Photovoltaic power station power prediction method based on multi-modal data

Abstract

The present application relates to the technical field of photovoltaic power station power prediction, and more particularly to a photovoltaic power station power prediction method based on multi-modal data, comprising: acquiring multi-modal data of a region where a photovoltaic power station is located, respectively performing feature extraction on the multi-modal data, determining a prediction available power not constrained by a power grid according to environmental features and power station operation features; determining a prediction power demand according to the environmental features and power grid demand features, combining a grid-connected capacity and a historical limited generation state to determine a grid-allowed grid-connected power; taking a smaller value of the prediction available power and the grid-allowed grid-connected power as an initial prediction result; performing link-by-link convergence correction according to the stability of each feature in the current scenario, and outputting a photovoltaic available power prediction value, an actual grid-connected power prediction value and a limited power prediction value. The present application fuses multi-modal data and introduces a power grid consumption constraint, so that the prediction result reflects photovoltaic power generation capacity and power grid accommodation capacity at the same time, and the prediction accuracy and reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power plant power prediction technology, and in particular to a photovoltaic power plant power prediction method based on multimodal data. Background Technology

[0002] Photovoltaic power generation is significantly affected by meteorological factors such as solar irradiance and cloud movement, exhibiting marked randomness and volatility. Existing photovoltaic power prediction methods mainly rely on numerical weather prediction data and historical power data, but they do not fully utilize multimodal information such as satellite cloud images and sky images, making it difficult to accurately capture the impact of local meteorological changes such as cloud movement on power generation. Furthermore, most existing prediction methods only focus on the photovoltaic power plant's own power generation capacity, neglecting the limiting effect of grid-side electricity demand, grid-connected capacity, and historical power curtailment conditions on actual grid-connected power. This leads to discrepancies between predicted results and actual grid-connected power, making it difficult to provide accurate grid-connected power references for grid dispatch.

[0003] Chinese Patent Publication No. CN117791548A discloses a method for predicting the power generation of a photovoltaic power station, which is implemented through four steps: data acquisition and processing, selection of representative stations, monitoring visualization and prediction model construction, and monitoring and prediction visualization. This invention's prediction method is beneficial for fully exploring the relationships between various types of data by aggregating distributed photovoltaic power station output, equipment parameters, and meteorological data, and for using algorithms to apply the data in depth, thereby improving the prediction level of photovoltaic power station output. However, the method has the following problem: it does not consider the constraints of grid-side electricity demand on photovoltaic power station power generation, leading to inaccurate prediction results. Summary of the Invention

[0004] To address this issue, the present invention provides a photovoltaic power plant power prediction method based on multimodal data, which overcomes the problem in the prior art that does not consider the constraints of grid-side electricity demand on photovoltaic power plant power generation, resulting in inaccurate prediction results.

[0005] To achieve the above objectives, the present invention provides a photovoltaic power plant power prediction method based on multimodal data, comprising: Acquire multimodal data of the area where the photovoltaic power station is located, including environmental factor data, photovoltaic power station operation data, and grid-side demand data; The multimodal data are subjected to feature extraction processing to obtain the environmental features, power plant operation features, and power grid demand features at the corresponding target prediction time. The predicted power generation capacity of the photovoltaic power station under conditions unconstrained by the power grid is determined based on the environmental characteristics and the operating characteristics of the photovoltaic power station. Based on the environmental characteristics and the power grid demand characteristics, the predicted power demand on the power grid side is determined, and the power grid absorption constraint parameters are determined based on the predicted power demand, grid-connected capacity, and historical power generation restriction status. The initial photovoltaic power generation prediction result is determined based on the photovoltaic power generation prediction result and the grid absorption constraint parameter. Based on the environmental change status, power plant operation status, and grid absorption status corresponding to the target prediction time, the stability of the corresponding characteristics on the initial photovoltaic power plant power prediction result is determined, and the initial photovoltaic power plant power prediction result is corrected according to the stability to obtain the final photovoltaic power plant power prediction result. Output the final predicted power of the photovoltaic power plant within the target prediction time, including the predicted value of the photovoltaic power generation capacity, the predicted value of the actual grid-connected power, and the predicted value of the confined power.

