Cascade hydropower station water level prediction method and system based on water balance and machine learning

Abstract

The application discloses a cascade hydropower station water level prediction method and system based on water balance and machine learning, and belongs to the technical field of cascade hydropower station water level prediction. Reservoir storage data of an upstream reservoir in a predetermined period is acquired, and the discharge flow of the upstream reservoir is calculated; a conversion coefficient is determined based on historical reservoir storage data, the reservoir inflow of a downstream reservoir area is estimated in combination with the discharge flow, the upstream average water level of the downstream reservoir area is obtained based on a gross water head regression model, a water consumption rate regression model and a tail water level regression model, and error identification is performed on the calculated upstream average water level in combination with an error regression model. The application uses the water balance principle and multiple regression models to calculate the upstream average water level of the downstream reservoir area, more comprehensively captures various factors influencing the water level change, and provides more reliable water level data support for the operation and dispatching of the cascade hydropower station.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of cascade hydropower station water level prediction, and particularly relates to a cascade hydropower station water level prediction method and system based on water balance and machine learning. BACKGROUND

[0002] In the super-short-term scheduling calculation of the cascade hydropower station, there are mainly the problems of dynamic storage capacity, rainfall in the reservoir area, imbalance of the cascade hydropower station in and out of the reservoir, time lag of the flow between the cascade hydropower stations, non-constant flow in the downstream, water level jacking of the Gezhouba hydropower station on the downstream water level of the Three Gorges, and non-convergence of the calculation. In view of these problems, the super-short-term water level prediction is carried out through machine learning, deep learning and other means.

[0003] At present, the traditional short-term water level prediction model adopts the standard curves derived in the water regulation system, specifically including: the Three Gorges water level-storage capacity curve, the Three Gorges outflow-downstream power station-tail water level curve, the Three Gorges gross water head water consumption rate curve; the Gezhouba water level-storage capacity curve, the Gezhouba downstream discharge-tail water level curve, and the Gezhouba gross water head water consumption rate curve. However, the traditional short-term water level prediction model has the following defects: most of the above curves are derived based on relatively stable storage capacity settings or under conventional working conditions. However, the dynamic storage capacity dynamically changes with the water level, backwater range and other factors. The standard curve is difficult to reflect the complex change of the dynamic storage capacity in real time, so that the influence of the dynamic storage capacity cannot be accurately considered when the water level is predicted, resulting in a large deviation between the prediction result and the actual water level condition, especially in the period of frequent fluctuations of the reservoir water level, the error is more obvious. SUMMARY

[0004] The application provides a cascade hydropower station water level prediction method and system based on water balance and machine learning to solve the technical problems in the background.

[0005] The application adopts the following technical scheme: a cascade hydropower station water level prediction method based on water balance and machine learning, comprising the following steps:

[0006] The historical conversion coefficient is determined based on the historical period outflow of the upstream reservoir and the historical period inflow of the downstream water area, and the upstream preset average water level of the downstream water area in the current period is set ;

[0007] The outflow of the upstream reservoir in the current period is calculated according to the water storage data of the upstream reservoir in the current period , the outflow is converted into the inflow of the downstream reservoir area in the current period by using the historical conversion coefficient ; and the inflow is calculated based on the upstream preset average water level and the inflow​ The gross head-water consumption rate curve was obtained by fitting;

[0008] The gross head of the downstream reservoir area is predicted using the gross head regression model , based on gross head The water consumption rate of the downstream reservoir area is obtained using the water consumption rate regression model ; Based on water consumption rate Calculate the outflow flow of the downstream reservoir area , and use the water level regression model to predict the average water level of the downstream reservoir area ;

[0009] According to gross head and average water level Calculate the estimated average water level upstream of the downstream reservoir area ; Set the upstream average water level and upstream estimated average water level Perform precision analysis:

[0010] If the accuracy requirement is met, the upstream estimated average water level is output If not satisfied, use the upstream estimated average water level Replace the upstream preset average water level , repeat the above steps until the accuracy requirements are met.

