Intelligent analysis system and method for new energy photovoltaic power prediction data

Abstract

The invention relates to the technical field of photovoltaic power prediction, in particular to an intelligent analysis system and method for new energy photovoltaic power prediction data, and the system comprises a data collection and processing module, a prediction modeling module and an energy management module which are connected in sequence. The data acquisition and processing module acquires operation data of a photovoltaic power station, and establishes a hierarchical data structure comprising a real-time data layer DRT, a historical data layer DRH and a prediction data layer DRP. The prediction modeling module constructs a real-time generated power model and a generated power prediction model based on the hierarchical data structure; and the energy management module optimizes the tracking type photovoltaic array angle and the charging and discharging power of the energy storage system according to the prediction result. Through accurate short-time power prediction and an intelligent energy management strategy, power fluctuation of photovoltaic power generation is remarkably reduced, system stability and economic benefits are improved, and the method is suitable for intelligent operation management of a large-scale photovoltaic power station and a micro-grid system.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power prediction, and particularly relates to an intelligent analysis system and method for new energy photovoltaic power prediction data. Background Technique

[0002] Due to the randomness and intermittency of solar radiation intensity, photovoltaic power generation has significant volatility and uncertainty characteristics. Currently, photovoltaic power stations generally adopt the method of configuring energy storage systems to suppress power fluctuations, forming a grid-connected mode of photovoltaic + energy storage.

[0003] However, this mode still faces severe control difficulties in actual operation: the energy storage system often can only respond passively without accurate power prediction, resulting in a lag in charge and discharge strategies and being unable to effectively suppress rapid power fluctuations; the capacity of the energy storage system is limited and the cost is high. How to maximize the use efficiency of the energy storage system while ensuring power fluctuation suppression is still a key issue faced by the industry; for complex power stations with both fixed and tracking photovoltaic arrays configured, the response characteristics of the two arrays to weather changes are different, and it is difficult for conventional control strategies to achieve overall optimization; existing photovoltaic power prediction technologies mainly focus on long time scales (day-ahead, intraday), and the accuracy of short-term prediction (15 minutes to 1 hour) is insufficient, unable to provide timely and effective decision support for the photovoltaic + energy storage system.

[0004] Therefore, the development of a high-precision short-term power prediction technology and the realization of intelligent collaborative control of photovoltaic and energy storage systems based on this have become the key to solving the problem of fluctuation control in the photovoltaic + energy storage grid-connected mode. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent analysis system and method for new energy photovoltaic power prediction data. By constructing a hierarchical data structure and a multi-dimensional prediction model, the problems of insufficient accuracy of traditional photovoltaic power prediction, unsuitable prediction period for real-time control, and suboptimal coordination between photovoltaic and energy storage are solved. High-precision prediction of the power generation power of a photovoltaic power station in the next 15 minutes to 1 hour and intelligent collaborative control of the photovoltaic-energy storage system based on the prediction results are realized, significantly improving the operation stability and economic benefits of the photovoltaic power station.

[0006] In the first aspect, the present invention provides an intelligent analysis system for new energy photovoltaic power prediction data, which is used to accurately predict the power generation power of a photovoltaic power station in the target area in the next 15 minutes to 1 hour, and then optimize the coordinated operation of the photovoltaic power station and the energy storage system based on the prediction results; the system includes: a data acquisition and processing module, a prediction and modeling module, and an energy management module, which are connected in sequence.

[0007] The data acquisition and processing module is used to collect the operation data of the photovoltaic power station in the target area, perform standardization processing and outlier removal on the operation data, generate a standardized data set, and establish a hierarchical data structure DR based on the standardized data set.

[0008] The operation data of the photovoltaic power station in the target area includes: real-time total power generation, fixed photovoltaic array power generation, tracking photovoltaic array power generation, and energy storage system power.

[0009] The prediction modeling module is used to perform feature extraction and prediction modeling based on the hierarchical data structure DR, construct a real-time power generation model using DRT data, construct a preliminary power generation prediction model based on DRH data and the real-time power generation model; combine DRT and DRH data to construct a final power generation prediction model.

[0010] The energy management module is used to optimize the angle of the tracking photovoltaic array and control the charge and discharge power of the energy storage system in real time based on the prediction results of the final power generation prediction model, so as to reduce the fluctuation of the output power on the grid side.

[0011] Furthermore, the sampling period of the data acquisition and processing module is set to 1 minute. The hierarchical data structure DR includes: real-time data layer DRT, historical data layer DRH, and prediction data layer DRP. The data of each layer is composed of a sequence with a length of 1440.

[0012] The real-time data layer DRT stores the standardized data set of the most recent 1440 minutes, denoted as the DRT sequence, which is used to construct the actual power generation model; the data of the historical data layer DRH is aggregated and transformed from DRT data, which is used to construct the preliminary power generation prediction model; the data of the prediction data layer DRP is obtained by converting DRH data.

[0013] The method for the data acquisition and processing module to perform standardization processing on the operation data is: , where is the original data of the real-time data layer DRT, and are the maximum and minimum values of the data respectively, is the standardized DRT data.

[0014] Outlier detection is performed using a sliding time window , where and are the mean and standard deviation of the data within the time window respectively. The window length is 12 hours, which is used to identify abnormal power generation caused by sudden weather changes.

