Reservoir power generation flow calculation method based on coupling of physical method and deep learning

Abstract

The invention discloses a reservoir power generation flow calculation method based on coupling of a physical method and deep learning. The reservoir power generation flow calculation method comprises the steps of collecting reservoir historical operation characteristic data; decomposing the data into a periodic term, a trend term and a residual term by adopting a BFAST method; the method comprises the following steps: constructing and training an LSTM deep learning model by taking a period term, a trend term and a residual term of reservoir water level and inflow as input variables and taking a residual term of generated output as a target variable, so that the model learns a time sequence relationship between the generated output residual and the input variables; predicting a generated output residual error by using the trained model, and adding the predicted residual error item with the period item and the trend item of the corresponding time period to obtain the total generated output of the reservoir; and based on a hydroenergy power generation conversion formula, the reservoir power generation output is converted into power generation flow. According to the reservoir power generation prediction method, the nonlinear dynamic characteristics in the reservoir power generation process can be effectively captured, the accuracy of reservoir power generation output prediction can be improved, and the reservoir power generation prediction method is used for optimizing the water energy utilization efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of water conservancy projects and intelligent scheduling, and specifically relates to a reservoir power generation flow calculation method based on the coupling of physical methods and deep learning. Background Art

[0002] As an important facility for regulating water resources in a river basin, the rational operation of reservoirs can effectively improve the efficiency of water resource utilization and ensure the stable operation of the power grid. Accurately predicting the power generation output of reservoirs is of great significance for optimizing power grid dispatching, improving water energy utilization and achieving sustainable energy development.

[0003] Due to climate change, precipitation uncertainty and complex operation and scheduling rules, traditional reservoir power generation prediction methods such as physical statistical models, regression analysis and time series analysis methods (such as ARIMA) have limitations in dealing with highly nonlinear, multivariate coupled hydrological processes, and it is difficult to fully capture the dynamic characteristics of the reservoir system, resulting in limited prediction accuracy. In recent years, the application of deep learning technology in the field of hydrology and water conservancy engineering has made significant progress, especially LSTM neural networks. Due to its special gating mechanism, LSTM can effectively capture the dependencies of long time series data and has powerful time series data modeling capabilities. Summary of the invention

[0004] The present invention aims to improve the accuracy of reservoir power generation output prediction and proposes a reservoir power generation flow calculation method based on the coupling of physical methods and deep learning, which can be used to optimize the efficiency of water energy utilization.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: The reservoir power flow calculation method based on the coupling of physical method and deep learning includes: Step 1: Collect historical reservoir operation characteristic data, including reservoir water level, inflow, power generation output and power generation flow, and perform data cleaning, remove outliers and fill in missing data; Step 2: Use the BFAST method to decompose the reservoir water level, inflow and power output data to obtain the periodic term, trend term and residual term data of the reservoir water level, inflow and power output; Step 3: Taking the periodic term, trend term and residual term of reservoir water level and inflow as input variables and the residual term of power generation output as the target variable, a deep learning model is constructed and trained to enable the model to learn the temporal relationship between the residual term of power generation output and the input variables; Step 4: Use the trained deep learning model to predict the power generation output residual, and add the predicted residual term to the period term and trend term of the corresponding period to obtain the total power generation output of the reservoir; Step 5: Based on the hydropower conversion formula, convert the total power generation output of the reservoir into power generation flow.

[0006] Furthermore, the historical operation characteristic data of the reservoir collected in step 1 are daily-scale data to ensure the temporal resolution and data consistency of the model.

[0007] Furthermore, the principle of using the BFAST method in step 2 to decompose the power generation output, reservoir water level and inflow data is: BFAST (Breaks For Additive Season and Trend) is a statistical method for time series analysis, mainly used to detect and monitor structural changes (breakpoints) in time series, especially in data with seasonal and trend components.

[0008] The mathematical expression of BFAST is:

[0009] Among them, Y t is the measured value at time, T t is the trend term, S t is the periodic term, e t is the residual term; The trend term is piecewise linear with breakpoints set to… , and is defined as ,but:

[0010] This means that there is a linear trend term between every two adjacent breakpoints, and there are a total of m breakpoints. j For one of the breakpoints, According to the intercept and slope of the continuous linear model, the amplitude direction of the mutation between the detected breakpoints and the slope of the gradient can be deduced. The amplitude of the mutation at the breakpoint is given by T t At two adjacent breakpoints and The difference at is derived, namely:

[0011] The gradient slopes before and after the breakpoint are and , the intercepts before and after the breakpoint are and .