[0011] In a further embodiment, the historical conversion coefficient is determined as follows:

[0012] Get the upstream reservoir in historical periods respectively Outbound traffic and downstream water areas in historical periods Inbound traffic ,in, ;

[0013] The historical conversion factor The calculation formula is as follows:

[0014] ;

[0015] Correspondingly, the downstream reservoir area is Inbound traffic The calculation formula is as follows:

[0016] .

[0017] In a further embodiment, the current period The water storage data at least includes: Inbound traffic , total output and abandoned water flow ;

[0018] Correspondingly, the outbound flow The calculation process is as follows:

[0019] The following energy conservation formula is constructed based on the water flow and the sum of the output:

[0020] Where, is the coefficient of the intermediate variable, Indicates that the upstream reservoir is in the current period The head of water;

[0021] Given the intermediate variable coefficient An initial value , and substitute it into the above energy conservation formula to get the outbound flow Initial value of ;

[0022] Based on the initial value Query the gross head-water consumption rate curve to obtain the new coefficient , the new coefficient Substitute into the above energy conservation formula;

[0023] Repeat the above steps for the Iterations, the calculation formula is:

[0024] ;

[0025] Each calculation is , the new coefficients are obtained through the upstream reservoir gross head-water consumption rate curve , continue iterating until ,in, For the The outbound flow obtained by iteration, is the accuracy threshold.

[0026] In a further embodiment, the gross head The calculation process is as follows:

[0027] Get the downstream reservoir area in the current period Water storage data, including inflow , total output and abandoned water flow ;

[0028] The downstream reservoir area in the current period The water storage data is input into the gross head regression model to obtain the gross head , the expression of the gross head regression model is as follows:

[0029] ;

[0030] wherein, is the intercept top of the downstream reservoir area, represents the inflow of the reservoir and the relative relationship coefficient of the total output , represents the nonlinear influence coefficient of the inflow of the reservoir , represents the interaction coefficient of the abandoned water flow and the total output .

[0031] In further embodiments, the expression form of the water consumption rate regression model is as follows:

[0032] ;

[0033] wherein, is the intercept top of the downstream reservoir area, and are influence coefficients.

[0034] In further embodiments, the prediction process of the average water level of the downstream reservoir area includes:

[0035] Based on the water consumption rate , the outflow of the downstream reservoir area is calculated using the following formula: :

[0036] ;

[0037] wherein, is the total output of the downstream reservoir area in the current period , is the change in water demand of the downstream reservoir area in the current period , is the length of the current period ;

[0038] A periodic influence factor is introduced to construct a water level regression model, and the expression form is as follows:

[0039] ;

[0040] wherein, is the intercept top of the downstream reservoir area, , , , and are influence coefficients, It is the total duration from the historical period to the current period t.

[0041] In a further embodiment, the estimated average water level upstream of the downstream reservoir is The calculation formula is as follows:

[0042] ;

[0043] Where, is the weight coefficient, is a constant term.

[0044] In a further embodiment, the accuracy requirement is expressed as: , is a pre-set difference.

[0045] In a further embodiment, a water level prediction error regression model is constructed, and the historical water storage data of the upstream reservoir and the downstream water area are input into the water level prediction error regression model to obtain the upstream budget water level error value. ;

[0046] The revised upstream estimated average water level is : .