[0015] The method for standardizing DRH data is: ; where, is the DRH data before standardization, and are the mean and standard deviation of the DRH data, is the DRH data after standardization; The standardization method for DRP data is: , where, is the DRP data before standardization, and are the minimum and maximum values of the DRH data after standardization.

[0016] The data acquisition and processing module updates the historical data using the formula: ; The mapping from DRH data to DRP data is expressed as: ; where, and represent the new DRH data and the old DRH data respectively, represents the mean of the DRT data; represents the average of the difference between the new DRH data and the old DRH data; is the trend term, representing the solar radiation intensity trend coefficient, determined based on the fitting error of historical light data. [[ID= forty]]

[0017] Furthermore, the prediction and modeling module constructs a power generation model using DRT data, including: establishing a real-time total system power model based on DRT data: ; where, is the real-time total system power at time t, in minutes, is the power generation of the fixed photovoltaic array at time, and the fixed photovoltaic array refers to photovoltaic modules with a fixed installation angle; is the power generation of the tracking photovoltaic array at time, and the tracking photovoltaic array can optimize the angle according to the sun position; is the power of the energy storage system at time, negative during charging and positive during discharging, and the maximum charge and discharge power does not exceed 40% of the rated power.

[0018] The prediction and modeling module constructs a preliminary power generation prediction model based on DRH data as: ; where, is the preliminary prediction result of the photovoltaic power generation at , is the prediction duration, represents the base power generation of the most recent 15 minutes, reflecting the background light level of the local area, ; is the peak power amplitude, representing the power fluctuation amplitude caused by the change of solar illumination angle within the most recent 15 minutes, is the phase angle, reflecting the intraday distribution characteristics of light intensity; is the power random fluctuation component affected by cloud cover and atmospheric transparency factors at

[0019] The prediction modeling module constructs the final power generation prediction model by combining DRT and DRH data: ; where, is the final prediction result of the photovoltaic power generation at time, is the prediction trend term caused by the change of solar radiation intensity at is the prediction error term at time, satisfying the normal distribution

[0020] Further, the method for the energy management module to optimize the angle of the tracking photovoltaic array and control the power of the energy storage system includes: When time, if the predicted power generation is greater than the current actual photovoltaic power generation, it indicates that the lighting conditions will improve. Then, optimize the angle of the tracking photovoltaic array to improve the power generation efficiency; at the same time, set the charging power of the energy storage system according to the power difference to store the excess electric energy generated at future times.

[0021] When time, if the predicted power generation is less than the current actual photovoltaic power generation, it indicates that the lighting conditions will deteriorate. Then, increase the charging power of the energy storage system to store the current excess electric energy, and the upper limit of the charging power does not exceed 95% of the rated power of the energy storage system.

[0022] The energy management module monitors the total power of the real-time system in real time and keeps it stable within the time interval [t, t + 15], and the volatility does not exceed 10%.

[0023] Further, the prediction modeling module establishes an evaluation index system for the prediction model: Mean absolute error: , used to evaluate the overall accuracy of the prediction model, where is the actually measured photovoltaic power generation.

[0024] Root mean square error: , which is used to evaluate the sensitivity of the prediction model to light changes.

[0025] Prediction accuracy: , which is used to evaluate the relative accuracy of the prediction model under different light intensities.

[0026] The prediction modeling module dynamically optimizes the prediction model parameters based on the evaluation results of the evaluation index system: When MAE > 10%, recalculate the solar radiation intensity trend coefficient ; When RMSE > 15%, update 's value to adapt to cloud changes and weather fluctuations; When ACC < 85%, adjust 's prediction period.

[0027] Furthermore, the method for the prediction modeling module to calculate the power random fluctuation component is as follows: ; where is the average deviation of the photovoltaic power generation at the same time within the last 24 hours, reflecting the daily light pattern; is the standard deviation of the photovoltaic power generation within the last 24 hours, reflecting the volatility of cloud cover and atmospheric transparency; when , take , indicating that the cloud fluctuation factor is given priority under highly unstable weather conditions.

[0028] Furthermore, the method for the prediction modeling module to calculate the power change trend term is as follows: ; where is the photovoltaic power generation at the current moment , is the photovoltaic power generation 24 hours before the moment , and the difference between the two reflects the daily change pattern of the solar altitude angle and light intensity.

[0029] When the calculated is greater than the historical maximum change rate, take the historical maximum change rate value to avoid over-prediction caused by extreme weather changes.

[0030] Furthermore, the method for the prediction modeling module to calculate the prediction error term is as follows: , where is the maximum prediction error within the last 7 days, is the characteristic time constant, with a value of 12; when >60 minutes, that is, when the forecast duration exceeds 1 hour, The values are: ,in is the error attenuation coefficient, which ranges from [0.01, 0.05] and reflects the decrease in prediction accuracy caused by uncertainty factors as the prediction time increases.

[0031] Furthermore, the prediction modeling module has an adaptive optimization mechanism: When the MAE is greater than 12% in three consecutive evaluations, the model retraining mechanism is triggered to readjust the mapping relationship between light intensity and power generation power.

