Pumped storage power station upstream small watershed flow prediction method
By using a flow prediction model based on meteorological data in a small watershed upstream of the pumped storage power station, the problem that traditional methods cannot accurately predict is solved, and high-precision flow prediction is achieved, which adapts to the complexity needs of the power station.
Patent Information
- Application Number
- CN202411435451.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-15
- Publication Date
- 2024-11-15
AI Technical Summary
The traditional small watershed flow prediction method for upstream pumped storage power plants cannot accurately deal with the problems of complexity and lack of data in remote areas, resulting in low prediction accuracy and inability to meet the accuracy requirements of the power plants.
The flow prediction model based on meteorological data is adopted, including the meteorological-water storage recursive computing layer, the first fully connected computing layer of water storage-runflow, and the total runoff calculation layer. Through recursive computing and linear processing, combined with neural network training, small watersheds are predicted. traffic.
It improves the accuracy of flow prediction, adapts to the particularity of small watershed upstream of pumped storage power stations, does not rely on high-precision hydrological cycle data, and meets the requirements of the power station for accurate flow prediction.
Smart Images

Figure CN118964960A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for predicting flow of a small watershed upstream of a pumped storage power station, and is applicable to the field of hydrological prediction. Background Art
[0002] Pumped storage power stations use electricity during the off-peak period to pump water into the reservoir, and release water to generate electricity during the peak period, thereby balancing the power load and improving energy efficiency. The operating efficiency and economic benefits of pumped storage power stations are directly affected by the water from upstream.
[0003] Accurately calculating the flow in the small watershed upstream of a pumped-storage power station is crucial for the dispatch and optimal operation of the power station. Traditional flow prediction methods are based on differential equation calculations based on the physical process of hydrology, which usually require accurate inputs. High-precision hydrological and meteorological stations cannot be deployed in remote areas, and components of the hydrological cycle such as evaporation, infiltration, and groundwater flow are difficult to measure. This means that traditional flow prediction methods cannot cope with the particularity and complexity of the small watershed upstream of a pumped-storage power station, and the prediction accuracy is not high, which cannot meet the requirements of pumped-storage power stations for flow prediction accuracy. Summary of the invention
[0004] The technical problem to be solved by the present invention is: in view of the above-mentioned problems, a method for predicting the flow rate of a small watershed upstream of a pumped storage power station is provided.
[0005] The technical solution adopted by the present invention is: a method for predicting flow in a small watershed upstream of a pumped storage power station, characterized in that it includes: Obtain the current meteorological data series of the small watershed upstream of the pumped storage power station; Input the current meteorological data sequence into the trained flow prediction model to obtain the flow prediction value; The traffic prediction model includes: The meteorological-water storage recursive operation layer is used to obtain the water storage increment of the small watershed through recursive operation according to the meteorological data sequence; The first fully connected operation layer of water storage-runoff is used to process the water storage increment of the small watershed through linear operations to obtain the hydrodynamic characteristic sequence; The second fully connected operation layer of water storage and runoff is used to process the hydrodynamic characteristic sequence through linear operations to obtain the multi-runoff flow sequence; The runoff volume calculation layer is used to sum the elements of multiple runoff flow sequences to obtain the flow prediction value.
[0006] The meteorological-water storage recursive operation layer includes: Among them, s(n) is the water storage increment of the small watershed in the nth unit time; For the The increment of water storage in a small watershed per unit time; r i (nk) is the i-th element of the meteorological data sequence of the nk-th unit time; w i,k is the k-th order i-th weighted update coefficient; For the The recursive coefficient is N, the update order is M, and the recursive order is M.
[0007] The first fully connected water storage-runoff operation layer includes: Among them, λ j is the jth element of the hydrodynamic characteristic sequence; ξ 1,j is the j-th output weight coefficient of the first fully connected layer; s is the water storage increment of the small watershed; b 1,j is the j-th output bias coefficient of the first fully connected layer; j is a positive integer greater than or equal to 1 and less than or equal to L, and L is the length of the hydrodynamic feature sequence.