[0047] A cascade power station water level prediction system based on water balance and machine learning, used to implement the cascade power station water level prediction method described above, comprising:

[0048] The first module is set to determine the historical conversion coefficient based on the historical outflow of the upstream reservoir and the historical inflow of the downstream water area, and set the downstream water area in the current period. Upstream preset average water level ;

[0049] The second module is set to be based on the upstream reservoir in the current period The outflow of the upstream reservoir is calculated based on the water storage data , using the historical conversion coefficient to convert the outbound flow Converted into downstream reservoir area in the current period Inbound traffic ; Based on the upstream preset average water level and inbound traffic The gross head-water consumption rate curve was obtained by fitting;

[0050] The third module is set up to use the gross head regression model to predict the gross head of the downstream reservoir area. , based on gross head The water consumption rate of the downstream reservoir area is obtained using the water consumption rate regression model ; Based on water consumption rate Calculate the downstream reservoir area of the delivery flow , and predict the average water level of the downstream reservoir area by using the water level regression model ;

[0051] The fourth module is configured to calculate the upstream estimated average water level of the downstream reservoir area according to the gross water head and the average water level ; the upstream preset average water level and the upstream estimated average water level are subjected to accuracy analysis:

[0052] If the accuracy requirement is met, the upstream estimated average water level is output ; if not, the upstream estimated average water level is used to replace the upstream preset average water level , and the above steps are repeated until the accuracy requirement is met.

[0053] The beneficial effects of the present application: The present application comprehensively considers many factors closely related to water level changes, such as the discharge flow of the upstream reservoir, the inflow of the downstream reservoir, the total output of each reservoir, the abandoned water flow, and the periodic influencing factors, etc. By incorporating these factors into different calculation links and corresponding regression models (such as gross water head regression model, water consumption rate regression model, water level regression model, etc.), the complex mechanism of water level change can be more comprehensively and meticulously described, and compared with the traditional water level prediction method which only considers single or few factors, the accuracy of water level prediction of cascade hydropower stations can be significantly improved. For example, when calculating the gross water head, the relative relationship between the total output and the inflow, the nonlinear influence of the inflow, and the interaction between the abandoned water flow and the total output are considered, which makes the prediction of the gross water head more in line with the actual physical process, and thus lays a more accurate foundation for subsequent calculation of other parameters based on the gross water head and final water level prediction.

[0054] The present application carries out each link calculation based on the water balance principle, from determining the historical conversion coefficient to accurately convert the flow relationship between upstream and downstream, to calculating the inflow and outflow of each reservoir area and considering the change of water demand, etc., all of which strictly follow the basic hydrological law of water balance. This helps to accurately grasp the water balance of each reservoir area in cascade hydropower stations, so as to realize the rational allocation and efficient utilization of water resources. For example, by accurately predicting the water level change and the corresponding flow condition, the power station management personnel can more reasonably arrange the power generation plan, meet the power generation demand while avoiding the waste of water resources, and better cope with the water quantity change in different periods (such as dry season and wet season), to ensure the sustainable utilization of water resources.

[0055] ​​Finally, the present invention performs precision analysis based on both the upstream preset average water level and the upstream estimated average water level, and continuously approximates the actual water level by iteratively updating the preset average water level. This iterative mechanism adjusts and optimizes each round of calculation based on the results of the previous round, ensuring that the final output water level estimate is as close to the actual water level as possible while meeting pre-set accuracy requirements, effectively reducing prediction errors. This is particularly suitable for complex hydraulic systems such as cascade power stations, where water levels are affected by the interaction of multiple dynamic factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of the water level prediction method for cascade power stations in Example 1. DETAILED DESCRIPTION

[0057] The present invention will be further described below with reference to the embodiments and accompanying drawings.

[0058] Example 1

[0059] like Figure 1 As shown, this embodiment discloses a method for predicting water levels of cascade power stations based on water balance and machine learning, comprising the following steps:

[0060] The historical conversion coefficient is determined based on the outflow of the upstream reservoir in the historical period and the inflow of the downstream water area in the historical period. Upstream preset average water level In this embodiment, the upstream reservoir is taken as the Three Gorges Reservoir, and the downstream water area is taken as the Gezhouba Dam. Therefore, the Three Gorges Reservoir-Gezhouba Dam constitutes a cascade power station.