[0006] As a preferred technical solution for photovoltaic power plant power prediction methods based on multimodal data, the environmental factor data includes solar irradiance, ambient temperature, wind speed, and satellite cloud images; The photovoltaic power plant operation data includes historical output power, inverter operating status, and available power plant capacity; The grid-side demand data includes historical projected electricity demand, grid-connected capacity, and historical power rationing status.

[0007] As a preferred technical solution for photovoltaic power plant power prediction methods based on multimodal data, feature extraction processing is performed on the multimodal data, including: The multimodal data are sorted according to time identifiers to construct environmental numerical data sequences, satellite cloud image sequences, photovoltaic power plant operation data sequences, and grid-side demand data sequences. Environmental features are extracted from the environmental numerical data sequence and satellite cloud image sequence. The environmental features include solar irradiance variation sequence, ambient temperature variation sequence, cloud coverage ratio, cloud displacement direction, and cloud displacement amount. The operation characteristics of the photovoltaic power plant are extracted from the operation data sequence of the photovoltaic power plant, including the inverter status sequence and the change sequence of the available capacity of the power plant. The grid demand features are extracted from the grid-side demand data sequence, which include historical projected electricity demand sequence, grid-connected capacity, and historical power restriction status sequence.

[0008] As a preferred technical solution for a photovoltaic power plant power prediction method based on multimodal data, the step of determining the predicted generating power of the photovoltaic power plant under grid-free conditions based on the environmental characteristics and the operating characteristics of the photovoltaic power plant includes: Based on the cloud coverage ratio, cloud displacement direction, and cloud displacement amount, determine the predicted cloud coverage ratio corresponding to the predicted time when the cloud reaches the area where the photovoltaic power station is located and the target prediction time. The predicted solar irradiance corresponding to the target prediction time is determined based on the solar irradiance change sequence, and the predicted solar irradiance is corrected based on the predicted cloud cover ratio to obtain the effective irradiance at the target prediction time. The available status of the equipment is determined based on the inverter state sequence, and the available capacity coefficient corresponding to the target prediction time is determined based on the power station available capacity change sequence. The predicted power generation capacity of the photovoltaic power station under grid-free conditions is determined based on the effective irradiance and the available capacity factor.

[0009] As a preferred technical solution for photovoltaic power plant power prediction methods based on multimodal data, the step of determining the predicted electricity demand on the grid side according to the environmental characteristics and the grid demand characteristics includes: Determine the predicted ambient temperature corresponding to the target prediction time based on the ambient temperature change sequence; The predicted electricity demand at the target prediction time is determined based on the historical predicted electricity demand sequence and the predicted ambient temperature.

[0010] As a preferred technical solution for photovoltaic power plant power prediction methods based on multimodal data, the step of determining grid absorption constraint parameters based on the predicted electricity demand, grid-connected capacity, and historical power curtailment status includes: The basic absorption coefficient is determined based on the comparison between the predicted electricity demand and the grid-connected capacity; The risk level of issuance restriction at the target prediction time is determined based on the historical issuance restriction status sequence. The basic absorption coefficient is corrected according to the aforementioned power generation restriction risk level to obtain the power grid absorption constraint parameters.

[0011] As a preferred technical solution for a photovoltaic power plant power prediction method based on multimodal data, the step of determining the initial photovoltaic power plant power prediction result based on the photovoltaic power generation prediction result and the grid absorption constraint parameters includes: The predicted generating power is constrained by the grid absorption constraint parameters to obtain the initial photovoltaic power plant power prediction results; The grid absorption constraint parameter is the proportion of the predicted power that can be connected to the grid, and the initial photovoltaic power plant power prediction result is equal to the product of the predicted power and the grid absorption constraint parameter. Wherein, when the predicted power generation is less than or equal to the allowable grid-connected power, the predicted power generation is determined as the actual grid-connected power prediction value; Conversely, the predicted generating power is limited according to the grid absorption constraint parameters, and the limited power value is determined as the actual grid-connected power prediction value.

