A distributed photovoltaic output prediction method and system based on decoupling and DTW clustering of clearance models

Abstract

The application discloses a distributed photovoltaic output prediction method and system based on a clearance model decoupling and DTW clustering, and relates to the technical field of new energy power generation. The method calculates theoretical output under ideal clear sky conditions by constructing a clearance model, extracts a cloudiness attenuation factor by using a seasonal decoupling method, and performs clustering analysis on a normalized clear sky index sequence by using a K-medoids clustering model based on dynamic time warping (DTW) distance measurement, so as to finally identify a typical weather scene and construct a multi-season and multi-scenario photovoltaic output time sequence model library. The application effectively separates seasonal trends and random fluctuations, can accurately identify output curves that are similar in shape but time-displaced, and provides a more reliable basis for power grid dispatching.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation technology, specifically to a distributed photovoltaic power output modeling method based on headroom model decoupling and DTW clustering. Background Technology

[0002] Distributed photovoltaic (PV) systems exhibit significant intermittency and volatility. Existing modeling methods typically use mixed data from across the year, neglecting the significant differences in meteorological conditions across seasons (such as high temperatures in summer and haze in winter), which limits model accuracy. Furthermore, traditional clustering methods (such as K-means combined with Euclidean distance) are insufficient in capturing the morphological similarity of time series data and cannot effectively handle the timeline misalignment caused by cloud movement. Summary of the Invention

[0003] The technical problem this invention aims to solve is to overcome the shortcomings of existing distributed photovoltaic (PV) power output modeling methods, such as ignoring seasonal characteristics, low accuracy in matching temporal fluctuations, and poor robustness in scene recognition. It provides a distributed PV power output prediction method and system based on headroom model decoupling and DTW clustering. By integrating the physical mechanism of PV power output with data-driven clustering, eliminating the influence of meteorological amplitude differences through seasonal component decoupling, and improving scene recognition accuracy using an improved clustering algorithm, this invention constructs a multi-seasonal, multi-scene power output model library adapted to different regions and meteorological conditions. This significantly improves the accuracy and engineering applicability of distributed PV power output modeling, providing reliable support for PV power prediction, grid absorption analysis, and optimized scheduling.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0005] First, this invention provides a distributed photovoltaic power output prediction method based on headroom model decoupling and DTW clustering, comprising the following steps:

[0006] (2) Collect historical output data of distributed photovoltaic power, meteorological environment data and operation data of the transformer area, complete the data missing value filling, outlier removal and standardization preprocessing, and construct a standardized modeling dataset.

[0007] (3) Based on the solar position algorithm, atmospheric attenuation model and photovoltaic module temperature correction model, a net air output model is constructed to calculate the theoretical photovoltaic net air output sequence under ideal clear sky conditions and extract the seasonal deterministic component of output.

[0008] (4) Based on the theoretical and measured power output of the net air, a cloud cover attenuation factor is constructed, and the clear sky index (CSI) is introduced to complete the data normalization process, eliminate the difference in amplitude between seasons and time periods, and retain the characteristics of random meteorological fluctuations.

[0009] (5) Using dynamic time warping (DTW) as the distance metric, the K-medoids clustering algorithm is used to iteratively cluster the normalized clear sky index sequence. The elbow rule and the profile coefficient are combined to determine the optimal number of clusters and identify typical weather power output scenarios.

[0010] (6) Statistically analyze the probability and characteristic indicators of each typical scenario in the four seasons, establish a multi-dimensional power output model library including seasonal type, scenario category, probability weight, and power output characteristics, and complete the accurate modeling of distributed photovoltaic power output.

[0011] Furthermore, the net air output model described in step (2) calculates the solar zenith angle, atmospheric mass and clear sky index, and combines the photovoltaic module power temperature coefficient and NOCT temperature model to obtain the seasonal benchmark output without cloud interference, thereby achieving physical decoupling between deterministic trends and random fluctuations.

