence relationship between the elements in the array, and the time corresponding to each element is each time node between the current time node and the prediction time node, and these time nodes correspond to the time nodes in the rainfall characteristics and run through the calculation process of the evaporation amount characteristics.

[0048] 3) For the inflow data of the reservoir:

[0049] The inflow data of the reservoir in the present application only considers the river data having a flow-in relationship with the reservoir, and considers that the rainfall around the river and / or the rainfall in the area around the reservoir also affects the water level of the reservoir through the river, i.e., the rainfall first falls into the river and then is brought into the water level by the river.

[0050] In a possible implementation manner of the present application, the river information associated with the reservoir is first determined, and the association here refers to the river water flowing into the reservoir. The historical inflow data and the historical meteorological data of the river information in the time period corresponding to the current time node and the predicted time node are obtained, the historical inflow data here is the average of the inflow data in the historical time period corresponding to the previous N2 years, and the historical meteorological data is the meteorological data in the historical time period corresponding to the previous N1 years, wherein N1≥N2, and the initial value of N2 is 1; for example, the time period between the current time node and the preset time node is August 3-5, and the obtained historical inflow data can be the average of the inflow flow of August 3-5 in the past 1 year, and the historical meteorological data is also the meteorological data of August 3-5 of the previous year. Here, the unit is year, because the meteorological data or flow data at the same time node in different years in the same region are similar, and have certain reference value compared with other time.

[0051] Further, the similarity calculation is performed on the historical meteorological data and the meteorological information, and the similarity calculation process at least includes rainfall similarity calculation and temperature and humidity similarity calculation, which are also the two parameters with the greatest influence on the river reservoir water level in the meteorological data, and the similarity calculation process here can be realized by the existing technology or algorithm, which will not be described herein. If the result of the similarity calculation is greater than or equal to the preset similarity threshold, the historical flow data is determined as the inflow flow feature, and if the result of the similarity calculation is less than the preset similarity threshold, N2=N2+1 is performed, the purpose is to find a more similar year of meteorological data, and the similarity calculation is performed again, until the result of the similarity calculation is greater than or equal to the preset similarity threshold and / or the value of N2 is greater than 5, and the inflow flow feature is obtained.

[0052] 4) For the outflow data of the reservoir:

[0053] The application obtains the water scheduling scheme between the current time node and the predicted time node, extracts any one or more of the irrigation water, the power generation water, the navigation water, the city water supply, the discharge flow and the leakage loss in the water scheduling scheme, and the corresponding discharge flow data. Then, the discharge flow data corresponding to the same time are superimposed and then arranged in a positive sequence according to the time sequence, that is, the discharge flow characteristics of the reservoir are obtained.

[0054] Finally, the rainfall characteristics, the evaporation characteristics, the inflow characteristics and the outflow characteristics extracted through the above four processes are used to construct the external influence characteristics.

[0055] In step 103, the correlation between the external influence characteristics and the reservoir water level is analyzed, and the weight of the external influence characteristics is assigned according to the analysis result.

[0056] The external influence characteristics input into the water level prediction model for water level prediction mainly include the rainfall characteristics, the evaporation characteristics, the inflow characteristics and the outflow characteristics, so when the correlation analysis is performed, the relationship between the four characteristics and the reservoir water level is analyzed. Since the influence factors have certain regularity on the reservoir water level, the historical water level data Q [q1, q2, …, qn], the historical rainfall characteristics R [r1, r2, …, rn], the historical evaporation characteristics E [e1, e2, …, ei], the historical inflow characteristics D [d1, d2, …, dn] and the historical outflow characteristics F [f1, f2, …, fn] corresponding to a plurality of time nodes between the current time node and the predicted time node are used to perform the aforementioned correlation analysis process to calculate the correlation between the reservoir water level and the external influence characteristics. The correlation can be used to represent the influence degree of each external influence characteristic on the reservoir water level. In this way, no matter what model is used in the water level prediction model in the application, as long as the water level prediction model itself has a certain prediction accuracy, the influence degree of each influence factor can be considered when predicting the future water level of the reservoir, so that the prediction accuracy is further improved, and the applicability of the application is improved. It should be noted that the historical corresponding time nodes (for example, the current date is August 2, and the water level on August 5 is predicted, then the historical data of August 3-5 of the previous year is used for correlation analysis) are selected because the data at the same time of different years has higher similarity than the data at other time points, and the calculated result is more reliable.

