A method, device and equipment for dynamic power distribution of a mountain photovoltaic power station inverter

Abstract

The application discloses a mountain photovoltaic power station inverter dynamic power distribution method, device and equipment. The method comprises the following: obtaining historical operation data of each photovoltaic array inverter of the mountain photovoltaic power station, and synthesizing corresponding array total power curve; reading real-time power data and operation state information of each inverter according to a preset period, and generating a real-time power time sequence segment; performing similarity matching on the real-time power time sequence segment and the array total power curve, determining the historical date with the highest similarity and the corresponding best matching time point according to the matching result, determining the predicted distribution ratio at the current time based on the reference power curve; dynamically correcting the predicted distribution ratio to obtain the target power distribution ratio of each inverter at the next time; generating power adjustment instructions of each inverter according to the target power distribution ratio and issuing the power adjustment instructions to the corresponding inverter for execution. The method can realize real-time and differentiated dynamic distribution of power of each inverter under complex terrain.

Description

Technical Field

[0001] This invention relates to the field of new energy power technology, and in particular to a method, device and equipment for dynamic power distribution of inverters in mountain photovoltaic power stations. Background Technology

[0002] In recent years, the photovoltaic industry has experienced explosive growth, and its application scenarios have rapidly expanded from traditional flat and open areas such as deserts, wastelands, and solar-fishery hybrid systems to complex terrains such as hills, mountains, and basins. This shift in geographical environment has further amplified the component mismatch problem within photovoltaic arrays, posing unprecedented challenges to the power distribution of inverters. In complex terrains, it is not only necessary to cope with rapid weather changes, but also to focus on solving the problem of spatial non-uniformity of sunlight caused by the terrain.

[0003] However, existing inverter power allocation methods generally suffer from the following drawbacks, which limit the power generation efficiency and operation and maintenance benefits of mountain photovoltaic power plants: (1) Existing power allocation methods are usually based on global meteorological data. The total power of the entire array or power station is predicted by similar day theory or neural network. Then, the power is allocated to each inverter according to the rated capacity ratio or a fixed coefficient. This method ignores the difference in actual output capacity of different inverters in the same array due to local terrain and shading. This results in a serious mismatch between the allocation command and the actual lighting conditions. The overall power generation efficiency is lower than the design value. At the same time, it accelerates the aging of inverters and increases operation and maintenance costs.

[0004] (2) Existing forecasting and allocation methods rely heavily on external data such as real-time irradiance meters, cloud maps, and weather stations. However, many mountain power stations have complex terrain and lack complete meteorological data collection conditions, especially lacking independent meteorological data for each array, which makes it difficult to effectively construct similar day models or neural network models, or significantly reduces forecast accuracy.

[0005] (3) Existing allocation methods mostly use complex models such as clustering and deep neural networks. The training and prediction processes involve a large amount of computation, making it difficult to achieve a fast response on local edge devices and failing to meet the real-time adjustment requirements of mountain power stations for complex lighting environment changes.

[0006] Therefore, existing technologies cannot perform real-time dynamic power allocation for the different lighting patterns of each inverter, and cannot effectively solve the problem of uneven power distribution caused by the complex terrain of mountain power plants, thus restricting the improvement of power generation efficiency and the extension of equipment life. Summary of the Invention

[0007] The technical problem to be solved by this invention is to address the above-mentioned shortcomings of the prior art by proposing a method, device, and equipment for dynamic power allocation of inverters in mountainous photovoltaic power stations. This method can realize real-time and differentiated dynamic allocation of power for each inverter under complex terrain.

[0008] In a first aspect, the present invention provides a dynamic power distribution method for an inverter in a mountain photovoltaic power station, the method comprising the following steps: Historical operating data of each photovoltaic array inverter in the mountain photovoltaic power station is obtained, a reference power curve for each inverter is generated based on the historical operating data, and the corresponding total power curve of the array is synthesized based on the reference power curves. Read the real-time power data and operating status information of each inverter according to a preset cycle, update the sliding window buffer based on the real-time power data and operating status information; generate the real-time total power sequence of the photovoltaic array based on the time-series data in the sliding window buffer, and finally obtain the real-time power time-series segment. The real-time power timing segment is matched with the total power curve of the array. Based on the matching results, the historical date with the highest similarity and the corresponding best matching time point are determined. Based on the benchmark power curve, the power allocation ratio of each inverter at the next moment of the best matching time point on the historical date is determined as the predicted allocation ratio at the current moment. Based on the operating status information and the change in the number of online inverters, the predicted allocation ratio is dynamically corrected to obtain the target power allocation ratio of each inverter in the next moment. Based on the target power allocation ratio, power adjustment commands for each inverter are generated and sent to the corresponding inverters for execution, so as to achieve dynamic power allocation for the inverters of the mountain photovoltaic power station.

[0009] Furthermore, a reference power curve for each inverter is generated based on historical operating data, including: Perform validity verification on historical operational data; Extract historical power curves from verified historical operating data; The historical power curves are preprocessed to generate the reference power curves for each inverter; Validating historical operational data includes: The power data of each inverter when the output power is greater than zero is selected from the historical operation data, and the selected power data is collected into daily power data on a daily basis. Daily power data is subjected to data integrity and data rationality checks. The conditions for passing the data integrity check are: the number of data sampling points on the day is not less than a preset first threshold; the conditions for passing the data rationality check are: in the non-zero power data on the day, the ratio of the frequency of any identical power value to the total amount of non-zero data on the day is not higher than a preset second threshold, and the cumulative power value on the day is positive. When both the data integrity check and the data rationality check for the day pass, the historical operational data for that day is deemed to be qualified.

[0010] Furthermore, real-time power data and operating status information of each inverter are read at preset intervals, and the sliding window buffer is updated based on the real-time power data and operating status information, specifically including: Each inverter is allocated a sliding window buffer of fixed length, the length of which is greater than the preset minimum matching length; Real-time power data and operating status information of each inverter are read according to a preset sampling period; the real-time power data package contains power, and the operating status information includes inverter identifier, timestamp and operating status. For each read of real-time power data and operating status information, the data at the current moment is stored at the tail of the corresponding inverter's sliding window buffer, while the data at the head of the buffer corresponding to the earliest moment is removed, thereby updating the sliding window buffer to ensure that the sliding window buffer always stores a fixed number of consecutive sampling moments up to the current moment.

[0011] Furthermore, based on the time-series data within the sliding window buffer, a real-time total power sequence of the photovoltaic array is generated, ultimately yielding a real-time power time-series segment, specifically including: Based on the operating status of each inverter and the sampling time and corresponding power value stored in the sliding window buffer according to the sampling time sequence, the real-time validity of each inverter is judged and the valid inverters are selected. The conditions for determining that an inverter is valid include: the inverter is in the power generation state and the communication state is normal, and the power value of the consecutive preset number of sampling times including the current time in the sliding window buffer is not all zero. For each sampling moment stored in the sliding window buffer, the power values ​​of the selected effective inverters at the sampling moment are time-aligned and accumulated to obtain the total power of the matrix at the sampling moment. The total power of the matrix at each sampling moment is then arranged according to the sampling time sequence to form the original real-time total power sequence. The original real-time total power sequence was smoothed by a Savitzky-Golay filter, which eliminated high-frequency noise while preserving the timing trend characteristics, thus obtaining a real-time power timing segment.

[0012] Furthermore, the real-time power time series segments are matched with the total power curve of the array using similarity matching, specifically including: Based on the fact that each inverter in the set of inverters marked as valid at the current time has complete historical power data in the corresponding time period of the historical date, a candidate set of historical dates is selected from the historical benchmark curve library. If the candidate historical date set is not empty, a sliding window method is used to extract a historical segment with the same duration as the real-time power time series segment from the total power curve of each candidate historical date. The distance metric between the real-time power time series segment and each historical segment is calculated using Euclidean distance or dynamic time warping distance. Among all candidate historical dates and all starting index positions, the historical segment with the smallest distance metric value is selected as the optimal matching result, and the candidate historical date and end time index corresponding to the optimal matching result are recorded.

[0013] Furthermore, the power allocation ratio of each inverter at the next moment after the optimal matching time point on the historical date is determined as the predicted allocation ratio for the current moment, specifically including: If the distance metric value corresponding to the best matching time point is less than or equal to the preset similarity threshold, the match is determined to be successful, and the power allocation ratio of each inverter corresponding to the next moment of the best matching time point on the historical date is extracted from the historical allocation ratio benchmark library as the predicted allocation ratio for the current moment. If the distance metric value corresponding to the best matching time point is greater than the preset similarity threshold, the matching is determined to be unsuccessful. The real-time power of each effective inverter at the current moment is obtained, the actual allocation ratio at the current moment is calculated, and the actual allocation ratio is used as the predicted allocation ratio at the current moment.

