Long-lead-time photovoltaic power forecasting method based on power reconstruction and timing constraints
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-12
- Publication Date
- 2026-08-11
AI Technical Summary
[0003]现有研究通常使用NWP关键气象要素作为功率预测的输入，难以解决NWP预报误差对功率预测精度的影响
[0032] Through the above design scheme, the present invention can bring the following beneficial effects: a long-foresight photovoltaic power prediction method based on power reconfiguration and timing constraints, which fully considers the output characteristics of photovoltaic power for prediction, has clear physical meaning, higher practical value, and higher accuracy, and is suitable for long-foresight photovoltaic power cluster prediction; it can also evaluate other photovoltaic power predictions.
Smart Images
Figure CN117239724B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of photovoltaic power prediction technology, and in particular relates to a long-foresight photovoltaic cluster power prediction method based on power reconfiguration and time series constraints. Background Technology
[0002] To reduce carbon emissions and mitigate the environmental crisis, the installed capacity of new energy sources has been increasing year by year, with photovoltaic (PV) power generation growing rapidly due to the convenient access to solar resources. Accurate forecasting of new energy power and maximizing economic benefits have always been hot topics in the power system. However, due to the complex and variable nature of daily weather processes, PV output exhibits extremely high uncertainty, posing a severe challenge to the stable operation of the power grid. Accurate PV power forecasting is crucial for its grid connection. Because the reliability of long-term numerical weather forecasts decreases with increasing time scale, the autocorrelation of PV power gradually decreases, making it difficult to effectively predict PV power at high time scales. The gradually increasing installed capacity of PV has brought enormous challenges to large-scale PV grid connection. To ensure the safe and stable operation of the power system, it is necessary to break through the limitations of longer-term PV power forecasting.
[0003] Existing studies typically use key meteorological elements of the National Photovoltaic Power Spectrum (NWP) as input for power forecasting, which makes it difficult to address the impact of NWP forecast errors on power forecast accuracy. Forecasting methods usually consider predicting the power itself without effectively extracting predictable information, further limiting the improvement of photovoltaic power forecast accuracy over high timescales.
[0004] Therefore, a new technical solution is urgently needed in the existing technology to solve the above problems. Summary of the Invention
[0005] The technical problem to be solved by this invention is to provide a photovoltaic power prediction method for long-term forecast based on power reconfiguration and timing constraints, which comprehensively considers the daily correlation characteristics of electricity and the mapping characteristics between irradiance and electricity, has higher practical value and higher accuracy, and can accurately predict photovoltaic output over a long period of time.
[0006] The long-term photovoltaic power prediction method based on power reconfiguration and time-series constraints includes the following steps, which are performed sequentially.
[0007] Step 1: Calculation of Electricity Using Approximate Integration Method
[0008] Given power and irradiance at a 15-minute resolution, the area enclosed by a line graph connecting 96 discrete power points daily and the time axis is used to approximate the daily electricity generation and daily irradiance. The formulas for calculating daily photovoltaic electricity generation and irradiance are as follows:
[0009]
[0010] Where: z is the daily electricity consumption or daily irradiance; i is the time of day; y(i) is the power value corresponding to each time point;
[0011] Step 2: Decomposition of daily electricity consumption and irradiance based on variational mode decomposition (VMD)
[0012] The photovoltaic daily electricity and irradiance sequences obtained in step one are decomposed using VMD, which decomposes the non-stationary and irregular sequences into multiple stationary and regular component sequences, and extracts the intrinsic features of the electricity and irradiance sequences.