[0012] As a preferred technical solution for photovoltaic power plant power prediction methods based on multimodal data, the stability of the corresponding features with respect to the initial photovoltaic power plant power prediction results is determined, including: The stability of environmental characteristics is determined based on the solar irradiance variation sequence and cloud displacement. The stability of the power plant's operating characteristics is determined based on the inverter state sequence and the power plant's available capacity change sequence. The stability of the power grid demand characteristics is determined based on the predicted electricity demand and the historical power rationing sequence. Each stability level is used to characterize the reliability of the corresponding feature for the initial photovoltaic power plant power prediction results under the current prediction scenario.

[0013] As a preferred technical solution for photovoltaic power plant power prediction methods based on multimodal data, the initial photovoltaic power plant power prediction result is corrected according to the stability level to obtain the final photovoltaic power plant power prediction result, including: When the stability of any feature is lower than a preset threshold, the contribution ratio of that feature to the initial photovoltaic power plant power prediction result is reduced, and the remaining features are used for correction to obtain the final photovoltaic power plant power prediction result.

[0014] Compared with existing technologies, the advantages of this invention lie in its ability to construct a complete method for predicting both the potential power generation of photovoltaic (PV) power and the actual grid-connected power by integrating multimodal information such as meteorological data, PV power plant operation data, and grid-side demand data. By using the grid-side projected electricity demand as a constraint, the power prediction results simultaneously reflect both PV power generation capacity and grid acceptance capacity, directly addressing the actual grid-connected power that the grid can absorb. Simultaneously, through joint analysis of cloud cover ratio, cloud displacement direction, and displacement amount, satellite cloud image information is transformed into quantifiable irradiance correction criteria, enabling early response to sudden changes in irradiance caused by cloud movement. Furthermore, this invention performs segmented corrections based on a quantitative assessment of the stability of three types of characteristics: environment, equipment, and grid. When a characteristic is in a state of drastic fluctuation, the intermediate quantity corresponding to that characteristic converges towards a steady-state benchmark, effectively avoiding the misleading influence of a single characteristic anomaly on the final prediction result and improving the reliability of the prediction results in complex scenarios. While improving the accuracy of PV power prediction, this invention is particularly suitable for the actual needs of grid dispatch for accurate prediction of grid-connected power in the context of high-proportion PV grid integration. Attached Figure Description

[0015] Figure 1 This is a flowchart of a photovoltaic power plant power prediction method based on multimodal data, as described in an embodiment of the present invention. Figure 2 This is a logical schematic diagram of the photovoltaic power plant power prediction method based on multimodal data according to an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0017] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0018] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0019] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0020] Please see Figure 1 and Figure 2 As shown, this invention provides a photovoltaic power plant power prediction method based on multimodal data, comprising: Step S1: Obtain multimodal data of the area where the photovoltaic power station is located. The multimodal data includes environmental factor data, photovoltaic power station operation data, and grid-side demand data. Step S2: Perform feature extraction processing on the multimodal data to obtain the environmental features, power plant operation features, and power grid demand features at the corresponding target prediction time. Step S3: Determine the predicted power generation capacity of the photovoltaic power station under conditions where it is not constrained by the power grid, based on the environmental characteristics and the operating characteristics of the photovoltaic power station. Step S4: Determine the predicted electricity demand on the grid side based on the environmental characteristics and the grid demand characteristics, and determine the grid absorption constraint parameters based on the predicted electricity demand, grid-connected capacity and historical power restriction status. Step S5: Determine the initial photovoltaic power plant power prediction result based on the photovoltaic power generation prediction result and the grid absorption constraint parameters; Step S6: Based on the environmental change status, power plant operation status and grid absorption status corresponding to the target prediction time, determine the stability of the corresponding features on the initial photovoltaic power plant power prediction result, and correct the initial photovoltaic power plant power prediction result according to the stability to obtain the final photovoltaic power plant power prediction result. Step S7: Output the final photovoltaic power plant prediction power within the target prediction time, including the photovoltaic power generation prediction value, the actual grid-connected power prediction value, and the restricted power prediction value.