[0014] Furthermore, based on operational status information and changes in the number of online inverters, the predicted allocation ratio is dynamically adjusted, specifically including: If any inverter is detected to change from a normal state to a fault or shutdown state, the predicted allocation ratio of the faulty inverter is set to zero, and the original allocation weight of the faulty inverter is normalized and redistributed according to the current predicted allocation ratio of the remaining normal inverters; if all inverters are in an abnormal state, the equal allocation strategy is executed. If any inverter is detected to have recovered from a fault or shutdown state to a normal state, a recovery observation window of a preset duration is started. During the window period, an equal distribution strategy is implemented for each inverter. After the window ends, if there are a preset number of consecutive non-zero sample values ​​in the buffer of the inverters that have recovered to normal, the inverters that have recovered to normal are included in the similarity matching and proportional allocation process. If the number of inverters recorded in the historical baseline curve library is greater than or equal to the number of currently active inverters, then the ratio value corresponding to the current active inverter identifier is extracted from the historical allocation ratio baseline library, the invalid inverter ratio is removed, and the extracted ratio value is normalized so that the sum of the ratio values ​​is 1. If the number of currently active inverters in real time is greater than the number of records in the historical baseline curve library, it is determined that the historical data is insufficient to support similarity matching, and an equal distribution strategy is implemented for each active inverter in real time. When the running time of a newly added inverter reaches the preset time, the running data of the newly added inverter is included in the offline curve fitting process to update the historical baseline curve library and the historical allocation ratio baseline library.

[0015] Furthermore, based on the target power allocation ratio, power adjustment commands are generated for each inverter and sent to the corresponding inverter for execution, specifically including: The target active power value of each inverter is determined based on the target power allocation ratio of each inverter and the total power target value of the photovoltaic power station. Convert the target active power value into a power regulation command that conforms to the communication protocol specifications of each inverter; The power adjustment command is sent to the corresponding inverter to control each inverter to perform power output according to the target power allocation ratio at the next scheduling time.

[0016] Furthermore, the historical power curves are preprocessed to generate reference power curves for each inverter, specifically including: Identify the effective power generation start point, effective power generation end point, and the boundary point corresponding to the maximum power in the historical power curve; Based on the time sequence of the effective power generation start point, boundary point, and effective power generation end point, the historical power curve is divided into an ascending segment and a descending segment; the ascending segment is the power data from the effective power generation start point to the boundary point, and the descending segment is the power data from the boundary point to the effective power generation end point. Outlier points are iteratively removed from the rising segment until the power sequence of the rising segment satisfies the monotonically non-decreasing characteristic. Detect trend inflection points in the downward segment and extract trend control points that reflect the power change trend; Based on the processed data points of the rising segment and the trend control points extracted from the falling segment, interpolation and smoothing fitting are performed to obtain the baseline power curve.

[0017] Furthermore, based on the time sequence of the effective power generation start point, boundary point, and effective power generation termination point, the historical power curve is divided into an ascending segment and a descending segment, specifically including: Determine the first time point corresponding to the effective power generation start point and the third time point corresponding to the effective power generation end point; Traverse the historical power curves to identify the global maximum power value, and determine a unique boundary point as the second moment based on the occurrence of the global maximum power value; According to the chronological order of the time series, the power data sequence between the first and second time points is divided into the rising segment, and the power data sequence between the second and third time points is divided into the falling segment. Specifically, determining the unique boundary point based on the occurrence of the global maximum power value includes: if there is only one sampling point in the historical power curve with a power value equal to the global maximum power value, the time corresponding to the sampling point is determined as the boundary point; if there are multiple sampling points in the historical power curve with power values ​​equal to the global maximum power value, the time corresponding to the sampling point that appears first in the time series is selected as the boundary point; if there is a continuous range of sampling points in the historical power curve with power values ​​fluctuating within the preset tolerance range of the global maximum power value and the duration exceeding a preset threshold, the starting time of the continuous sampling points is selected as the boundary point.

[0018] Furthermore, outlier removal is performed iteratively on the rising segment, specifically including: Calculate the first difference of the current power sequence in the rising segment; Detect negative values ​​in the first-order difference values, identify the next sampling point corresponding to the negative value as an outlier, and remove it from the current power sequence; The first-order difference of the power sequence after removing outliers is recalculated, and the detection and removal process is repeated until all the first-order difference values ​​of the current power sequence are greater than or equal to zero, or the preset maximum number of iterations is reached.

[0019] Furthermore, trend inflection point detection is performed on the downward segment to extract trend control points reflecting the power change trend, specifically including: Calculate the first difference of the power sequence in the falling segment and the relative rate of change between adjacent sampling points; When the number of consecutive sampling points in the falling segment exceeds the preset consecutive frame number threshold, and the relative change rate of the consecutive sampling points is less than the preset relative change rate threshold, the first sampling point in the consecutive sampling points is determined as the power drop start point. The signs of the relative change rates before and after each sampling point in the descending segment are detected. If the signs of the relative change rates before and after the sampling point are opposite, and their absolute values ​​are both greater than the relative change rate threshold, then the sampling point is determined to be a trend inflection point. The power data at the end of the descent phase is detected, and the sampling point where the value first drops to a preset zero value close to the threshold is determined as the tail stabilization point; The indexes of the starting point of the power drop, the turning point of the trend, and the tail stabilization point are summarized, and after deduplication, they are sorted in ascending order of time to generate a set of trend control points.

[0020] Furthermore, based on the processed rising segment data points and the trend control points extracted from the falling segment, interpolation and smoothing fitting are performed to obtain the baseline power curve, specifically including: The data points in the rising segment after removing outliers, and the trend control points extracted from the falling segment, are combined to form the benchmark fitting point set. Based on the set of benchmark fitting points, interpolation is performed on the positions in the original power curve other than the benchmark fitting points to generate a reconstructed power sequence. Akima interpolation is applied sequentially to the reconstructed power sequence to generate continuous curves, and Savitzky-Golay filters are used to denoise the continuous curves to obtain the reference power curves.

[0021] In a second aspect, the present invention provides a dynamic power distribution device for an inverter in a mountain photovoltaic power station, the device comprising: The first data processing unit is used to acquire historical operating data of each photovoltaic array inverter in the mountain photovoltaic power station, generate a reference power curve for each inverter based on the historical operating data, and synthesize the corresponding total power curve of the array based on each reference power curve. The second data processing unit is used to read the real-time power data and operating status information of each inverter according to a preset cycle, update the sliding window buffer based on the real-time power data and operating status information, generate the real-time total power sequence of the photovoltaic array based on the time-series data in the sliding window buffer, and finally obtain the real-time power time-series segment. The matching prediction unit is connected to the first data processing unit and the second data processing unit respectively. It is used to receive the total power curve of the array and the real-time power timing segment; perform similarity matching between the real-time power timing segment and the total power curve of the array; determine the historical date with the highest similarity and the corresponding optimal matching time point based on the matching result; and determine the power allocation ratio of each inverter at the next moment of the optimal matching time point on the historical date based on the reference power curve, as the predicted allocation ratio at the current moment. The correction unit, connected to the matching prediction unit, is used to dynamically correct the predicted allocation ratio based on the operating status information and the change in the number of online inverters, so as to obtain the target power allocation ratio of each inverter in the next moment. The generation unit, connected to the correction unit, is used to generate power adjustment commands for each inverter according to the target power allocation ratio and send them to the corresponding inverters for execution, so as to realize dynamic power allocation of inverters in mountain photovoltaic power stations.

[0022] Thirdly, the present invention provides an electronic device including a memory and a processor. The memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the dynamic power distribution method for mountain photovoltaic power station inverters according to the first aspect.

[0023] This method can achieve differentiated power allocation based on the historical operating characteristics of each inverter. Without relying on external meteorological monitoring equipment, it can respond dynamically in real time to the non-uniformity of sunlight caused by complex mountainous terrain, thereby improving the overall power generation efficiency of the photovoltaic array and extending the service life of the inverter.

[0024] This invention constructs an individualized benchmark model by extracting historical power curves from each inverter, uses a sliding window for real-time acquisition and filtering to obtain the current power time-series segment, and predicts the power distribution ratio for the next moment based on historical similarity matching. Combined with dynamic corrections based on the number and operating status of online inverters, it generates and issues power adjustment commands. This enables accurate identification and rapid tracking of the differentiated lighting patterns of each inverter, effectively solving the problem of uneven power distribution in mountainous power plants caused by complex terrain and limited meteorological data collection conditions. Specific beneficial effects are as follows: (1) Fully consider terrain differences to improve power generation efficiency and equipment lifespan. In view of the problem that the existing method of allocating according to rated capacity or fixed coefficient leads to misalignment between command and actual illumination, the present invention extracts the historical reference power curve of each inverter and obtains the power allocation ratio of the next moment of the historical best matching moment based on similarity matching. This ratio naturally includes the output capacity difference caused by factors such as mountainous terrain and local shading. At the same time, dynamic correction is made in combination with the change of the number of online inverters, so that the allocation command is highly consistent with the actual illumination and operating status of each inverter, effectively solving the problem of uneven distribution, improving the overall power generation efficiency, avoiding the accelerated aging of inverters due to load mismatch, and reducing operation and maintenance costs.