[0013] Step 3: Time series forecasting based on a multiple linear regression model
[0014] The decomposed electrical charge and the corresponding irradiance component at the center frequency are input into a multiple linear regression model. This model combines electrical charge time-series extrapolation and the mapping between irradiance and electrical charge to predict electrical charge over a long-term forecast period. The formula for the multiple linear regression model is:
[0015] Y = β0 + β1X1 + β2X2 + ... + β n X n +ε (2)
[0016] Where: Y represents the output variable, X1, X2, ..., X n Let β0, β1, β2, ..., β be the input variables. n ε represents the unknown parameter, and ε represents the error term;
[0017] Step 4: Energy-Power Reconfiguration Based on Timing Constraints
[0018] The start and end points of daily photovoltaic power output are fitted with a sine function to create a sine curve above the time axis. The area enclosed by this sine curve and the time axis is approximately equal to the daily electricity consumption. In other words, the daily electricity consumption is reconstructed into the predicted power based on the distribution pattern of photovoltaic power output. The formula for the sine function is:
[0019]
[0020] Where y is the reconstruction power, a1 is the amplitude, and w is the angular frequency. This is the initial phase;
[0021] The starting and ending points of daily photovoltaic power output are fitted with a downward-opening quadratic curve using the two-point equation of a quadratic curve. The area enclosed by the quadratic curve and the time axis represents the daily electricity consumption. The formula for the two-point equation of the quadratic curve is:
[0022] y = a²·(x-x1)(x-x2) (4)
[0023] Where y is the reconfiguration power, a2 is the quadratic coefficient, x1 is the starting point of daily photovoltaic power output, and x2 is the ending point of daily photovoltaic power output;
[0024] Based on the distribution characteristics of photovoltaic power output, the predicted electricity is reconstructed into predicted power under time constraints.
[0025] Based on steps one through four, establish simulation input quantities, analyze the measured data of the electric field, and determine the total installed capacity of the electric field; the sampling interval for power and NWP data is 15 minutes; perform simulation calculations to obtain the long-term photovoltaic power cluster prediction results.
[0026] Using the accuracy index RMSE P MAPE, R 2 The standard formula for evaluating the accuracy index of long-term photovoltaic power cluster prediction results is as follows:
[0027]
[0028]
[0029]
[0030] Among them, RMSE P RMSE is the root mean square error of power prediction, Cap is the cluster installed capacity, N is the number of test samples, and P is the value of P. i This represents the actual value of photovoltaic power generation. This is the predicted value of photovoltaic power generation.
[0031] The predicted power calculated by the model is compared with the measured power using the error evaluation standards (5), (6), and (7) to calculate the error and obtain the prediction accuracy.
[0032] Through the above design scheme, the present invention can bring the following beneficial effects: a long-foresight photovoltaic power prediction method based on power reconfiguration and timing constraints, which fully considers the output characteristics of photovoltaic power for prediction, has clear physical meaning, higher practical value, and higher accuracy, and is suitable for long-foresight photovoltaic power cluster prediction; it can also evaluate other photovoltaic power predictions. Attached Figure Description
[0033] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:
[0034] Figure 1 This is a framework diagram of the long-foresight photovoltaic power prediction method based on power reconfiguration and timing constraints of the present invention.
[0035] Figure 2 This diagram illustrates the daily and seasonal characteristics of photovoltaic power in the long-foresight photovoltaic power prediction method based on power reconfiguration and timing constraints of this invention.
[0036] Figure 3This diagram illustrates the comparison between predicted and actual values of the long-foresight photovoltaic power prediction method based on power reconfiguration and timing constraints according to the present invention. Detailed Implementation
[0037] A long-term photovoltaic power prediction method based on power reconfiguration and time-series constraints, combined with Figures 1-3 This includes the following steps:
[0038] Step 1) Convert the power sequence and irradiance sequence into daily electricity sequence and daily irradiance energy sequence, respectively.
[0039] Step 2) Decompose the electrical charge sequence and the irradiation energy sequence to extract various features of the time series.
[0040] Step 3) Predict each electrical component, accumulate and reconstruct to obtain the predicted electrical quantity.
[0041] Step 4) Based on the distribution characteristics of photovoltaic power output, reconstruct the predicted electricity into predicted power under time constraints.
[0042] Specifically, step 1 uses the approximate integration method to calculate the amount of electricity.
