Multi-model fusion multi-reservoir interval universal incoming water forecasting system

Abstract

The invention relates to the technical field of hydrological forecasting, and particularly discloses a multi-model fusion multi-reservoir interval universal incoming water forecasting system, which comprises a data standardization processing module, a short-term incoming water forecasting module, a medium and long-term incoming water forecasting module and a redisk analysis module, and is characterized in that the short-term incoming water forecasting module obtains short-term incoming water of each section; the data standardization processing module obtains surface rainfall and surface evaporation data; the medium-and-long-term incoming water forecasting module obtains medium-and-long-term incoming water of each section; and the redisk analysis module analyzes and compares a forecast result stored during future water forecast and historical measured data stored by regular processing. According to the universal incoming water forecasting method and system based on multi-model fusion and multi-reservoir interval, short-term, medium-term and long-term incoming water forecasting of multiple reservoirs, intervals and sections can be quickly adapted, the adaption cost of model application is reduced, and the efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological forecasting, and in particular to a water inflow forecasting system that is universal for multiple reservoir intervals and integrates multiple models. Background Art

[0002] In the existing technology, the water inflow forecast system generally uses historical runoff data or future meteorological data and rainfall data as input to predict the water inflow of the reservoir. The prediction results can be used to support flood control scheduling of the reservoir during the flood season and water resource allocation during the dry season, providing strong data support for the final implementation of the reservoir scheduling plan. The water inflow forecast can be divided into two categories according to the time scale: short-term water inflow forecast and medium- and long-term water inflow forecast.

[0003] The average flow rate of the Lijiang River hydrological station in Guilin is 133m over the years 3 / s, the minimum monthly average flow measured over the years is 10.8m 3 / s, the minimum monthly average flow is 5.9m 3 / s, and the flow on the driest day is only 3.8m 3 / s. In recent years, with the rapid development of economy, the population of Guilin has increased rapidly. The total amount of water taken from Lijiang River by Guilin city has reached 600,000 m per day. 3 , equivalent to 7m 3 / s, the city's water use has exceeded the minimum monthly average flow of the Li River of 5.9m 3 / s, the water diversion safety of Guilin and the people along the two banks of the river is seriously threatened. Due to the low water volume in the Lijiang River Basin during the dry season, the water transport and navigation conditions of many scenic spots distributed within the Lijiang River Basin will be severely affected. In the winter when the water volume is low, the most famous Baili Lijiang River Scenic Area sometimes only has 10-20km of waterways to support navigation. The lack of water in the rivers in the Lijiang River Basin in winter reduces the navigable mileage for tourism, and in severe cases even leads to the suspension of river navigation. The lack of water in the Lijiang River has greatly affected the beautiful natural landscape and good tourism image within the Lijiang River. It is urgent to reasonably forecast the flow of each reservoir, section, and section in the Guilin Lijiang River Basin to support the water replenishment work of the Lijiang River.

[0004] Research on the water inflow forecast system was carried out as early as 2018-2019 (Research and Application of Ecological Dispatching of Lijiang Reservoir Group, Fan Fuxin, 2018.6); However, the proposed solution has certain limitations and the following problems need to be solved urgently:

[0005] 1. The historical water inflow forecast research for the Lijiang River Basin only considers the medium- and long-term runoff forecast. The medium- and long-term water inflow forecast has an overall prediction of the water inflow trend of the Guilin section in the future period, which can provide a general basis and direction for the comprehensive water replenishment plan based on the year, month, and ten days; however, it cannot provide an effective real-time basis for water replenishment in the scenario where water replenishment is needed for cruise ships due to insufficient water. For this scenario, the short-term water inflow forecast can provide a basis for water replenishment for the sections that urgently need water replenishment in a short period of time. The short-term water inflow forecast for the Lijiang River Basin is a topic that has not been studied before.

[0006] 2. The medium- and long-term water inflow forecast model of the Lijiang River Basin is affected by the lack of original input data. The forecast model only considers the impact of historical runoff series on the forecast series. The water inflow forecast model is greatly affected by meteorological factors related to runoff (such as rainfall, temperature, etc.), and only considering the impact of historical runoff on future runoff is rather one-sided.

[0007] 3. Historical research on water inflow forecasts in the Lijiang River Basin has only made medium- and long-term forecasts for the Guilin section. No detailed analysis and research has been made on the water inflow forecasts for the reservoirs upstream of the Guilin Station (Qingshitan Reservoir, Xiaorongjiang Reservoir, Fuzikou Reservoir, and Chuanjiang Reservoir), the Guilin to Siku section, and the Guilin to Yangshuo section. It is difficult to accurately replenish water for each section based on the water inflow forecast for the Guilin section alone.

[0008] 4. Previous research on the Lijiang River's water inflow forecast has been limited to the analysis, forecasting and verification of runoff based on the compiled data of previous years. It has failed to make real-time forecasts of future runoff based on meteorological factors, and to review and assess the accuracy of previous forecast results in real time. Past research lacks certain practical value in actual applications.

[0009] 5. During the process of removing hazards and reinforcing the upstream reservoir, the reservoir cannot carry out normal regulation and storage. In this scenario, there is a lack of research and analysis on the water inflow forecast for the section. Summary of the invention

[0010] The present invention aims to solve at least one of the above-mentioned technical problems and provide a water inflow forecasting system that is universal for multiple reservoir intervals and integrates multiple models.