[0025] (2) Eliminating dependence on external data and enhancing adaptability to complex environments. Addressing the problem of similarity day models or neural networks failing due to the complex terrain and lack of comprehensive meteorological data collection in mountainous power plants, this invention eliminates the need for external data such as real-time irradiance meters, cloud maps, and weather stations. It generates a baseline curve solely using the inverter's own historical operating data and performs similarity matching based on real-time power sequences. This method, driven by internal operating data, fundamentally overcomes the bottleneck of missing or insufficient external meteorological data, greatly improving the model's applicability and accuracy in predictive allocation within complex mountainous environments.

[0026] (3) Lightweight computation to meet real-time edge adjustment requirements. Addressing the issues of high computational cost and difficulty in rapid response on local edge devices for complex models such as clustering and deep neural networks, this invention employs a sliding window buffer mechanism and a temporal segment similarity matching algorithm to circumvent complex model training processes and high computational consumption. This solution not only has low computational cost and rapid response, but can also be deployed on local edge devices to achieve millisecond-level / second-level real-time computation and command issuance, perfectly meeting the stringent requirements of mountain power stations for real-time adjustment of rapidly changing weather and complex lighting environments.

[0027] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0028] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the detailed description of exemplary embodiments with reference to the accompanying drawings, in which: Figure 1 This is a schematic diagram of a dynamic power distribution method for inverters in mountain photovoltaic power stations provided in an embodiment of the present invention; Figure 2 This is a dynamic power distribution framework diagram of an inverter for a mountain photovoltaic power station provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of data acquisition and sliding window maintenance provided in an embodiment of the present invention; Figure 4 This is a similarity matching diagram provided in an embodiment of the present invention; Figure 5 A flowchart of inverter power ratio allocation logic provided for embodiments of the present invention; Figure 6 This is a schematic diagram of a dynamic power distribution device for an inverter in a mountain photovoltaic power station, provided in an embodiment of the present invention. Figure 7 A framework diagram of an electronic device provided in an embodiment of the present invention.

[0029] Reference numerals: 10, first data processing unit; 20, second data processing unit; 30, matching prediction unit; 40, correction unit; 50, generation unit; 100, processor; 200, memory. Detailed Implementation

[0030] It is understood that the specific embodiments and accompanying drawings described herein are merely for explaining the invention and are not intended to limit the invention.

[0031] It is understood that, without conflict, the various embodiments and features in the embodiments of the present invention can be combined with each other.

[0032] It is understood that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, while the parts unrelated to the present invention are not shown in the drawings.

[0033] It is understood that each unit or module involved in the embodiments of the present invention may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple units or modules may be integrated into one entity structure.

[0034] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this invention may occur in a different order than that marked in the accompanying drawings.

[0035] It is understood that the flowcharts and block diagrams of this invention illustrate the possible architecture, functions, and operations of systems, apparatuses, devices, and methods according to various embodiments of this invention. Each block in the flowchart or block diagram may represent a unit, module, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagram and flowchart can be implemented using a hardware-based system to achieve the specified function, or using a combination of hardware and computer instructions.

[0036] It is understood that the units and modules involved in the embodiments of the present invention can be implemented by software or by hardware. For example, the units and modules can be located in a processor.

[0037] Example 1: like Figure 1 As shown, this embodiment proposes a dynamic power allocation method for inverters in mountainous photovoltaic power plants, aiming to solve the power allocation problem in scenarios with complex terrain and varying photovoltaic module orientations and tilt angles. Due to the undulating mountainous terrain, the photovoltaic arrays exhibit significant differences in illumination conditions, shading, and temperature distribution. Traditional fixed-capacity or simple equal-sharing strategies are ill-suited to dynamic operating conditions such as cloud movement and localized shading changes, easily leading to power imbalances among inverters, reduced power generation efficiency, and even equipment overload. Therefore, this embodiment constructs a benchmark power curve library by mining historical inverter operating data, utilizes real-time power time series similarity matching to predict future power changes, and combines this with dynamic adjustment of the allocation ratio based on online inverter status. This achieves accurate and rapid power allocation among inverters in complex mountainous environments, effectively improving the overall power generation efficiency and operational stability of the plant. The dynamic power allocation method for inverters in mountainous photovoltaic power plants provided in this embodiment includes steps S1 to S5.

[0038] Step S1: Obtain historical operating data of each photovoltaic array inverter in the mountain photovoltaic power station, generate a reference power curve for each inverter based on the historical operating data, and synthesize the corresponding total power curve of the array based on each reference power curve.

[0039] As a specific implementation method, the reference power curves for each inverter are generated based on historical operating data. This includes: first, validating the historical operating data and filtering out usable data; then, extracting historical power curves from the valid historical operating data, performing appropriate preprocessing on the historical power curves, and finally generating the reference power curves for each inverter. Specifically, the validity verification of the historical operating data is performed according to the following steps: First, based on the filter condition that the inverter output power is greater than zero, the corresponding historical operating data is extracted, and the extracted power values ​​are collected daily to form a daily dataset.

[0040] Secondly, for each day's dataset, completeness and reasonableness checks are performed sequentially: (i) Integrity verification requires that the total number of sampling points on the day is not less than the preset first threshold to ensure that the coverage of data in the time dimension meets the specified standards; (ii) Reasonableness verification, with the following dual constraints: Among the non-zero power values ​​of the day, the frequency of the same power value must not exceed the preset second threshold to prevent data stagnation or duplicate recording. The cumulative power value for the day must be positive to exclude situations where the effective power is zero or negative due to data anomalies.

[0041] Finally, the historical operating data for that day can only be considered valid if both of the above checks pass; if either check fails, the data for that day should be considered invalid.

[0042] As a specific implementation method, the historical power curve is preprocessed to generate the reference power curve for each inverter, including the following steps: First, the effective power generation start point, effective power generation end point, and the boundary point corresponding to the maximum power value on the historical power curve are identified. Then, based on the chronological order of the effective power generation start point, boundary point, and effective power generation end point, the historical power curve is divided into an ascending segment and a descending segment; wherein, the ascending segment refers to the power data from the effective power generation start point to the boundary point, and the descending segment refers to the power data from the boundary point to the effective power generation end point. For the ascending segment, an iterative outlier removal process is performed until the remaining power sequence within the ascending segment satisfies the monotonically non-decreasing characteristic. For the descending segment, trend inflection point detection is performed to extract trend control points that reflect the power change trend. Finally, based on the processed ascending segment data points and the trend control points extracted from the descending segment, interpolation and smoothing fitting are performed to generate the reference power curve for the inverter.

[0043] In one specific implementation, when dividing the historical power curve into rising and falling segments based on the chronological order of the effective power generation start point, boundary point, and effective power generation termination point in the time series, the first time point corresponding to the effective power generation start point and the third time point corresponding to the effective power generation termination point are first determined. Then, the historical power curve is traversed to identify the global maximum power value, and based on the occurrence of this global maximum power value, a unique boundary point time is determined and designated as the second time point. On this basis, according to the chronological order of the time series, the power data sequence between the first and second time points is divided into rising segments, and the power data sequence between the second and third time points is divided into falling segments. The specific rules for determining the unique dividing point time based on the occurrence of the global maximum power value are as follows: If there is only one sampling point in the historical power curve with a power value equal to the global maximum power value, then the time corresponding to that sampling point is determined as the dividing point time; if there are multiple sampling points in the historical power curve with power values ​​equal to the global maximum power value, then the time corresponding to the sampling point that appears first in the time series is selected as the dividing point time; if there is a continuous sampling point in the historical power curve whose power values ​​fluctuate within the preset tolerance range of the global maximum power value, and the duration of this fluctuation exceeds a preset threshold, then the starting time of this continuous sampling point is selected as the dividing point time. Through the above methods, it is ensured that the dividing point time can be uniquely and clearly determined under any circumstances, thus making the division between the rising and falling segments deterministic and operable.

[0044] As a specific implementation method, outlier removal is performed iteratively on the rising segment, with the following steps: First, calculate the first-order difference value of the current power sequence. Second, detect all negative values ​​in the first-order difference value and identify the next sampling point corresponding to each negative value as an outlier, then remove these outliers from the current power sequence. Then, for the new power sequence obtained after removing outliers, recalculate its first-order difference value and repeat the aforementioned process of detecting negative values, identifying outliers, and removing them. The above iterative operation should continue until either of the following termination conditions is met: all first-order difference values ​​of the current power sequence are greater than or equal to zero, or the number of iterations performed has reached the preset maximum number of iterations. After the above termination conditions are met, the iterative removal process ends, and the final power sequence is taken as the valid data sequence of the rising segment.