[0032] When RMSE>15% and ACC<85%, data quality assessment was re-performed to remove abnormal data points caused by rainy weather or equipment failure.

[0033] The model is fully evaluated every 24 hours, and the prediction period and model parameters are adaptively adjusted based on the evaluation results to adapt to seasonal changes in the sun's trajectory and sunshine duration.

[0034] The model optimization adopts a sliding time window mechanism with a window length of 10080 minutes to capture periodic weather patterns.

[0035] In a second aspect, the present invention provides an intelligent analysis method for new energy photovoltaic power prediction data, the method comprising the following steps: Step 1: Collect operating data of the photovoltaic power station in the target area, standardize the operating data and remove outliers to generate a standardized data set, and establish a hierarchical data structure DR based on the standardized data set; Step 2: Perform feature extraction and predictive modeling based on the hierarchical data structure DR, use DRT data to build a real-time power generation model, and build a preliminary power generation prediction model based on the DRH data and the real-time power generation model; and build a final power generation prediction model by combining the DRT and DRH data. Step three: Based on the prediction results of the final power generation prediction model, optimize the tracking photovoltaic array angle and control the charging and discharging power of the energy storage system in real time to reduce the fluctuation of the grid-side output power.

[0036] The beneficial effects of the present invention are as follows: The present invention realizes high-precision prediction of the power generation of a photovoltaic power station in the next 15 minutes to 1 hour by constructing a three-layer hierarchical data structure and integrating a multi-dimensional prediction model. Based on this, an intelligent photovoltaic-energy storage collaborative control strategy is developed, which significantly reduces the volatility of the output power on the grid side (controlled within 10%), while optimizing the angle of the tracking photovoltaic array and the charge and discharge management of the energy storage system, improving the overall power generation efficiency of the system and the utilization rate of the energy storage, reducing the power curtailment loss and the loss of energy storage equipment, and greatly enhancing the economic benefits and operation stability of the photovoltaic + energy storage grid-connected system, providing reliable technical support for the grid connection and consumption of large-scale photovoltaic power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the composition of an intelligent analysis system for new energy photovoltaic power prediction data of the present invention; Figure 2 It is a flowchart of an intelligent analysis method for new energy photovoltaic power prediction data of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be described clearly and completely below. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0039] Embodiment 1 As Figure 1 shown, it is a schematic diagram of the composition of an intelligent analysis system for new energy photovoltaic power prediction data of the present invention, which is used to accurately predict the power generation of a photovoltaic power station in the target area in the next 15 minutes to 1 hour, and then optimize the coordinated operation of the photovoltaic power station and the energy storage system based on the prediction results.

[0040] The system includes: a data acquisition and processing module, a prediction and modeling module, and an energy management module that are connected in sequence; this intelligent analysis system can be deployed in the central control room of the photovoltaic power station and is connected to each data acquisition device and control device through a field bus; the system hardware can adopt an industrial-grade server configuration, such as an Intel Xeon E5 processor, more than 16GB of memory, and solid-state drive storage, to ensure high-speed data processing capabilities and system stability; at the software level, a modular architecture design is adopted, and data interaction between modules is carried out through standardized interfaces, which is convenient for system upgrade and maintenance.

[0041] The data acquisition and processing module is used to collect the operation data of the photovoltaic power station in the target area, perform standardized processing and outlier rejection on the operation data, generate a standardized data set, and establish a hierarchical data structure DR based on the standardized data set; The operation data of the target area photovoltaic power station includes: real-time total power generation, fixed photovoltaic array power generation, tracking photovoltaic array power generation, and energy storage system power.

[0042] The data acquisition and processing module of the prior art obtains raw data through the illuminance sensor, temperature sensor, wind speed sensor, and photovoltaic inverter communication interface installed on site; in a large photovoltaic power station, dozens or even hundreds of sensors may be configured simultaneously, distributed in different areas to capture local meteorological changes. However, the present invention only focuses on the real-time total power generation, fixed photovoltaic array power generation, tracking photovoltaic array power generation, and energy storage system power, and infers external factors such as weather from these data, greatly simplifying the complexity of the operation. The target area can be divided into multiple sub-areas for data acquisition, and each sub-area is configured with an independent sensor network. For example, a 20MW photovoltaic power station is divided into 4 sub-areas of 5MW each, and each sub-area independently measures the power generation parameter. This partitioned acquisition method helps to improve the fault tolerance of the system and the accuracy of targeted prediction. For a large group of photovoltaic power stations across regions, it can also be partitioned according to geographical location and grid connection method for acquisition to adapt to the meteorological differences in different regions.

[0043] The sampling period of the data acquisition and processing module is set to 1 minute, and the hierarchical data structure DR includes: real-time data layer DRT, historical data layer DRH, and prediction data layer DRP. The data of each layer is composed of a sequence with a length of 1440.

[0044] Using a minute-level sampling period can capture the impact of rapid cloud changes on light intensity. For example, under cloudy weather conditions, the change in light intensity caused by cloud movement usually occurs within a few minutes, and minute-level sampling can effectively record this change characteristic. Selecting 1440 as the sequence length corresponds to 24 hours (1440 minutes) of the whole day, ensuring that the system can capture the complete daily cycle change pattern.