[0061] According to the upstream reservoir in the current period The outflow of the upstream reservoir is calculated based on the water storage data , using the historical conversion coefficient to convert the outbound flow Converted into downstream reservoir area in the current period Inbound traffic ; Based on the upstream preset average water level and inbound traffic The gross head-water consumption rate curve is obtained by fitting. It is worth noting that the gross head-water consumption rate curve in this embodiment is fitted using the following methods: Ordinary Least Squares (OLS), Nonlinear Least Squares (NLS), etc.

[0062] The gross head of the downstream reservoir area is predicted using the gross head regression model , based on gross head The water consumption rate of the downstream reservoir area is obtained using the water consumption rate regression model ; based on water consumption rate Calculate the downstream reservoir area of the discharge flow , and predict the average water level of the downstream reservoir area using the water level regression model ;

[0063] According to the gross water head And the average water level The upstream estimated average water level of the downstream reservoir area is calculated ; The upstream preset average water level And the upstream estimated average water level Accuracy analysis:

[0064] If the accuracy requirement is met, the upstream estimated average water level is output ; If not, the upstream estimated average water level Substitute the upstream preset average water level , repeat the above steps until the accuracy requirement is met. Further, the accuracy requirement is represented as: , The difference is preset.

[0065] In further embodiments, the determination process of the historical conversion coefficient is as follows:

[0066] The upstream reservoir discharge flow In the historical period And the downstream water area in the historical period The inflow of the reservoir , wherein ;

[0067] The calculation formula of the historical conversion coefficient As follows:

[0068] ;

[0069] Correspondingly, the calculation formula of the inflow of the downstream reservoir area in the current period As follows:

[0070] .

[0071] For example, on the basis of defining the current period as t, the upstream reservoir discharge flow In the historical period , the discharge flow In the historical period , the discharge flow In the historical period And the downstream water area in the historical period The discharge flow​ , historical period Outbound traffic , historical period Outbound traffic ,Right now .

[0072] therefore, .

[0073] By adopting the above-mentioned method for determining the historical conversion coefficient, the inherent correlation and variation patterns between upstream and downstream flows can be well reflected, closely fitting the actual hydrological characteristics of the basin where the cascade hydropower stations are located. Calculations are performed using long-term historical data, covering multiple time nodes from the past to the current period, and can capture trends in flow changes over time, such as seasonal changes and interannual changes. This means that the determined historical conversion coefficient is not a static fixed value, but a dynamic parameter that integrates the characteristics of different time stages, which is more in line with the natural dynamic variation characteristics of river flow. Subsequently, when this coefficient is used to convert the inflow flow to the downstream reservoir during the current period, it can more accurately reflect the actual possible flow conditions, thereby improving the accuracy of related work such as flow forecasting and water level forecasting.

[0074] In a further embodiment, the current period The water storage data at least includes: Inbound traffic , total output and abandoned water flow ;

[0075] Correspondingly, the outbound flow The calculation process is as follows:

[0076] The following energy conservation formula is constructed based on the water flow and the sum of the output:

[0077] Where, is the coefficient of the intermediate variable, Indicates that the upstream reservoir is in the current period By incorporating key factors such as the reservoir's inflow, outflow, power generation output, and gross head into the framework of energy conservation, the calculation of outflow is theoretically more rigorous and reliable, accurately reflecting the energy conversion and water volume balance in the actual water flow process. This fundamentally ensures that the calculation results are consistent with actual physical conditions and lays a solid theoretical foundation for subsequent water level prediction and other related work based on this flow.

[0078] Given the intermediate variable coefficient An initial value , and substitute it into the above energy conservation formula to get the outbound flow Initial value of ;

[0079] Based on the initial value Query the gross head-water consumption rate curve to obtain the new coefficient , the new coefficient Substitute into the above energy conservation formula;

[0080] Repeat the above steps for the Iterations, the calculation formula is:

[0081] ;

[0082] Each calculation is , the new coefficients are obtained through the upstream reservoir gross head-water consumption rate curve , continue iterating until ,in, For the The outbound flow obtained by iteration, is the accuracy threshold.