[0011] In order to achieve the above purpose, a multi-model fusion multi-reservoir interval universal water inflow forecast system is proposed, including a data standardization processing module, a short-term water inflow forecast module, a medium- and long-term water inflow forecast module and a re-analysis module, wherein:

[0012] The data standardization processing module processes the rainfall, evaporation and flow data reported hourly by the measuring station, and outputs the processing results to the short-term water inflow forecast module, the medium- and long-term water inflow forecast module and the replay module. The data processing includes:

[0013] (1) Convert monitoring point data into surface data: Design a station weight table to store the weights of the stations within the reservoir range, find the corresponding weights, and then calculate the weighted sum of the rainfall and evaporation values ​​of the tuples in the list according to the weights to obtain hourly surface rainfall and surface evaporation data;

[0014] The input data of the medium- and long-term water forecast module is processed and converted into the average surface rainfall data of each ten-day and each month according to the hourly surface rainfall data. The calculation formula is:

[0015]

[0016] where p m Indicates the average rainfall per decade / month, p t is the accumulated rainfall data within t hours reported at the t hour, and N is the total number of hours in the decade or month;

[0017] The calculation formula for processing the average measured flow data of each ten-day / month period for medium- and long-term water inflow forecast and reply is:

[0018]

[0019] where q m Indicates the average rainfall per decade / month, q t is the instantaneous flow rate at the tth hour, in m 3 / s, N is the total number of hours in the decade or month;

[0020] (4) Design a database table structure to store the average flow data of each reservoir / section every ten days or every month, calculate the flow data regularly through a scheduled task, and save the data in the medium- and long-term water inflow forecast report data table;

[0021] The short-term water inflow forecast module calculates the water inflow of each reservoir based on the geographical location of the river and reservoir, and uses the output results of the data standardization processing module as input. Through the conventional reservoir scheduling algorithm or data fusion under the risk prevention and reinforcement conditions, the reservoir outflow is obtained, and the water inflow of the interval is superimposed to obtain the water inflow of each section;

[0022] The medium- and long-term water inflow forecast module uses the forecast factor as input, obtains the forecast value through BP neural network and K-nearest neighbor algorithm respectively, and weighted averages the BP neural network forecast value and the K-nearest neighbor algorithm forecast value to forecast the medium- and long-term water inflow of the reservoir and the section. According to the water inflow of the reservoir and the section, and whether the risk removal and reinforcement working condition scenario is configured, the medium- and long-term conventional scheduling module and the risk removal and reinforcement processing unit are combined to calculate the discharge flow, and the section water inflow is superimposed to finally calculate the section water inflow;

[0023] The replay analysis module analyzes and compares the forecast results saved during future water forecasting and the historical measured data that is regularly processed and saved, and evaluates and analyzes the correlation and error between the forecast flow series and the measured flow series.

[0024] Preferably, the control area of ​​each station is calculated by the Thiessen polygon method, and the weight of a single station is obtained by dividing the control area of ​​the station by the total control area of ​​all stations. The control area of ​​the polygon where the i-th station is located is ΔA i , the total control area is A, then the weight of the station is And meet Where n is the number of monitoring stations, and the calculation of surface rainfall is the weighted average of the monitored rainfall at each monitoring station in the area:

[0025] P=w1p1+w2p2+...+w m p m

[0026] Where m is the number of stations in the area, P is the surface rainfall, and p is the rainfall monitored by the station.

[0027] Preferably, the data standardization processing module includes a general data processing unit, which obtains the forecast rainfall sequence, evaporation sequence and initial flow reported by the radar station, calculates the surface rainfall and surface evaporation data within the control range of the reservoir or interval, and inputs the surface data into the Xin'anjiang runoff model to calculate the forecast flow of the reservoir or interval.

[0028] Preferably, the short-term water inflow forecast and medium- and long-term water inflow forecast modules include a hazard removal and reinforcement working condition processing unit, which saves three types of information, namely, whether each reservoir has been de-risked, the start and end time of the hazard removal and reinforcement, and the discharge method, through a hash table data structure; and according to the configuration of the reservoir, for the reservoir that has been de-risked and reinforced, if the discharge type is fixed discharge, the discharge flow is the flow at the moment before the start of the reservoir hazard removal and reinforcement; if the discharge type is natural runoff, the discharge flow is set to the reservoir inflow during that period.

[0029] Preferably, the short-term water inflow forecast module includes a section short-term water inflow forecast calculation unit. The section short-term water inflow forecast calculation unit uses the water inflow forecast results of the upstream reservoir of the section. For reservoirs that are undergoing hazard removal and reinforcement, the predicted inflow of the reservoir is input into the hazard removal and reinforcement working condition processing module. Otherwise, a conventional reservoir scheduling algorithm is used to obtain the discharge flow of each reservoir, and the interval water inflow is superimposed to finally obtain the section flow.

[0030] Preferably, the reservoir discharge is calculated by the Muskingum model, and the calculation equation is as follows:

[0031] Q2=C0I2+C1I1+C2Q1,where

[0032] Preferably, the BP neural network includes a model input layer, a hidden layer and an output layer. The input data is a time series of a prediction factor vector for a period of time to be predicted. For each vector, the data is first normalized, that is, the data of each dimension is normalized to between 0 and 1; then, the normalized vector is input to the input layer of the neural network, and each characteristic element in the vector is operated by a neuron y=Wx+b, where W and b are weights and biases, respectively, and then propagated to the neuron nodes of the hidden layer. After each hidden layer neuron node receives the output of each input layer neuron node, it is propagated to the neuron nodes of the output layer in the same way through the activation function y=1 / (1+exp(-x)). The output layer outputs the result through the same activation function, and the output value of the neural network is denormalized to obtain the predicted runoff.

[0033] Preferably, the K-nearest neighbor algorithm calculates the Euclidean distance between the normalized prediction factor vector and the vector in the historical data set, sorts them in a max-heap, takes out the topK smallest vectors, and for the topK smallest vectors, calculates the weighted average of their flow rates according to the sorting results and the weights to obtain the predicted flow rate.