[0045] As a specific implementation method, the following steps are taken to detect trend inflection points and extract trend control points reflecting power change trends during the descent segment: First, calculate the first-order difference value of the power sequence during the descent segment, and simultaneously calculate the relative rate of change between adjacent sampling points. Second, during the descent segment, if the number of consecutive sampling points exceeds a preset consecutive frame threshold, and the relative rate of change of these consecutive sampling points is less than a preset relative rate of change threshold, then the first sampling point among these consecutive sampling points is determined as the power drop start point. Third, perform sign detection on each sampling point in the descent segment. If the relative rate of change of a sampling point at the previous moment has the opposite sign to the relative rate of change at the next moment, and the absolute values ​​of both relative rates of change are greater than the relative rate of change threshold, then this sampling point is determined as a trend inflection point. Fourth, detect the power data at the end of the descent segment, and determine the sampling point where the power value first decreases to a preset zero value close to the threshold as the tail stabilization point. Finally, the indices of the power drop start point, trend turning point, and tail stabilization point determined above are summarized. The summarized indices are deduplicated and sorted in ascending order of time. The resulting set of indices is the set of trend control points.

[0046] As a specific implementation method, this application employs the following steps when performing benchmark power curve fitting: First, the rising segment data points after outlier removal are merged with the trend control points extracted from the falling segment data to form a benchmark fitting point set. Second, based on this benchmark fitting point set, numerical interpolation is performed on all positions in the original power curve except for the aforementioned benchmark fitting points to generate a fully reconstructed power sequence. Third, for the reconstructed power sequence, the Akima interpolation method is first applied to generate a continuous interpolation curve; then, the Savitzky-Golay filter is applied to this continuous curve for smoothing and denoising to filter out high-frequency disturbances, ultimately obtaining the benchmark power curve. The above steps are performed sequentially, with interpolation and smoothing based on the benchmark fitting point set, and the denoising process does not alter the overall trend of the fitted curve.

[0047] Step S2: Read the real-time power data and operating status information of each inverter according to a preset cycle, and update the sliding window buffer based on the real-time power data and operating status information; generate the real-time total power sequence of the photovoltaic array based on the time-series data in the sliding window buffer, and filter and denoise the real-time total power sequence to obtain the real-time power time-series segment.

[0048] As a specific implementation method, this method reads the real-time power data and operating status information of each inverter according to a preset period, and updates the sliding window buffer based on the above information. The specific implementation process is as follows: First, a sliding window buffer of fixed length is allocated to each inverter. The buffer length must be strictly greater than the preset minimum matching length. Then, data from each inverter is read according to a preset sampling period. The real-time power data includes at least active power, and the operating status information includes the inverter identifier, timestamp, and operating status. For each piece of data read, the information at the current moment is stored at the tail of the corresponding inverter's sliding window buffer, while the earliest time data stored at the head of the buffer is removed, thus completing the dynamic update of the buffer. This method ensures that the sliding window buffer always stores time-series data corresponding to a fixed number of consecutive sampling moments, with the current moment as the cutoff time.

[0049] As a specific implementation method, the process of generating the real-time total power sequence of a photovoltaic array based on time-series data within a sliding window buffer, and finally obtaining a real-time power time-series segment, is as follows: First, based on the operating status of each inverter and the power values ​​stored in the buffer according to the sampling time sequence, the real-time validity of each inverter is determined, and valid inverters are selected. The conditions for determining validity include that the inverter is in power generation mode and communication is normal, and that the power values ​​of a consecutive preset number of sampling times containing the current moment in the buffer are not all zero. Second, for each sampling moment stored in the buffer, the power values ​​of the selected valid inverters at that moment are time-aligned and accumulated to obtain the total power of the array at that moment. Then, the total power of the array at each sampling moment is arranged according to the sampling time sequence to form the original real-time total power sequence. Finally, a Savitzky-Golay filter is used to smooth the original real-time total power sequence, eliminating high-frequency noise while preserving the time-series trend characteristics, and finally obtaining the real-time power time-series segment.

[0050] Step S3: Perform similarity matching between the real-time power timing segment and the total power curve of the array. Based on the matching results, determine the historical date with the highest similarity and the corresponding optimal matching time point. Based on the benchmark power curve, determine the power allocation ratio of each inverter at the next moment after the optimal matching time point on the historical date, and use it as the predicted allocation ratio for the current moment.

[0051] As a specific implementation method, when performing similarity matching between real-time power time-series segments and the total power curve of the array, the first step is to filter candidate historical date sets from the historical benchmark curve library based on the condition that each inverter in the set of inverters marked as valid at the current time has complete historical power data within the corresponding time period of the historical date. If the candidate historical date set is not empty, a historical segment with the same duration as the real-time power time-series segment is extracted from the total power curve of the array for each candidate historical date using a sliding window method, and the distance metric between the real-time power time-series segment and each historical segment is calculated using Euclidean distance or dynamic time warping distance. Finally, among all candidate historical dates and all starting index positions, the historical segment with the smallest distance metric is selected as the optimal matching result, and the candidate historical date and end time index corresponding to the optimal matching result are recorded.

[0052] As a specific implementation method, when determining the power allocation ratio of each inverter at the next moment after the optimal matching time point on a historical date as the predicted allocation ratio for the current moment, the specific implementation steps are as follows: First, compare the distance metric value corresponding to the optimal matching time point with a preset similarity threshold. If the distance metric value is less than or equal to the preset similarity threshold, the match is considered successful. At this time, the power allocation ratio of each inverter recorded at the next moment after the optimal matching time point on a historical date is extracted from the historical allocation ratio benchmark library and used as the predicted allocation ratio for the current moment. Conversely, if the distance metric value is greater than the preset similarity threshold, the match is considered unsuccessful. In this case, historical data is no longer relied upon; instead, the real-time power of each effective inverter at the current moment is obtained, the actual allocation ratio at the current moment is calculated, and this actual allocation ratio is used as the predicted allocation ratio for the current moment.

[0053] Step S4: Based on the operating status information and the change in the number of online inverters, dynamically correct the predicted allocation ratio to obtain the target power allocation ratio of each inverter at the next moment.

[0054] Specifically, the predicted allocation ratio is dynamically adjusted based on the operating status information and changes in the number of online inverters, following these rules: First, if any inverter is detected to change from a normal state to a fault or shutdown state, the predicted allocation ratio of the faulty inverter is set to zero, and its original allocation weight is normalized and redistributed according to the current predicted allocation ratio of the remaining normal inverters; if all inverters are in an abnormal state, an equal allocation strategy is applied to all inverters. Second, if any inverter is detected to recover from a fault or shutdown state to a normal state, a recovery observation window of a preset duration is started from the recovery time, and an equal allocation strategy is uniformly applied to all inverters during this window period; after the window period expires, the inverter can only be reintroduced into the similarity matching and ratio allocation process if there are a preset number of consecutive non-zero sample values ​​in the buffer of the recovered inverter. Third, if the number of inverters recorded in the historical baseline curve library is greater than or equal to the current number of active inverters, the ratio value corresponding to each active inverter identifier is extracted from the historical allocation ratio baseline library. The ratios corresponding to failed inverters are removed, and the extracted active ratio values ​​are normalized to ensure that the sum of all ratio values ​​is 1. Fourth, if the current number of active inverters is greater than the number recorded in the historical baseline curve library, the historical data is deemed insufficient to support similarity matching, and an equal distribution strategy is applied to each active inverter. Once the cumulative runtime of newly added inverters reaches a preset duration, their runtime data is included in the offline curve fitting process to update the historical baseline curve library and the historical allocation ratio baseline library. These four scenarios are implemented according to their applicable conditions and are not mutually exclusive, but at any given time, only the currently triggered condition applies.

[0055] Step S5: Based on the target power allocation ratio, generate power adjustment commands for each inverter and send them to the corresponding inverters for execution, so as to realize dynamic power allocation of inverters in mountain photovoltaic power stations.

[0056] As a specific implementation method, power regulation commands for each inverter are generated based on the target power allocation ratio and sent to the corresponding inverters for execution. Specifically, this includes: in the calculation stage, determining the target active power value for each inverter based on the target power allocation ratio of each inverter and the total power target value of the photovoltaic power station; in the command conversion stage, converting these target active power values ​​into power regulation commands conforming to the communication protocol specifications of each inverter, ensuring that the converted command format and numerical range meet the parsing requirements of the corresponding inverter; and in the command issuance stage, sending the aforementioned power regulation commands to the corresponding inverters, which receive and execute them to control their power output to strictly adhere to the allocated ratio at the next scheduling time, thereby achieving coordinated power regulation across the entire station.

