A Day-ahead Photovoltaic Forecasting Method and Device Based on Time-Period Clustering of Meteorological Sensitivity
By performing time-segmented clustering and matching of photovoltaic power generation data with meteorological data, the problem of low prediction accuracy of photovoltaic power generation in existing technologies has been solved, achieving higher prediction accuracy and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHONGYAODA DIGITAL ENERGY ECOLOGICAL TECH (ZHEJIANG) CO LTD
- Filing Date
- 2026-01-13
- Publication Date
- 2026-05-26
AI Technical Summary
Existing photovoltaic power generation forecasting methods struggle to effectively handle nonlinear and non-stationary meteorological characteristics, resulting in low forecast accuracy and error accumulation, particularly in long-term continuous day forecasts.
A time-segmented clustering method based on meteorological data is adopted. By segmenting historical meteorological data and photovoltaic power generation data into multiple time periods and classifying sensitive meteorological types, an unsupervised clustering algorithm is used to cluster the data into typical meteorological type centers. The prediction results are then corrected by combining meteorological change trends and solar irradiance integrals to improve accuracy.
It improves the accuracy and reliability of photovoltaic power generation forecasting, alleviates the limitations of nonlinear meteorological data processing, reduces the cumulative error of multi-day forecasts, and enhances the stability of forecast results.
Smart Images
Figure CN121529549B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of photovoltaic power generation prediction, specifically relating to a day-ahead photovoltaic prediction method and device based on meteorological sensitivity time-segmented clustering. Background Technology
[0002] Cluster-based day-ahead photovoltaic forecasting refers to the process of performing meteorologically sensitive clustering on weather factors affecting photovoltaic power generation, and then searching through historical datasets for the time period most similar to the forecast day's weather to match power output, thereby overcoming the limitations of traditional power forecasting models in handling nonlinear meteorological characteristics.
[0003] Photovoltaic power generation exhibits significant intermittency and volatility. As the proportion of photovoltaic power generation connected to the grid continues to increase, it will inevitably pose challenges to the power balance of the power grid. Constructing virtual power plants with trading as the core to provide power auxiliary services, relying on accurate photovoltaic power forecasting is the key to solving the problem. Good photovoltaic forecasting is of great significance for optimizing power generation plans, enhancing grid stability, and reducing grid operating costs.
[0004] Current physical models such as PVLIB (Photovoltaic Library) and multi-meteorological parameter prediction technologies struggle to adapt quickly to nonlinear meteorological conditions when dealing with photovoltaic data that exhibits significant fluctuations and intermittency. For example, patent CN119401384A proposes a photovoltaic output power prediction and adjustment device and method based on meteorological parameters. Furthermore, single neural network models such as LSTM (Long Short-Term Memory) accumulate prediction errors over long consecutive days, resulting in low prediction accuracy. For instance, patent CN119885808A proposes a photovoltaic power prediction method based on bidirectional LSTM. Summary of the Invention
[0005] To overcome the shortcomings of existing technologies in handling nonlinear meteorological characteristics, this invention provides a day-ahead photovoltaic (PV) forecasting method and device based on meteorological sensitivity time-segmented clustering. Specifically, it involves historical time-series data segmentation and meteorological characteristic sensitivity clustering, achieving high reliability and accuracy in PV day-ahead power forecasting under different weather conditions.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0007] The first aspect: Provides a day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data, including the following steps:
[0008] Historical meteorological data, solar irradiance data, and photovoltaic power generation data are divided into multiple time periods along the time dimension of photovoltaic power generation. Based on the meteorological data, each sub-time period is classified into sensitive meteorological types.
[0009] For each sub-period, based on meteorological data, each moment in the sub-period is clustered into different typical meteorological type cluster centers to obtain solar irradiance data and photovoltaic power generation data clustered under each typical meteorological type cluster center in the sub-period;
[0010] The meteorological data and solar irradiance data for the forecast date are divided into sub-periods, and the sensitive meteorological types of each sub-period are classified. The set of the closest historical dates is found based on the sensitive meteorological types of each sub-period of the forecast date.