[0034] Preferably, the selection of the prediction factor is based on the following basis: using the correlation coefficient to measure the correlation between the prediction factor and the runoff, the correlation coefficient calculation formula is:

[0035]

[0036] Among them, ρ represents the correlation coefficient between a prediction factor and runoff q, It represents the multi-year average value of a factor. represents the average sample runoff value, l is the number of runoff sample years,

[0037] The correlation coefficient between each factor x and the runoff q is obtained by calculation, and the correlation coefficient values ​​are arranged in order from large to small. The factors with the highest correlation coefficients are selected as the preliminary factors. Then, the mutual correlation of the preliminary factors is analyzed pairwise according to the above formula, and the factors with large correlation coefficient values ​​and small pairwise correlation are selected.

[0038] Preferably, the evaluation indicators of the repetitive analysis module include coefficient of certainty, absolute error, relative error and pass rate, where:

[0039] The certainty coefficient mainly describes the degree of consistency between the predicted flow process and the measured flow process. Its calculation formula is:

[0040]

[0041] where y c (i) is the forecast flow value at the i-th moment, y0(i) is the measured flow value at the i-th moment, is the mean of the measured sequence, n is the length of the forecast sequence;

[0042] The absolute error is the difference between the predicted value and the measured value at each time point. The calculation formula is:

[0043] absErr i =y c (i)-y0(i)(1.2)

[0044] The relative error is the proportion of the absolute error to the measured value, and its calculation formula is:

[0045]

[0046] The pass rate is the cumulative number of forecast pass times divided by the total number of forecast periods. The calculation formula is:

[0047]

[0048] Among them, isHege() is a function that receives the measured sequence and the forecast sequence as input, and calculates and returns whether the forecast result at that time point is qualified according to different time scale requirements. If qualified, it returns 1, otherwise it returns 0.

[0049] The beneficial effects are as follows: compared with the prior art, the water inflow forecast system of the present invention, which integrates real-time and forecast data, fills the gap in the short-term water inflow forecast of each reservoir section of the Lijiang River in historical research; improves and optimizes the structure of the medium- and long-term forecast model to adapt it to multi-dimensional data input (such as rainfall, temperature, historical runoff, etc.) to achieve more accurate and practical forecast results; supplements the water inflow forecast of each reservoir and section in the upper reaches of the Lijiang River, and provides a basis for how each reservoir adjusts water; adds a historical review module to the historical forecast results, provides a credible basis for the accuracy of the forecast, and assists users in making more accurate water adjustment decisions; for the water inflow forecast of the section when the reservoir loses its regulating function under the condition of hazard removal and reinforcement, a configurable solution is proposed, which allows the configuration of the start time of hazard removal and reinforcement, as well as the discharge method of the reservoir during this period, and the water inflow forecast of the section is carried out under this discharge condition. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings, wherein:

[0051] Figure 1 It is a flow chart of the general short-term water inflow forecast module for a single reservoir or interval;

[0052] Figure 2 This is the structural diagram of the Xinanjiang model;

[0053] Figure 3 It is a schematic diagram of the processing flow of the danger removal and reinforcement working condition processing unit;

[0054] Figure 4 It is a schematic diagram of the processing flow of the cross-section short-term water inflow forecast calculation unit;

[0055] Figure 5 This is a schematic diagram of the process of calculating the interval water inflow in the short-term water inflow forecast in the review analysis module.

[0056] Figure 6 Schematic diagram of the general medium- and long-term water inflow forecast processing flow for a single reservoir or interval DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] It should be noted that when a component is referred to as being "fixed to" another component, it may be directly on the other component or there may be a central component. When a component is considered to be "connected to" another component, it may be directly connected to the other component or there may be a central component at the same time. When a component is considered to be "set on" another component, it may be directly set on the other component or there may be a central component at the same time. When a component is referred to as being "set in the middle", it does not only mean being set in the middle, as long as it is not set at both ends within the range defined by the middle. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0060] The present invention discloses a water inflow forecast system universal for multiple reservoir intervals with multiple models fusion, including a data standardization processing module, a short-term water inflow forecast module, a medium- and long-term water inflow forecast module, and a re-analysis module, wherein:

[0061] The data standardization processing module processes the rainfall, evaporation and flow data reported hourly by the measuring station, and outputs the processing results to the short-term water inflow forecast module, the medium- and long-term water inflow forecast module and the replay module. The data processing content includes:

[0062] (1) Convert monitoring point data into surface data: Design a station weight table to store the weights of the stations within the reservoir range, find the corresponding weights, and then calculate the weighted sum of the rainfall and evaporation values ​​of the tuples in the list according to the weights to obtain hourly surface rainfall and surface evaporation data;

[0063] Process the input data of the medium- and long-term water forecast module, and convert the hourly surface rainfall data into the average surface rainfall data for each decade and each month: the average rainfall for each decade / month is calculated by accumulating the rainfall per hour in the period of the decade / month, and then dividing it by the total number of hours in the period. The calculation formula is:

[0064]

[0065] where p m Indicates the average rainfall per decade / month, p t It is the accumulated rainfall data within t hours reported at the t hour, and N is the total number of hours in the decade or month.

[0066] Process the average measured flow data for each ten-day / month for medium- and long-term water forecasting. According to the hourly flow data reported by the flow measuring station, it is converted into hourly runoff, and then the runoff of all hours in the ten-day / month is superimposed, and then divided by the number of seconds in the ten-day / month to obtain the average flow of the ten-day / month. The calculation formula is:

[0067]

[0068] where q m Indicates the average rainfall per decade / month, q t is the instantaneous flow rate at the tth hour, in m 3 / s, N is the total number of hours in the decade or month.