[0057] To facilitate understanding of the present invention, the method in this embodiment will be described in more detail below with reference to a specific example. For example... Figure 2 As shown, this example completes five main steps in sequence: historical data validity verification, power curve fitting, real-time curve construction, similarity matching, and state switching judgment. First, historical data validity verification is performed to filter out valid historical power generation data from the inverters. Next, power curve fitting is carried out, constructing a baseline characteristic curve for each inverter through curve segmentation, outlier removal, feature point identification, and interpolation smoothing. Then, real-time data from the power plant is collected and a sliding window buffer is maintained to complete inverter validity judgment, generating a smoothed matrix of real-time power segments. Next, similarity matching is performed between the real-time power segments and the historical baseline curve library, predicting the inverter power allocation ratio based on the matching results. Finally, considering inverter commissioning / decommissioning, fault recovery, and other operating conditions, the allocation ratio is dynamically adjusted and state switching is handled. The specific steps are as follows: (1) Validation of historical data: The verification of historical data validity follows these steps: First, all historical data belonging to each inverter array stored on the data acquisition board must be read. The sampling precision can be set by the data acquisition device to once every 5 minutes or every 10 minutes. The core fields collected should include at least the date, array number, inverter number, inverter operating status, and power value. Second, for each inverter, historical power data during its power generation operation (specifically, "grid-connected" and "dispatch" states) should be selected. Furthermore, for each inverter power curve generated after screening, the following three validity checks must be performed sequentially: First, the power curve must have no fewer than 60 data points on any given day (calculated at one sampling point every 5 minutes). This requirement aims to ensure at least 5 consecutive hours of valid data records on any given day. Second, the power value that appears repeatedly on the power curve must not exceed 10 times. If there are power values ​​that remain unchanged for a long period, it is presumed that there may be an anomaly in data acquisition, and the curve should be excluded. Third, the sum of the power values ​​at all time points covered by the power curve must be greater than zero. All of the above conditions must be met simultaneously for the set of historical power data to be considered valid. If any of the above conditions are not met, the verification should be considered unsuccessful, and the data should be discarded and cannot be used as the basis for power generation statistics, equipment operation efficiency assessment, or subsequent data report preparation.

[0058] (2) Power curve fitting: This step involves segmenting, outlier removal, trend identification, and smoothing of the daily power curves that have passed historical data validity verification, generating baseline characteristic curves for each inverter on different dates. Specifically, it includes the following sub-steps: (2.1): Curve segmentation For inverter i, the effective power sequence P(t) on date d k), k=0,1,…,N-1 (increasing by time), identify the point P with the maximum power for the day. max and the corresponding time t peak Using this peak point as the boundary, the curve is divided into two segments (an ascending segment and a descending segment): Ascending phase: from the first effective power generation point t of the day start to t peak This reflects the stage of gradually increasing irradiation.

[0059] Falling segment: from t peak Until the effective power generation end point t of the day end This reflects the stage where radiation gradually weakens or is affected by clouds.

[0060] If there are multiple peaks, take the first global maximum value point; if the maximum power is continuously flat (multiple consecutive points are close to the maximum value), take the leftmost point as the segment point.

[0061] (2.2): Removal of outliers in the rising segment (removal of power drop caused by shadow transients)

[0062] In mountain power plants, there may be occasional brief drops in power during the rising phase due to the rapid sweep of local vegetation or cloud shadows. These points are not part of the normal upward trend of irradiance and should be excluded.

[0063] Step A1: Calculate the first-order difference of the rising power sequence (subtracting the previous point from the later point): ΔP(t) k )=P(t k+1 )-P(t k ), k=0,1,…,m-2 where m is the number of points in the ascending segment.

[0064] Step A2: Detect negative values ​​(ΔP < 0) in the difference values. If negative values ​​exist, it indicates the presence of local descent points. Record the next point P(t) corresponding to these descent points. k+1 The position index of ).

[0065] Step A3: Remove all marked falling points (i.e., points where the power value drops abruptly) from the rising segment. Recalculate the differences for the new sequence after the removal.

[0066] Step A4: Repeat steps A2 to A3 until all the difference values ​​of the current rising segment sequence are greater than or equal to 0 (i.e., the sequence is monotonically non-decreasing), and record the original index (relative to the original curve) of the final valid data points.

[0067] To optimize the monotonicity of the data sequence (e.g., the natural trend of the power decline segment), the iterative elimination process in this embodiment is executed a maximum of 3 times. This process gradually approaches the ideal monotonic state by repeatedly eliminating outliers in the sequence that do not conform to the monotonic trend (such as fluctuations caused by interference). If the sequence still does not reach complete monotonicity after 3 iterations, considering that excessive iteration may reduce the sequence fidelity by eliminating too much valid data, or that subsequent iterations may not significantly improve the monotonicity, the sequence after 3 iterations is retained as the final processing result.

[0068] (2.3): Detection of the turning point of the downward trend (identifying the starting point of rapid drop caused by cloud cover)

[0069] The natural decrease in power during the descent phase is gradual (due to the reduced solar altitude angle), but cloud cover causes a rapid drop in power. The locations of these drop points are significant for identifying the "abrupt response" in subsequent similarity matching and need to be extracted as feature points.

[0070] Step B1: Calculate the first difference of the power sequence in the falling segment.

[0071] ΔP(t k and relative rate of change: , k=0,1,...,m'-2.

[0072] in, It is the relative rate of change; For a moment The power value; For a moment The power value; Let m' be a very small positive number (to avoid the denominator being zero), and m' be the number of points in the descending segment.

[0073] Step B2: Set the relative rate of change threshold η th (Currently set to -0.01, the setting can be modified according to the operating conditions). When η(t) k )<η th When the threshold is negatively exceeded, "cumulative number of consecutive exceedance frames" is initiated. An accumulation counter cnt is defined, which increments once for each consecutive point that meets the condition; otherwise, the counter is reset to zero. If cnt ≥ K th (K th If set to 3 (the setting value can be modified according to the working conditions during specific implementation), then the index of the first point of the continuous drop segment is recorded as the feature point.

[0074] Step B3: If the signs of the two relative rates of change before and after a certain point are opposite (i.e., from decreasing to increasing, or from increasing to decreasing) and their absolute values ​​are both greater than η. th If so, then record that point as a turning point.

[0075] Step B4: The rate of change of power at the end of the falling segment usually tends to level off. If there is a power value close to 0, record the first point close to 0 at the end as a valid tail index.

[0076] Step B5: Summarize all feature point indices obtained in steps B2, B3, and B4, remove duplicates, and sort them in ascending order.

[0077] (2.4): Interpolation and smooth fitting: Step C1: Merge the valid index set retained after outlier removal in the rising segment, the feature point index set identified by trend inflection point detection in the falling segment, and other points in the original curve that were not removed, to construct a complete index space of the same length as the original curve (the position of missing index is defined as the interpolation point).

[0078] Step C2: Use linear interpolation to initially fill the missing positions, the formula is:

[0079] in: The power value at the location to be interpolated; The power value at the effective point on the right boundary; The power value at the effective point on the left boundary; t left The left boundary of the valid point; t right It is the right boundary of the valid point.

[0080] Step C3: For the complete sequence after linear interpolation, apply Akima interpolation to generate a continuous curve (ensuring natural curvature), and then use a Savitzky-Golay filter (in this example, the window length is set to 7 and the polynomial order is set to 2) to remove high-frequency noise, and obtain the final smooth reference power curve P`i,d(t).

[0081] (3) Real-time curve construction: like Figure 3 As shown, this step is responsible for periodically reading the real-time power and status data of each inverter from shared memory, maintaining the sliding window buffer, determining the validity of the inverters, calculating the real-time power summation curve of the array, and smoothing the curve using the same standard as historical curves to provide high-quality real-time segments for similarity matching. Specifically, it includes the following processes: (3.1): Periodically read shared memory data: In this embodiment, real-time data from all inverters is read from the shared memory at a fixed period of Ts = 5 minutes (consistent with the historical data sampling period). Each data entry includes: inverter ID, timestamp, active power, and operating status. The data in the shared memory is refreshed in real time by the underlying data acquisition board to ensure that the latest collected values ​​are obtained during the reading process.

[0082] (3.2): Sliding window buffer maintenance: A fixed-size sliding buffer is allocated to each inverter to store its power and status data for the most recent L sampling periods (in this embodiment, the buffer length is set to L=8). Since similarity matching requires L consecutive sampling points to form a reliable real-time power segment, if the number of valid data points in the buffer is less than the minimum matching length L... min (In this embodiment, L is taken as L) min When the similarity is 5, similarity matching is not triggered, and the equal distribution strategy is used directly. In specific implementation, other strategies can also be used, such as distribution according to the number of PV branches.

[0083] The sliding buffer works as follows: after each new set of data is read, the data is placed at the tail of the corresponding inverter buffer, while the oldest set of data at the head of the buffer is removed (first-in, first-out). Therefore, the buffer always stores the time-series data of the most recent L sampling times as real-time power segments for similarity matching.