[0057] Specifically, the correlation analysis process is as follows: first, the mean and variance corresponding to Q, R, E, D and F are calculated, then the covariance between Q and R, Q and E, Q and D, Q and F is calculated based on the mean corresponding to Q, R, E, D and F, and the standard deviation between Q and R, Q and E, Q and D, Q and F is calculated based on the variance corresponding to Q, R, E, D and F; then the ratio between the covariance and the standard deviation is determined as the correlation coefficient between Q and R, Q and E, Q and D, Q and F, and a correlation coefficient matrix is generated, which has four elements, each element corresponding to the correlation degree of different external influence characteristics and reservoir water level; finally, the elements in the correlation coefficient matrix are normalized, and the values of the elements are assigned to the weights of the external influence characteristics.

[0058] Step 104: inputting the water level time series data and the external influence characteristics with weights into the pre-trained water level prediction model to obtain an initial predicted water level.

[0059] After the execution of the foregoing three steps is completed, the water level time series data and the external influence characteristics with weights are input into the pre-trained water level prediction model to output an initial predicted water level. In one example, since the external influence characteristics in the present application are subjected to correlation analysis and are assigned different weights based on the analysis results to represent different influence degrees, the water level prediction model herein has a high probability of using several types of models commonly used in the current technology, and the requirement for the model is lower, making the adaptability of the present application higher; at the same time, the process of feature extraction is pre-positioned, which can also improve the training time and training difficulty of the model, making the model more lightweight.

[0060] Step 105: calculating first internal influence data of the reservoir water level according to the water quality information and geological information of the reservoir inflow river and the inflow flow of the reservoir.

[0061] In the present application, factors other than the foregoing several external influence characteristics that may affect the reservoir water level are roughly divided into two categories: one is the first internal influence data that has a positive influence on the reservoir water level, i.e., that can make the reservoir water level possibly rise, and the other is the second internal influence data that has a negative influence on the reservoir water level, i.e., that can make the reservoir water level possibly fall. Then, the two categories of data are used to correct the foregoing predicted initial predicted water level, so as to achieve a higher precision or accuracy of the water level prediction result based on a certain precision of the prediction result.

[0062] The first internal influence data in the present application is obtained by water quality information, geological information and inflow of the reservoir inflow river. Specifically, the historical average inflow of the inflow river in the dry season is obtained, and the predicted inflow of the inflow river in the time period corresponding to the current time node to the preset time node is calculated according to the historical average inflow, then the rainfall data of the corresponding area of the inflow river is extracted in the rainfall dynamic cloud map, and the actual inflow is obtained according to the rainfall data and the predicted inflow. This step is equivalent to taking the inflow in the dry season as the basic flow, and the inflow of the inflow river without the influence of rainfall is determined, and then the rainfall data is applied to the basic inflow to obtain the actual inflow. In this way, compared with directly selecting the inflow in the rainy season or not limiting the inflow in the dry season, the difference in inflow caused by different rainfall is avoided. Then the vegetation coverage in the preset range on both sides of the inflow river is obtained, and when the vegetation coverage is higher than the preset coverage threshold, the physical entity content of the inflow river under the influence of the rainfall data is predicted through the vegetation coverage, wherein the physical entity at least includes any one or more of leaves, branches and trunks, that is, the amount of physical entity such as branches and trunks that the river carries in the time period is predicted through the vegetation coverage; and when the vegetation coverage is lower than the preset coverage threshold, the physical entity content is taken as zero, and the first product between the river sediment content (which can be calculated according to the geological data of the river by using existing algorithm, or is a known quantity) and the actual inflow, and the second product between the physical entity content and the actual inflow are added to obtain the non-water mass quality, that is, the amount of silt and physical entity, carried into the reservoir by the inflow river; finally, the first influence data of the inflow river on the water level of the reservoir is calculated through the non-water mass quality and the storage capacity of the reservoir.

[0063] That is, the first influence data here reflects the content of other substances that the inflow river brings into the reservoir except the outflow, which may affect the water level of the reservoir.

[0064] Step 106, measure the underground water level data and underground geological data of the reservoir, and determine the second internal influence data of the reservoir water level according to the measurement results.

[0065] According to the geographical location of the reservoir, the bottom of some reservoirs is composed of substances with poor water absorption capacity such as rocks, and the bottom of some reservoirs is composed of substances with strong water absorption capacity such as soil, for example, natural reservoirs such as Hu Bo. Therefore, the underground conditions of the reservoir also have a certain influence on the water level of the reservoir, and these influences are represented by the second internal influence data.