[0083] The calculation results are gradually adjusted through the above-mentioned iterative mechanism to move towards a more accurate direction. In each iteration, the next round of calculations will be optimized based on the information obtained from the previous round of calculations (such as the outflow flow and the corresponding coefficient), making full use of the inherent relationship between the various parameters and the laws reflected by historical data (reflected by the gross head-water consumption rate curve), continuously narrowing the gap between the calculated results and the actual values, and finally achieving the accuracy threshold. requirements to obtain relatively accurate outbound flow values.

[0084] In a further embodiment, the gross head The calculation process is as follows:

[0085] Get the downstream reservoir area in the current period Water storage data, including inflow , total output and abandoned water flow ;

[0086] The downstream reservoir area in the current period The water storage data is input into the gross head regression model to obtain the gross head , the expression of the gross head regression model is as follows:

[0087] ;

[0088] Where, is the intercept of the downstream reservoir area, which is taken as -20m~20m; is the relative relationship coefficient of the inflow and the power sum , which is generally taken as -0.5~0.5, and when it is negative, it means that the influence of the power sum change on the gross water head is dominant, so that the gross water head more presents a downward trend with the changes of the two; is the nonlinear influence coefficient of the inflow , which is taken as 0 to 10; is the interaction coefficient of the abandoned water flow and the power sum , which is taken as -2 to 2, and if the synergy of the abandoned water flow and the power sum makes the gross water head downward trend obvious, it is negative.

[0089] The above technical solution fully considers the multiple important water storage related data of the downstream reservoir area in the current period , i.e. the inflow , the power sum and the abandoned water flow . These parameters reflect the actual situation of the water balance, energy conversion and flow state of the downstream reservoir area from different angles, and they are related to each other and jointly affect the size of the gross water head of the reservoir area. For example, the inflow directly determines the inflow of the reservoir area, which will affect the rising amplitude of the water level and then affect the gross water head; the power sum reflects the water and energy consumed in the power generation process, which is closely related to the water level of the reservoir area and the water level difference (i.e. the gross water head) between the upstream and downstream; the abandoned water flow changes the water balance of the reservoir area and also indirectly affects the value of the gross water head. By including these key factors in the calculation of the gross water head, the real hydraulic state of the downstream reservoir area can be more comprehensively and accurately reflected, and the calculated gross water head result is more in line with the actual operation condition.

[0090] In addition, not only the factors themselves are included, but also the relative relationship and interaction between different factors are reflected through the coefficients in the model . For example , the relative relationship coefficient of the inflow and the power sum reflects the correlation degree of their influence on the gross water head when the inflow and the power sum change; , the interaction coefficient of the abandoned water flow and the power sum means that the comprehensive influence on the gross water head is different under different combinations of the abandoned water flow and the power sum. This detailed consideration of the relationship between factors makes the calculation of the gross water head be able to capture the influence of the coordinated changes of various parameters under complex working conditions, and compared with the method of simply considering each factor in isolation, it can more accurately determine the actual value of the gross water head.

[0091] Correspondingly, the expression of the water consumption rate regression model is as follows:

[0092] ;

[0093] In the formula, is the intercept top of the downstream reservoir area, and are influence coefficients.

[0094] The predicted process of the average water level of the downstream reservoir area includes:

[0095] Based on the water consumption rate , the outflow of the downstream reservoir area is calculated by using the following formula :

[0096] ;

[0097] In the formula, is the total output of the downstream reservoir area in the current period , is the water demand change of the downstream reservoir area in the current period , and the time length of the current period ;

[0098] The water level regression model is constructed by introducing periodic influence factors, and the expression is as follows:

[0099] ;

[0100] In the formula, is the intercept top of the downstream reservoir area, , , , and are influence coefficients, and the total time length from the historical period to the current period t.