[0045] The real-time data layer DRT stores the standardized data set of the most recent 1440 minutes, denoted as the DRT sequence, for constructing the actual power generation model; the data of the historical data layer DRH is aggregated and transformed from the DRT data, for constructing the preliminary power generation prediction model; the data of the prediction data layer DRP is obtained by converting the DRH data.

[0046] In terms of data storage, DRT data is stored in a cache to ensure millisecond-level read and write speeds; DRH data is stored in a relational database such as MySQL or PostgreSQL for easy structured query; DRP data, as intermediate calculation results, is temporarily stored in a memory data structure. The system also regularly backs up DRT and DRH data to a distributed storage system to ensure data security and traceability. In actual deployment, redundant servers and storage devices can be configured, and a primary-backup switching mechanism can be adopted to ensure the continuous operation of the system 7×24 hours.

[0047] The method by which the data acquisition and processing module standardizes the operation data is as follows: , where is the raw data of the real-time data layer DRT, and are the maximum and minimum values of the data respectively, is the standardized DRT data. In regions with obvious seasonal changes, there may be significant differences in the maximum power generation in summer and winter; for a photovoltaic power station in a certain eastern region of China, the maximum power in summer can reach 95% of the rated capacity, while in winter it can only reach about 75%; the system ensures the rationality of data standardization by dynamically updating and parameter values; for newly built power stations, the system will collect data for at least 7 days during the initial operation stage to establish an initial maximum and minimum value range.

[0048] An outlier detection is performed using a sliding time window , where and are the mean and standard deviation of the data within the time window respectively, and the window length is 12 hours, which is used to identify abnormal power generation caused by sudden weather changes.

[0049] For the data points initially identified as abnormal, the system will further compare and analyze them with the data at adjacent time points to reduce the misjudgment rate. For example, when it is detected that the power suddenly drops at a certain moment, the system will query the data of the illuminance sensor at the same moment to confirm whether it is caused by cloud cover. In the actual application of a photovoltaic power station in a certain northwestern region, the system successfully identified the abnormal power generation data caused by sandstorms through this mechanism, avoiding the wrong training of the prediction model.

[0050] The method for standardizing DRH data is as follows: ; where is the DRH data before standardization, and are the mean and standard deviation of the DRH data, is the standardized DRH data; the method for standardizing DRP data is as follows: , where is the DRP data before standardization, and are the minimum and maximum values of the DRH data after standardization.

[0051] The data acquisition and processing module updates the historical data using the formula: ; The weight coefficients 0.8 and 0.2 avoid the excessive influence caused by temporary fluctuations. For example, on the first sunny day after rainy days, if new data is fully used for updating, the prediction model may be overly sensitive to the increased light; while using the weighted update method can smooth this sudden change. According to the power station operation environment, these weights can be adjusted appropriately. For example, power stations with stable light in desert areas can increase the weight of new data, while areas with variable weather can appropriately increase the weight of historical data.

[0052] The prediction modeling module is used to perform feature extraction and prediction modeling based on the hierarchical data structure DR, construct a real-time power generation model using DRT data, and construct a preliminary power generation prediction model based on DRH data and the real-time power generation model; combine DRT and DRH data to construct a final power generation prediction model. In terms of hardware configuration, this module can be deployed on a server with GPU acceleration function, significantly improving the speed of complex feature extraction and model training. For example, using NVIDIA Tesla V100 GPU, the model training time can be shortened from the hour level of traditional CPU to the minute level, meeting the real-time prediction requirements.

[0053] The mapping from DRH data to DRP data is expressed as: ; where and represent the new DRH data and the old DRH data respectively, represents the mean value of DRT data; represents the average value of the difference between the new DRH data and the old DRH data; is the trend term, representing the solar radiation intensity trend coefficient, determined based on the fitting error of historical light data.

[0054] The energy management module is used to optimize the angle of the tracking photovoltaic array and control the charging and discharging power of the energy storage system in real time based on the prediction results of the final power generation prediction model, so as to reduce the fluctuation of the power output on the grid side. The energy management module is directly connected to the SCADA system at the power station site and sends control instructions through standard industrial communication protocols (such as Modbus or IEC 61850). This module adopts a hierarchical control strategy. First, it calculates the optimal control target, and then decomposes it into specific execution instructions. The module also integrates a safety constraint check mechanism to ensure that all control instructions meet the equipment operation limit conditions. The system provides an operator interface, allowing manual intervention in the automatic control process to adapt to the operation requirements in special situations.

[0055] The prediction modeling module uses DRT data to construct a power generation model, including: establishing the real-time total system power based on DRT data Model: ; where is the real-time total system power at time t, in minutes, is the power generation of the fixed photovoltaic array at time, and the fixed photovoltaic array refers to the photovoltaic modules with a fixed installation angle; is the power generation of the tracking photovoltaic array at time, and the tracking photovoltaic array can be optimized and adjusted according to the sun position; is the power of the energy storage system at time, negative during charging and positive during discharging, and the maximum charging and discharging power does not exceed 40% of the rated power; the system will optimize the parameters according to the characteristics of different types of photovoltaic modules. For example, for a power station using a mixture of monocrystalline and polycrystalline modules, the system will establish the power generation characteristic models of the two types of modules respectively to improve the accuracy of the total power prediction.