[0012] Seasonal breakpoints may occur at different times than the breakpoints detected in the trend term. Let seasonal breakpoints be , there are p seasonal breakpoints in total, j’ For one of the breakpoints, , and define , then for , assuming:

[0013] Where: s is the seasonal period, represents season i and j’ The seasonal effect parameter is, therefore, When s consecutive seasons S t The sum of is exactly zero, which ensures that the seasonal effects over all seasons (including season 0) sum to zero, which is a common constraint in seasonal models to avoid over-parameterization.

[0014] The seasonal term can be reformulated as:

[0015] when t In the i Season, t =1, otherwise 0, so if t In season 0 (i.e., the base season), d t,i - d t,0 =- 1; For other seasons, t In season i ≠0, d t,i - d t,0 = 1, d t,i d t,0 are seasonal dummy variables. d t,i Indicates whether time 𝑡 belongs to season 𝑖, d t,0 Indicates whether time 𝑡 belongs to season 0.

[0016] The power generation output of a reservoir usually shows a certain periodicity, which is mainly due to the influence of multiple factors. First, the seasonal changes in hydrological conditions directly affect the inflow and water level of the reservoir. For example, in the flood season, the rainfall increases significantly, resulting in an increase in inflow, which in turn increases the power generation output of the reservoir; while in the dry season, the inflow decreases and the power generation output decreases accordingly. Secondly, the seasonal fluctuations in electricity demand will also affect the power generation output of the reservoir. In summer, especially in hot weather, the demand for electricity increases significantly, and the reservoir will be dispatched according to the changes in demand to maximize power generation. In addition, the dispatching strategy and operation management of the reservoir also have an important impact on the periodic characteristics of power generation output, especially in the cascade reservoir system, the linkage effect between different reservoirs may further affect its power generation output cycle. Therefore, understanding the mechanism of these factors is crucial to accurately simulate the changes in the power generation output of the reservoir. The output residual is related to the reservoir water level and inflow, but because the output residual is discontinuous, decomposing the reservoir water level and inflow into season, trend and residual terms for simulation can better capture the contingency of the residual.

[0017] Furthermore, the deep learning model in step 3 is an LSTM model.

[0018] The principle of the LSTM deep learning model is: The LSTM deep learning model mainly maintains its state and regulates the information flow in the cell state structure through an input gate, an output gate, and a forget gate. Its functions are mainly realized through the following formula: Forget gate: determines which past information should be forgotten in the current time step. The calculation formula is as follows:

[0019] in, f t Represents the output of the forget gate, which is a vector between 0 and 1, indicating the proportion of retention or forgetting, W f represents the weight matrix of the forget gate, h t-1 is the hidden state of the previous time step, x t is the input of the current time step, b f is the bias vector of the forget gate, Represents the sigmoid function, which compresses the output between 0 and 1.

[0020] Input gate: The input gate determines what new information will be added to the cell state. It is divided into two steps: generating input gate coefficients and candidate memory cell states.

[0021] Input gate coefficient:

[0022] in, W i represents the weight matrix of the input gate, b i represents the bias vector of the input gate, [ℎ 𝑡−1 ,𝑥 𝑡 ] is the hidden state ℎ at the previous moment 𝑡−1 and the current input 𝑥 𝑡 splicing.

[0023] Candidate memory cell status:

[0024] Finally, the calculation formula for the input gate is:

[0025] in, i t represents the output of the input gate, represents the candidate cell state, represents the new information of the current time step, C t is the cell state at time t, C t-1 is the cell state at time t-1; W c is the weight matrix of the candidate cell states.

[0026] Output gate: The output gate controls what information is extracted from the cell state as the output for the current time step and updates the hidden state:

[0027] Among them, t represents the output of the output gate; h t Represents the hidden state of the current time step, which serves as the input for the next time step; I is the weight matrix of the output gate; 𝑜 is the bias term of the output gate; tanh is the hyperbolic tangent activation function, which compresses the value to between [−1,1].