[0008] The second fully connected water storage-runoff operation layer includes: Among them, q m is the mth runoff flow in the multi-runoff flow sequence; ξ 2,m,j b is the weighted coefficient of the mth output and jth input of the second fully connected layer; 2,m,j is the j-th input bias coefficient of the m-th output of the second fully connected layer; m is a positive integer greater than or equal to 1 and less than or equal to P, and P is the length of the multi-path flow sequence.
[0009] The meteorological data series includes the total precipitation per unit time, the mean temperature per unit time and the mean humidity per unit time.
[0010] The training of the traffic prediction model includes: Obtain meteorological data series within Z unit time, water increment of pumped storage power station reservoir, and measured flow of small watershed; The water increment of the pumped storage power station reservoir within Z unit time is used as the measured water increment of the small basin, and the meteorological data sequence within Z unit time is combined to conduct initial training on the meteorological-water storage recursive operation layer; The flow prediction model is trained as a whole based on the meteorological data series within Z unit time and the measured flow in the small watershed.
[0011] The objective function used in the initial training is: Among them, L1 is the objective function used in the initial training, s x is the water storage increment of the small watershed calculated by the meteorological-water storage recursive operation layer corresponding to the x-th unit time, s α,x is the measured water storage increment of the small watershed corresponding to the x-th unit time.
[0012] The objective function used for the overall training is: Among them, L2 is the objective function used in the overall training, q ,x is the predicted flow corresponding to the x-th unit time, q ,α,x is the measured flow rate of the small watershed corresponding to the x-th unit time, is the mean of the measured flow in Z small watersheds.
[0013] A storage medium stores a computer program that can be executed by a processor, characterized in that when the computer program is executed, the steps of the method for predicting the flow rate of a small watershed upstream of a pumped-storage power station are implemented.
[0014] A device for predicting flow in a small watershed upstream of a pumped-storage power station comprises a memory and a processor, wherein the memory stores a computer program that can be executed by the processor, and wherein the device is characterized in that when the computer program is executed, the steps of a method for predicting flow in a small watershed upstream of a pumped-storage power station are implemented.
[0015] The beneficial effect of the present invention is as follows: the present invention performs flow prediction based on a meteorological data sequence through a flow prediction model, and the flow prediction model includes a meteorological-water storage recursive operation layer, a water storage-runoff first fully connected operation layer, a water storage-runoff second fully connected operation layer and a runoff total amount calculation layer. The water storage increment of a small watershed is obtained according to the meteorological data sequence through the meteorological-water storage recursive operation layer, the water storage increment of the small watershed is processed through the water storage-runoff first fully connected operation layer to obtain a hydrodynamic characteristic sequence, the hydrodynamic characteristic sequence is processed through the water storage-runoff second fully connected operation layer to obtain a multi-runoff flow sequence, and the elements of the multi-runoff flow sequence are summed through the runoff total amount calculation layer to obtain a flow prediction value.
[0016] The present invention integrates the recursive operation in the time domain with the weighted operation of each factor in the meteorological data sequence to form a meteorological-water storage recursive operation layer, constructs a system-level calculus relationship between each factor in the meteorological data sequence (precipitation, temperature and humidity) and the water storage increment of the small watershed, and effectively converts the complex parameter calibration of the hydrological physical process into parameter training of the neural network discretization operation, thereby simplifying the prediction process and improving the prediction accuracy.
[0017] The present invention regards each runoff in a small watershed as each neuron of a fully connected neural network structure. There is no need to specifically find out which runoff belongs to surface runoff and which runoff belongs to underground runoff. Instead, the fuzzy training mechanism of the neural network is utilized. It only needs to let the neural network go through appropriate training links to predict the accurate flow rate based on the current small watershed water storage increment.
[0018] The present invention uses a neural network to perform recursive calculations and analysis on the meteorological data of the small watershed upstream of the pumped-storage power station to predict the flow. Compared with traditional technologies, it does not need to establish complex hydrological physical differential equations, and does not rely on high-precision hydrological cycle data such as evaporation, infiltration and groundwater flow. It can adapt to the particularity and complexity of the small watershed upstream of the pumped-storage power station only through machine learning training, has good accuracy, and meets the requirements of the pumped-storage power station for flow prediction accuracy.