[0011] Calculate the solar irradiance integral for each sub-period of the predicted date, and retrieve the historical sub-periods with the same meteorological change trend and the closest solar irradiance integral from the corresponding sub-periods of the historical date set;
[0012] Based on meteorological data, each time point in each sub-period of the forecast day is clustered into different typical meteorological type cluster centers. The historical photovoltaic power generation power with the same typical meteorological type cluster center and the closest solar irradiance in the corresponding sub-period is obtained from the historical date set. The historical photovoltaic power generation power that meets the smoothing requirements is used to replace the historical photovoltaic power generation power at the corresponding time in the historical sub-period to obtain the final day-ahead photovoltaic forecast result for the forecast day.
[0013] Several alternative methods are provided below, but they are not intended as additional limitations on the overall solution above. They are merely further additions or optimizations. Provided there are no technical or logical contradictions, each alternative method can be combined individually with respect to the overall solution above, or multiple alternative methods can be combined with each other.
[0014] Preferably, the process of dividing the photovoltaic power generation period into multiple time periods in the time dimension includes: dividing the photovoltaic power generation period according to the time characteristics of photovoltaic power generation.
[0015] Preferably, the step of classifying the sensitive meteorological types of each sub-time period after segmentation based on meteorological data includes:
[0016] The number of times with the same weather type in a sub-period is counted, and the weather types with a number greater than or equal to the number threshold are taken as the sensitive weather types of the sub-period; if there is no number greater than or equal to the number threshold, the weather type of the sub-period is set to no weather sensitive type.
[0017] As a preferred approach, cluster centers for typical meteorological types are set, and a center-based unsupervised clustering method is adopted to cluster each time period in the sub-period into different meteorological type cluster centers based on the meteorological data.
[0018] Preferably, the step of finding the set of closest historical dates based on the sensitive weather type to which each sub-period of the predicted day belongs includes:
[0019] The sensitive weather types of each sub-period of the forecast day are coded, and the coded values are combined in chronological order to obtain the weather type sequence of the forecast day;
[0020] The sensitive weather types of each sub-period of a historical date are encoded, and the encoded values are combined in chronological order to obtain a sequence of weather types for historical dates;
[0021] The similarity between the predicted daily weather type sequence and the weather type sequences of each historical date is measured. Historical dates with similarity greater than the date similarity threshold are selected to obtain the historical date set.
[0022] Preferably, the meteorological change trends are the same, as determined below:
[0023] The weather types for all times within the sub-periods of the forecast day are encoded, and the encoded values are combined in chronological order to obtain the weather type sequence for the sub-periods of the forecast day.
[0024] The weather types of all times within a sub-period of a historical date are encoded, and the encoded values are combined in chronological order to obtain a sequence of weather types for each sub-period of a historical date.
[0025] The similarity between the meteorological type sequence of the predicted day and the meteorological type sequence of the corresponding historical date in the historical date set is measured. Historical sub-periods with similarity greater than the time period similarity threshold are taken as sub-periods with the same meteorological change trend.
[0026] Preferably, satisfying the smoothness requirement includes:
[0027] The current search time in the current search sub-period of the predicted day is recorded as the replacement time, and the historical photovoltaic power generation obtained from the historical date set is recorded as the replacement power;
[0028] The replacement range is determined based on the historical photovoltaic power generation of the two moments adjacent to the replacement moment in the corresponding historical sub-period.
[0029] If the replacement power is within the replacement range, then the replacement power meets the smoothing requirement; otherwise, the replacement power does not meet the smoothing requirement.
[0030] Preferably, obtaining the final day-ahead photovoltaic forecast result includes:
[0031] Take the historical photovoltaic power generation at each moment in the replaced historical sub-period as the predicted value of photovoltaic power generation at the corresponding moment on the prediction day;
[0032] Set the predicted photovoltaic power generation for times outside the photovoltaic power generation period to the default value;
[0033] By sequentially piecing together the photovoltaic power generation forecast values for each moment of the forecast day, a complete day-ahead photovoltaic forecast result is obtained.
[0034] The second aspect: provides a day-ahead photovoltaic forecasting device based on time-segmented clustering of meteorological data, including a processor and a memory storing a number of computer instructions, wherein the computer instructions, when executed by the processor, implement the steps of the day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data.
[0035] This invention provides a day-ahead photovoltaic forecasting method and device based on meteorological sensitivity time-segment clustering. Compared with the prior art, this invention effectively alleviates the limitations of traditional photovoltaic power forecasting methods in processing nonlinear and non-stationary meteorological data, while reducing the cumulative error generated by forecasting multiple consecutive days. In the field of day-ahead photovoltaic forecasting, the overall forecasting accuracy is significantly improved. Attached Figure Description
[0036] Figure 1 This is a flowchart of the day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data according to the present invention. Detailed Implementation
[0037] 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.