[0066] In one possible implementation of the present application, to obtain the second internal influence data, first, the underground water level data Q uand underground geological data, where the underground geological data refers to the water absorption index k of the underground of the reservoir, and the second internal influence data is calculated by the following formula:

[0067]

[0068] wherein, λ is the second internal influence data, Q0 is the initial predicted water level, P c is the edge pressure of the bottom of the reservoir, P u is the underground water level pressure, and R is a shape parameter of the bottom of the reservoir, which takes different values according to different shapes of the bottom of the reservoir.

[0069] Step 107, based on the first internal influence data and the second internal influence data, the initial predicted water level is corrected by a water level correction model to obtain a final predicted water level.

[0070] After obtaining the first internal data and the second internal data, the application corrects the predicted water level by using the two data, and has the following two correction methods:

[0071] Method 1: correcting the water level prediction model. Specifically, the particle swarm optimization algorithm is used to optimize the hyperparameters of the water level prediction model, where the hyperparameters refer to parameters of the model itself such as the number of convolution kernels. The water level prediction model is updated by the optimization result, and the updated water level prediction model is trained again to obtain a water level correction model. The initial predicted water level, the first internal influence data and the second internal influence data are input into the water level correction model, and the final predicted water level is output. The process of hyperparameter optimization and model training again is because the process of obtaining the final predicted water level is considered to be a process predicted from the initial predicted water level, and the only influence factors in the process are the first internal influence data and the second internal influence data. Hyperparameter optimization makes the model more suitable for new influence factors, and secondary training ensures the prediction accuracy of the new model. The process of particle swarm optimization of hyperparameters and the process of secondary training of the model can be realized by existing schemes, and the application will not be repeated here.

[0072] In one example, the water level prediction model can be implemented by using a LSMT model, a Seq2Seq model or other models with prediction ability.

[0073] The second way is to correct the initial predicted water level. Specifically, the positive influence parameter and the negative influence parameter are obtained by quantifying the influence degree of the first internal influence data and the second internal influence data on the water level of the reservoir, the quantification process can also be realized by the existing technology or algorithm, the product of the initial predicted water level and the positive influence parameter and the negative influence parameter is calculated, and finally the final predicted water level is determined according to the product. In this way, the model is not needed any more, but the initial predicted water level is adjusted by the influence data itself, so that the initial predicted water level is slightly raised or lowered, and the final predicted water level with higher accuracy is obtained.

[0074] Through any one of the above two correction methods, the final predicted water level with high precision and accuracy is obtained.

[0075] Based on the same inventive concept, the application also provides a water level prediction device for water conservancy projects, which has the structure as shown in Figure 2 The water level prediction device for water conservancy projects in the application has the structure as shown in

[0076] Figure 2 The water level prediction device for water conservancy projects in the application has the structure as shown in Figure 2 The water level prediction device for water conservancy projects in the application has the structure as shown in

[0077] In a possible implementation manner of the application, the processor is used for reading the water level time series data of the reservoir at intervals of a preset time, obtaining the external influence data of the water level of the reservoir, and performing feature extraction on the external influence data to obtain external influence features, then performing correlation analysis on the external influence features and the water level of the reservoir, and assigning weights to the external influence features according to the analysis result; inputting the water level time series data and the external influence features carrying the weights into a pre-trained water level prediction model to obtain an initial predicted water level; calculating the first internal influence data of the water level of the reservoir through the inflow of the reservoir according to the water quality information and the geological information of the inflow river of the reservoir; measuring the underground water level data and the underground geological data of the reservoir, and determining the second internal influence data of the water level of the reservoir through the measurement result; correcting the initial predicted water level through a water level correction model based on the first internal influence data and the second internal influence data to obtain a final predicted water level.

[0078] The various embodiments herein are described with progression in application, but one of ordinary skill in the art will understand that each of the embodiments described can be combined with each of the other embodiments to describe various permutations of the present application. Each embodiment, however, is focused on different aspects of the application, and thus, the skilled artisan will understand that the embodiments are not mutually exclusive, and that the various aspects of the embodiments can be combined with each other.

[0079] The devices and methods provided by the present application are one-to-one correspondence, therefore, the device also has the beneficial technical effects similar to the method corresponding to it. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the device will not be repeated here.

[0080] Those skilled in the art will appreciate that embodiments of the present application can be a method, apparatus, system, or computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including disks, diskettes, hard disk drives, CD-ROMs, and the like) embodying computer-readable program code.