[0101] In the above formula, then used to reflect the periodic influence of time, t is the current time, and T is a complete period length (for example, if one year is taken as a period, T is the time unit conversion value corresponding to one year; if a month is taken as a period, T is the time length conversion value corresponding to a month, etc.). Correspondingly, is the coefficient for measuring the influence degree of the periodic influence on the average water level, which is determined by fitting the water level data corresponding to different periodic time points. For example, the water level is usually higher in the flood season, and this seasonal change feature can be captured by the sine function and the coefficient.

[0102] Therefore, the estimated average water level of the upstream of the downstream reservoir area The calculation formula is as follows:

[0103] ;

[0104] Where, is the weight coefficient, reflecting the average water level of the downstream reservoir area Estimating the average water level upstream From the perspective of physical process, the average water level It is closely related to factors such as the water balance between upstream and downstream, the continuity of water flow, and the hydraulic characteristics of the reservoir area. The degree of its impact on the upstream water level will vary depending on the specific conditions of the reservoir area. For example, if it is related to the topography of the reservoir area, for relatively flat terrain and regular reservoir morphology, the relationship between the upstream and downstream water levels is relatively simple and linear. The value of is between 0.5 and 1.5, which means that the downstream average water level has a relatively direct and important impact on the estimated upstream average water level. However, if the reservoir is located in a mountainous area with a large terrain and a winding river channel, the energy loss and water level change rules in the process of water flow are complex. The value range of may be narrowed, for example, around 0.1 to 0.8, indicating that the influence of the downstream average water level is relatively limited, and it is necessary to combine other factors to determine the upstream water level. If it is related to the water flow characteristics, when the water flow speed is fast and the water flow inertia is obvious, the speed and amplitude of the downstream water level change transmitted to the upstream will also be affected, thereby affecting For example, in some cascade hydropower station reservoirs on mountain rivers with turbulent water flow, the impact of downstream water level changes on upstream is relatively delayed and weak due to the rapid flow of water through the reservoir area. It may take a smaller value; while for reservoirs with slow water flow and similar to lakes, water level changes are transmitted relatively promptly and have a greater impact. It may take a larger value, ranging from 0.2 (turbulent water flow) to 1.8 (slow water flow and open reservoir area).

[0105] It is a constant term, and its value depends on the initial conditions of the reservoir area, the relationship with surrounding water bodies, etc. It is generally between 1 meter and 3 meters.

[0106] The following data illustrates that the water level prediction accuracy of this embodiment is higher than that of existing technologies. Table 1 shows the error distribution statistics of the Gezhouba water level data predicted using traditional methods. The overall error is relatively large. Specifically, the traditional prediction method is based on the water balance curve, using the Three Gorges water level and storage capacity curve, the Three Gorges outflow-downstream power station-tailwater level curve, and the Three Gorges gross head water consumption rate curve; the Gezhouba water level and storage capacity curve, the Gezhouba outflow tailwater level curve, and the Gezhouba gross head water consumption rate curve to check values.

[0107] Table 1 Gezhouba traditional method water level prediction error distribution

[0108]

[0109] It can be seen that the calculation effect of the traditional model on Gezhouba is poor, because the unit output change of Gezhouba is more complex, and the error is caused by the uncertainty of the relationship between the Three Gorges outflow and the Gezhouba inflow, which leads to that the water level prediction of Gezhouba is not suitable for using the traditional model.

[0110] In combination with Table 2, Table 2 is the water level data error distribution statistics of Gezhouba predicted by the method.

[0111] Table 2 Gezhouba method water level prediction error distribution

[0112]

[0113] According to Table 2, the prediction error of the method is smaller than the traditional prediction error, and the error distribution range is significantly smaller, which achieves a significant prediction effect and significantly improves the accuracy.