[0056] The preliminary power generation prediction model constructed by the prediction modeling module based on DRH data is: ; where is the preliminary prediction result of the photovoltaic power generation at , is the prediction duration, represents the basic power generation in the last 15 minutes, reflecting the background light level in the local area, ; is the peak power amplitude, representing the power fluctuation amplitude caused by the change of the sun illumination angle in the last 15 minutes, is the phase angle, reflecting the daily distribution characteristics of the light intensity; is The power random fluctuation component affected by cloud cover and atmospheric transparency factors at that time; The prediction modeling module constructs a final power generation prediction model by combining DRT and DRH data: ; where is the final prediction result of the photovoltaic power generation at time, is the predicted trend term caused by the change of solar radiation intensity at time, is the prediction error term at

[0057] The method for the energy management module to optimize the angle of the tracking photovoltaic array and control the power of the energy storage system includes: When time, if the predicted power generation is greater than the current actual photovoltaic power generation, it indicates that the lighting conditions will improve, then optimize the angle of the tracking photovoltaic array to improve the power generation efficiency; at the same time, set the charging power of the energy storage system according to the power difference to store the excess electric energy generated in the future.

[0058] The angle adjustment of the tracking photovoltaic array adopts a step-by-step strategy to avoid mechanical wear caused by frequent large adjustments. The system calculates the optimal inclination angle and azimuth angle according to the predicted future lighting direction and gradually adjusts them in 3 - 5 steps. The charging power setting of the energy storage system adopts a gradient strategy. When the predicted power increases significantly, reserve part of the capacity to cope with possible over-generation situations. For example, if the predicted power increases by 20% in the next 30 minutes, the current charging power is set to 70% of the difference, leaving 30% of the capacity to cope with uncertainties.

[0059] When time, if the predicted power generation is less than the current actual photovoltaic power generation, it indicates that the lighting conditions will deteriorate, then increase the charging power of the energy storage system to store the current excess electric energy, and the upper limit of the charging power does not exceed 95% of the rated power of the energy storage system; if the power is predicted to continue to decline within the next 45 minutes, the system will complete most of the charging tasks within the first 15 minutes and then gradually reduce the charging power in the next 30 minutes to prepare for possible discharge demands.

[0060] The energy management module monitors the total power of the real-time system in real time and keeps it stable within the time interval [t, t + 15], with the volatility not exceeding 10%. When the short-term volatility reaches 7%, an alarm is issued to allow the operator to intervene in advance; when it reaches 9%, the emergency control program is automatically started to ensure that the hard limit of not exceeding 10% is not exceeded.

[0061] The prediction modeling module establishes an evaluation index system for the prediction model: Mean absolute error: , used to evaluate the overall accuracy of the prediction model, where is the actual measured photovoltaic power generation power; Root mean square error: , used to evaluate the sensitivity of the prediction model to illumination changes; Prediction accuracy: , used to evaluate the relative accuracy of the prediction model under different light intensities; The prediction modeling module dynamically optimizes the prediction model parameters based on the evaluation results of the evaluation index system: When MAE>10%, recalculate the solar radiation intensity trend coefficient ; When RMSE>15%, update to adapt to cloud changes and weather fluctuations; When ACC < 85%, adjust prediction period.

[0062] The predictive modeling module calculates the random fluctuation component of power The method is: ;in, The average deviation of photovoltaic power generation in the same period within the last 24 hours, reflecting the sunlight pattern; is the standard deviation of photovoltaic power generation in the last 24 hours, reflecting the fluctuation of cloud cover and atmospheric transparency; When, take , indicating that cloud fluctuation factors are given priority in highly unstable weather conditions.

[0063] The predictive modeling module calculates power change trend items The method is: ;in, For the current moment The photovoltaic power generation power, For the moment The difference between the photovoltaic power generation in the previous 24 hours reflects the diurnal variation of the solar altitude angle and light intensity. When the calculated When it is greater than the historical maximum rate of change, the historical maximum rate of change is taken to avoid over-forecasting caused by extreme weather changes.

[0064] The predictive modeling module calculates the prediction error term The method is: ,in, is the maximum forecast error in the last 7 days, is the characteristic time constant, which takes a value of 12; when >60 minutes, that is, when the forecast duration exceeds 1 hour, The values are: ,in is the error attenuation coefficient, which ranges from [0.01, 0.05] and reflects the decrease in prediction accuracy caused by uncertainty factors as the prediction time increases.

[0065] It should be noted that the predictive modeling module has an adaptive optimization mechanism: When the MAE is greater than 12% in three consecutive evaluations, the model retraining mechanism is triggered to readjust the mapping relationship between light intensity and power generation; When RMSE>15% and ACC<85%, re-evaluate the data quality and remove abnormal data points caused by rainy weather or equipment failure; The model is fully evaluated every 24 hours, and the prediction period and model parameters are adaptively adjusted based on the evaluation results to adapt to seasonal changes in the sun's trajectory and sunshine duration; Model optimization uses a sliding time window mechanism with a window length of 10,080 minutes to capture periodic weather patterns. The adaptive optimization mechanism adopts a layered architecture, including three levels: real-time fine-tuning, daily optimization, and periodic reconstruction. Real-time fine-tuning is performed once an hour, mainly adjusting the forecast parameters; daily optimization is performed every 24 hours to re-evaluate the model structure and parameters; periodic reconstruction is performed every 7 days to comprehensively update the basic model parameters and algorithm selection. The system also sets a trigger mechanism for special weather events. For example, when meteorological changes such as rainfall and strong winds are detected, model adjustments are automatically triggered to adapt to environmental changes in advance. A backup strategy is adopted during the model optimization process to retain the model parameters of the last three versions. When the new parameters are not effective, it can quickly fall back to the historical optimal version.