[0069] (4) Design a database table structure (medium- and long-term water inflow forecast and return data table) to save the average flow data of each reservoir / section every ten days or every month, and calculate the flow data according to method (3) through a scheduled task (triggered every ten days) and save it in the medium- and long-term water inflow forecast and return data table;

[0070] The short-term water inflow forecast module calculates the water inflow of each reservoir based on the geographical location of the river and reservoir, and uses the output results of the data standardization processing module as input. Through the conventional reservoir scheduling algorithm or data fusion under the risk prevention and reinforcement conditions, the reservoir outflow is obtained, and the water inflow of the interval is superimposed to obtain the water inflow of each section;

[0071] The medium- and long-term water inflow forecast module uses the forecast factor as input, obtains the forecast value through BP neural network and K-nearest neighbor algorithm respectively, and weighted averages the BP neural network forecast value and the K-nearest neighbor algorithm forecast value to forecast the medium- and long-term water inflow of reservoirs and sections. Based on the water inflow of reservoirs and sections, combined with the medium- and long-term conventional dispatching model, the cross-section water inflow is obtained by superposition calculation;

[0072] The replay analysis module analyzes and compares the forecast results saved during future water forecasting and the historical measured data that is regularly processed and saved, and evaluates and analyzes the correlation and error between the forecast flow series and the measured flow series.

[0073] Specifically, the Lijiang River Basin is used as one of the embodiments for illustration in this application. The short-term water inflow forecast module first divides the reservoirs, intervals and sections in the Lijiang River Basin according to the geographical locations of the rivers and reservoirs using the Thiessen polygon method. The division results are as follows: Reservoirs upstream of Guilin Station: Qingshitan, Fuzikou, Chuanjiang, Xiaorongjiang; The interval includes: Guilin to Siku interval, and the interval from Guilin to Yangshuo is divided into: Guilin to Mopanshan, Mopanshan to Caoping, Caoping to Yangdi, Yangdi to Xingping, and Xingping to Yangshuo. The flow of Chaotian Station also flows to Yangshuo through the river. The flow of Chaotian Station is calculated by evolving the discharge flow of the Sianjiang Reservoir through the Muskingum model and superimposing the flow from the Sianjiang to Chaotian interval.

[0074] The short-term water inflow forecast module first calculates the water inflow of each reservoir above the Guilin section, obtains the reservoir outflow through the conventional reservoir dispatching algorithm or data fusion under the risk prevention and reinforcement conditions, and superimposes the water inflow of the interval to obtain the water inflow of each station (section). Figure 1As shown in the figure, the short-term water inflow forecast module includes a general data processing unit for executing conventional reservoir scheduling algorithms. The general data processing unit obtains the forecast rainfall sequence, evaporation sequence and initial flow reported by the radar station, calculates the surface data within the control range of the reservoir or interval, and inputs the surface data into the Xin'anjiang runoff model to calculate the forecast flow of the reservoir or interval. Among them, the Xin'anjiang model was proposed by Professor Zhao Renjun of Hohai University in 1963. The model modeling is based on the principle of full storage and runoff in humid areas. The Xin'anjiang model is a decentralized conceptual model and has been widely used in humid and semi-humid areas of my country. This scheme adopts the Xin'anjiang model of three water source divisions, and its main features are three divisions, namely, division of units, division of water sources, and division of stages. Unit division is to divide the whole basin into many units, mainly to consider the impact of uneven rainfall distribution, and secondly to consider the differences and changes in underlying surface conditions; water source division is to divide runoff into three components, namely surface, soil, and underground. The confluence speeds of the three water sources are different, with the surface being the fastest and the underground being the slowest; stage division is to divide the confluence process into the slope confluence stage and the river network confluence stage. The reason is that the confluence characteristics of the two stages are different. On the slope, the confluence speeds of various water sources are different, but there is no such difference in the river network. The model structure diagram is as follows Figure 2 As shown in the figure, since each module of the Xinanjiang model has a specific implementation method, there is no major optimization and modification in this application module, so it will not be described again here.

[0075] like Figure 3 As shown, the short-term water inflow forecast and medium- and long-term water inflow forecast modules also include a hazard removal and reinforcement working condition processing unit. The hazard removal and reinforcement working condition processing unit uses a hash table data structure to save three types of information: whether each reservoir has been de-risked, the start and end time of the hazard removal and reinforcement, and the discharge method. According to the configuration of the reservoir, for the reservoir that has been de-risked and reinforced, if the discharge type is fixed discharge, the discharge flow is the flow at the moment before the reservoir hazard removal and reinforcement begins; if the discharge type is natural runoff, the discharge flow is set to the reservoir inflow during that period.

[0076] In addition, if Figure 4 As shown, the short-term water inflow forecast module includes a short-term water inflow forecast calculation unit for the section. The short-term water inflow forecast calculation unit for the section uses the forecast results of the upstream reservoir mouth, the Sichuan River, and the interval from the reservoir to the section. For the reservoirs that have been reinforced, the forecast inflow of the reservoir is input into the reinforcement working condition processing module. Otherwise, the conventional reservoir scheduling algorithm is used to obtain the discharge flow of each reservoir, and the interval water inflow is superimposed to finally obtain the section flow. Specifically, the reservoir discharge flow is calculated using the Muskingum model, and its calculation equation is as follows:

[0077] Q2=C0I2+C1I1+C2Q1,where

[0078] In the above formula, the flow rate of the lower section in the second period is equal to the flow rate of the upper section in the second period multiplied by the coefficient C0 plus the flow rate of the upper section in the first period multiplied by C1, plus the flow rate of the lower section in the first period multiplied by the coefficient C2. The K value is equivalent to the time for the water flow to propagate from the upper section to the lower section, x is the flow weight coefficient, and Δt is the calculation period. For a continuous flow sequence, the evolved flow rate of the lower section can be calculated based on the iterative recursion of the above formula over time. The Muskingum module mainly calculates the result of the evolution of the flow rate of the upper section to the flow rate of the lower section at each moment based on the above formula. Its purpose is to ensure that at the same moment, the composition of the flow of this section is the result of the evolution of the upstream flow.