[0028] The training and prediction process of this model not only takes into account the long-term seasonal cycle changes and trends, but also effectively deals with short-term fluctuations and random noise; this model provides a powerful tool for power generation output error simulation, helps to reveal the evolution law of output errors, and provides a more accurate reference basis for scheduling optimization.

[0029] The key parameters for improving the learning performance of the LSTM model mainly include: the number of hidden layers, training rounds, the number of neurons in the hidden layer, the learning rate, and the batch size. The selection of these parameters should be adjusted according to the characteristics of the dataset, or determined through experiments and optimization to obtain the best model performance.

[0030] Furthermore, the performance of the power generation residual simulation in step 4 is evaluated by the systematic bias BIAS and the correlation coefficient R To evaluate, the calculation formula is:

[0031] Among them, X s1 is the simulated value of power generation residual, X o1 is the measured value of the power generation output residual, and are the means of the simulated and measured values ​​of power generation output residual, and are the time series numbers of the measured values.

[0032] Furthermore, the performance of the power generation flow simulation in step 5 is evaluated by the efficiency coefficient NSE and the relative error RE, which are calculated as follows:

[0033] Among them, X s2 is the simulated value of power generation flow, X o2 is the measured value of power generation flow, and are the means of the simulated and measured values ​​of power generation flow, and are the time series numbers of the measured values.

[0034] Furthermore, in step 5, based on the hydropower conversion formula, the predicted total power generation output of the reservoir is converted into power generation flow Q:

[0035] Where P is the power output, is the efficiency of the generator, is the density of water, is the acceleration due to gravity, and H is the hydraulic head, which refers to the difference between the water level upstream and downstream of the reservoir.

[0036] Beneficial results of the present invention: The reservoir power generation prediction method of the present invention can effectively capture the nonlinear dynamic characteristics in the reservoir power generation process, can improve the accuracy of reservoir power generation output prediction, and is used to optimize water energy utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is the decomposition result of the power output data of the BFAST method of the present invention; wherein, Figure 2 (a) is the measured value of power generation output. Figure 2 (b) is the periodic term after the power output data is decomposed. Figure 2 (c) is the trend item after the power output data is decomposed. Figure 2 (d) is the residual term after the decomposition of power output data; Figure 3 is the simulation result of the residual error of power generation in the embodiment, where: Figure 3 (a) is a comparison chart of the measured residuals and simulated residuals during the rate period. Figure 3 (b) is a scatter plot of the measured residuals and simulated residuals within the rate period. Figure 3 (c) is a comparison chart of the measured residuals and simulated residuals during the validation period. Figure 3 Middle (d) is the scatter plot of measured residuals and simulated residuals during the validation period; Figure 4 The power generation flow simulation result of the embodiment; Figure 5 is the reservoir water level data decomposition result in the embodiment; wherein, Figure 5 (a) is the periodic term after the reservoir water level data is decomposed. Figure 5 (b) is the trend item after the reservoir water level data is decomposed. Figure 5 (c) is the residual term after decomposition of reservoir water level data; Figure 6 is the decomposition result of the inflow data in the embodiment; wherein, Figure 6 (a) is the periodic term after the inflow data is decomposed. Figure 6 (b) is the trend term after the inflow data is decomposed. Figure 6 (c) is the residual term after the inflow data is decomposed. DETAILED DESCRIPTION

[0038] The technical method of the present invention is further described below in conjunction with the accompanying drawings and specific examples.

[0039] Example

[0040] This embodiment uses the Ertan Reservoir in the Yalong River Basin as the research reservoir, and the research period is from January 1, 2002 to December 31, 2018. The reservoir power flow calculation method based on the coupling of physical methods and deep learning is as follows: Figure 1 , including the following steps: Step 1: Collect historical reservoir operation characteristic data, including reservoir water level, inflow, power generation output and power generation flow, and perform data cleaning, remove outliers and fill in missing data.

[0041] The collected historical reservoir operation characteristic data are daily scale data.

[0042] Step 2: Use the BFAST method to decompose the reservoir water level, power generation output and inflow data respectively, and obtain the periodic term, trend term and residual term of the reservoir water level, power generation output and inflow data respectively. The periodic term reflects the changes in the power generation capacity of the reservoir in different seasons, while the trend term shows the overall change trend of the reservoir power generation output over time. The residual term mainly represents the noise caused by changes in reservoir water level, inflow or manual operation in the short term.