[0066] The present invention was implemented in a 20MW photovoltaic power station in East China. The power station was equipped with a 5MW / 10MWh lithium battery energy storage system. The photovoltaic modules included a 15MW fixed installation and a 5MW dual-axis tracking installation.

[0067] Before implementation, the power station experienced severe power fluctuations under cloudy weather conditions, with the maximum fluctuation rate reaching 35% within 15 minutes, frequently triggering power-rationing orders from the power grid dispatching. The average annual power-rationing loss reached 8.3% of the total power generation.

[0068] After applying the system of the present invention, through high-precision short-term power prediction and intelligent photovoltaic-storage collaborative control, the 15-minute power volatility of the power station under cloudy weather conditions is controlled within 9.2%, significantly lower than the control target of 10%. Especially during sunrise and sunset, by optimizing the angle of the tracking photovoltaic array, the power generation duration is effectively extended, and the average daily utilization hours of the power station are increased. The system also reasonably arranges the charging and discharging plans of the energy storage system according to the power prediction results, which not only suppresses short-term fluctuations but also participates in the peak shaving service of the power grid, increasing additional income.

[0069] The one-year operation data shows that the average accuracy rate of the system for 15-minute photovoltaic power prediction reaches 92.5%, the MAE is controlled within 7.3%, and the RMSE is 10.8%, significantly better than the industry average level. By reducing curtailment losses and improving power generation efficiency, the system realizes an annual power generation increase of 4.2%. At the same time, the additional income created by the energy storage system participating in the grid auxiliary service accounts for 3.5% of the total income of the power station. The overall investment payback period is less than 2 years, with significant economic and social benefits.

[0070] Embodiment 2 As Figure 2 shown, it is a schematic flow diagram of an intelligent analysis method for new energy photovoltaic power prediction data of the present invention, which is implemented based on the intelligent analysis system for new energy photovoltaic power prediction data in Embodiment 1. The method includes the following steps: Step 1, collect the operation data of the photovoltaic power station in the target area, perform standardization processing and outlier removal on the operation data to generate a standardized data set, and establish a hierarchical data structure DR based on the standardized data set.

[0071] The system configures a timed task to trigger data collection once every 1 minute to ensure the consistency of data timestamps. After the collection is completed, the data is transmitted to the central processing server through a pre-configured communication network (such as industrial Ethernet or 4G / 5G wireless network). For large-scale photovoltaic power stations, a distributed edge computing architecture can be adopted to perform preliminary data processing on-site and then transmit it to the central server to reduce communication load and latency.

[0072] The system first performs an integrity check on the received data, fills in the missing data points using the interpolation method, and then processes it using the standardization formula in the present invention. For abnormal weather conditions (such as sandstorms, heavy rains, strong convective weather, etc.), the system will automatically adjust the outlier determination threshold to improve the accuracy of abnormal data identification. The standardized and outlier-removed data is organized according to the hierarchical data structure described in Embodiment 1, stored in the system database, and an index is established to accelerate subsequent queries and processing.

[0073] Step 2: Perform feature extraction and predictive modeling based on the hierarchical data structure DR, use DRT data to build a real-time power generation model, and build a preliminary power generation prediction model based on DRH data and the real-time power generation model; and build a final power generation prediction model by combining DRT and DRH data.

[0074] In this step, the system first reads the latest DRT and DRH data from the database, and determines the prediction duration parameters based on the current system time and the predicted target time. The feature extraction process uses parallel computing technology to simultaneously process data features from multiple time windows, improving computational efficiency. The system updates real-time power generation model parameters, including the power characteristics of fixed PV arrays, tracking PV arrays, and energy storage systems, based on an adaptive cycle (typically every five minutes).

[0075] To construct the preliminary power generation forecast model, the system employs a two-stage process: first, it determines the baseline power generation and peak power amplitude based on historical data from the same period. Then, it calculates the phase angle and random power fluctuation component, incorporating today's meteorological data. The system provides an interactive parameter adjustment interface, allowing professionals to intervene and adjust key model parameters as needed.

[0076] During the final forecast model construction phase, the system employs differentiated strategies based on the forecast duration: for ultra-short-term forecasts under 15 minutes, real-time data and recent trends are prioritized; for short-term forecasts of 15-60 minutes, historical data from the same period and meteorological trends are considered. After each forecast is completed, the system compares the predicted results with actual power generation data, calculates the forecast error, and records it for subsequent model optimization.

[0077] Step three: Based on the prediction results of the final power generation prediction model, optimize the tracking photovoltaic array angle and control the charging and discharging power of the energy storage system in real time to reduce the fluctuation of the grid-side output power.