[0084] (3.3): Real-time validity assessment of the inverter: To eliminate invalid data caused by communication interruptions, system downtime, etc., it is necessary to determine the validity of each inverter at the current moment.

[0085] Step D1: Status Filtering The inverter is considered to have the basic conditions to participate in power distribution only when the inverter is in the power generation state (grid-connected, dispatched) and the communication state is normal, and the power value is written to the corresponding position in the buffer; otherwise, 0 is written to the corresponding position in the buffer.

[0086] Step D2: Power continuity judgment: The inverter processed in step D1 is further examined in its sliding window buffer for the most recent consecutive L - L. min The active power value of each sampling point (which can be set according to the operating conditions; in this embodiment, the last 3 points of the sliding window are used for power zero determination) is used. If the power of these 3 points is 0, the inverter is determined to be an invalid inverter.

[0087] (3.4): Calculation of Real-Time Power Sum Curve of the Square Array

[0088] For the current time tc Based on the real-time power values ​​of all valid inverters (taking the latest power value from the buffer of each inverter, i.e., the tail data), the total power of the array is calculated: Indicates the current moment The sum of the real-time power of all effective inverters, that is, the total real-time power of the array at that moment; Indicates the current moment The power value; Simultaneously, by utilizing the data from multiple time points stored in the buffer, a matrix total power sequence of the most recent L sampling points can be generated: .

[0089] L represents the number of sampling points; This represents the time of the earliest sampling point in the sequence, that is, the time L-1 sampling intervals backward from the current time; This represents a real-time total power sequence of length L.

[0090] S35: Employs a Savitzky-Golay filter for... Perform a smoothing process to obtain the smoothed real-time summed power sequence. (Smoothed total power time series).

[0091] (4) Similarity matching: like Figure 4 As shown, this step constructs a smooth real-time matrix total power segment based on the real-time curve. The total power curve of the array is matched with the historical benchmark curve library constructed by power curve fitting to find the most similar historical date and corresponding position, and then the power allocation ratio of each inverter at the next moment is extracted from the historical allocation ratio library. This step specifically includes the following sub-steps: (4.1): Historical dates can be used for filtering. From the historical benchmark curve library, a set of candidate dates matching the current operating conditions is selected. Selection criteria: Step E1, Inverter Validity Constraint: For the set of inverters marked as valid at the current moment (Valid), it is required that all these inverters have complete historical power data in the candidate historical dates. If, in a certain historical date, the entire day's historical data of any currently valid inverter is 1e-6 (or all 0, depending on the data filling method), then that historical date is removed from the candidate set.

[0092] Step E2: If the candidate historical date set is empty after the above filtering, this step ends and directly switches to the equal distribution strategy, skipping the subsequent segment distance calculation, optimal matching location and ratio prediction.

[0093] (4.2): Segment distance calculation: Based on the obtained real-time smoothed total power time series Distance calculations are performed between each candidate historical date and all possible time segments of equal length.

[0094] Step F1: Real-time segment representation: Q = [q1,q2,…,q L ], where q k This is the smoothed total power value at the k-th sampling point (5-minute interval).

[0095] Step F2: Historical Segment Traversal: For each candidate historical date d, its total power curve is Hd(t) (smoothed, with the same sampling rate as real time). Slide a window of length L on this curve, with starting indices s from 0 to ML (M being the total number of sampling points for that day), to obtain the historical segment C. d,s = [H d (t s ), H d (t s+1 ),…, H d (t s+L-1 )).

[0096] Step F3, Distance Metric: Calculate Q and each C using Euclidean distance or Dynamic Time Warping (DTW). d,s The similarity distance is calculated using Euclidean distance (which requires less computation). Indicate Q and The Euclidean distance between them is used to measure the similarity between the two; the smaller the value, the more similar they are. This represents the power value at the k-th sampling point of the query sequence Q; express The power value at the kth sampling point.

[0097] (4.3): Optimal matching location: Among all candidate historical dates and all starting indices s, find the matching pair with the smallest distance (d ,s ):

[0098] Record the historical date d corresponding to the optimal match. and the ending index e=s +L-1 (i.e., the index of the last sample point of the historical segment that the current real-time segment is aligned with).

[0099] (4.4): Proportion Prediction: Based on the minimum distance Dist min Whether the value is less than a preset threshold (the threshold is 100,000 in this embodiment, depending on the operating conditions) determines how to obtain the inverter power allocation ratio at the next moment.

[0100] Case 1: Dist min ≤ Threshold (Match successful)

[0101] Based on the historical date d of the best match And with the ending index e, determine the index next = e + 1 for the next moment (corresponding to the next sampling moment in the real-time stream). Extract the historical date d from the historical allocation ratio benchmark library. The allocation ratio ri(d) of each inverter at index next time. (, next). These ratios are used as the power prediction ratios for the next moment of the current real-time data segment and output to step S5 for subsequent constraint correction.

[0102] Case 2: Dist min Threshold (Match failed, no sufficiently similar historical patterns)

[0103] At this point, there is no curve similar to the historical curve. To avoid incorrect allocation, the allocation ratio of the real-time data at the last moment is used as the prediction ratio for the next moment. That is: Based on the current time t c Real-time power Pi(t) of each inverter c ), calculate the actual allocation ratio at the current moment: This represents the actual power distribution ratio of the i-th inverter at the current moment, that is, the proportion of the inverter's power to the total effective power.

[0104] The ratio ri curr As the prediction ratio for the next time step, it is output to the constraint correction.

[0105] (5) State transition judgment: like Figure 5 As shown, this step is based on the predicted proportion ri. pred / ri currBy combining the real-time monitoring of inverter operating status and quantity changes, the final allocation ratio is dynamically adjusted to ensure that the allocation scheme is safe and effective under any operating condition.

[0106] (5.1) Inverter state switching processing: Monitor the operating status of each inverter in real time, and when the status changes, handle the allocation ratio according to the following rules.

[0107] Step G1: Status changes from normal to abnormal (fault / shutdown): If an inverter changes from normal to abnormal, its allocation ratio is forcibly set to 0, and its original ratio weight is normalized and redistributed according to the current ratio weights of the remaining effective inverters. Let r be the predicted ratio vector of the N inverters before the fault. pred =[r1,r2,…,rN], and satisfying If the k-th inverter fails, the updated proportional vector r new for: If all fail, the distribution will be carried out according to the equal distribution strategy.

[0108] Step G2: The status changes from abnormal to normal: To prevent sudden power surges in newly recovered inverters due to transient instability (such as voltage rebuilding or MPPT instability), a recovery observation window is established. Starting from the recovery time, the observation window is considered "completed" only when three consecutive non-zero sample values ​​exist in the buffer. Before the window completes its sliding, the allocation strategy switches to an equal-sharing strategy. Once complete, the inverter is reintegrated into the normal similarity matching and proportional allocation process.

[0109] (5.2) Inverter quantity switching processing: When power plants are expanded, inverters are added or removed, or some inverters are offline for extended periods, discrepancies may occur between the real-time inverter count and the count in the historical baseline database. This step addresses these discrepancies based on the direction of the discrepancy.

[0110] (5.2.1) The number of inverters in historical data is greater than or equal to the number of inverters in real-time data: In this case, we still first match based on similarity and look up N from the historical ratio benchmark database. hist A proportional vector r hist Then, only the ratio value corresponding to the currently valid inverter ID is retained, and the ratio values ​​of other inverters are set to 0. The retained ratio values ​​are then normalized so that their sum is 1. This represents the final allocation ratio of the i-th inverter after screening and normalization; This represents the allocation ratio value of the i-th inverter in the historical ratio vector; This represents the sum of the corresponding proportional values ​​of all real-time active inverters in the historical proportional vector; Where i is the real-time valid inverter ID. If the real-time valid inverter is empty, the equal distribution strategy will be followed.

[0111] (5.2.2) The number of inverters in historical data < the number of inverters in real-time data

[0112] Once N is detected real >N hist If the new inverter has no historical similarity matching, proportional prediction will no longer be used (because the new inverter has no historical benchmark curve and cannot participate in the matching). All real-time effective inverters will be treated as equals, and an equal distribution strategy will be used. After the new inverter has been running for a sufficient number of days (7 days in this embodiment), its data will be included in the offline curve fitting process, the historical benchmark curve library and the proportional benchmark library will be updated, and then similarity matching can be used normally.