[0079] In this embodiment, by accessing the data standardization processing module for the scenario of multiple stations within the reservoir or interval, the reservoir (or interval) to which the station belongs can be directly associated in the system and its weight can be configured, so that the station data can be accessed as input data for the water inflow forecast without restarting and modifying the application, thereby improving the scalability of station data access. The data standardization processing module mainly processes the monitoring point data reported within each reservoir or interval and converts it into surface data. The rainfall and evaporation data of the reservoir or interval control surface are multiplied by the detection value and weight of each monitoring point within the control range, and the sum of the weights of the stations in the basin is guaranteed to be 1. The data standardization processing module designs a station weight table to store the weights corresponding to the stations within the reservoir interval. For a reservoir or interval, the data standardization processing module receives a tuple sequence, that is, the data reported by the station, whose tuple attributes are <station code, evaporation, rainfall, time>. Then, the processing logic is first grouped by time, and the tuples of the same time are assembled into a list. In the list, according to the station value of each tuple, the corresponding weight is found, and then the rainfall value and evaporation value of the tuple in the list are weighted and calculated according to the weight to obtain the surface rainfall and surface evaporation data. The above operation is performed on the list corresponding to each time, and the surface rainfall and evaporation data can be calculated. Specifically, the station weight table is as follows:

[0080] Field Name Field Tags Data Types Remark Primary Key Id id int Primary key id Station code stcd varchar Hydrological station code Station name stnm varchar Name of hydrological station River basin code bsnm varchar The code of the reservoir or section where the measuring station is located Weight value weight double The weight of the watershed where the station is located

[0081] The control area of ​​each station in the basin is calculated by the Thiessen polygon method. In the Thiessen polygon method, all adjacent rain gauges are connected into triangles, and perpendicular bisectors are drawn for each side of these triangles. The polygon formed by the perpendicular bisectors around each rain gauge is the Thiessen polygon. The weight of a single station is obtained by dividing the control area of ​​the station by the total control area of ​​all stations. The control area of ​​the polygon where the i-th station is located is ΔA i , the total control area is A, then the weight of the station And meet Where n is the number of monitoring stations. The calculation of surface rainfall is the weighted average of the monitored rainfall at each monitoring station in the area:

[0082] P=w1p1+w2p2+...+w m p m

[0083] Where m is the number of stations in the area, P is the surface rainfall, and p is the rainfall monitored by the station.

[0084] In this embodiment, Figure 6 The calculation flow diagram of the medium- and long-term water inflow forecast module is shown. The medium- and long-term water inflow forecast module takes the forecast factor as input, obtains the forecast value through the BP neural network and the K-nearest neighbor algorithm respectively, and weighted averages the BP neural network forecast value and the K-nearest neighbor algorithm forecast value to forecast the medium- and long-term water inflow of the reservoir and the section. Then, according to the water inflow of the reservoir and the section, and whether the user configures the reservoir to be in the state of risk removal and reinforcement, combined with the output results of the risk removal and reinforcement working condition module and the medium- and long-term conventional scheduling module, the section water inflow is superimposed, and the cross-section water inflow can be calculated. The medium- and long-term conventional scheduling module determines the amount of water discharged from the reservoir based on the reservoir scheduling diagram, river irrigation water demand, water replenishment requirements, and other requirements.

[0085] Among them, the selection of prediction factors is based on the following basis: the correlation coefficient is used to measure the correlation between the prediction factor and runoff. The correlation coefficient calculation formula is:

[0086]

[0087] Among them, ρ represents the correlation coefficient between a prediction factor and runoff q, It represents the multi-year average value of a factor. represents the average sample runoff value, l is the number of runoff sample years,

[0088] By calculating the correlation coefficient between each factor x and runoff q, the correlation coefficient values ​​are arranged in order from large to small, and the factors with the highest correlation coefficient are selected as the primary factors. Then, according to the above formula, the mutual correlation of the primary factors is analyzed pairwise, and the factors with large correlation coefficient values ​​and small pairwise correlation are selected. Finally, through calculation, the forecast factors of rainfall, temperature, and runoff in the previous period are selected, and the forecast factors are input into the BP neural network and K-nearest neighbor algorithm.

[0089] Among them, the BP neural network includes a model input layer, a hidden layer and an output layer. The input data is the forecast factor (rainfall, temperature, runoff in the previous period) vector time series of the forecast period. For each vector, the data is first normalized, that is, the data of each dimension is normalized to between 0 and 1; then, the normalized vector is input to the input layer of the neural network, and each characteristic element in the vector is operated by the neuron y=Wx+b, where W and b are weights and biases respectively, and then propagated to the neuron nodes of the hidden layer. Each hidden layer neuron node receives the output of each input layer neuron node. After that, it is propagated to the neuron nodes on the output layer in the same way through the activation function y=1 / (1+exp(-x)). The output layer outputs the result through the same activation function. The output value of the neural network is denormalized to obtain the predicted runoff. The BP neural network in this embodiment takes into account the factors affecting the runoff, that is, it is considered that the factors of rainfall, temperature, and runoff in the previous period have a greater impact on the runoff in the next period, and these characteristics are independent of each other and do not have dependencies. Compared with previous studies that only consider the single factor of historical runoff characteristics, the factors considered in this embodiment are more detailed. The training algorithm of the neural network uses the reverse diffusion algorithm to train the initialization parameters of each neuron, that is, W, b (weight and bias). In order to avoid overfitting of the BP neural network, the input used to calculate the error uses the predicted value and the measured value of the validation set respectively. When the error is less than a certain threshold, the training is stopped and the BP neural network parameters are saved.