[0012] Among them, the solar zenith angle satisfies:

[0013] ,

[0014] In the formula: The zenith angle of the sun. The latitude is the local latitude. The solar declination angle, It is the hour angle.

[0015] The formula for calculating net air output is:

[0016] ,

[0017] In the formula: Rated power, This represents the actual irradiance. The reference irradiance under standard test conditions. For power temperature coefficient, For the component operating temperature, T STC This is the reference temperature under standard test conditions.

[0018] Furthermore, the clear sky index normalization mentioned in step (3) involves making a ratio between the measured per-unit power output and the theoretical per-unit power output of the net air, completely removing the influence of seasonal factors such as solar altitude angle and sunshine duration, so that data from different seasons and different time periods have a homogeneous clustering basis.

[0019] The clear sky index is defined as follows:

[0020] ,

[0021] In the formula: To measure the per-unit force value, This is the per-unit value of net air output.

[0022] Furthermore, the DTW distance metric described in step (4) can achieve nonlinear time axis alignment of photovoltaic power output time series, effectively capturing the power output offset characteristics caused by cloud movement; the K-medoids algorithm uses actual data points as cluster centers, significantly improving robustness to outliers and measurement noise.

[0023] The DTW distance calculation formula is as follows:

[0024] ,

[0025] in, This represents the photovoltaic power output value corresponding to the k-th matching position in the original time series X after dynamic warping and matching. Represents the original time series sequence Y, and The photovoltaic output power value corresponding to the kth position of the optimal match; K and m represent the total number of optimal matching points on the DTW dynamic warping path and the original fixed sampling dimension of a single photovoltaic output time series sample, respectively.

[0026] Furthermore, the typical weather scenarios mentioned in step (5) include five categories: sunny, cloudy, overcast, light rain, and heavy rain. The output characteristics include peak power, fluctuation rate, ramp rate, and duration. The model library can be directly used for photovoltaic forecasting, distribution network carrying capacity assessment, and scheduling optimization.

[0027] On the other hand, the present invention provides an electronic system including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0028] Finally, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0029] By adopting the above technical solution, compared with the prior art, the present invention has the following technical effects:

[0030] (1) By decoupling through the net model, seasonal trends and random fluctuations are effectively separated, which improves the physical interpretability of the model.

[0031] (2) Using DTW distance measurement, it is possible to accurately identify output curves that are similar in shape but misaligned in time (such as the scenario of rapid cloud movement).

[0032] (3) A seasonal scenario library was constructed, which can more accurately reflect the spatiotemporal distribution characteristics of photovoltaic power output and provide a more reliable basis for grid dispatch. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0035] Figure 2 Flowchart for constructing the clearance model.

[0036] Figure 3 This is a schematic diagram of seasonal decoupling.

[0037] Figure 4 The flowchart is for K-medoids+DTW clustering analysis.

[0038] Figure 5 This is a typical weather scene recognition result image. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] As shown in Figure 1, a distributed photovoltaic power output prediction method based on headroom model decoupling and DTW clustering includes the following steps:

[0041] (1) Collect historical power output data and meteorological environment data of distributed photovoltaic power generation, and perform data preprocessing;

[0042] (2) Based on the solar position algorithm and atmospheric attenuation model, a photovoltaic net air output model is constructed, and the theoretical net air output sequence is calculated;

[0043] (3) Based on the net air volume theory, seasonal decoupling is performed, cloud cover attenuation factor is calculated and data normalization is completed through clear sky index;

[0044] (4) Clustering was performed using DTW distance metric and K-medoids algorithm to identify typical weather power output scenarios;

[0045] (5) Statistical analysis of scenario probabilities and characteristic indicators to construct a multi-season, multi-scenario photovoltaic power output model library.

[0046] The specific implementation process of each step of the present invention will be further described in detail below:

[0047] Step 1: Data acquisition and preprocessing.