[0019] The water storage increment of a small watershed is different from the hydrodynamic characteristic sequence and is a key parameter with actual physical significance. Therefore, the meteorological-water storage recursive operation layer in the present invention needs to be effectively trained first to ensure that the calculation of the water storage increment of the small watershed can approach the true value, and then the overall training of the entire neural network is carried out; in the two training stages, different objective functions are appropriately set as the difference between the predicted output and the actual output of the training measurement model to improve the training effect of the neural network. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 This is a structural diagram of the traffic prediction model in Example 1. DETAILED DESCRIPTION
[0021] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0022] Embodiment 1: This embodiment is a method for predicting flow in a small watershed upstream of a pumped storage power station, which is characterized by comprising: S1. Obtain the current meteorological data sequence of the small watershed upstream of the pumped storage power station. In this example, the meteorological data sequence is a sequence of three elements, including: total precipitation per unit time, mean temperature per unit time, and mean humidity per unit time.
[0023] For the entire system of a small watershed, precipitation is the main source of water in the small watershed, while temperature and humidity directly affect the amount of water lost by evaporation. By controlling these three physical quantities, we can understand the overall changes in the runoff of the small watershed.
[0024] S2. Input the meteorological data sequence obtained in step S1 into the trained flow prediction model to obtain a flow prediction value.
[0025] In this embodiment, the flow prediction model is constructed based on a recursive operation neural network. The model includes a meteorological-water storage recursive operation layer and a water storage-runoff operation sub-network. The meteorological-water storage recursive operation layer is used to obtain the water storage increment of the small watershed through recursive operation according to the meteorological data sequence; the water storage-runoff operation sub-network is used to predict the flow according to the water storage increment of the small watershed through multi-layer fully connected network operation (see Figure 1 ).
[0026] In this example, the storage-runoff operation subnetwork includes: the first fully connected storage-runoff operation layer, the second fully connected storage-runoff operation layer and the total runoff calculation layer, wherein the first fully connected storage-runoff operation layer is used to process the small watershed storage increment through linear operations to obtain a hydrodynamic feature sequence; the second fully connected storage-runoff operation layer is used to process the hydrodynamic feature sequence through linear operations to obtain a multi-runoff flow sequence; the total runoff calculation layer is used to sum the elements of the multi-runoff flow sequence to obtain the predicted flow.
[0027] The expression of the meteorological-water storage recursive operation layer in this embodiment is: Among them, s(n) is the water storage increment of the small watershed in the nth unit time, For the The increment of water storage in a small watershed per unit time, r i (nk) is the i-th element of the meteorological data sequence of the nk-th unit time, w i,k is the k-th order i-th weighted update coefficient, For the The recursive coefficient is N, the update order and M the recursive order.
[0028] In this example, the expression of the first fully connected operation layer of water storage-runoff is: Among them, λ j is the jth element of the hydrodynamic characteristic sequence, ξ 1,j is the j-th output weighting coefficient of the first fully connected layer, s is the water storage increment of the small watershed, b 1,j is the j-th output bias coefficient of the first fully connected layer, j is a positive integer greater than or equal to 1 and less than or equal to L, and L is the length of the hydrodynamic feature sequence; In this embodiment, the expression of the second fully connected operation layer of water storage-runoff is: Among them, q m is the mth runoff flow in the multi-runoff flow sequence, ξ 2,m,j is the weighted coefficient of the jth input of the mth output of the second fully connected layer, b 2,m,j is the jth input bias coefficient of the mth output of the second fully connected layer, where m is a positive integer greater than or equal to 1 and less than or equal to P, and P is the length of the multi-path flow sequence; The expression of the total runoff calculation layer in this embodiment is: Among them, q is the predicted flow.