[0090] The K-nearest neighbor algorithm calculates the Euclidean distance between the normalized forecast factor (rainfall, temperature, runoff in the previous period) vector (normalized) and the vector in the historical data set (the number is optional, that is, the value of K), and sorts it in a large heap, takes out the top K smallest vectors, and for the top K smallest vectors, calculates the weighted average of their flow according to the sorting result according to the weight (the higher the ranking, the higher the weight, and the sum of all weights is 1) to obtain the forecast flow. Specifically, the formula for calculating the forecast flow using the K-nearest neighbor algorithm is as follows. It is assumed that the top K data items in the Euclidean distance have been found from the historical data set through the characteristic factor vector of the flow to be forecasted, and their corresponding flows are y1, y2, ..., y k , then the calculation formula for the flow rate to be predicted y′ is:

[0091]

[0092] Where W jThe calculation method of is calculated using the method suggested in the literature (Upmanu Lall, Ashish Sharma. A nearest neighbor bootstrap for resampling hydrologic time series. Water Resource Research, 1996), that is,

[0093]

[0094] Among them, j = 1, 2, ..., k. Through the weight formula, it can be seen that among the topK data items, the smaller the Euclidean distance between its characteristic factor and the characteristic factor of the flow to be predicted, the higher the weight is given when calculating the flow. The following example uses the K-nearest neighbor algorithm to predict the flow. Suppose the forecast flow of a certain decade is y pred , whose characteristic factor vector is assumed to be x pred (average rainfall = 3.0, average temperature = 35.2), assuming that the historical data set contains the following data: x1 (average rainfall = 3.2, average temperature = 36.0, flow = 23), x2 (average rainfall = 3.5, average temperature = 34.0, flow = 28), x3 (average rainfall = 0.5, average temperature = 33.0, flow = 18). By normalizing the eigenvector, the result is the normalized vector of the eigenfactor of the data to be predicted: Similarly, the eigenvectors of x1 to x3 can be obtained as (0.9, 1), (1, 0.33), and (0, 0) respectively. Assuming that the value of k is 2, by calculating the normalized x pred The Euclidean distance from x1 to x3 shows that pred The closest top2 vectors are x1 and x2. According to formula (2.2), we get According to formula (2.1), we can calculate y pred =W1*23+W2*28=24.67.

[0095] In this embodiment, the medium- and long-term water forecast module uses the weighted average calculation method of the BP neural network forecast value and the K-nearest neighbor algorithm forecast value to obtain the final result. The reason for combining these two methods is that the BP neural network has a certain uncertainty in the forecast of extreme data (such as extreme rainfall or temperature, etc.), but has a higher accuracy for conventional inputs. The K-nearest neighbor algorithm is a deterministic algorithm. When the historical data set is certain, the result calculated by it will not change. Combining these two algorithms, the medium- and long-term forecast system has better forecast accuracy under normal circumstances, and at the same time, takes into account the stability of the forecast in the case of extreme input data.

[0096] In this implementation, a re-analysis module is also included. The re-analysis module analyzes and compares the forecast results saved during the water inflow forecast (including short-term and medium- and long-term water inflow forecasts) and the historical measured data that is regularly processed and saved, so as to understand the accuracy of the water inflow forecast and make timely adjustments to the water inflow forecast. Taking the water inflow forecast re-analysis of the Lijiang River Basin as an example, the re-analysis module includes a water inflow forecast result table and a water inflow forecast re-analysis data table, where the water inflow forecast result table is as follows:

[0097]

[0098] The pre-order data table is as follows:

[0099]

[0100] As shown in the table above, the forecast data of the re-analysis module is centrally saved in the forecast result table. For each forecast, the forecast unique identifier (call_id), forecast type (predict_type) (marking the forecast as a short-term water inflow forecast or a medium- to long-term water inflow forecast), forecast flow (predict_flow) and time (time) corresponding to each time period, and the reservoir / interval / section number corresponding to the forecast and other data are saved. The time interval for each data of this forecast varies according to the forecast type. For example, the time interval for short-term water inflow forecast data is 1 hour, and the time interval for medium- to long-term water inflow forecast data is 10 days (for example, the first ten days are represented by x month 1 of xxxx year, the middle ten days are represented by x month 11 of xxxx year, and the last ten days are represented by x month 21 of xxxx year). For the acquisition of re-dispatch data, because the time interval of the data reported by each flow measuring station is hourly, an additional timer (triggered every ten days) is designed to call the monitoring data service every ten days to obtain the hourly flow data within a certain time range, and the re-dispatch measured (reverse) flow data is obtained from the water inflow forecast and re-dispatch data table. For the medium- and long-term water inflow forecast, the average measured flow data is completed in the data standardization processing module and regularly updated to the forecast and re-dispatch data table through the scheduled task; the short-term water inflow forecast is updated in this table according to the report of the measuring station. After data cleaning (abnormal data smoothing, missing data completion, etc.), the average flow rate every ten days or every month is calculated, and the average value is saved in the medium- and long-term water inflow forecast and re-dispatch data table according to the measuring station type. The flow type includes the instantaneous flow of the reservoir outflow and inflow (type is hourly data), the average flow (type is ten-day or monthly data), and the flow through the section. The calculation method of the indicators for the re-analysis is given in formula (1.1~1.4). Its qualification standard is based on the short-term, medium-term and long-term qualitative and quantitative accuracy assessment defined in the "GBT22482-2008 Hydrological Information Forecast Specification". The specific parameter values ​​are not repeated here. Only the source and storage form of the data to be analyzed are described here. The data table used in the re-analysis module is used as a table to store intermediate measured data. When the re-analysis function is used, it avoids directly requesting monitoring and processing from the monitoring site, and directly reads data from the library table for assessment, which improves the efficiency of re-analysis; the data standardization processing module starts the thread through the background timer to regularly trigger, process and save the monitoring data, which effectively supports the analysis ability and performance of the re-analysis module.