[0113] This embodiment provides a dynamic power allocation method for inverters in mountainous photovoltaic power stations, aiming to solve the problems of power allocation imbalance and efficiency reduction under complex terrain and dynamic cloud shadow shading. The method offered in this embodiment has significant advantages in terms of low cost and ease of deployment. It can operate using only existing data acquisition boards and inverter communication data, requiring no additional hardware investment. The baseline curve library is stored in a CSV file, requiring minimal computation and can be directly implemented at the edge controller, facilitating engineering promotion. In specific implementation, the method first acquires and verifies historical operating data of the inverters. Through curve segmentation, iterative removal of outliers, identification of trend inflection points, and interpolation smoothing, a baseline power curve for each inverter and a total power curve for the array are constructed, forming a historical illumination pattern library. Secondly, a sliding window buffer is used to collect power and status data of each inverter in real time. After validity screening and filtering for noise reduction, a real-time total power segment of the array is generated. Subsequently, this real-time segment is matched with the historical total power curve for similarity. The power allocation ratio for the next moment is predicted based on the minimum distance metric. If the match fails, the current real-time ratio is used as a substitute. Based on this, and considering operating conditions such as inverter failure shutdowns, recovery observation periods, and changes in the number of online inverters, the predicted ratio is dynamically corrected using a normalized redistribution or equal-sharing strategy to ensure the safety and reliability of the allocation scheme. Finally, the target power of each inverter is calculated based on the corrected target ratio, and adjustment commands are issued to achieve accurate, rapid, and stable dynamic power coordination allocation under complex operating conditions. By dynamically matching historical similar illumination patterns, the method provided in this embodiment achieves fine-grained power allocation based on forward-looking predictions, effectively avoiding equipment overload and idleness, and bringing the actual power generation close to the theoretical design value. Compared to traditional timed or constant-proportion allocation methods, the method provided in this embodiment not only has stronger robustness and real-time performance but also significantly improves the overall power generation efficiency of mountain power plants.

[0114] Example 2: like Figure 6 As shown, this embodiment provides a dynamic power distribution device for inverters applied to mountain photovoltaic power plants, aiming to optimize power distribution strategies by combining historical data analysis with real-time operating conditions. The device mainly includes a first data processing unit 10, a second data processing unit 20, a matching prediction unit 30, a correction unit 40, and a generation unit 50. These units work collaboratively to achieve efficient and stable operation of the power plant.

[0115] Specifically, the first data processing unit 10 is responsible for historical data mining and benchmark establishment. This unit first comprehensively acquires historical operating data from each photovoltaic array inverter within the mountain photovoltaic power station and executes a rigorous data governance process, eliminating outliers or missing records through validity checks to ensure the quality of the data source. Based on this, it extracts the individual historical power curves of each inverter and generates benchmark power curves that reflect the equipment's operating characteristics through preprocessing. Finally, the benchmark curves of each inverter are combined into the total power curve of the array, providing a solid historical reference benchmark for subsequent prediction and matching.

[0116] The second data processing unit 20 focuses on real-time data acquisition and processing. It reads the real-time power data and operating status information of each inverter according to a preset cycle, and continuously updates the data using sliding window buffer technology to ensure the continuity and timeliness of the time-series data. This unit further processes the data in the buffer to generate the real-time total power sequence of the photovoltaic array, and performs filtering and noise reduction operations to eliminate random fluctuation interference, outputting a smooth and reliable real-time power time-series segment.

[0117] The matching prediction unit 30, acting as a bridge between historical and real-time data, is connected to the first data processing unit 10 and the second data processing unit 20, respectively. It receives the historically generated total power curve of the array and the current real-time power time series segment. Using a similarity matching algorithm, it searches the historical database for the historical date with the highest similarity and its corresponding optimal matching time point. Based on the matching result, the unit calls upon the historical reference power curve to calculate the inverter power allocation ratio for the next historical moment and maps it to the predicted allocation ratio for the current moment.

[0118] Considering the complexity of the mountainous environment and the actual condition of the equipment, the correction unit 40 is connected to the matching prediction unit 30 and is responsible for dynamically optimizing the prediction results. This unit comprehensively considers the real-time operating status information of the inverters and the dynamic changes in the number of online inverters, and adjusts and corrects the predicted allocation ratio in real time. This process effectively avoids allocation deviations caused by equipment maintenance or failures, and finally outputs the target power allocation ratio of each inverter that conforms to the actual operating conditions of the current power station.

[0119] Finally, the generation unit 50 is connected to the correction unit 40, responsible for generating and issuing execution commands. Based on the determined target power allocation ratio, this unit generates precise power adjustment commands and transmits them to the corresponding inverters for execution. Through this series of processes, the device achieves dynamic and scientific power allocation for the inverters in the mountain photovoltaic power station, effectively improving the overall power generation efficiency and operational stability of the power station.

[0120] The apparatus in this embodiment is capable of performing the method in Embodiment 1.

[0121] Example 3: like Figure 7 As shown, this embodiment provides an electronic device equipped with a memory 200 and a processor 100. The memory 200 is used to store computer programs, and the processor 100 is used to call and run the computer programs stored in the memory 200, thereby executing the dynamic power distribution method for the inverter of the mountain photovoltaic power station described in Embodiment 1.

[0122] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A dynamic power distribution method for inverters in mountainous photovoltaic power stations, characterized in that, Includes the following steps: Historical operating data of each photovoltaic array inverter in the mountain photovoltaic power station is obtained, a reference power curve for each inverter is generated based on the historical operating data, and the corresponding total power curve of the array is synthesized according to the reference power curves. Real-time power data and operating status information of each inverter are read at a preset cycle, and the sliding window buffer is updated based on the real-time power data and the operating status information; based on the time series data in the sliding window buffer, the real-time total power sequence of the photovoltaic array is generated, and finally the real-time power time series segment is obtained. The real-time power timing segment is matched with the total power curve of the array. Based on the matching result, the historical date with the highest similarity and the corresponding best matching time point are determined. Based on the benchmark power curve, the power allocation ratio of each inverter at the next moment of the best matching time point on the historical date is determined as the predicted allocation ratio at the current moment. Based on the operating status information and the change in the number of online inverters, the predicted allocation ratio is dynamically corrected to obtain the target power allocation ratio of each inverter at the next moment. Based on the target power allocation ratio, power adjustment commands for each inverter are generated and sent to the corresponding inverters for execution, so as to realize dynamic power allocation for the inverters of the mountain photovoltaic power station.

2. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 1, characterized in that, Based on the historical operating data, a reference power curve for each inverter is generated, including: The validity of the historical operation data is verified. Extract historical power curves from verified historical operating data; The historical power curves are preprocessed to generate reference power curves for each inverter; The validity verification of the historical operational data specifically includes: The power data of each inverter when its output power is greater than zero is selected from the historical operating data, and the selected power data is collected into daily power data on a daily basis. The daily power data is subjected to data integrity verification and data rationality verification; wherein, the conditions for passing the data integrity verification are: the number of data sampling points on the day is not less than a preset first threshold; the conditions for passing the data rationality verification are: in the non-zero power data on the day, the ratio of the frequency of any identical power value to the total non-zero data on the day is not higher than a preset second threshold, and the cumulative power value on the day is positive; When both the data integrity check and the data rationality check for the day pass, the historical operation data for that day is deemed to be qualified.

3. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 1, characterized in that, The step of reading real-time power data and operating status information of each inverter at a preset period, and updating the sliding window buffer based on the real-time power data and the operating status information, specifically includes: A fixed-length sliding window buffer is allocated to each inverter, the length of which is greater than a preset minimum matching length; Real-time power data and operating status information of each inverter are read according to a preset sampling period; wherein, the real-time power data package contains power, and the operating status information includes inverter identifier, timestamp and operating status; For each read of real-time power data and operating status information, the data at the current moment is stored at the tail of the corresponding inverter's sliding window buffer, while the data at the head of the buffer corresponding to the earliest moment is removed, thereby updating the sliding window buffer to ensure that the sliding window buffer always stores time-series data of a fixed number of consecutive sampling moments up to the current moment.

4. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 1, characterized in that, The process of generating a real-time total power sequence of the photovoltaic array based on the time-series data within the sliding window buffer, and ultimately obtaining a real-time power time-series segment, specifically includes: Based on the operating status of each inverter and the sampling time and corresponding power value stored in the sliding window buffer according to the sampling time sequence, the real-time validity of each inverter is determined, and valid inverters are selected. The conditions for determining that an inverter is valid include: the inverter is in the power generation state and the communication state is normal, and the power value of the consecutive preset number of sampling times including the current time in the sliding window buffer is not all zero. For each sampling moment stored in the sliding window buffer, the power values ​​of the selected effective inverters at the sampling moment are time-aligned and accumulated to obtain the total power of the array at the sampling moment. The total power of the array at each sampling moment is arranged according to the sampling time sequence to form the original real-time total power sequence. The original real-time total power sequence is smoothed by a Savitzky-Golay filter to eliminate high-frequency noise while preserving the timing trend characteristics, thus obtaining a real-time power timing segment.

5. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 1, characterized in that, The step of performing similarity matching between the real-time power time series segment and the total power curve of the array specifically includes: Based on the fact that each inverter in the set of inverters marked as valid at the current time has complete historical power data in the corresponding time period of the historical date, a candidate set of historical dates is selected from the historical benchmark curve library. If the candidate historical date set is not empty, a sliding window method is used to extract a historical segment with the same duration as the real-time power time series segment from the total power curve of each candidate historical date matrix. The distance metric between the real-time power time series segment and each of the historical segments is calculated using Euclidean distance or dynamic time warping distance. Among all the candidate historical dates and all starting index positions, the historical segment with the smallest distance metric value is selected as the optimal matching result, and the candidate historical date and end time index corresponding to the optimal matching result are recorded.

6. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 1, characterized in that, Determining the power allocation ratio of each inverter at the next moment after the optimal matching time point on the historical date, as the predicted allocation ratio for the current moment, specifically includes: If the distance metric value corresponding to the optimal matching time point is less than or equal to the preset similarity threshold, the matching is determined to be successful, and the power allocation ratio of each inverter corresponding to the next moment of the optimal matching time point on the historical date is extracted from the historical allocation ratio benchmark library as the predicted allocation ratio for the current moment. If the distance metric value corresponding to the optimal matching time point is greater than the preset similarity threshold, the matching is determined to be unsuccessful. The real-time power of each effective inverter at the current moment is obtained, the actual allocation ratio at the current moment is calculated, and the actual allocation ratio is used as the predicted allocation ratio at the current moment.

7. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 1, characterized in that, The dynamic correction of the predicted allocation ratio based on the operating status information and changes in the number of online inverters specifically includes: If any inverter is detected to change from a normal state to a fault or shutdown state, the predicted allocation ratio of the faulty inverter is set to zero, and the original allocation weight of the faulty inverter is normalized and redistributed according to the current predicted allocation ratio of the remaining normal inverters; if all inverters are in an abnormal state, the equal allocation strategy is executed. If any inverter is detected to have recovered from a fault or shutdown state to a normal state, a recovery observation window of a preset duration is started. During the window period, an equal distribution strategy is implemented for each inverter. After the window ends, if there are a preset number of consecutive non-zero sample values ​​in the buffer of the inverters that have recovered to normal, the inverters that have recovered to normal are included in the similarity matching and proportional allocation process. If the number of inverters recorded in the historical baseline curve library is greater than or equal to the number of currently active inverters, then the ratio value corresponding to the current active inverter identifier is extracted from the historical allocation ratio baseline library, the invalid inverter ratio is removed, and the extracted ratio value is normalized so that the sum of the ratio values ​​is 1. If the number of currently active inverters in real time is greater than the number of records in the historical baseline curve library, it is determined that the historical data is insufficient to support similarity matching, and an equal distribution strategy is implemented for each active inverter in real time. When the running time of a newly added inverter reaches the preset time, the running data of the newly added inverter is included in the offline curve fitting process to update the historical baseline curve library and the historical allocation ratio baseline library.

8. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 1, characterized in that, The step of generating power adjustment commands for each inverter based on the target power allocation ratio and issuing them to the corresponding inverters for execution specifically includes: The target active power value of each inverter is determined based on the target power allocation ratio of each inverter and the total power target value of the photovoltaic power station. The target active power value is converted into a power regulation command that conforms to the communication protocol specifications of each inverter; The power adjustment command is sent to the corresponding inverter to control each inverter to perform power output according to the target power allocation ratio at the next scheduling time.

9. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 2, characterized in that, The preprocessing of the historical power curves to generate reference power curves for each inverter specifically includes: Identify the effective power generation start point, effective power generation end point, and the boundary point corresponding to the maximum power in the historical power curve; Based on the time sequence of the effective power generation start point, the boundary point, and the effective power generation end point, the historical power curve is divided into an ascending segment and a descending segment; wherein, the ascending segment is the power data from the effective power generation start point to the boundary point, and the descending segment is the power data from the boundary point to the effective power generation end point; The rising segment is iteratively eliminated for outliers until the power sequence of the rising segment satisfies the monotonically non-decreasing characteristic. The downward segment is subjected to trend inflection point detection, and trend control points reflecting the power change trend are extracted. Based on the processed rising segment data points and the trend control points extracted from the falling segment, interpolation and smoothing fitting are performed to obtain the benchmark power curve.

10. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 9, characterized in that, The step of dividing the historical power curve into an ascending segment and a descending segment based on the time sequence of the effective power generation start point, the boundary point, and the effective power generation end point specifically includes: Determine the first time point corresponding to the effective power generation start point and the third time point corresponding to the effective power generation end point; The historical power curves are traversed to identify the global maximum power value, and a unique dividing point moment is determined as the second moment based on the occurrence of the global maximum power value. According to the chronological order of the time series, the power data sequence located between the first time point and the second time point is divided into the rising segment, and the power data sequence located between the second time point and the third time point is divided into the falling segment; Specifically, determining the unique dividing point time based on the occurrence of the global maximum power value includes: if there is only one sampling point in the historical power curve with a power value equal to the global maximum power value, the time corresponding to the sampling point is determined as the dividing point time; if there are multiple sampling points in the historical power curve with power values ​​equal to the global maximum power value, the time corresponding to the sampling point that appears first in the time series is selected as the dividing point time; if there is a continuous range of sampling points in the historical power curve with power values ​​fluctuating within a preset tolerance range of the global maximum power value and the duration exceeding a preset threshold, the starting time of the continuous sampling points is selected as the dividing point time.

11. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 9, characterized in that, The iterative removal of outliers in the rising segment specifically includes: Calculate the first-order difference of the current power sequence of the rising segment; Detect negative values ​​in the first-order difference values, identify the next sampling point corresponding to the negative value as an outlier, and remove it from the current power sequence; The first-order difference of the power sequence after removing outliers is recalculated, and the detection and removal process is repeated until all the first-order difference values ​​of the current power sequence are greater than or equal to zero, or the preset maximum number of iterations is reached.

12. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 9, characterized in that, The step of detecting trend inflection points in the descending segment and extracting trend control points reflecting power change trends specifically includes: Calculate the first-order difference of the falling power sequence and the relative rate of change between adjacent sampling points; When the number of consecutive sampling points in the falling segment exceeds a preset consecutive frame number threshold, and the relative change rate of the consecutive sampling points is less than a preset relative change rate threshold, the first sampling point in the consecutive sampling points is determined as the power drop start point. The sign of the relative rate of change before and after each sampling point in the descending segment is detected. If the signs of the relative rate of change before and after the sampling point are opposite, and the absolute values ​​of each are greater than the relative rate of change threshold, then the sampling point is determined to be a trend inflection point. The power data at the end of the falling segment is detected, and the sampling point where the value first drops to a preset zero value close to the threshold is determined as the tail stabilization point; The indices of the power drop start point, the trend turning point, and the tail stabilization point are summarized, deduplicated, and sorted in ascending order by time to generate the trend control point set.

13. The dynamic power distribution method for inverters in mountain photovoltaic power stations according to claim 9, characterized in that, The process of interpolating and smoothing the data points from the rising segment and the trend control points extracted from the falling segment to obtain the baseline power curve specifically includes: The rising segment data points after outlier removal and the trend control points extracted from the falling segment are combined to form a set of benchmark fitting points. Based on the set of benchmark fitting points, interpolation is performed on the positions in the original power curve other than the benchmark fitting points to generate a reconstructed power sequence. The reconstructed power sequence is sequentially subjected to Akima interpolation to generate a continuous curve, and the continuous curve is then denoised using a Savitzky-Golay filter to obtain the reference power curve.

14. A dynamic power distribution device for an inverter in a mountain photovoltaic power station, characterized in that, include: The first data processing unit is used to acquire historical operating data of each photovoltaic array inverter in the mountain photovoltaic power station, generate a reference power curve for each inverter based on the historical operating data, and synthesize the corresponding total power curve of the array according to each reference power curve. The second data processing unit is used to read the real-time power data and operating status information of each inverter according to a preset period, update the sliding window buffer based on the real-time power data and the operating status information, generate the real-time total power sequence of the photovoltaic array based on the time-series data in the sliding window buffer, and finally obtain the real-time power time-series segment. A matching prediction unit, connected to the first data processing unit and the second data processing unit respectively, is used to receive the total power curve of the array and the real-time power time series segment; perform similarity matching between the real-time power time series segment and the total power curve of the array; determine the historical date with the highest similarity and the corresponding optimal matching time point based on the matching result; and determine the power allocation ratio of each inverter at the next moment of the optimal matching time point on the historical date based on the benchmark power curve, as the predicted allocation ratio at the current moment. The correction unit, connected to the matching prediction unit, is used to dynamically correct the predicted allocation ratio based on the operating status information and the change in the number of online inverters, so as to obtain the target power allocation ratio of each inverter at the next moment. The generation unit, connected to the correction unit, is used to generate power adjustment commands for each inverter according to the target power allocation ratio and send them to the corresponding inverters for execution, so as to realize dynamic power allocation of the inverters in the mountain photovoltaic power station.

15. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the dynamic power distribution method for mountain photovoltaic power station inverters according to any one of claims 1 to 13.

Images