[0043] The decomposition results of power generation are as follows: Figure 2 As shown in (a), (b), (c), and (d), Figure 2 (a) is the measured value of power generation output Y t , Figure 2 (b) is the periodic term after the power output data is decomposed S t , Figure 2 (c) is the trend item after the power output data is decomposed T t , Figure 2 (d) is the residual term after the power output data is decomposed e t . Through the analysis of the data of Ertan Reservoir from 2002 to 2017, it was found that significant mutations occurred in 2007 and 2014: the mutation in 2007 was mainly manifested in an increase in the amplitude of output fluctuations, which may be related to the fluctuation of power demand and the performance of peak and frequency regulation functions; the mutation in 2014 may be related to the commissioning of downstream cascade power stations, resulting in a decrease in Ertan power generation output, although the trend item is still rising. In addition, as the complexity of scheduling increases, it gradually increases, which may be related to the joint scheduling of cascade power stations and fluctuations in water inflow.

[0044] like Figure 5 (a), (b), and (c) show the periodic term, trend term, and residual term after the decomposition of the reservoir water level data.

[0045] like Figure 6 (a), (b), and (c) show the periodic term, trend term, and residual term data after inflow decomposition, respectively. Figure 2-6 All data are daily scale data. Because the series is long, only the year is marked.

[0046] Step 3. After obtaining the periodic term, trend term and residual term, the periodic term, trend term and residual term of the reservoir water level and inflow are used as input variables, the residual term of power generation output is used as the target variable, and the rate period is 2002-2012. An LSTM deep learning model is constructed and trained to enable the model to learn the complex time series relationship between the power generation output residual and the input variables.

[0047] In this embodiment, the key parameters of the LSTM model learning performance are set as: number of hidden layers 4, training rounds 100, number of neurons 256, learning rate 0.0001, and batch size 32.

[0048] Step 4: Use the trained LSTM deep learning model to predict the power generation residual. The performance of the power generation residual simulation is measured by the systematic deviation BIAS and the correlation coefficient. R To evaluate, the calculation formula is:

[0049] Among them, X s1 is the simulated value of power generation residual, X o1 is the measured value of the power generation output residual, and are the means of the simulated and measured values ​​of power generation output residual, and are the time series numbers of the measured values.

[0050] The measured value is the residual term obtained by decomposition using the BFAST method, and the simulated value is the residual term obtained by LSTM simulation.

[0051] The simulation results within the period are as follows Figure 3 As shown in Figure a, the measured residuals and simulated residuals are generally consistent, but there are deviations at some peaks and valleys, which may be due to the limited ability of the model to capture extreme values. Figure 3 As shown in b, the correlation coefficient between the measured value and the simulated value is 0.87, and the fitting effect is good, but the deviation value is -47.35, indicating that the model has a certain underestimation phenomenon. The simulation results during the validation period are as follows Figure 3 As shown in Figure c, the model can capture the overall trend of the residuals, but the simulation of some larger peaks is not accurate enough. The scatter plot of the measured residuals and the simulated residuals during the validation period is shown in Figure Figure 3 As shown in (d), the deviation in the validation period is reduced to -26.97, and the correlation coefficient remains at 0.87, indicating that the model performance has improved.

[0052] The power generation output residual mainly reflects the random noise caused by factors such as water level fluctuations, inflow changes or human intervention in the short term. The LSTM deep learning model has a strong ability to process time series information and can effectively identify complex time-dependent patterns in output errors. As a key input variable, inflow can reflect the dynamic evolution of climate and hydrological conditions in the basin and directly affect the short-term fluctuations in power generation output. The water level of a reservoir not only reflects the water storage state of the reservoir, but is also closely related to the dispatching rules, reflecting the regulation capacity and water resource utilization efficiency of the reservoir at different operation stages, and reflecting the periodicity and trend change characteristics of the reservoir operation. By taking the periodic term, trend term and noise term (i.e., residual term) of inflow and water level as input, the LSTM deep learning model can fully capture the long-term and short-term dependencies of these time series data, thereby deeply mining the nonlinear association and time series characteristics between the output residual and the input variables.