[0101] Specifically, the re-analysis module mainly analyzes the correlation and error between the forecasted flow sequence and the measured flow sequence. The input of this module is the forecasted flow sequence and the measured flow sequence (or the reversed flow sequence). Its evaluation indicators include the coefficient of certainty, absolute error, relative error and qualified rate, among which:

[0102] The certainty coefficient mainly describes the degree of consistency between the predicted flow process and the measured flow process. Its calculation formula is:

[0103]

[0104] where y c (i) is the forecast flow value at the i-th moment, y0(i) is the measured flow value at the i-th moment, is the mean of the measured sequence, n is the length of the forecast sequence;

[0105] The absolute error is the difference between the predicted value and the measured value at each time point. The calculation formula is:

[0106] absErr i =y c (i)-y0(i)

[0107] The relative error is the proportion of the absolute error to the measured value, and its calculation formula is:

[0108]

[0109] The pass rate is the cumulative number of forecast pass times divided by the total number of forecast periods. The calculation formula is:

[0110]

[0111] Among them, isHege() is a function that receives the measured sequence and the forecast sequence as input, and calculates and returns whether the forecast result at that time point is qualified according to different time scale requirements. If qualified, it returns 1, otherwise it returns 0.

[0112] In the replay analysis module, since there is no measured flow in the interval, in order to evaluate the accuracy of the short-term water inflow forecast of the interval, this module proposes a method of calculating the water inflow of the interval in the short-term water inflow forecast by reverse-calculating the evolutionary flow. The specific calculation process is as follows Figure 5 As shown. Since water flow needs time to evolve, the flow rate of the interval at the same time cannot be calculated by subtracting the outflow flow of the reservoir from the flow rate of the section at the same time. Therefore, this calculation method takes into account the evolution process after the reservoir is discharged, and uses the Muskingum evolution equation to calculate the flow rate of each reservoir outflow reaching the section at each time point.

[0113] It can be seen from the above embodiments that the water inflow forecasting system of the present invention has the following advantages compared with the prior art:

[0114] 1. This plan fills the gap in historical research on short-term water inflow forecasts for various reservoirs, sections, and sections of the Lijiang River. The relevant results of short-term water inflow forecasts have been successfully applied to the "Digital Twin Lijiang River Water Resources Allocation "Four Forecasts"" system. The short-term water inflow forecast results for reservoirs and sections are good. According to the results of historical flood assessments, the average certainty coefficient is about 0.8.

[0115] 2. This solution improves and optimizes the structure of the medium- and long-term water forecast model based on the Lijiang River Basin. For the first time, it proposes a model that adapts to multi-dimensional data input (rainfall, temperature, historical runoff, etc.) in the Lijiang River Basin, and optimizes the parameter calibration strategy of the BP neural network, so that the model fits better during the training process using historical data. This achievement has been successfully applied to the "Digital Twin Lijiang River Water Resources Allocation "Four Forecasts"" system. The review results show that the qualitative and quantitative pass rates of the forecast are above 70%.

[0116] 3. This solution proposes a standardized processing module for access data. For scenarios with multiple stations within a reservoir or interval, the reservoir (or interval) to which the station belongs and its weight can be directly configured in the database table, and the station data can be accessed as input data for inflow forecasting without restarting or modifying the application, thereby improving the scalability of station data access.

[0117] 4. This plan rationally processes forecast data and historical measured data, and develops a re-analysis module to assess the accuracy of water inflow forecasts for each reservoir, interval, and section in the Lijiang River Basin.

[0118] 5. This solution provides a configuration method for discharge during the reservoir reinforcement period. By coordinating with various modules, it realizes the water inflow forecast of the section during the reservoir reinforcement period.

[0119] The above achievements have been applied to the "Digital Twin Lijiang River Water Resources Allocation "Four Prediction"" system

[0120] The above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Any modification or equivalent substitution that does not depart from the spirit and scope of the present invention shall be included in the scope of the technical solution of the present invention.

Claims

1. A water inflow forecast system that integrates multiple models and multiple reservoirs, characterized in that: It includes data standardization processing module, short-term water inflow forecast module, medium- and long-term water inflow forecast module and re-analysis module. The data standardization processing module processes the rainfall, evaporation and flow data reported hourly by the measuring station, and outputs the processing results to the short-term water inflow forecast module, medium- and long-term water inflow forecast module and re-analysis module. Data processing includes: (1) Convert monitoring point data into surface data: Design a station weight table to store the weights of the stations within the reservoir range, find the corresponding weights, and then calculate the weighted sum of the rainfall and evaporation values ​​of the tuples in the list according to the weights to obtain hourly surface rainfall and surface evaporation data; (2) Process the input data of the medium- and long-term water forecast module and convert the hourly surface rainfall data into the average surface rainfall data for each decade and each month. The calculation formula is: where p m Indicates the average rainfall per decade / month, p t is the accumulated rainfall data within t hours reported at the t hour, and N is the total number of hours in the decade or month; (3) Processing the average measured flow data for each ten-day / month period for medium- and long-term water inflow forecast and reporting, the calculation formula is: where q m Indicates the average rainfall per decade / month, q t is the instantaneous flow rate at the tth hour, in m 3 / s, N is the total number of hours in the decade or month; (4) Design a database table structure to store the average flow data of each reservoir / section every ten days or every month, and calculate the flow data through a scheduled task, and save the data in the medium- and long-term water inflow forecast report data table; the short-term water inflow forecast module calculates the water inflow of each reservoir based on the geographical location of the river and reservoir, and uses the output results of the data standardization processing module as input. The reservoir outflow is obtained through conventional reservoir scheduling algorithms or data fusion under risk prevention and reinforcement conditions, and the water inflow of the interval is superimposed to obtain the water inflow of each section; The medium- and long-term water inflow forecast module uses the forecast factor as input, obtains the forecast value through BP neural network and K-nearest neighbor algorithm respectively, and weighted averages the BP neural network forecast value and the K-nearest neighbor algorithm forecast value to forecast the medium- and long-term water inflow of reservoirs and sections. Based on the water inflow of reservoirs and sections, combined with the medium- and long-term conventional dispatching model, the cross-section water inflow is obtained by superposition calculation; The replay analysis module analyzes and compares the forecast results saved during future water forecasting and the historical measured data that is regularly processed and saved, and evaluates and analyzes the correlation and error between the forecast flow series and the measured flow series.