[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention.
[0039] Example 1:
[0040] To overcome the limitations of existing technologies in handling nonlinear meteorological characteristics, this embodiment provides a day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data, such as... Figure 1 As shown, it includes the following steps:
[0041] Step 1: Divide historical meteorological data, solar irradiance data, and photovoltaic power generation data into multiple time periods along the time dimension of photovoltaic power generation. Based on the meteorological data, classify each sub-time period into sensitive meteorological types.
[0042] Historical meteorological data (such as text information like sunny, cloudy, and showery weather), solar irradiance data, and historical photovoltaic power generation data of the photovoltaic power station are acquired. To improve matching accuracy, this embodiment divides the photovoltaic power generation period according to its time-specific characteristics. For example, the photovoltaic power generation period can be divided into three sub-periods, corresponding to the initial power generation period, the peak power generation period, and the power generation decline period, respectively. The meteorological sensitivity and power generation characteristics are different in each sub-period.
[0043] Specifically, the effective photovoltaic (PV) power generation period (t1 to t2) is determined by using 00:00:00 as the starting point and 23:59:59 as the ending point. Historical meteorological data, solar irradiance data, and PV power generation data within this effective power generation period are then divided into N sub-periods to capture the characteristic changes in PV power generation across different periods. Here, N is a positive integer representing the number of sub-periods. This division is based on the time-specific characteristics of PV power generation throughout the day (e.g., initial power ramp-up, steady-state power generation, power decline, etc., or initial power ramp-up, steady-state power generation, power cliff, fluctuating decline, etc.). By processing these sub-periods, a more accurate meteorological-power correspondence can be established within each sub-period, thereby improving prediction accuracy.
[0044] In this embodiment, the sensitive weather types include all weather types from historical weather data (such as sunny, cloudy, light, thunderstorm, etc.). The sensitive types for each sub-period are categorized based on a set threshold (e.g., 6). Specifically, the number of times with the same weather type within a sub-period is counted, and weather types with a count greater than or equal to the threshold are designated as the sensitive weather types for that sub-period. If no count is greater than or equal to the threshold, the weather type for that sub-period is set to "no sensitive weather type".
[0045] Step 2: For each sub-period, based on the meteorological data, each moment in the sub-period is clustered into different typical meteorological type cluster centers to obtain the solar irradiance data and photovoltaic power generation data clustered under each typical meteorological type cluster center in the sub-period.
[0046] This embodiment sets typical meteorological types as commonly used classifications in the photovoltaic power generation industry for assessing the impact of meteorology on the performance and lifespan of energy equipment: extreme efficiency meteorology, turbulent fluctuation meteorology, mild attenuation meteorology, and stress suppression meteorology. Corresponding cluster centers are then established based on these classifications. A center-based unsupervised clustering method (such as the K-Means clustering algorithm) is employed to cluster each time point within a sub-period into different meteorological type cluster centers based on the meteorological data. The timestamps under each cluster center for each sub-period are recorded, forming corresponding solar irradiance datasets and photovoltaic power generation datasets.
[0047] Here is an example of the clustering results for a sub-period:
[0048] Sub-period d1: The times under the typical meteorological type cluster center C1 are ti, tj, ..., the solar irradiances are ri, rj, ..., and the photovoltaic power generation is pi, pj, ...; The dates under the typical meteorological type cluster center C2 are tk, tl, ..., the solar irradiances are rk, rl, ..., and the photovoltaic power generation is pk, pl, ...; The times under the typical meteorological type cluster center C3 are tm, tn, ..., the solar irradiances are rm, rn, ..., and the photovoltaic power generation is pm, pn, ....
[0049] Step 3: Divide the meteorological data and solar irradiance data of the forecast date into sub-periods and classify the sensitive meteorological types of each sub-period. Find the set of the closest historical dates based on the sensitive meteorological types of each sub-period of the forecast date.
[0050] The logic for segmenting meteorological and solar irradiance data for the forecast date into sub-periods is the same as the logic for segmenting meteorological and solar irradiance data for historical dates into sub-periods; and the logic for classifying the sensitive meteorological types of each sub-period of the forecast date is the same as the logic for classifying the sensitive meteorological types of each sub-period of historical dates. This embodiment will not elaborate further.