[0053] Step 5: Based on the hydropower conversion formula, convert the predicted total power generation output of the reservoir into power generation flow Q:

[0054] Wherein, P is the power output, i.e. the total power output mentioned above; is the efficiency of the generator, is the density of water, is the acceleration due to gravity, H is the hydraulic head, which refers to the difference between the water level upstream and downstream of the reservoir; The performance of the power generation flow simulation is evaluated by the efficiency coefficient (NSE) and relative error (RE), which are calculated as follows:

[0055] Among them, X s2 is the simulated value of power generation flow, X o2 is the measured value of power generation flow, and are the means of the simulated and measured values ​​of power generation flow, and are the time series numbers of the measured values.

[0056] The simulation results of power generation flow are as follows: Figure 4As shown in the figure, the simulation effect of this method is relatively good during the rate period and the verification period. Although the power generation flow has large fluctuations during the day, the deep learning model can capture the output residual more accurately, reflecting the volatility under the action of daily regulation. By adding the simulated residual term to the periodic term and the trend term, the power generation output result that is closer to the measured value is finally obtained. This method achieved a higher NSE (Nash efficiency coefficient) and RE (relative error) during the rate period, specifically: NSE is 0.91, RE is -0.05; and during the verification period, although the model performance declined slightly, NSE was 0.80, RE was -0.03, but it still showed a relatively ideal simulation effect.

[0057] The above results show that the method proposed in the present invention can effectively decompose and extract the periodic fluctuations, trend evolution and residual characteristics of the reservoir power generation flow, and use the deep learning model to fully explore the nonlinear dynamic relationship between the power generation output residual and key influencing factors such as reservoir water level and inflow. The power generation output sequence data simulated by the method of the present invention and the reconstructed power generation output sequence not only better reproduce the fluctuation characteristics and stage-by-stage change trends under the daytime scheduling of the reservoir, but also significantly improve the simulation and prediction accuracy of the power generation output under complex scheduling conditions. This method fully utilizes the advantages of multiple decompositions and deep learning models.

[0058] The above is only a preferred embodiment of the present invention. The protection scope of the present invention is not limited to the above embodiment. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.

Claims

1. A reservoir power flow calculation method based on the coupling of physical methods and deep learning, characterized in that: include: Step 1: Collect historical reservoir operation characteristic data, including reservoir water level, inflow, power generation output and power generation flow, and perform data cleaning, remove outliers and fill in missing data; Step 2: Use the BFAST method to decompose the reservoir water level, inflow and power output data to obtain the periodic term, trend term and residual term data of the reservoir water level, inflow and power output; Step 3: Using the periodic term, trend term and residual term of reservoir water level and inflow as input variables and the residual term of power generation output as the target variable, a deep learning model is constructed and trained to enable the model to learn the temporal relationship between the residual term of power generation output and the input variables; Step 4: Use the trained deep learning model to predict the power generation output residual, and add the predicted residual term to the period term and trend term of the corresponding period to obtain the total power generation output of the reservoir; Step 5: Based on the hydropower conversion formula, convert the total power generation output of the reservoir into power generation flow.

2. The reservoir power generation flow calculation method based on physical method and deep learning coupling according to claim 1 is characterized in that: The reservoir historical operation characteristic data collected in step 1 are daily scale data.

3. The reservoir power generation flow calculation method based on physical method and deep learning coupling according to claim 1 is characterized in that: The deep learning model described in step 3 is an LSTM model.

4. The reservoir power generation flow calculation method based on physical method and deep learning coupling according to claim 1 is characterized in that: The performance of the power generation residual simulation in step 4 is measured by the systematic deviation and correlation coefficient R To evaluate, the calculation formula is: ； Among them, X s1 is the simulated value of power generation residual, X o1 is the measured value of the power generation output residual, and are the means of the simulated and measured values ​​of power generation output residual, and are the time series numbers of the measured values.

5. The reservoir power generation flow calculation method based on physical method and deep learning coupling according to claim 1 is characterized in that: The performance of the power generation flow simulation in step 5 is evaluated by the efficiency coefficient NSE and the relative error RE, which are calculated as follows: ； Among them, X s2 is the simulated value of power generation flow, X o2 is the measured value of power generation flow, and are the means of the simulated and measured values ​​of power generation flow, and are the time series numbers of the measured values.

Images