2. According to claim 1, a multi-model fusion multi-reservoir interval universal water inflow forecast system is characterized in that: The control area of ​​each station is calculated by the Thiessen polygon method, and the weight of a single station is obtained by dividing the control area of ​​the station by the total control area of ​​all stations. The control area of ​​the polygon where the i-th station is located is ΔA i , the total control area is A, then the weight of the station is And meet Where n is the number of monitoring stations, and the calculation of surface rainfall is the weighted average of the monitored rainfall at each monitoring station in the area: P=w1p1+w2p2+...+w m p m Where m is the number of stations in the area, P is the surface rainfall, and p is the rainfall monitored by the station.

3. The multi-model fusion multi-reservoir interval universal water inflow forecast system according to claim 1 is characterized in that: The data standardization processing module includes a general data processing unit, which obtains the forecast rainfall sequence, evaporation sequence and initial flow reported by the radar station, calculates the surface rainfall and surface evaporation data within the control range of the reservoir or interval, and inputs the surface data into the Xin'anjiang runoff model to calculate the forecast flow of the reservoir or interval.

4. The multi-model fusion multi-reservoir interval universal water inflow forecast system according to claim 1 is characterized in that: The short-term water inflow forecast and medium- and long-term water inflow forecast modules include a hazard removal and reinforcement working condition processing unit. The hazard removal and reinforcement working condition processing unit saves three types of information, namely, whether each reservoir has been de-risked, the start and end time of the hazard removal and reinforcement, and the discharge method, through a hash table data structure. According to the configuration of the reservoir, for a reservoir that has been de-risked and reinforced, if the discharge type is fixed discharge, the discharge flow is the flow at the moment before the start of the reservoir hazard removal and reinforcement; if the discharge type is natural runoff, the discharge flow is set to the reservoir inflow during that period.

5. The multi-model fusion multi-reservoir interval universal water inflow forecast system according to claim 4 is characterized in that: The short-term water inflow forecast module includes a section short-term water inflow forecast calculation unit. The section short-term water inflow forecast calculation unit uses the water inflow forecast results of the upstream reservoir of the section. For reservoirs that have been reinforced, the predicted inflow of the reservoir is input into the hazard removal and reinforcement working condition processing module. Otherwise, a conventional reservoir scheduling algorithm is used to obtain the discharge flow of each reservoir, and the interval water inflow is superimposed to finally obtain the section flow.

6. The multi-model fusion multi-reservoir interval universal water inflow forecast system according to claim 3 is characterized in that: The reservoir discharge is calculated by the Muskingum model, and the calculation equation is as follows: Q2 = C0I2 + C1I1 + C2Q1, where 7. The multi-model fusion multi-reservoir interval universal water inflow forecast system according to claim 1 is characterized in that: The BP neural network includes a model input layer, a hidden layer and an output layer. The input data is a time series of a prediction factor vector for a period of time to be predicted. For each vector, the data is first normalized, that is, the data of each dimension is normalized to between 0 and 1; then, the normalized vector is input to the input layer of the neural network, and each characteristic element in the vector is operated by a neuron y=Wx+b, where W and b are weights and biases, respectively, and then propagated to the neuron nodes of the hidden layer. After each hidden layer neuron node receives the output of each input layer neuron node, it is propagated to the neuron nodes of the output layer in the same way through an activation function y=1 / (1+exp(-x)). The output layer outputs the result through the same activation function, and the output value of the neural network is denormalized to obtain the predicted runoff.

8. The multi-model fusion multi-reservoir interval universal water inflow forecast system according to claim 1 is characterized in that: The K-nearest neighbor algorithm calculates the Euclidean distance between the normalized prediction factor vector and the vector in the historical data set, sorts them in a large heap, takes out the top K smallest vectors, and calculates the weighted average of their flow rates according to the sorting results and the weights to obtain the predicted flow rate.

9. The multi-model fusion multi-reservoir interval universal water inflow forecast system according to claim 1 is characterized in that: The selection of the prediction factor is based on the following basis: the correlation coefficient is used to measure the correlation between the prediction factor and the runoff. The correlation coefficient calculation formula is: Among them, ρ represents the correlation coefficient between a prediction factor and runoff q, It represents the multi-year average value of a factor. represents the average sample runoff value, l is the number of runoff sample years, The correlation coefficient between each factor x and the runoff q is obtained by calculation, and the correlation coefficient values ​​are arranged in order from large to small. The factors with the highest correlation coefficients are selected as the preliminary factors. Then, the mutual correlation of the preliminary factors is analyzed pairwise according to the above formula, and the factors with large correlation coefficient values ​​and small pairwise correlation are selected.

10. The multi-model fusion multi-reservoir interval universal water inflow forecast system according to claim 1, characterized in that: The evaluation indicators of the re-analysis module include the coefficient of certainty, absolute error, relative error and qualified rate, among which: The certainty coefficient mainly describes the degree of consistency between the predicted flow process and the measured flow process. Its calculation formula is: where y c (i) is the predicted flow value at the i-th moment, y0(i) is the measured flow value at the i-th moment, is the mean of the measured sequence, n is the length of the forecast sequence; The absolute error is the difference between the predicted value and the measured value at each time point. The calculation formula is: absErr i =y c (i)-y0(i) The relative error is the proportion of the absolute error to the measured value, and its calculation formula is: The pass rate is the cumulative number of forecast pass times divided by the total number of forecast periods. The calculation formula is: Among them, isHege() is a function that receives the measured sequence and the forecast sequence as input, and calculates and returns whether the forecast result at that time point is qualified according to different time scale requirements. If qualified, it returns 1, otherwise it returns 0.

Images