[0048] Collect historical output data of distributed photovoltaic power, total horizontal irradiance, ambient temperature, wind speed, cloud cover, atmospheric pressure and other meteorological data in the target area; preprocess the collected data, including linear interpolation to complete missing values, outlier removal, and data standardization to eliminate amplitude differences caused by different installed capacities and measurement units, and form a standardized modeling dataset.

[0049] Step 2: Construct a net air output model and calculate the seasonal deterministic components.

[0050] like Figure 2 As shown, the clearance model construction process is divided into three parts: input module, core calculation module, and output module. The specific implementation process is as follows:

[0051] The input module takes in the standardized dataset preprocessed in step 1, which mainly includes geographical parameters of photovoltaic installations, time-series meteorological data, and basic photovoltaic module parameters. The geographical parameters include the local latitude, time zone, and longitude / latitude of the photovoltaic site; the meteorological data includes hourly total horizontal irradiance and ambient temperature; and the module parameters include fixed parameters such as rated power, power temperature coefficient, standard test temperature, and NOCT operating temperature of the photovoltaic modules, providing raw data support for the entire process of net airflow output calculation.

[0052] The core calculation module is the main part of the clearance model. It completes five major steps in sequence: solar position calculation, atmospheric parameter calculation, irradiance correction, component temperature correction, and theoretical output calculation. The specific execution logic is as follows:

[0053] Based on the installation location of the photovoltaic modules, the local latitude and time zone are determined, and the solar zenith angle is calculated using a solar position algorithm:

[0054] ,

[0055] In the formula: The zenith angle of the sun. The latitude is the local latitude. The solar declination angle, It is the hour angle.

[0056] By combining the atmospheric quality model and the Liu-Jordan clear sky index model, the irradiance sequence under clear sky conditions was calculated; based on the NOCT module temperature model and power temperature coefficient, the theoretical power output under ideal clear sky conditions was calculated.

[0057] ,

[0058] In the formula: Rated power, This represents the actual irradiance. The reference irradiance under standard test conditions. For power temperature coefficient, For the component operating temperature, T STC This is the reference temperature under standard test conditions.

[0059] The output module outputs a time-by-time theoretical photovoltaic power output sequence. This sequence is only affected by the solar orbit, seasonal changes, and the inherent atmospheric attenuation characteristics. It is a seasonal deterministic component of photovoltaic power output, completely eliminating the interference of random factors such as clouds and short-term weather fluctuations. This provides a benchmark sequence for subsequent cloud attenuation factor calculation, data normalization, and cluster analysis.

[0060] Step 3: Seasonal decoupling and clear sky index normalization.

[0061] like Figure 3 As shown, through a three-level processing flow of "cloud cover attenuation factor calculation - clear sky index extraction - data standardization", the seasonal deterministic component and the meteorological random component in photovoltaic power output are completely decoupled, while eliminating the amplitude differences in different seasons and time periods and retaining the core fluctuation characteristics.

[0062] Input the theoretical net clearance output sequence calculated in step 2. Compared with the measured photovoltaic power output sequence after step 1 preprocessing ; Output power P based on the theory of clear space pv (t) is the deterministic baseline component; calculate the cloud cover attenuation factor:

[0063] ,

[0064] Introducing the Clear Sky Index (CSI) to normalize data and eliminate seasonal and time-of-day amplitude differences:

[0065] ,

[0066] In the formula: To measure the per-unit force value, This is the per-unit value of net air output.

[0067] The CSI sequences were denoised using a 15-minute moving average and then normalized using min-max to obtain a normalized time-series dataset for clustering.

[0068] Step 4: Typical scene recognition based on DTW-Kmedoids.

