Distributed photovoltaic ultra-short-term output prediction method based on space-time prediction factor preselection

Abstract

The invention relates to a distributed photovoltaic ultra-short-term output prediction method based on time-space prediction factor preselection, and the method comprises the following steps: carrying out the online scene recognition based on a distributed photovoltaic output power time sequence obtained in real time, and enabling the scene to comprise a clear sky scene and a climbing scene; obtaining corresponding space-time prediction factors by adopting corresponding space-time prediction factor pre-selection modes based on the different scenes obtained through identification; constructing a prediction model based on the selected space-time prediction factor, and obtaining photovoltaic prediction data; and coordinating and merging the photovoltaic prediction data of different scenes to form a final prediction result. Compared with the prior art, the method has the advantages of high prediction precision, high prediction efficiency and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed photovoltaics, and relates to a distributed photovoltaic output prediction method, and in particular to a distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors. Background Art

[0002] The effective use of distributed photovoltaics is an important means to promote energy production and low-carbon transformation. However, since its power is affected by environmental and meteorological conditions, it has great volatility and randomness. These characteristics make distributed photovoltaics have adverse effects on the safety of distribution networks, power quality, and system stability after being connected to the grid. Therefore, the accurate prediction of distributed photovoltaic power is of great significance to the optimization of the use of multi-level power grids.

[0003] According to the prediction time scale, photovoltaic power prediction can be divided into medium- and long-term prediction (1 day to 1 year), short-term prediction (6 hours to 1 day) and ultra-short-term prediction (0 to 6 hours). Since distributed photovoltaic systems do not have the inertial rotation of traditional thermal power generation, their output can fluctuate by more than 80% when blocked by clouds. When large-scale distributed photovoltaic power generation is connected to the grid, high-frequency fluctuations in distributed photovoltaic power generation will seriously affect the frequency stability of the power grid. In this regard, how to make a refined prediction of ultra-short-term distributed photovoltaic power has become an important issue that needs to be considered in the current distribution network planning.

[0004] There are currently two main methods for ultra-short-term distributed photovoltaic power forecasting. One is the mechanism-driven method, which uses meteorological information and photovoltaic system parameters to establish a forecast model based on physical principles. The other is the spatiotemporal forecasting method based on spatial reference points, which establishes a forecast model by mining the relationship between different spatial reference points (photovoltaic power stations or irradiance sensors) and the target power station measurement data. Commonly used methods include statistical models, machine learning algorithms, and fusion algorithms. Domestic and foreign scholars have conducted relevant research on the above two methods. For example, the paper "Large-scale distributed user photovoltaic spatial clustering method based on spatial correlation" (Jiao Tianli, Zhang Jianmin, Li Xiong, et al. Automation of Electric Power Systems, 2019, 43(21): 97-102) clusters the spatial correlation of the output of multiple users based on commercial meteorological data such as numerical weather forecast (NWP) services, and then predicts the power of the entire photovoltaic cluster; the paper "Distributed photovoltaic output prediction based on spatial correlation" (Zhang Jia'an, Wang Kunyue, Chen Jian, et al. Electric Power Construction, 2020, 41(3): 47-53.) constructs the output correlation model of distributed photovoltaic and centralized photovoltaic under various weather conditions based on Copula function; the paper "Expected output calculation based on inverse distanceweighting and its application in anomaly detection of distributed photovoltaic power stations" (Y.Shi, W.He, J.Zhao, et al. Journal of Cleaner Production, 2020, 253: 119965.) The inverse distance weighted algorithm based on meteorological data obtained the expected output of the target site. However, due to cost considerations, distributed sites often do not purchase NWP services and lack the simultaneous meteorological data information required for prediction. In addition, due to factors such as operation and maintenance level and component technology, it is generally difficult to obtain the specific parameters of hardware facilities for distributed photovoltaics. Physical modeling based on mechanism-driven can only use simplified models and empirical parameters, and the prediction accuracy is relatively low. The literature "Very short term irradiance forecasting using the lasso" (D. Yang, Z. Ye, L. Lim, Z. Dong. Solar Energy, 2015, 114: 314–326.) uses the lasso algorithm to establish a spatiotemporal prediction model for distributed photovoltaic networks. The spatial correlation between different photovoltaic power stations is screened through parameter regularization and assigned different weights.However, this type of method often only considers the isotropy between power stations. When cloud occlusion occurs frequently, the spatiotemporal correlation between power stations will change rapidly. It is difficult to maintain good accuracy and generalization in different scenarios using fixed-pattern spatiotemporal prediction factors. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors with high prediction accuracy and high prediction efficiency.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors comprises the following steps:

[0008] Performing online scene recognition based on the distributed photovoltaic output power time series acquired in real time, wherein the scenes include clear sky scenes and slope climbing scenes;

[0009] Based on the different scenarios obtained by identification, corresponding spatiotemporal prediction factors are obtained by pre-selecting corresponding spatiotemporal prediction factors;

[0010] Building a prediction model based on the selected spatiotemporal prediction factors to obtain photovoltaic prediction data;

[0011] The photovoltaic prediction data of different scenarios are coordinated and merged to form the final prediction results.

[0012] Furthermore, a clear sky index is constructed based on the distributed photovoltaic output power time series, and online scene recognition is achieved based on the clear sky index.

[0013] Furthermore, the clear sky index is obtained by normalizing the power data of the photovoltaic power station according to the rated capacity of the photovoltaic site.

[0014] Furthermore, the online scene recognition based on the clear sky index is specifically implemented as follows:

[0015] If the cumulative difference in clear sky index between any two PV sites in the region is less than the clear sky index threshold within the target window period, it is considered to be a clear sky scene;

[0016] If the cumulative difference in clear sky index between a pair of PV sites in the area is greater than the clear sky index threshold, it is considered that the current scenario is a climbing scenario.

[0017] Furthermore, for the clear sky scene, the spatiotemporal prediction factor preselection method includes:

[0018] The current time is t, intercept the observation window T obs Internal photovoltaic power station iTime series of clear sky index;

[0019] Taking the original time point as the initial position, the clear sky index sequence of the photovoltaic power station is shifted left and right to calculate the predicted target photovoltaic power station S i Different photovoltaic power station S after translation j Pearson correlation coefficient of the clear sky index series

[0020] According to the preset correlation coefficient threshold κ, the target photovoltaic power station S is predicted i The spatiotemporal predictors are preselected to satisfy Photovoltaic power station S j gather

[0021] Select Collection In The maximum time shift is taken as the optimal time shift The final pre-selected spatiotemporal prediction factor under the clear sky scenario is expressed as:

[0022]

[0023] Among them, x is the element in the clear sky index time series, the subscript represents the time, and the superscript j represents the photovoltaic power station S j .

[0024] Furthermore, for the hill climbing scenario, the spatiotemporal prediction factor preselection method includes:

[0025] Given a pre-selected time window t obs , record all the PV plants S i The neighboring photovoltaic power station S that satisfies the cumulative difference of clear sky index is greater than the clear sky index threshold j , build a collection of spatial sites

[0026] For different photovoltaic power plants Record the time when the peak of the clear sky index sequence occurs respectively

[0027] Calculate the photovoltaic power station separately and the target photovoltaic power station S i Peak time The time shift The final pre-selected spatiotemporal prediction factor in the climbing scenario is expressed as:

[0028]

[0029] Among them, x is the element in the clear sky index time series, the subscript represents the time, and the superscript j represents the photovoltaic power station S j .

[0030] Furthermore, the spatiotemporal prediction factor is dynamically updated based on a given update time interval ΔT.

[0031] Furthermore, the prediction model adopts an elastic network as a basic prediction model, establishes a global prediction framework for different scenarios, and updates the parameters of the basic prediction model in a rolling manner.

[0032] Furthermore, in the rolling update, for the clear sky scene, when the time from the last update exceeds the fixed update time T update For a hill climbing scene, it is updated when a hill climbing scene is identified.

[0033] The present invention also provides a computer-readable storage medium, comprising one or more programs for execution by one or more processors of an electronic device, wherein the one or more programs include instructions for executing the distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors as described above.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. The present invention obtains better matching spatiotemporal prediction factors through online scene recognition pre-selection, which enhances the stability and timeliness of the model in different prediction scenarios, thereby improving the prediction accuracy of distributed photovoltaic ultra-short-term output.