[0081] It should also be noted that the terms "comprising", "containing", or any other similar term are intended to encompass non-exclusive inclusions, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements, but can also include other elements not expressly listed, or other elements inherent in such process, method, article, or apparatus. Without more limitations, the element defined by the phrase "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0082] The above merely provides embodiments of the present application, and is not intended to limit the present application. The present application can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting water levels in a water conservancy project, wherein the water conservancy project includes at least a reservoir, characterized in that, The method includes: Read the reservoir's water level time series data at preset time intervals; External impact data of reservoir water level is acquired, and features are extracted from the external impact data to obtain external impact features, which are then stored in the blockchain. Correlation analysis is then performed on the external impact features and reservoir water level, and weights are assigned to the external impact features based on the analysis results. The external impact data includes at least rainfall data, evaporation data, inflow data, and outflow data. The water level time series data and the external influence features with weights are input into a pre-trained water level prediction model to obtain an initial predicted water level. The water level prediction model adopts a time series model. Based on the water quality and geological information of the rivers flowing into the reservoir, the first internal impact data of the reservoir water level is calculated using the actual inflow rate. The water quality information includes at least the content of physical substances, and the geological information includes at least the content of river sediment. The calculation of the first internal impact data of the reservoir water level includes: determining the non-aquatic material mass brought into the reservoir by the inflow river based on a first product between the river sediment content and the actual inflow rate, and a second product between the physical substance content and the actual inflow rate; and calculating the first impact data of the inflow river on the reservoir water level based on the non-aquatic material mass and the reservoir capacity. The groundwater level and underground geological data of the reservoir are measured, and the second internal influence data of the reservoir water level is determined based on the measurement results; wherein, the underground geological data includes at least the water absorption index, and the second internal influence data is determined by the following formula: in, This is the second internal influence data, where k is the water absorption index. This is the current water level data. For groundwater level data, The pressure at the bottom edge of the reservoir, R represents the groundwater pressure, and R represents the shape parameter of the bottom of the reservoir. Based on the first internal influence data and the second internal influence data, the initial predicted water level is corrected using a water level correction model to obtain the final predicted water level. This includes: optimizing the hyperparameters of the water level prediction model using a particle swarm optimization algorithm; updating the water level prediction model based on the optimization results and performing secondary training on the updated water level prediction model to obtain a water level correction model; and inputting the initial predicted water level, the first internal influence data, and the second internal influence data into the water level correction model to perform secondary prediction of the initial predicted water level to obtain the final predicted water level.

2. The water level prediction method for water conservancy projects according to claim 1, characterized in that, The method further includes: Read the geographical location data of the reservoir and obtain the meteorological data of the geographical location data between the current time point and the predicted time point; Rainfall data and temperature and humidity data are extracted from the meteorological data. Feature extraction is performed on the rainfall data to obtain the rainfall characteristics of the reservoir at the predicted time node. Feature extraction is also performed on the temperature and humidity data. Combined with the measured water temperature data, the evaporation characteristics of the reservoir at the predicted time node are obtained. Identify the river information associated with the reservoir, and based on the river information and the meteorological information corresponding to the river information between the current time node and the predicted time node, determine the inflow characteristics of the reservoir; Obtain the water dispatch plan for the reservoir between the current time node and the predicted time node, and extract the outflow characteristics of the reservoir according to the water dispatch plan; Based on the rainfall characteristics, evaporation characteristics, inflow characteristics, and outflow characteristics, external influence characteristics of the reservoir water level are constructed.

3. The water level prediction method for water conservancy projects according to claim 2, characterized in that, The process of extracting / determining the external influence features includes: When the geographic location data is administrative division data, obtain a dynamic cloud map of rainfall between the current time node and the predicted time node, and dynamically add the temperature and humidity data to the dynamic cloud map of rainfall. The reservoir area is identified from the administrative division data, and the corresponding reservoir area is extracted from the rainfall dynamic cloud map to obtain the rainfall dynamic cloud map of the reservoir area. In the dynamic cloud map of rainfall corresponding to the reservoir area, several sampling areas are set according to spatial location and / or rainfall distribution, and a corresponding two-dimensional array is generated for each sampling area. The first dimension element of the two-dimensional array is used to indicate the rainfall time series data corresponding to the current sampling area, and the second dimension element of the two-dimensional array is used to indicate the temperature and humidity time series data corresponding to the current sampling area. Extract the first dimension elements from all two-dimensional arrays, and then superimpose the rainfall data at the same time to output the rainfall characteristics; Extract the second dimension elements from all two-dimensional arrays, and after averaging the temperature and humidity data at the same time, output the temperature and humidity characteristics; The current water level data of the reservoir is obtained from the water level time series data, and the current water level data is updated using the rainfall characteristics; Based on the water temperature data and the temperature and humidity characteristics, the temperature influence coefficient α and humidity influence coefficient β are determined. Then, using the updated current water level data and the horizontal surface area of ​​the reservoir, the evaporation characteristics of the reservoir are calculated using the following formula: in, The evaporation rate is represented by α, the temperature influence coefficient, and β, the humidity influence coefficient. The wind speed function is derived from meteorological data, and S is the horizontal surface area of ​​the reservoir. This is the current water level data. The current water vapor pressure, For water temperature data, Temperature characteristics, Humidity characteristics To adjust the parameters, i represents any time node between the current time node and the predicted time node, and n represents the total number of time nodes between the current time node and the predicted time node.