[0069] like Figure 4As shown, a typical scene recognition method combining DTW distance metric and K-medoids clustering algorithm is used. This method automatically classifies the normalized clear sky index sequence by iteratively optimizing the cluster centers. The complete execution flow is as follows:

[0070] Using Dynamic Time Warping (DTW) as the distance metric, the similarity between two photovoltaic power output time series curves is calculated:

[0071] ,

[0072] Where, x p(k) y represents the photovoltaic power output value corresponding to the k-th matching position in the original time series X after dynamic warping and matching. q(k) Represents the original time series sequence Y, and The photovoltaic output power value corresponding to the kth position of the optimal match; K and m represent the total number of optimal matching points on the DTW dynamic warping path and the original fixed sampling dimension of a single photovoltaic output time series sample, respectively.

[0073] The K-medoids clustering algorithm was used to iteratively cluster the standardized CSI sequence. The optimal number of clusters K=5 was determined by combining the elbow rule, silhouette coefficient and Calinski-Harabasz exponent, and five typical weather power output scenarios, namely sunny, cloudy, overcast, light rain and heavy rain, were identified.

[0074] Step 5: Build a multi-seasonal, multi-scenario output model library.

[0075] The system is divided into four seasons: spring (March-May), summer (June-August), autumn (September-November), and winter (December-February). The occurrence probability, average peak power, volatility, ramp-up rate, and duration of each typical scenario are statistically analyzed in different seasons. A scenario-probability pair model is constructed with scenario probability as the weight, and finally a multi-dimensional photovoltaic power output model library containing seasonal, scenario, probability, and power output characteristics is formed.

[0076] Figure 5 The normalized power output characteristic curves and corresponding cluster centers of five typical weather scenarios are shown. The core power output characteristics of each scenario are as follows:

[0077] Sunny day scenario (5.1 in the figure): It is characterized by high irradiance and stable output. The output curve has a standard single peak shape, with the peak value close to 1.0 and small fluctuation range. It is the most stable photovoltaic output scenario.

[0078] Cloudy scenario (Figure 5.2): It is characterized by large irradiance fluctuations and unstable output. The output curve has multiple peaks and troughs with large fluctuations, reflecting the intermittent blocking of solar radiation by clouds.

[0079] Cloudy day scenario (5.3 in the figure): It is characterized by low irradiance and stable low output. The output curve is generally flat with a peak value of about 0.4 and no obvious spikes. It is a scenario with low and stable photovoltaic output.

[0080] Light rain scenario (5.4 in the figure): It is characterized by extremely low radiation and weak output. The peak value of the output curve is about 0.2. The overall level is low and the fluctuation range is small.

[0081] Heavy rain scenario (5.5 in the figure): It is characterized by almost no radiation and scattered micro-output. The output curve is close to 0 most of the time, with only a few scattered micro-output points during the intervals between rainfall. It is the worst scenario for photovoltaic output.

[0082] The model was validated using the mean absolute percentage error (MAPE) and the coefficient of determination (R²). The results showed that the MAPE was ≤3.2% for sunny scenes, ≤7.8% for cloudy scenes, ≤11.5% for rainy scenes, and R² was ≥0.9 for all scenes. The modeling accuracy meets the requirements for engineering applications.

[0083] Example 2: This example provides an electronic system including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the method proposed in the above example.

[0084] Example 3: This example provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method proposed in the above examples.

[0085] It should be noted that the electronic system can use terminal devices such as desktop computers, laptops, or cloud servers. Furthermore, terminal devices include, but are not limited to, processors and memory. For example, terminal devices can also include input / output devices, network access devices, and buses.

[0086] Furthermore, the processor can be a central processing unit (CPU). Of course, depending on the actual use, other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be used. The general-purpose processor can be a microprocessor or any conventional processor, etc., and this application does not limit it in this regard.

[0087] Furthermore, the memory can be an internal storage unit of the terminal device, such as the hard disk or RAM of the terminal device, or an external storage device of the terminal device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD), or flash memory card (FC) equipped on the terminal device. In addition, the memory can also be a combination of the internal storage unit and the external storage device of the terminal device. The memory is used to store computer programs and other programs and data required by the terminal device. The memory can also be used to temporarily store data that has been output or will be output. This application does not limit this.