[0114] Example 2

[0115] Based on the cascade hydropower station water level prediction method disclosed in Example 1, the following steps are further disclosed to improve the prediction accuracy, a water level prediction error regression model is constructed, historical period water storage data of the upstream reservoir and the downstream water area are input into the water level prediction error regression model to obtain the upstream budget water level error value .

[0116] The corrected upstream estimated average water level : .

[0117] The water level prediction error regression model constructed in this embodiment is a regression model based on machine learning, which can be a decision tree model, a random forest model, a gradient boosting regression tree, etc.

[0118] Example 3

[0119] This embodiment discloses a cascade hydropower station water level prediction system based on water balance and machine learning, which is used to realize the cascade hydropower station water level prediction method described in Example 1 and Example 2, comprising: a first module configured to determine a historical conversion coefficient based on the historical period outflow of the upstream reservoir and the historical period inflow of the downstream water area, and set the upstream preset average water level of the downstream water area in the current period . ;

[0120] A second module is configured to determine the upstream estimated average water level of the downstream water area in the current period The outflow of the upstream reservoir is calculated based on the water storage data , using the historical conversion coefficient to convert the outbound flow Converted into downstream reservoir area in the current period Inbound traffic ; Based on the upstream preset average water level and inbound traffic The gross head-water consumption rate curve was obtained by fitting;

[0121] The third module is set up to use the gross head regression model to predict the gross head of the downstream reservoir area. , based on gross head The water consumption rate of the downstream reservoir area is obtained using the water consumption rate regression model ; Based on water consumption rate Calculate the outflow flow of the downstream reservoir area , and use the water level regression model to predict the average water level of the downstream reservoir area ;

[0122] The fourth module is set up according to the gross head and average water level Calculate the estimated average water level upstream of the downstream reservoir area ; Set the upstream average water level and upstream estimated average water level Perform precision analysis:

[0123] If the accuracy requirement is met, the upstream estimated average water level is output If not satisfied, use the upstream estimated average water level Replace the upstream preset average water level , repeat the above steps until the accuracy requirements are met.

Claims

1. A water level prediction method for cascade power stations based on water balance and machine learning, characterized by: The following steps are involved: The historical conversion coefficient is determined based on the outflow of the upstream reservoir in the historical period and the inflow of the downstream water area in the historical period. Upstream preset average water level ; According to the upstream reservoir in the current period The outflow of the upstream reservoir is calculated based on the water storage data , using the historical conversion coefficient to convert the outbound flow Converted into downstream reservoir area in the current period Inbound traffic ; Based on the upstream preset average water level and inbound traffic The gross head-water consumption rate curve was obtained by fitting; The gross head of the downstream reservoir area is predicted using the gross head regression model , based on gross head The water consumption rate of the downstream reservoir area is obtained using the water consumption rate regression model ; Based on water consumption rate Calculate the outflow flow of the downstream reservoir area , and use the water level regression model to predict the average water level of the downstream reservoir area ; According to gross head and average water level Calculate the estimated average water level upstream of the downstream reservoir area ; Set the upstream average water level and upstream estimated average water level Perform precision analysis: If the accuracy requirement is met, the upstream estimated average water level is output If not satisfied, use the upstream estimated average water level Replace the upstream preset average water level Repeat the above steps until the accuracy requirement is met; the gross head The calculation process is as follows: Get the downstream reservoir area in the current period Water storage data, including inflow , total output and abandoned water flow ; The downstream reservoir area in the current period The water storage data is input into the gross head regression model to obtain the gross head , the expression of the gross head regression model is as follows: ； Where, is the intercept top of the downstream reservoir area, Indicates inbound flow Total output The relative relationship coefficient, Indicates inbound flow The nonlinear influence coefficient of Indicates the discarded water flow Total output The interaction coefficient of The water consumption rate regression model is expressed as follows: ； Where, is the intercept top of the downstream reservoir area, and All are impact coefficients; The average water level of the downstream reservoir The prediction process includes: Based on water consumption rate , the outflow flow of the downstream reservoir area is calculated using the following formula: : ； Where, For the downstream reservoir area in the current period The total output of For the downstream reservoir area in the current period Changes in water demand, Current period duration; The water level regression model is constructed by introducing periodic influencing factors, and its expression is as follows: ； Where, is the intercept top of the downstream reservoir area, 、 、 、 and are the influence coefficients, It is the total duration from the historical period to the current period t.