4. The water level prediction method for water conservancy projects according to claim 3, characterized in that, The process of extracting / determining the external influence features also includes: Obtain the historical data and historical meteorological data of the river information within the time period corresponding to the current time node and the predicted time node, wherein the historical data is the data from the period preceding the predicted time node. The average of the data stored in the database within the historical time period corresponding to each year, wherein the historical meteorological data refers to the data stored in the previous period. Meteorological data for the historical time period corresponding to each year, among which... ≥ And the The initial value is 1; The historical meteorological data and the meteorological information are subjected to similarity calculation, which includes at least rainfall similarity calculation and temperature and humidity similarity calculation. When the similarity calculation result is greater than or equal to a preset similarity threshold, the historical inbound data is determined as the inbound traffic feature; and when the similarity calculation result is less than the preset similarity threshold, then execution is performed. And perform similarity calculation again until the result of the similarity calculation is greater than or equal to a preset similarity threshold and / or The value is greater than 5; The water dispatching scheme identifies the outflow from the reservoir, which includes at least one or more of the following: irrigation water, power generation water, navigation water, urban water supply, discharge flow, and leakage loss during the time period corresponding to the current time node and the predicted time node. After performing time-series processing on the outbound flow, the outbound flow characteristics are obtained.

5. The water level prediction method for water conservancy projects according to claim 2, characterized in that, A correlation analysis was performed on the external influence characteristics and the reservoir water level, including: In the water level time series data, historical water level data Q[q1,q2,…,qn] corresponding to several time nodes between the current time node and the predicted time node are extracted, and the historical rainfall features R[r1,r2,…,rn], historical evaporation features E[e1,e2,…,ei], historical inflow features D[d1,d2,…,dn], and historical outflow features F[f1,f2,…,fn] corresponding to the historical water level data are obtained respectively. Calculate the mean and variance of Q, R, E, D, and F respectively, and calculate the covariance between Q and R, Q and E, Q and D, and Q and F based on the mean of Q, R, E, D, and F respectively. Calculate the standard deviation between Q and R, Q and E, Q and D, and Q and F based on the variance of Q, R, E, D, and F respectively. The ratio of the covariance to the standard deviation is determined as the correlation coefficient between Q and R, Q and E, Q and D, and Q and F, and a correlation coefficient matrix is ​​generated. After normalizing the elements in the correlation coefficient matrix, the values ​​of each element are assigned as the weights of the external influence features.

6. The water level prediction method for water conservancy projects according to claim 1, characterized in that, The calculation process for the first internal influence data includes: Obtain the historical average inflow of the river into the reservoir during the non-rainy season, and calculate the predicted inflow of the river into the reservoir during the time period from the current time node to the preset time node based on the historical average inflow. Rainfall data for the corresponding area of ​​the inflow river is extracted from the dynamic rainfall cloud map, and the actual inflow is obtained by summing the rainfall data and the predicted inflow. Obtain the vegetation coverage rate within a preset range on both sides of the river flowing into the reservoir; When the vegetation coverage rate is higher than a preset coverage threshold, the physical entity content of the inflow river under the influence of the rainfall data is predicted by the vegetation coverage rate. The physical entity includes at least one or more of leaves, branches and tree trunks. When the vegetation coverage rate is lower than the preset coverage threshold, the physical entity content is set to zero.

7. The water level prediction method for water conservancy projects according to claim 1, characterized in that, The correction to the initial predicted water level also includes: Based on the initial predicted water level, the positive influence parameter corresponding to the first internal influence data and the negative influence parameter corresponding to the second internal influence data are determined, wherein the influence parameter is used to indicate the degree of influence of the first internal influence data and / or the second internal influence data on the upward and downward fluctuation of the initial predicted water level. Calculate the product between the initial predicted water level and the positive and negative influence parameters, and determine the final predicted water level based on the product.

8. A water level prediction device for water conservancy projects, characterized in that, The device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to execute a water level prediction method for hydraulic engineering according to any one of claims 1-7.

Images