[0021] In practice, solar irradiance, ambient temperature, and wind speed are collected through meteorological monitoring stations deployed at the photovoltaic power station site. The sampling cycle is consistent with the power station's data acquisition system, typically on the order of minutes, for example, set to 1 minute. Satellite cloud images are obtained from meteorological satellite remote sensing data service systems, including but not limited to infrared and visible light cloud images from geostationary meteorological satellites (such as Himawari and FY-4). The spatial and temporal resolution of satellite cloud images varies depending on the data source, with spatial resolution ranging from 0.5 to 4 kilometers and an update cycle of 10 to 30 minutes.

[0022] Photovoltaic power plant operation data is acquired through the power plant's monitoring and data acquisition equipment. Historical output power is the actual measured power at the power plant's grid connection point, while inverter operating status and available power plant capacity are read from the inverter and power plant monitoring equipment.

[0023] Historical projected electricity demand refers to the load forecast data of the target area within the historical forecast period. It is obtained by acquiring historical load forecast results generated by the power grid dispatching department and represents the changes in regional electricity demand over different time periods. Grid-connected capacity is the maximum allowable transmission capacity of the grid node corresponding to the photovoltaic power station's grid connection. It can be obtained through the power station's grid connection agreement, substation operating parameters, or the power grid dispatching system, and represents the capacity limit of the photovoltaic power station's power transmission to the grid. Historical power curtailment status refers to the power curtailment status data of the photovoltaic power station during its historical operation, resulting from grid absorption capacity, dispatching restrictions, or grid connection constraints. It is obtained through the photovoltaic power station's historical operating records, dispatching instruction records, and power curve data, and represents the grid's restriction on photovoltaic power output during historical periods.

[0024] The release times and effective time ranges of the aforementioned multimodal data vary. Satellite cloud images are released several minutes to tens of minutes later than the acquisition time, numerical weather predictions are released much earlier than their prediction times, while power plant operation data is acquired in real time.

[0025] The target prediction time is a future point in time when power needs to be predicted. It is determined according to the power plant prediction application scenario and grid dispatch requirements. It can be set as a single future time point or a continuous future time series, such as 1 hour or 4 hours after the current time. For ultra-short-term power prediction scenarios, the target prediction time can be set at fixed time intervals to multiple nodes from several minutes to several hours after the current time. For day-ahead short-term power prediction scenarios, the target prediction time can cover the entire operating period of the next day at fixed intervals.

[0026] Since the release time, effective time window, and target prediction time of the above different data are not consistent, it is necessary to use the target prediction time as a benchmark to perform time alignment on each modality of data, and select data whose release time is earlier than the target prediction time and effectively covers that time to participate in feature extraction.

[0027] For environmental numerical data and satellite cloud images, the time stamp is usually the data acquisition time; for photovoltaic power plant operation data, the time stamp is the recording time of power, status, and capacity; for grid-side demand data, the time stamp of the expected electricity demand is the predicted time period, and the grid-connected capacity and historical power curtailment status are based on the current grid connection agreement and historical records. Since the time stamps of different data may not completely overlap on the absolute time axis, for each target prediction time, data within a fixed time window is extracted backward from that time as a baseline. The data falling within that time window are arranged in ascending order according to their respective time stamps to form various data sequences.

[0028] Environmental features were extracted from the sorted environmental numerical data sequence and satellite cloud image sequence. Among them, the solar irradiance variation sequence was obtained by calculating the difference between irradiance values ​​between adjacent time markers.

[0029] The ambient temperature variation sequence is constructed in the same way as the solar irradiance variation sequence, and is obtained by subtracting the ambient temperature values ​​of adjacent time points.

[0030] The cloud coverage ratio is determined by dividing the target image area corresponding to the photovoltaic power station in the satellite cloud image into grids. Specifically, the target image area is divided into multiple grid units, and the proportion of the number of grid units covered by clouds to the total number of grid units is counted. This proportion is the cloud coverage ratio. Generally, the granularity of grid division is 3×3 to 10×10.

[0031] The direction and amount of cloud displacement are obtained by block matching of two satellite cloud images at adjacent time points. A reference block containing cloud texture features is selected in the cloud image at the previous time point. The similarity (e.g., mean square error) between the reference block and the corresponding search area in the cloud image at the next time point is calculated. The displacement between the position with the highest similarity and the original position of the reference block is the amount of cloud displacement, and the direction of this displacement is the direction of cloud displacement.