[0043] Given the power and irradiance at a 15-minute resolution, the daily electricity consumption and daily irradiance can be approximated by the area enclosed by a line graph connecting 96 discrete power points per day and the time axis. The calculation method for electricity consumption and irradiance can be expressed by equation (1):
[0044]
[0045] Where: z is the daily electricity consumption or daily irradiance; i is the time of day; y(i) is the power value corresponding to each time point;
[0046] Step 2 employs VMD-based decomposition of daily electricity consumption and irradiance.
[0047] In the VMD decomposition process, variables are decomposed into multiple effective amplitude-modulated and frequency-modulated sub-signals with finite bandwidths, exhibiting sparsity in the reconstruction of the input signal. The main goal of VMD is to decompose the input signal into many sub-signals, commonly referred to as modes. This is an adjustable method that can predict modes while balancing their inaccuracies. This technique is based on three fundamental theorems: Wiener filtering, Hilbert transform, and analytic signals. VMD utilizes the alternating direction method to improve the aggregation bandwidth of modes, which is achieved through a combination of multiple methods. This invention uses VMD to decompose photovoltaic daily electricity and irradiance sequences, decomposing non-stationary, irregular sequences into multiple stationary, regular component sequences, and extracting multiple intrinsic features of the electricity and irradiance sequences.
[0048] Step 3 uses a multiple linear regression model for time series forecasting.
[0049] This invention converts power into electrical quantity for time-series extrapolation prediction. The invention decomposes the electrical quantity and inputs the corresponding irradiance energy component at the center frequency into a multiple linear regression model, combining electrical quantity time-series extrapolation and irradiance energy-electricity mapping to predict electrical quantity over a long-term prediction period. The principle formula of the multiple linear regression model is Equation (2):
[0050] Y = β0 + β1X1 + β2X2 + ... + β n X n +ε (2)
[0051] Where: Y represents the output variable, X1, X2, ..., X n Let β0, β1, β2, ..., β be the input variables. n Let represent the unknown parameter, and ε represent the error term.
[0052] Step 4 employs a time-constrained charge-power reconfiguration method.
[0053] It is known that the start and end points of daily photovoltaic power output can be fitted with a sine function to form a sine curve above the time axis. The area enclosed by this sine curve and the time axis is approximately equal to the daily electricity consumption. That is, the daily electricity consumption can be reconstructed into the predicted power according to the distribution law of photovoltaic power output. The sine function formula is shown in (3):
[0054]
[0055] Where y is the reconstruction power, a1 is the amplitude, and w is the angular frequency. This is the initial phase.
[0056] Similarly, given the start and end points of daily photovoltaic power output, a downward-opening quadratic curve can be fitted using the two-point equation of a quadratic curve. The area enclosed by the quadratic curve and the time axis represents the daily electricity consumption. The two-point equation of the quadratic curve is shown in (4):
[0057] y = a²·(x-x1)(x-x2) (4)
[0058] Where y is the reconfiguration power, a2 is the coefficient of the quadratic term, x1 is the starting point of daily photovoltaic power output, and x2 is the ending point of daily photovoltaic power output.
[0059] Simulation calculations are performed on steps 1-4 to obtain the long-term photovoltaic power cluster prediction results;
[0060] To evaluate the quality of the model's predictions, RMSE was used. P MAPE, R 2 To evaluate the prediction results. The standard formula for the above accuracy indicators is expressed as:
[0061]
[0062]
[0063]
[0064] Among them, RMSE P RMSE (Root Mean Square Error) for power prediction, Cap is the cluster installed capacity, N is the number of test samples, and P is the power prediction value. i This represents the actual value of photovoltaic power generation. This represents the predicted photovoltaic power generation.
[0065] Input the simulation input quantity, and calculate the error between the predicted power calculated by the model and the measured power using the error evaluation standard formulas (5), (6), and (7) to obtain the prediction accuracy.
[0066] Specific example analysis
[0067] This invention analyzes measured data from a photovoltaic power plant cluster in Gansu Province as an example, with a sampling interval of 15 minutes. The total installed capacity of this photovoltaic power plant cluster is 419MW; the RMSE (Real-Time Sequence) index was selected as the evaluation index for the prediction results. P MAPE, R 2 .