[0051] When searching a set of historical dates, the search range is first determined (e.g., all dates in the past year). To facilitate matching, the sensitive weather types of each sub-period of the forecast date need to be encoded, and the encoded values are combined in chronological order to obtain the weather type sequence of the forecast date. Similarly, the sensitive weather types of each sub-period of the historical dates are encoded, and the encoded values are combined in chronological order to obtain the weather type sequence of the historical dates.
[0052] It should be noted that if multiple quantities exceed the quantity threshold in a sub-period, then multiple sensitive weather types are allowed in that sub-period (e.g., both cloudy and sunny weather can exist at the same time). Therefore, when encoding, the maximum number of sensitive weather types in all sub-periods is taken as the maximum number of bits in the code, and when the number of sensitive weather types in a sub-period is less than the maximum number, the code of the corresponding sub-period is padded with 0.
[0053] For example, the sensitive weather types for the first sub-period are cloudy and sunny, the sensitive weather type for the second sub-period is cloudy, the sensitive weather type for the third sub-period is rain, and the sensitive weather type for the fourth sub-period is no sensitive weather type. Furthermore, this embodiment sets the coding rules as follows: the coding value for cloudy is 1, the coding value for sunny is 2, the coding value for rain is 3, and the coding value for no sensitive weather type is 4. Therefore, the weather type sequence for the first sub-period is [1,2], the weather type sequence for the second sub-period is [1,0], the weather type sequence for the third sub-period is [3,0], and the weather type sequence for the fourth sub-period is [4,0]. Thus, the weather type sequence for a day containing these four sub-periods is [1,2,1,0,3,0,4,0].
[0054] After obtaining the meteorological type sequence, the similarity between the predicted day's meteorological type sequence and the meteorological type sequences of each historical date is measured. One or more historical dates with a similarity greater than a date similarity threshold (e.g., set to 80%; if no match is found, the date similarity threshold can be lowered) are selected to obtain a set of historical dates. The similarity calculation method can use Euclidean distance or Manhattan distance to calculate the distance and normalize it to obtain the similarity, or it can be directly measured using cosine similarity or similar methods. In this embodiment, both the predicted day and a historical date refer to a single day.
[0055] Step 4: Calculate the solar irradiance integral for each sub-period of the predicted date, and retrieve the historical sub-periods with the same meteorological change trend and the closest solar irradiance integral from the corresponding sub-periods of the historical date set.
[0056] The solar irradiance integral represents the solar irradiance integral value within a sub-period, that is, the total amount of solar radiation received within the sub-period. The calculation formula is as follows:
[0057]
[0058] In the formula, Represents the solar irradiance integral. and These represent the lower and upper limits of integration, i.e., the start and end times, respectively. Indicates at a point in time Solar irradiance intensity (instantaneous value) at the location. It represents a tiny change over time.
[0059] Since this application divides sub-time periods according to the characteristics of power generation periods, that is, each sub-time period has its own unique characteristics, this embodiment uses data from the corresponding time period in historical dates to predict each sub-time period of the prediction date, so as to ensure that the prediction results obtained have the same characteristics of power generation periods.
[0060] When determining the meteorological trend of two sub-periods, the meteorological types of all moments in the sub-period of the forecast date are first encoded, and the encoded values are combined in chronological order to obtain the meteorological type sequence of the sub-period of the forecast date. Similarly, the meteorological types of all moments in the sub-period of a historical date are encoded, and the encoded values are combined in chronological order to obtain the meteorological type sequence of the sub-period of the historical date. Since the number of moments in corresponding sub-periods is the same, and each moment has one and only one meteorological type, this embodiment can directly obtain the meteorological type sequence of the sub-periods according to the correspondence between meteorological types and encoded values.
[0061] Then, the similarity between the meteorological type sequence of the predicted day's sub-period and the meteorological type sequence of the corresponding sub-period in the historical date set is measured. Historical sub-periods with similarity greater than a time-period similarity threshold (e.g., set to 90%, which can be lowered if no match is found) are considered to have the same meteorological trend. The similarity calculation method can use Euclidean distance or Manhattan distance to calculate the distance and normalize it, or it can be directly measured using cosine similarity. Each sub-period of the predicted day can be matched with the closest historical sub-period, and subsequent steps are performed based on the matched historical sub-period.