[0032] The inverter state sequence is discretely encoded according to state type, mapping normal operation, derating operation, and shutdown to different numerical codes. If multiple abnormal states exist, they are encoded separately. The power plant available capacity change sequence is obtained by calculating the difference between the power plant available capacity between adjacent time markers.

[0033] Grid demand features are extracted from the sorted grid-side demand data sequence. Historical projected electricity demand sequences directly use load forecasts issued by the grid dispatching department. Grid-connected capacity is either a fixed value or a dynamic value that changes according to dispatching instructions, and is directly used as feature input. Historical power rationing status sequences are discretely encoded based on whether a power rationing event occurred at the corresponding time period; if a rationing event occurred, it is marked as the first encoded value; otherwise, it is marked as the second encoded value.

[0034] All of the above feature sequences end at the target prediction time. The sequence length is determined according to the time scale of the prediction task. For example, for short-term predictions of 1 to 4 hours, the sequence length can be set to 6 to 24 time steps.

[0035] Based on the extracted cloud coverage ratio, cloud displacement direction, and displacement amount, the cloud is assumed to maintain a constant displacement direction and displacement velocity within the current prediction time window. The cloud coverage ratio of each grid at the current moment is extrapolated to the target prediction time to obtain the predicted cloud coverage ratio at the target prediction time.

[0036] Based on the solar irradiance variation sequence, the predicted solar irradiance at the target prediction time is determined using a time-series extrapolation method. Then, the predicted solar irradiance is corrected based on the predicted cloud cover ratio. When the predicted cloud cover ratio is greater than a preset threshold, it is determined that the area where the photovoltaic power station is located is blocked by clouds, and the predicted solar irradiance at that time is reduced. The reduction coefficient is negatively correlated with the cloud cover ratio; the higher the cloud cover ratio, the smaller the reduction coefficient. The correspondence between the preset threshold, the reduction coefficient, and the cloud cover ratio is calibrated through regression analysis of the cloud cover ratio and measured irradiance in the historical operating data of the photovoltaic power station.

[0037] The available status of the equipment at the target prediction time is determined based on the inverter state sequence, and the available capacity coefficient at the target prediction time is determined based on the power plant available capacity change sequence. If the inverter state is shutdown at the target prediction time, the available capacity coefficient is zero; if it is operating normally, the available capacity coefficient is one; if it is operating at reduced capacity, the available capacity coefficient is the ratio of the current rated output power to the rated capacity.

[0038] The corrected effective irradiance is converted into theoretical power based on the conversion efficiency of the photovoltaic power plant, and then multiplied by the available capacity factor to obtain the predicted power generation under the condition of not being constrained by the grid.

[0039] It is understandable that there is a significant relationship between electricity demand and temperature. During the high-temperature period in summer, the concentrated release of cooling load leads to a surge in electricity demand; while during the low-temperature period in winter, heating load increases. Therefore, relying solely on historical electricity demand sequences for extrapolation is insufficient to accurately capture the impact of sudden temperature changes on load. By incorporating predicted ambient temperature into the historical electricity demand sequence for correction, the accuracy of electricity demand forecasting under conditions of drastic temperature changes can be improved.

[0040] When determining the predicted electricity demand, the predicted ambient temperature at the target prediction time is first extrapolated from the ambient temperature change sequence. Based on the historical predicted electricity demand sequence, an offset correction is made according to the change in the predicted ambient temperature relative to the corresponding historical temperature in the electricity demand sequence. When the predicted ambient temperature rises, the electricity demand will increase accordingly; conversely, it will decrease.

[0041] The specific correction method is determined in advance based on the relationship between temperature and electricity demand changes in the historical load data of the region. By performing linear regression on the historical load data and historical temperature data of the region, the sensitivity coefficient of electricity demand relative to temperature changes is obtained. This sensitivity coefficient is multiplied by the difference between the predicted temperature and the reference temperature as the correction amount for electricity demand.