2. The method for predicting water level of cascade power stations based on water balance and machine learning according to claim 1, characterized in that: The process of determining the historical conversion coefficient is as follows: Get the upstream reservoir in historical periods respectively Outbound traffic and downstream water areas in historical periods Inbound traffic ,in, ; The historical conversion factor The calculation formula is as follows: ； Correspondingly, the downstream reservoir area is Inbound traffic The calculation formula is as follows: 。 3. The method for predicting water level of cascade power stations based on water balance and machine learning according to claim 1, characterized in that: The current period The water storage data at least includes: Inbound traffic , total output and abandoned water flow ; Correspondingly, the outbound flow The calculation process is as follows: The following energy conservation formula is constructed based on the water flow and the sum of the output: Where, is the coefficient of the intermediate variable, Indicates that the upstream reservoir is in the current period The head of water; Given the intermediate variable coefficient An initial value , and substitute it into the above energy conservation formula to get the outbound flow Initial value of ; Based on the initial value Query the gross head-water consumption rate curve to obtain the new coefficient , the new coefficient Substitute into the above energy conservation formula; Repeat the above steps for the The calculation formula for iterations is: ； Each calculation is , the new coefficients are obtained through the upstream reservoir gross head-water consumption rate curve , continue iterating until ,in, For the The outbound flow obtained by iteration, is the accuracy threshold.

4. The method for predicting water level of cascade power stations based on water balance and machine learning according to claim 1, characterized in that: The estimated average water level upstream of the downstream reservoir area The calculation formula is as follows: ； Where, is the weight coefficient, is a constant term.

5. The method for predicting water level of cascade power stations based on water balance and machine learning according to claim 1, characterized in that: The accuracy requirement is expressed as: , is a pre-set difference.

6. The method for predicting water level of cascade power stations based on water balance and machine learning according to claim 1, characterized in that: Construct a water level prediction error regression model, input the historical water storage data of the upstream reservoir and downstream water area into the water level prediction error regression model, and obtain the upstream budget water level error value ; The revised upstream estimated average water level is : .

7. A cascade power station water level prediction system based on water balance and machine learning, used to implement the cascade power station water level prediction method according to any one of claims 1 to 6, characterized in that: include: The first module is set to determine the historical conversion coefficient based on the historical outflow of the upstream reservoir and the historical inflow of the downstream water area, and set the downstream water area in the current period. Upstream preset average water level ; The second module is set to be based on the upstream reservoir in the current period The outflow of the upstream reservoir is calculated based on the water storage data , using the historical conversion coefficient to convert the outbound flow Converted into downstream reservoir area in the current period Inbound traffic ; Based on the upstream preset average water level and inbound traffic The gross head-water consumption rate curve was obtained by fitting; The third module is set up to use the gross head regression model to predict the gross head of the downstream reservoir area. , based on gross head The water consumption rate of the downstream reservoir area is obtained using the water consumption rate regression model ; Based on water consumption rate Calculate the outflow flow of the downstream reservoir area , and use the water level regression model to predict the average water level of the downstream reservoir area ; The fourth module is set up according to the gross head and average water level Calculate the estimated average water level upstream of the downstream reservoir area ; Set the upstream average water level and upstream estimated average water level Perform precision analysis: If the accuracy requirement is met, the upstream estimated average water level is output ; If not satisfied, use the estimated average water level upstream Replace the upstream preset average water level , repeat the above steps until the accuracy requirements are met.

Images