[0062] Step 5: Based on the meteorological data, cluster each time point in each sub-period of the forecast day into different typical meteorological type cluster centers. Obtain the historical photovoltaic power generation power with the same typical meteorological type cluster center and the closest solar irradiance in the corresponding sub-period from the historical date set. Replace the historical photovoltaic power generation power of the corresponding time in the historical sub-period with the historical photovoltaic power generation power that meets the smoothing requirements to obtain the final day-ahead photovoltaic forecast result for the forecast day.
[0063] Since the entire historical sub-period spans a relatively long time, in order to further improve the prediction accuracy of the method of the present invention, this embodiment corrects the photovoltaic power generation at each moment in the historical sub-period matched in step 4. The logic of clustering each moment in each sub-period of the prediction day into different typical meteorological type cluster centers is the same as the logic of clustering each moment in each sub-period of the historical date into different typical meteorological type cluster centers, and will not be repeated in this embodiment.
[0064] For example, if the cluster center of the typical meteorological type to which the first moment in the first sub-period of the predicted day belongs is the extreme efficiency type of meteorology, then we will iterate through the first sub-period of each historical date from the historical date set to the moment when the extreme efficiency type of meteorology is clustered, obtain the historical photovoltaic power generation corresponding to the moment with the closest solar irradiance, record it as the replacement power, and record the first moment in the first sub-period as the replacement moment.
[0065] To correct any discrepancies in the smoothness of the original photovoltaic power generation data, this embodiment performs a smoothing requirement judgment on the retrieved replacement power. Specifically, a replacement range is formed based on the historical photovoltaic power generation of two times adjacent to the replacement time in the corresponding historical sub-period. If the replacement power is within the replacement range, the replacement power meets the smoothing requirement; otherwise, the replacement power does not meet the smoothing requirement.
[0066] It should be noted that if it is the first or last moment in a historical sub-period, the historical photovoltaic power generation of the last or first moment of the adjacent historical sub-period is used to form the replacement range. If there is no adjacent historical sub-period, 0 is used to form the replacement range.
[0067] Then, using the replacement power that meets the smoothing requirements, the original photovoltaic power generation power at the corresponding time in the historical sub-period (e.g., the first time in the historical sub-period matched with the first sub-period of the prediction day) is replaced with the replacement power.
[0068] Finally, the historical photovoltaic power generation at each moment in the replaced historical sub-period is taken as the photovoltaic power generation prediction value at the corresponding moment on the prediction day; the photovoltaic power generation prediction value at moments outside the photovoltaic power generation period on the prediction day is set as the default value; the photovoltaic power generation prediction values at each moment on the prediction day are spliced in sequence to obtain the complete day-ahead photovoltaic prediction result for the prediction day.
[0069] The method of this invention utilizes the dual similarity between meteorological characteristics and solar irradiance characteristics during power generation periods for power matching, and adopts a multi-layer matching correction mechanism to effectively alleviate the limitations of traditional forecasting methods in handling nonlinear and non-stationary meteorological data. At the same time, it reduces the cumulative error of multi-day continuous forecasts and improves the stability and reliability of forecast results to a certain extent.
[0070] Example 2:
[0071] This embodiment provides a day-ahead photovoltaic forecasting device based on time-segmented clustering of meteorological data, including a processor and a memory storing a number of computer instructions. When the computer instructions are executed by the processor, they implement the steps of the day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data.
[0072] For specific limitations on day-ahead photovoltaic forecasting devices based on time-segmented clustering of meteorological data, please refer to the limitations on day-ahead photovoltaic forecasting methods based on time-segmented clustering of meteorological data mentioned above, which will not be repeated here.
[0073] The memory and processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory stores a computer program that can run on the processor, which implements the method of the present invention by running the computer program stored in the memory.
[0074] The memory may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory stores the program, and the processor executes the program upon receiving an execution instruction.
[0075] The processor may be an integrated circuit chip with data processing capabilities. The aforementioned processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.
[0076] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0077] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.