[0029] It is worth noting that the flow rate predicted in this embodiment is the total flow rate of the small watershed upstream of the pumped-storage power station injected into the reservoir of the pumped-storage power station.
[0030] The training of the traffic prediction model in this embodiment includes the following steps: ①. Set the duration of the unit time. In this embodiment, the duration of the unit time is set to 6 hours.
[0031] ② Obtain the meteorological data series, the water increment of the pumped storage power station reservoir, and the measured flow rate of the small watershed within Z unit time (during which the pumped storage power station is suspended), where Z is an integer multiple of M×N.
[0032] In this embodiment, the recursion order M is set to 3, the update order N is set to 1, and Z is set to 900.
[0033] In addition, in this embodiment, the length L of the hydrodynamic characteristic sequence is a fuzzy parameter in the intermediate operation link of the fully connected neural network structure, which has no actual physical meaning. Usually, if the length is too small, it will lead to under-fitting of the neural network; if it is too large, it will lead to over-fitting of the neural network. Under the working conditions of this embodiment, it is set to 1024 after debugging.
[0034] The length P of the multi-runoff flow sequence is a vector sequence used to fit the surface and underground runoff of a small watershed. When its value is greater than the number of runoffs in the small watershed of the application environment, accuracy can be guaranteed. However, too large a value will also waste hardware resources. According to the working conditions of this embodiment, it is set to 128.
[0035] The remaining coefficients are all coefficients to be trained in the neural network training phase.
[0036] ③. The water increment of the pumped storage power station reservoir within Z unit time is used as the measured water storage increment of the small basin. Combined with the meteorological data sequence within Z unit time, the meteorological-water storage recursive operation layer is initially trained.
[0037] The objective function used in the initial training in this embodiment is: Among them, L1 is the objective function used in the initial training, s x is the water storage increment of the small watershed calculated by the meteorological-water storage recursive operation layer corresponding to the x-th unit time, s α,x is the measured water storage increment of the small watershed corresponding to the x-th unit time.
[0038] ④. Based on the meteorological data series within Z unit time and the measured flow in the small watershed, the flow prediction model is trained as a whole.
[0039] The objective function used in the overall training in this embodiment is: Among them, L2 is the objective function used in the overall training, q ,x is the predicted flow corresponding to the x-th unit time, q ,α,x is the measured flow rate of the small watershed corresponding to the x-th unit time, is the mean of the measured flow in Z small watersheds.
[0040] In this embodiment, during training, the target values of the L1 objective function and the L2 objective function should be set to 0. There are many specific training methods for neural networks, such as gradient descent algorithm and genetic algorithm, and the present invention does not limit the specific training method.
[0041] Embodiment 2: This embodiment is a storage medium on which a computer program that can be executed by a processor is stored. When the computer program is executed, the steps of the method for predicting flow in a small watershed upstream of a pumped-storage power station in Embodiment 1 are implemented.
[0042] Embodiment 3: This embodiment is a flow prediction device for a small watershed upstream of a pumped-storage power station, comprising a memory and a processor. The memory stores a computer program that can be executed by the processor. When the computer program is executed, the steps of the flow prediction method for a small watershed upstream of a pumped-storage power station in Embodiment 1 are implemented.
[0043] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.
[0044] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.
Claims
1. A method for predicting flow in a small watershed upstream of a pumped storage power station. It is characterized in that include: Obtain the current meteorological data series of the small watershed upstream of the pumped storage power station; Input the current meteorological data sequence into the trained flow prediction model to obtain the flow prediction value; The traffic prediction model includes: The meteorological-water storage recursive operation layer is used to obtain the water storage increment of the small watershed through recursive operation according to the meteorological data sequence; The first fully connected operation layer of water storage-runoff is used to process the water storage increment of the small watershed through linear operations to obtain the hydrodynamic characteristic sequence; The second fully connected operation layer of water storage and runoff is used to process the hydrodynamic characteristic sequence through linear operations to obtain the multi-runoff flow sequence; The total runoff calculation layer is used to sum the elements of multiple runoff flow sequences to obtain the flow prediction value.