[0036] 2. The present invention establishes a spatiotemporal prediction factor pre-selection mechanism, which greatly reduces the complexity of the model while ensuring the prediction accuracy, and can enhance the convergence speed of the model, thereby improving the prediction efficiency.

[0037] 3. The present invention establishes a global prediction framework to unify the prediction results in different scenarios, making them more timely. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of the flow chart of the global prediction framework in the present invention;

[0039] Figure 2 The geographical information of the distributed solar irradiance measurement in the simulation calculation of the embodiment of the present invention;

[0040] Figure 3 This is a comparison of the results of the 5-minute prediction time scale using DH4 as the target site in the simulation calculation of an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0042] Example 1

[0043] refer to Figure 1 As shown, this embodiment provides a distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors, comprising the following steps:

[0044] S1. Performing online scene recognition based on the real-time acquired distributed photovoltaic output power time series, wherein the scenes include clear sky scenes and slope climbing scenes;

[0045] S2, based on the different scenarios obtained by identification, a corresponding spatiotemporal prediction factor pre-selection method is used to obtain a corresponding spatiotemporal prediction factor;

[0046] S3, constructing a prediction model based on the selected spatiotemporal prediction factors to obtain photovoltaic prediction data;

[0047] S4. Coordinate and merge the photovoltaic prediction data of different scenarios to form the final prediction result.

[0048] The above method can be applied to a region with multiple distributed photovoltaic sites. Let Φ = {S1, S2, …, S N} is the set of distributed photovoltaic sites in the region, where S i Represents a single photovoltaic power station in the region, and N is the number of distributed photovoltaic power stations in the region. In step S1, the distributed photovoltaic power stations in the region can be divided into two different working scenarios: clear sky and climbing according to the fluctuation of the real-time time series of photovoltaic output. The specific steps are as follows:

[0049] 101) The current time is t, according to the rated capacity of the photovoltaic site The power data of the photovoltaic power station Perform normalization to obtain the clear sky index

[0050]

[0051] 102) If the cumulative difference in clear sky index between any two PV sites in the region is less than the threshold x within the target window period T s ,Right now It is considered that the current scene is clear sky;

[0052] 103) If there is a pair of photovoltaic sites in the region whose cumulative difference in clear sky index is greater than the threshold x s ,Right now It is considered that the current scene is a climbing scene;

[0053] 104) Repeat 101)-103) to implement a continuous scene recognition process based on the sliding window.

[0054] Since the photovoltaic output data is relatively flat in the clear sky scenario, the correlation analysis can be performed based on the historical power time series of each distributed photovoltaic site within the observation window to screen the appropriate prediction factors; however, the photovoltaic output fluctuation frequency is high in the ramp scenario, so a faster and more efficient pre-selection method is required to ensure the timeliness of the prediction factors. In this regard, in step S2, this method uses different spatiotemporal prediction factor pre-selection strategies for the clear sky scenario and the ramp scenario respectively.

[0055] For clear sky scenarios, the pre-selection steps of spatiotemporal prediction factors are:

[0056] 201) The current time is t, and the observation window T is intercepted obs Internal photovoltaic power station i Time series of clear sky index

[0057] 202) Taking the original time point as the initial position, the clear sky index sequence of the photovoltaic power station is shifted left and right, and the predicted target photovoltaic power station S is calculated respectively. i Different photovoltaic power station S after translation j The Pearson correlation coefficient of the clear sky index series between the two is used as a measure of spatial correlation at different time points of the power station, which is specifically expressed as:

[0058]

[0059] Where Cov(·) and Var(·) represent the covariance and variance calculations respectively, and δt=1,2,…,n is the time-shifted signal. i ,S j The greater the correlation between The closer the value of is to 1;

[0060] 203) According to the preset correlation coefficient threshold k (usually set to 0.8), the target power station S is predicted i The spatial predictors can be preselected to satisfy Power Station S j The set can be expressed as:

[0061]

[0062] 204) Select Collection In The maximum time shift is the optimal time shift and can be expressed as:

[0063]

[0064] 205) The final pre-selected spatiotemporal prediction factors under the clear sky scenario can be expressed as:

[0065]