Claims
1. A day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data, characterized in that, Includes the following steps: Historical meteorological data, solar irradiance data, and photovoltaic power generation data are divided into multiple time periods along the time dimension of photovoltaic power generation. Sensitive meteorological types are then categorized for each sub-time period based on the meteorological data. This categorization includes: counting the number of times with the same meteorological type within each sub-time period; selecting meteorological types with a number greater than or equal to a threshold as the sensitive meteorological type for that sub-time period; if no such type exists, the meteorological type for that sub-time period is set to "no meteorological sensitivity type." For each sub-period, the meteorological data is used to cluster each moment in the sub-period into different typical meteorological type cluster centers to obtain the solar irradiance data and photovoltaic power generation data clustered under each typical meteorological type cluster center in the sub-period. The typical meteorological type is used to characterize the impact of meteorology on the performance and lifespan of energy equipment. The meteorological data and solar irradiance data for the forecast date are divided into sub-periods, and the sensitive meteorological types of each sub-period are classified. The set of most similar historical dates is then searched based on the sensitive meteorological types of each sub-period of the forecast date. This process includes: encoding the sensitive meteorological types of each sub-period of the forecast date and combining the encoded values in chronological order to obtain a forecast date meteorological type sequence; encoding the sensitive meteorological types of each sub-period of historical dates and combining the encoded values in chronological order to obtain a historical date meteorological type sequence; measuring the similarity between the forecast date meteorological type sequence and each historical date meteorological type sequence, and selecting historical dates with a similarity greater than a date similarity threshold to obtain a historical date set. Calculate the solar irradiance integral for each sub-period of the predicted date, and retrieve the historical sub-periods with the same meteorological change trend and the closest solar irradiance integral from the corresponding sub-periods of the historical date set; Based on meteorological data, each time point in each sub-period of the forecast day is clustered into different typical meteorological type cluster centers. The historical photovoltaic power generation power with the same typical meteorological type cluster center and the closest solar irradiance in the corresponding sub-period is obtained from the historical date set. The historical photovoltaic power generation power that meets the smoothing requirements is used to replace the historical photovoltaic power generation power at the corresponding time in the historical sub-period to obtain the final day-ahead photovoltaic forecast result for the forecast day.
2. The day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data according to claim 1, characterized in that, The method of dividing the photovoltaic power generation period into multiple time periods in the time dimension includes: dividing the photovoltaic power generation period according to the time characteristics of photovoltaic power generation.
3. The day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data according to claim 1, characterized in that, Typical meteorological type cluster centers are set, and a center-based unsupervised clustering method is adopted to cluster each time period in the sub-period into different meteorological type cluster centers according to the meteorological data.
4. The day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data according to claim 1, characterized in that, The meteorological trends are the same, and the following conclusion is drawn: The weather types for all times within the sub-periods of the forecast day are encoded, and the encoded values are combined in chronological order to obtain the weather type sequence for the sub-periods of the forecast day. The weather types of all times within a sub-period of a historical date are encoded, and the encoded values are combined in chronological order to obtain a sequence of weather types for each sub-period of a historical date. The similarity between the meteorological type sequence of the predicted day and the meteorological type sequence of the corresponding historical date in the historical date set is measured. Historical sub-periods with similarity greater than the time period similarity threshold are taken as sub-periods with the same meteorological change trend.
5. The day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data according to claim 1, characterized in that, The requirement to satisfy smoothness includes: The current search time in the current search sub-period of the predicted day is recorded as the replacement time, and the historical photovoltaic power generation obtained from the historical date set is recorded as the replacement power; The replacement range is determined based on the historical photovoltaic power generation of the two moments adjacent to the replacement moment in the corresponding historical sub-period. If the replacement power is within the replacement range, then the replacement power meets the smoothing requirement; otherwise, the replacement power does not meet the smoothing requirement.
6. The day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data according to claim 1, characterized in that, The final day-ahead photovoltaic forecast result obtained includes: Take the historical photovoltaic power generation at each moment in the replaced historical sub-period as the predicted value of photovoltaic power generation at the corresponding moment on the prediction day; Set the predicted photovoltaic power generation for times outside the photovoltaic power generation period to the default value; By sequentially piecing together the photovoltaic power generation forecast values for each moment of the forecast day, a complete day-ahead photovoltaic forecast result is obtained.
7. A day-ahead photovoltaic forecasting device based on time-segmented clustering of meteorological data, comprising a processor and a memory storing a plurality of computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the day-ahead photovoltaic forecasting method based on time-segmented clustering of meteorological data as described in any one of claims 1 to 6.