[0078] The prediction results are converted into control objectives, generating a specific sequence of control instructions. For tracking photovoltaic arrays, the system calculates the optimal tracking angle and, taking into account mechanical execution time and power consumption, generates step-by-step angle adjustment instructions. For energy storage system control, the system comprehensively considers the battery's state of charge (SOC), remaining life, and charge and discharge efficiency, using a multi-objective optimization algorithm to calculate the optimal charge and discharge power setpoints.

[0079] The system also implements a hierarchical control strategy: for power fluctuation predictions less than 5%, the system mainly suppresses them by adjusting the angle of the tracking photovoltaic array; for medium fluctuations of 5% - 10%, the energy storage system is simultaneously called for auxiliary regulation; for large fluctuations greater than 10%, the system will activate the warning mechanism, and while calling all controllable resources, send warning messages to the operators. The control instructions are sent to the on-site execution devices through an encrypted channel, and a feedback mechanism is provided to ensure the accuracy and timeliness of the instruction execution.

[0080] The system also sets up an emergency handling process. When a communication interruption or equipment failure is detected, it can automatically switch to a preset safe operation mode to ensure the safe and stable operation of the power station. All control operations and system state changes are recorded in detail for post-event analysis and system optimization.

[0081] The specific implementation manners described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only the specific implementation manners of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent analysis system for new energy photovoltaic power prediction data, which is used to accurately predict the power generation of a photovoltaic power station in a target area in the next 15 minutes to 1 hour, and then optimize the coordinated operation of the photovoltaic power station and the energy storage system based on the prediction results; characterized in that, The system includes, connected in sequence: a data acquisition and processing module, a prediction and modeling module, and an energy management module; The data acquisition and processing module is configured to collect the operation data of a photovoltaic power station in a target area, perform standardization processing and outlier removal on the operation data, generate a standardized data set, and establish a hierarchical data structure DR based on the standardized data set; The operation data of the photovoltaic power station in the target area includes: real-time total power generation, fixed photovoltaic array power generation, tracking photovoltaic array power generation, and energy storage system power; The prediction and modeling module is configured to perform feature extraction and prediction modeling based on the hierarchical data structure DR, construct a real-time power generation model using DRT data, and construct a preliminary power generation prediction model based on DRH data and the real-time power generation model; combine DRT and DRH data to construct a final power generation prediction model; The energy management module is configured to, based on the prediction result of the final power generation prediction model, optimize the angle of the tracking photovoltaic array and control the charge and discharge power of the energy storage system in real time, so as to reduce the fluctuation of the power output on the grid side.

2. The intelligent analysis system for new energy photovoltaic power prediction data according to claim 1, characterized in that The sampling period of the data acquisition and processing module is set to 1 minute. The hierarchical data structure DR includes: a real-time data layer DRT, a historical data layer DRH, and a prediction data layer DRP. The data of each layer is composed of a sequence with a length of 1440; The real-time data layer DRT stores the standardized data set of the most recent 1440 minutes, denoted as the DRT sequence, which is used to construct an actual power generation model; the data of the historical data layer DRH is aggregated and transformed from DRT data, which is used to construct a preliminary power generation prediction model; the data of the prediction data layer DRP is obtained by converting DRH data; The method for the data acquisition and processing module to perform standardization processing on operation data is as follows: , where is the original data of the real-time data layer DRT, and are the maximum and minimum values of the data respectively, is the standardized DRT data; An outlier detection method using a sliding time window , where and are the mean and standard deviation of the data within the time window respectively. The window length is 12 hours and is used to identify abnormal power generation caused by sudden weather changes; The standardization method for DRH data is as follows: ; where is the DRH data before standardization, and are the mean and standard deviation of the DRH data, is the DRH data after standardization; The standardization method for DRP data is: , where is the DRP data before standardization, and are the minimum and maximum values of the DRH data after standardization; The data acquisition and processing module updates the historical data using the formula: ； The mapping from DRH data to DRP data is represented as: ; Among them, and represent new DRH data and old DRH data respectively, represents the mean value of DRT data; represents the average value of the difference between new DRH data and old DRH data; is a trend term, representing the solar radiation intensity trend coefficient, which is determined based on the fitting error of historical light data.

3. The intelligent analysis system for new energy photovoltaic power prediction data according to claim 2, wherein The prediction modeling module constructs a real-time power generation model using DRT data, including: establishing the total real-time system power based on DRT data model: wherein, is the total real-time system power at time t, in minutes, is the power generation power of the fixed photovoltaic array at time, and the fixed photovoltaic array refers to photovoltaic modules with a fixed installation angle; is the power generation power of the tracking photovoltaic array at time, and the tracking photovoltaic array can be optimized and adjusted according to the sun position; is the power of the energy storage system at time, negative during charging, positive during discharging, and the maximum charge and discharge power does not exceed 40% of the rated power; The prediction and modeling module constructs a preliminary power generation prediction model based on DRH data as: ; among them, is the preliminary prediction result of the photovoltaic power generation at ; is the prediction duration, represents the basic power generation power in the recent 15 minutes, reflecting the local area background light level, ; is the peak power amplitude, representing the power fluctuation amplitude caused by the change of the sunshine angle in the recent 15 minutes, is the phase angle, reflecting the intraday distribution characteristics of the light intensity; is the power random fluctuation component affected by the cloud cover and atmospheric transparency factors at The prediction and modeling module combines DRT and DRH data to construct a final power generation prediction model: ; wherein, is the final prediction result when the photovoltaic power generation is at . is the predicted trend term caused by the change in solar radiation intensity at is the prediction error term at , and satisfies the normal distribution .