2. The method for predicting flow in a small watershed upstream of a pumped storage power station according to claim 1, It is characterized in that The meteorological-water storage recursive operation layer includes: Among them, s(n) is the water storage increment of the small watershed in the nth unit time; For the The increment of water storage in a small watershed per unit time; r i (nk) is the i-th element of the meteorological data sequence of the nk-th unit time; w i,k is the k-th order i-th weighted update coefficient; For the order recursion coefficient; N is the update order; M is the recursion order.
3. The method for predicting flow in a small watershed upstream of a pumped storage power station according to claim 1, It is characterized in that The first fully connected water storage-runoff operation layer includes: Among them, λ j is the jth element of the hydrodynamic characteristic sequence; ξ 1,j is the j-th output weight coefficient of the first fully connected layer; s is the water storage increment of the small watershed; b 1,j is the j-th output bias coefficient of the first fully connected layer; j is a positive integer greater than or equal to 1 and less than or equal to L, and L is the length of the hydrodynamic feature sequence.
4. The method for predicting flow in a small watershed upstream of a pumped storage power station according to claim 1, It is characterized in that The second fully connected water storage-runoff operation layer includes: Among them, q m is the mth runoff flow in the multi-runoff flow sequence; ξ 2,m,j b is the weighted coefficient of the mth output and jth input of the second fully connected layer; 2,m,j is the j-th input bias coefficient of the m-th output of the second fully connected layer; m is a positive integer greater than or equal to 1 and less than or equal to P, and P is the length of the multi-path flow sequence.
5. The method for predicting flow in a small watershed upstream of a pumped storage power station according to claim 1, Features: The meteorological data series includes the total precipitation per unit time, the mean temperature per unit time and the mean humidity per unit time.
6. The method for predicting flow in a small watershed upstream of a pumped storage power station according to claim 1, It is characterized in that The training of the traffic prediction model includes: Obtain meteorological data series within Z unit time, water increment of pumped storage power station reservoir, and measured flow of small watershed; The water increment of the pumped storage power station reservoir within Z unit time is used as the measured water increment of the small basin, and the meteorological data sequence within Z unit time is combined to conduct initial training on the meteorological-water storage recursive operation layer; The flow prediction model is trained as a whole based on the meteorological data series within Z unit time and the measured flow in the small watershed.
7. The method for predicting flow in a small watershed upstream of a pumped storage power station according to claim 6, It is characterized in that The objective function used in the initial training is: Among them, L 1 is the objective function used in the initial training, s x is the water storage increment of the small watershed calculated by the meteorological-water storage recursive operation layer corresponding to the x-th unit time, s α,x is the measured water storage increment of the small watershed corresponding to the x-th unit time.
8. The method for predicting flow in a small watershed upstream of a pumped storage power station according to claim 6, It is characterized in that The objective function used for the overall training is: Among them, L 2 is the objective function used for the overall training, q ,x is the predicted flow corresponding to the x-th unit time, q ,α,x is the measured flow rate of the small watershed corresponding to the x-th unit time, is the mean of the measured flow in Z small watersheds.
9. A storage medium having stored thereon a computer program executable by a processor, Features: When the computer program is executed, the steps of the method for predicting flow in a small watershed upstream of a pumped-storage power station according to any one of claims 1 to 8 are implemented.
10. A device for predicting flow in a small watershed upstream of a pumped storage power station, comprising a memory and a processor, wherein the memory stores a computer program executable by the processor. Features: When the computer program is executed, the steps of the method for predicting flow in a small watershed upstream of a pumped-storage power station according to any one of claims 1 to 8 are implemented.
Citation Information
Patent Citations
Power station water level process prediction method and apparatus based on neural network model
CN113256005A
Rainfall-runoff space-time relation simulation method based on long short-term memory neural network
CN113837352A
Reservoir inflow runoff forecasting method and forecasting model construction method thereof
CN117034779A
Basin weather forecasting method and system based on weather big data
CN117348108A
Battery performance influence factor analysis method
CN117872167A