[0066] For the ramp-climbing scenario, the specific steps for preselecting spatiotemporal prediction factors are as follows:

[0067] 206) Given a pre-selected time window t obs , t obs <<T obs , record all the targets S i satisfy Nearby power station S j , build a collection of spatial sites

[0068] 207) For different power stations Record the time when the peak of the clear sky index sequence occurs respectively

[0069] 208) Calculate the adjacent power stations separately With the target power station S i Peak time The time shift is:

[0070]

[0071] 209) The final pre-selected spatiotemporal prediction factors in the climbing scenario can be expressed as:

[0072]

[0073] Given an update time interval ΔT, the above steps 201)-205) and 206)-209) will be repeatedly calculated, thereby dynamically updating the spatiotemporal prediction factors.

[0074] In step S3, based on the scene recognition information, the pre-selected spatiotemporal prediction factors are dynamically updated for model training. The specific steps are as follows:

[0075] 301) At the current time t, a spatiotemporal prediction model is built based on the spatiotemporal prediction factors newly preselected in step S2. The recent historical time series window T of each spatial identification site is intercepted. train , according to their respective optimal time shifts, they are translated and merged as feature vectors X train , the corresponding target power station is the training label Y train The present invention adopts elastic network as the basic prediction model, which can be expressed as:

[0076]

[0077] In the formula, Θ is the model parameter; λ is the regularization parameter, which is used to adjust the proportion of l1 and l2 norms; α is the model hyperparameter. In particular, when α=0, the elastic network model is transformed into ridge regression, and when α=1, the model is transformed into lasso regression. Therefore, the elastic network has both the stability of ridge regression and the coefficient shrinkage ability of lasso regression. All the above hyperparameters can be optimized using the k-fold algorithm.

[0078] 302) For different recognition scenarios, the prediction model will be automatically switched and its parameters will be updated in a rolling manner.

[0079] In step S4, a global prediction framework is constructed to coordinate and merge the predictions generated in different scenarios.

[0080] Since the PV output changes relatively slowly under clear sky scenarios, there is no need for high-frequency model iteration. This can be achieved by setting a fixed update time T update Update the model parameters. For the climbing scene, since the cloud movement is highly random, once the climbing scene occurs, the model parameters will be updated to improve the timeliness of the model.

[0081] In this example, a certain area of ​​1 km 2 The actual distributed solar energy monitoring area is used as a simulation example. The area consists of 8 distributed solar irradiation measurement stations with a data collection frequency of 1 minute. The specific geographical distribution information is as follows: Figure 2 In this embodiment, the DH4 site is selected as the target site, and the target site is continuously predicted on a 5-minute time scale through the historical and real-time output data of the surrounding power stations.

[0082] In this embodiment, the time series window length T for scene recognition is 10 minutes, and the scene recognition threshold x s = 10% · rated system capacity, observation window T for clear sky scenario prediction factor screening obs = 10 minutes, correlation coefficient threshold κ = 0.8, pre-selected time window t for the climbing scene obs = 1 minute, fixed update time T of the global prediction framework update = 30 minutes. Recent historical time series window T used for model training train =2 hours.

[0083] The model prediction performance was tested continuously on the 2014 full-year data set. The results are as follows: Figure 3As shown. In order to measure the performance of the prediction method proposed in the present invention, the most commonly used time series model, the long short-term memory network (LSTM), was selected as a baseline model for comparison. The prediction model proposed in the present invention achieved more accurate results on the annual data set, the prediction points converged more near the diagonal, and the variance was smaller. In this example, the average time for a single model training was 1.2s, which is significantly lower than the existing LSTM model (average training time of 3 minutes). In addition, in sunny, cloudy and rainy weather, the standard root mean square error of the prediction model of the present invention was 2.35%, 5.83%, and 4.92%, respectively. Compared with 3.19%, 8.34%, and 5.65% of the LSTM model, there are significant improvements.

[0084] In summary, it can be seen that the pre-selection of spatiotemporal prediction factors can effectively enhance the prediction performance of the model in different weather scenarios, while reducing the complexity of the model, which is of certain significance for practical engineering applications.