4. The intelligent analysis system for new energy photovoltaic power prediction data according to claim 3, wherein, The method for the energy management module to optimize the angle of the tracking photovoltaic array and control the charge and discharge power of the energy storage system in real time includes: When the predicted power generation is greater than the current actual photovoltaic power generation, indicating that the lighting conditions will improve, the angle of the tracking photovoltaic array is optimized to improve the power generation efficiency; at the same time, the charging power of the energy storage system is set according to the magnitude of the power difference to store the excess electric energy generated at future times; When the predicted power generation is less than the current actual photovoltaic power generation, it indicates that the lighting conditions will deteriorate, and then increase the charging power of the energy storage system to store the current excess electric energy, and the upper limit of the charging power does not exceed 95% of the rated power of the energy storage system; The energy management module monitors the real-time total power of the real-time system, keeps it stable within the time interval [t, t + 15], and the volatility does not exceed 10%.

5. The intelligent analysis system for new energy photovoltaic power prediction data according to claim 4, wherein The prediction and modeling module establishes an evaluation index system for the prediction model: Mean absolute error: , which is used to evaluate the overall accuracy of the prediction model, where is the actually measured photovoltaic power generation; Root Mean Square Error: , which is used to evaluate the sensitivity of the prediction model to light changes; Prediction accuracy: , which is used to evaluate the relative accuracy of the prediction model under different light intensities; The prediction and modeling module dynamically optimizes the prediction model parameters based on the evaluation results of the evaluation index system: When MAE > 10%, recalculate the solar radiation intensity trend coefficient ; When RMSE > 15%, update the value to adapt to cloud changes and weather fluctuations; When ACC < 85%, adjust the prediction period.

6. The intelligent analysis system for new energy photovoltaic power prediction data according to claim 5, wherein The prediction modeling module calculates the random power fluctuation component The method is as follows: ; among them, is the average deviation of the photovoltaic power generation at the same time period in the most recent 24 hours, reflecting the daily sunlight pattern; is the standard deviation of the photovoltaic power generation in the most recent 24 hours, reflecting the volatility of cloud cover and atmospheric transparency; when , take , indicating that the cloud fluctuation factor is given priority under highly unstable weather conditions.

7. The intelligent analysis system for new energy photovoltaic power prediction data according to claim 6, wherein The prediction modeling module calculates the power change trend term The method is as follows: ; wherein, is the photovoltaic power generation at the current moment , is the photovoltaic power generation 24 hours before the moment . The difference between the two reflects the diurnal variation law of the solar altitude angle and the light intensity; When the calculated is greater than the historical maximum change rate, the historical maximum change rate value is taken to avoid over-prediction caused by extreme weather changes.

8. The intelligent analysis system for new energy photovoltaic power prediction data according to claim 7, characterized in that The prediction modeling module calculates the prediction error term The method is as follows: , where is the maximum prediction error within the most recent 7 days, is the characteristic time constant, with a value of 12; when > 60 minutes, that is, when the prediction duration exceeds 1 hour, the value is: , where is the error attenuation coefficient, and its value range is [0.01, 0.05], which reflects the decrease in prediction accuracy caused by uncertain factors as the prediction duration increases.

9. The intelligent analysis system for new energy photovoltaic power prediction data according to claim 8, characterized in that The prediction and modeling module has an adaptive optimization mechanism: When the MAE is greater than 12% in three consecutive evaluations, trigger the model retraining mechanism and readjust the mapping relationship between light intensity and power generation; When RMSE > 15% and ACC < 85%, re-evaluate the data quality and remove abnormal data points caused by rainy weather or equipment failures; Perform a complete evaluation of the model every 24 hours, and adaptively adjust the prediction period and model parameters according to the evaluation results to adapt to seasonal changes in the sun's trajectory and sunshine duration; Model optimization uses a sliding time window mechanism with a window length of 10080 minutes to capture periodic weather patterns.

10. An intelligent analysis method for new energy photovoltaic power prediction data, which is used to execute the intelligent analysis system for new energy photovoltaic power prediction data described in any one of claims 1-9, and is characterized in that, The method includes the following steps: Step 1: Collect the operation data of the photovoltaic power station in the target area, perform standardization processing and outlier removal on the operation data to generate a standardized data set, and establish a hierarchical data structure DR based on the standardized data set; Step 2: Based on the hierarchical data structure DR, perform feature extraction and prediction modeling. Use DRT data to construct a real-time power generation power model, and based on DRH data and the real-time power generation power model, construct a preliminary power generation power prediction model; Combine DRT and DRH data to construct a final power generation power prediction model; Step 3: Based on the prediction results of the final power generation power prediction model, optimize the angles of the tracking photovoltaic arrays and control the charging and discharging power of the energy storage system in real time to reduce the fluctuations in the output power on the grid side.

Images