[0088] Furthermore, through this electronic system, any one of the methods in the above embodiments can be stored in the memory of the electronic system and loaded and executed on the processor of the terminal device for convenient use.

[0089] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A distributed photovoltaic power output prediction method based on headroom model decoupling and DTW clustering, characterized in that, Includes the following steps: Acquire historical power output data and corresponding meteorological environmental data of distributed photovoltaic power, and perform data cleaning and standardization. Construct a photovoltaic power output headroom model and calculate the headroom output sequence under ideal clear sky conditions; Based on the aforementioned net air output sequence, a seasonal decoupling method is used to extract the cloud cover attenuation factor, and a clear sky index is calculated to eliminate seasonal amplitude differences. A K-medoids clustering model based on dynamic time-warped distance metric is constructed to perform cluster analysis on the normalized clear sky index sequence; By identifying typical weather scenarios and combining them with seasonal distribution patterns, a photovoltaic power output time series model library is constructed, which includes seasonal types, scenario categories, probability weights, and power output characteristics, to perform real-time photovoltaic power output prediction.

2. The method as described in claim 1, characterized in that, The data cleaning and normalization process includes: S101. Data with irradiance less than 0 or greater than 1300 W / m² are excluded; S102, Remove data with output less than 0; S103. Convert the measured output power into a per-unit value based on the installed capacity.

3. The method as described in claim 1, characterized in that, The construction of the photovoltaic power output net clearance model specifically includes: S201. Calculate the solar zenith angle based on the solar position algorithm and calculate the atmospheric mass using the atmospheric mass model; S202. Calculate the clear sky index using the Liu-Jordan model and combine it with the module temperature correction coefficient to calculate the theoretical photovoltaic output under clear sky conditions. The formula for calculating net air output is: ， In the formula: Rated power, This represents the actual irradiance. The reference irradiance under standard test conditions. For power temperature coefficient, For the component operating temperature, T STC This is the reference temperature under standard test conditions.

4. The method as described in claim 1, characterized in that, The seasonal decoupling method is specifically as follows: S301, Calculate the cloud cover attenuation factor It is defined as the ratio of the measured per-unit output to the per-unit net output. S302, Calculate the Clear Sky Index It is defined as the ratio of the measured per-unit photovoltaic output to the theoretical per-unit output at the same time, and is used to characterize the degree of radiation attenuation under actual meteorological conditions relative to ideal clear-sky conditions.

5. The method as described in claim 1, characterized in that, The K-medoids clustering model is specifically as follows: Dynamic Time Warping (DTW) is used as the distance metric function to address the nonlinear offset of the photovoltaic power output curve on the time axis. We select actual data points as cluster centers and perform iterative clustering with the objective function of minimizing the sum of DTW distances from samples within a cluster to the center.

6. The method as described in claim 5, characterized in that, The DTW distance calculation formula is: ， in, This represents the photovoltaic power output value corresponding to the k-th matching position in the original time series X after dynamic warping and matching. Represents the original time series sequence Y with The photovoltaic output power value corresponding to the kth position of the optimal match; K and m represent the total number of optimal matching points on the DTW dynamic warping path and the original fixed sampling dimension of a single photovoltaic output time series sample, respectively.

7. The method as described in claim 1, characterized in that, The specific process of performing cluster analysis on the normalized clear sky index sequence is as follows: S401. Randomly select K actual sample points as the initial Medoids; S402. Calculate the DTW distance of all samples to Medoids and classify them accordingly; S403. Find the point within the cluster that minimizes the total distance and use it as the new Medoids; S404. Repeat the above process until convergence.

8. The method as described in claim 1, characterized in that, The model library construction includes statistically analyzing the probability of occurrence of each scenario in different seasons, which will be used for subsequent power grid dispatch simulations.

9. An electronic system comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the method of any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images