[0085] If the above method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of computer program instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0086] Example 2

[0087] This embodiment provides a distributed photovoltaic ultra-short-term output prediction device based on pre-selection of spatiotemporal prediction factors, including one or more processors, a memory and one or more programs stored in the memory, and the one or more programs include instructions for executing the line loss optimal segmentation point design method based on the deep optimization wolf pack algorithm as described in Example 1.

[0088] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0089] The above-mentioned computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0091] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0092] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors, characterized in that: The following steps are involved: Performing online scene recognition based on the distributed photovoltaic output power time series acquired in real time, wherein the scenes include clear sky scenes and slope climbing scenes; Based on the different scenarios obtained by identification, corresponding spatiotemporal prediction factors are obtained by pre-selecting corresponding spatiotemporal prediction factors; Building a prediction model based on the selected spatiotemporal prediction factors to obtain photovoltaic prediction data; The photovoltaic prediction data of different scenarios are coordinated and merged to form the final prediction results.

2. The distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors according to claim 1 is characterized in that: A clear sky index is constructed based on the distributed photovoltaic output power time series, and online scene recognition is achieved based on the clear sky index.

3. The distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors according to claim 2 is characterized in that: The clear sky index is obtained by normalizing the power data of the photovoltaic power station according to the rated capacity of the photovoltaic site.

4. The distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors according to claim 2 or 3, characterized in that: The specific steps of realizing online scene recognition based on the clear sky index are as follows: If the cumulative difference in clear sky index between any two PV sites in the region is less than the clear sky index threshold within the target window period, it is considered to be a clear sky scene; If the cumulative difference in clear sky index between a pair of PV sites in the area is greater than the clear sky index threshold, it is considered that the current scenario is a climbing scenario.

5. The distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors according to claim 2 is characterized in that: For the clear sky scene, the spatiotemporal prediction factor preselection method includes: The current time is t, intercept the observation window T obs Internal photovoltaic power station i Time series of clear sky index; Taking the original time point as the initial position, the clear sky index sequence of the photovoltaic power station is shifted left and right to calculate the predicted target photovoltaic power station S i Different photovoltaic power station S after translation j Pearson correlation coefficient of the clear sky index series According to the preset correlation coefficient threshold κ, the target photovoltaic power station S is predicted i The spatiotemporal predictors are preselected to satisfy Photovoltaic power station S j gather Select Collection In The maximum time shift is taken as the optimal time shift The final pre-selected spatiotemporal prediction factor under the clear sky scenario is expressed as: Among them, x is the element in the clear sky index time series, the subscript represents the time, and the superscript j represents the photovoltaic power station S j .

6. The distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors according to claim 2 is characterized in that: For the hill climbing scenario, the spatiotemporal prediction factor preselection method includes: Given a pre-selected time window t obs , record all the PV plants S i The neighboring photovoltaic power station S that satisfies the cumulative difference of clear sky index is greater than the clear sky index threshold j , build a collection of spatial sites For different photovoltaic power plants Record the time when the peak of the clear sky index sequence occurs respectively Calculate the photovoltaic power station separately and the target photovoltaic power station S i Peak time The time shift The final pre-selected spatiotemporal prediction factor in the climbing scenario is expressed as: Among them, x is the element in the clear sky index time series, the subscript represents the time, and the superscript j represents the photovoltaic power station S j .

7. The distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors according to claim 1 is characterized in that: The spatiotemporal prediction factors are dynamically updated based on a given update time interval ΔT.

8. The distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors according to claim 1 is characterized in that: The prediction model adopts elastic network as the basic prediction model, establishes a global prediction framework for different scenarios, and rolls over the parameters of the basic prediction model.

9. The distributed photovoltaic ultra-short-term output prediction method based on pre-selection of spatiotemporal prediction factors according to claim 8 is characterized in that: In the rolling update, for the clear sky scene, when the time from the last update exceeds the fixed update time T update For a hill climbing scene, it is updated when a hill climbing scene is identified.

10. A computer-readable storage medium, characterized in that: It includes one or more programs for execution by one or more processors of an electronic device, and the one or more programs include instructions for executing a distributed photovoltaic ultra-short-term output prediction method based on pre-selected spatiotemporal prediction factors as described in any one of claims 1-9.

Images