[0068] Table 1 Comparison of prediction results from different models
[0069] Tab.1 Comparison of prediction results of different models
[0070]
[0071]
[0072] The specific embodiments of the present invention have been described in detail, but are not limited to these embodiments. Any obvious modifications made by those skilled in the art based on the teachings of the present invention are within the scope of protection of the present invention.
Claims
1. A long-term photovoltaic power prediction method based on power reconfiguration and time-series constraints, characterized by: The steps are as follows, and the steps are performed in sequence. Step 1: Calculation of Electricity Using Approximate Integration Method Given power and irradiance at a 15-minute resolution, the area enclosed by a line graph connecting 96 discrete power points daily and the time axis is used to approximate the daily electricity generation and daily irradiance. The formulas for calculating daily photovoltaic electricity generation and irradiance are as follows: Where: z is the daily electricity consumption or daily irradiance; i is the time of day; y(i) is the power value corresponding to each time point; Step 2: Decomposition of daily electricity consumption and irradiance based on variational mode decomposition (VMD) The photovoltaic daily electricity and irradiance sequences obtained in step one are decomposed using VMD, which decomposes the non-stationary and irregular sequences into multiple stationary and regular component sequences, and extracts the intrinsic features of the electricity and irradiance sequences. Step 3: Time series forecasting based on a multiple linear regression model The decomposed electrical charge and the corresponding irradiance component at the center frequency are input into a multiple linear regression model. This model combines electrical charge time-series extrapolation and the mapping between irradiance and electrical charge to predict electrical charge over a long-term forecast period. The formula for the multiple linear regression model is: Y=β0+β1X1+β2X2+…+β n X n +e (2) Where: Y represents the output variable, X1, X2, ..., X n Let β0, β1, β2, ..., β be the input variables. n ε represents the unknown parameter, and ε represents the error term; Step 4: Energy-Power Reconfiguration Based on Timing Constraints The start and end points of daily photovoltaic power output are fitted with a sine function to create a sine curve above the time axis. The area enclosed by this sine curve and the time axis is approximately equal to the daily electricity consumption. In other words, the daily electricity consumption is reconstructed into the predicted power based on the distribution pattern of photovoltaic power output. The formula for the sine function is: Where y is the reconstruction power, a1 is the amplitude, and w is the angular frequency. This is the initial phase; The starting and ending points of daily photovoltaic power output are fitted with a downward-opening quadratic curve using the two-point equation of a quadratic curve. The area enclosed by the quadratic curve and the time axis represents the daily electricity consumption. The formula for the two-point equation of the quadratic curve is: y = a²·(x-x1)(x-x2) (4) Where y is the reconfiguration power, a2 is the quadratic coefficient, x1 is the starting point of daily photovoltaic power output, and x2 is the ending point of daily photovoltaic power output; Based on the distribution characteristics of photovoltaic power output, the predicted electricity is reconstructed into predicted power under time constraints.
2. The long-term photovoltaic power prediction method based on power reconfiguration and timing constraints according to claim 1, characterized in that: Based on steps one through four, establish simulation input quantities, analyze the measured data of the electric field, and determine the total installed capacity of the electric field; the sampling interval for power and NWP data is 15 minutes; perform simulation calculations to obtain the long-term photovoltaic power cluster prediction results.
3. The long-term photovoltaic power prediction method based on power reconfiguration and timing constraints according to claim 2, characterized in that: Using the accuracy index RMSE P MAPE, R 2 The standard formula for evaluating the accuracy index of long-term photovoltaic power cluster prediction results is as follows: Among them, RMSE P RMSE is the root mean square error of power prediction, Cap is the cluster installed capacity, N is the number of test samples, and P is the value of P. i This represents the actual value of photovoltaic power generation. This is the predicted value of photovoltaic power generation. The predicted power calculated by the model is compared with the measured power using the error evaluation standards (5), (6), and (7) to calculate the error and obtain the prediction accuracy.