[0042] Grid absorption constraints are used to quantify the grid's capacity to absorb photovoltaic (PV) power. It should be understood that when regional electricity demand cannot absorb the power generated by PV, the grid will issue a power curtailment signal to the PV power plant. With a fixed grid-connected capacity, the lower the electricity demand, the weaker the grid's absorption capacity. Grid-connected capacity represents the physical upper limit of the power that a PV power plant can transmit to the grid. When electricity demand is below this capacity, absorption capacity is limited; when electricity demand is above this capacity, grid absorption capacity is not constrained by this factor.

[0043] Therefore, the basic absorption coefficient is determined by comparing the predicted electricity demand with the grid-connected capacity. If the predicted electricity demand is greater than or equal to the grid-connected capacity, the basic absorption coefficient is at its maximum, indicating that the grid has sufficient absorption capacity. If the predicted electricity demand is less than the grid-connected capacity, the basic absorption coefficient is the ratio of the predicted electricity demand to the grid-connected capacity, which represents the proportion that the grid can absorb.

[0044] Based on this, historical power rationing conditions are introduced to correct the basic absorption coefficient. If power rationing events have occurred multiple times in the corresponding historical period at the target prediction time, it indicates that there are structural absorption constraints in that period. Even if the current predicted electricity demand is high, power rationing may still occur due to conservative grid dispatching strategies or transmission channel limitations.

[0045] Specifically, statistical analysis is performed on issuance restriction events in the historical issuance restriction status sequence that belong to the same seasonal and weekday type as the target prediction time. The frequency of issuance restriction events within this time period is calculated, i.e., the proportion of issuance restriction occurrences to the total sample size for that period. If this proportion is 0, the issuance restriction risk level is 0, and no downward adjustment is made to the basic absorption coefficient. If the proportion is greater than 0 and less than or equal to 10%, the issuance restriction risk level is 1, and the basic absorption coefficient is multiplied by a first correction coefficient, with a value ranging from 0.9 to 0.95. If the proportion is greater than 10% and less than or equal to 30%, the issuance restriction risk level is 2, and the basic absorption coefficient is multiplied by a second correction coefficient, with a value ranging from 0.8 to 0.9. If the proportion is greater than 30%, the issuance restriction risk level is 3, and the basic absorption coefficient is multiplied by a third correction coefficient, with a value ranging from 0.6 to 0.8. The specific values ​​of each correction coefficient are determined by regression analysis of the power plant's historical power generation restriction records and actual absorption coefficients, or by a limited number of tests.

[0046] After correction, the grid absorption constraint parameters are obtained, with values ​​ranging from 0.5 to 1. In actual power system operation, due to factors such as the minimum output requirements to ensure the safe and stable operation of the grid and the responsibility weight for renewable energy absorption, it is almost impossible for the grid to be completely unable to accept photovoltaic power (i.e., the absorption constraint parameters are close to 0). Setting the lower limit to 0.5 is closer to engineering reality, avoiding overly extreme power curtailment situations in the prediction results, while retaining sufficient parameter variation space to reflect different degrees of absorption constraints.

[0047] The actual grid-connectable power of a photovoltaic power plant is constrained by both its own power generation capacity and the grid's acceptance capacity. The grid absorption constraint parameter represents the proportion of the allowable grid-connected power to the generateable power. By multiplying and segmenting the data, the full absorption and restricted absorption scenarios are distinguished, which is in line with the operating logic of the grid's new energy output management.

[0048] In practice, this product is used as the upper limit of the allowed grid-connected power. If the predicted power output is less than or equal to this upper limit, it means that the power generation capacity of the power plant itself is already within the grid's acceptance boundary, and no additional reduction is needed; the power output is directly used as the actual grid-connected power. If the predicted power output is greater than this upper limit, it means that the grid cannot fully absorb the photovoltaic power generated, and power generation is limited according to the upper limit of the allowed grid-connected power; the excess portion is the restricted power.

[0049] The lower limit of the grid absorption constraint parameter is set to 0.5 instead of 0, based on the actual operating characteristics of the power system. Even in extreme power curtailment scenarios, due to factors such as the minimum output requirement to ensure the safe and stable operation of the grid and the responsibility weight for renewable energy absorption, the grid still has a minimum absorption guarantee level for grid-connected photovoltaic power plants, and there will be no situation where they cannot be absorbed at all. Setting the lower limit to 0.5 avoids extreme cases such as zero output or extremely low output in the prediction results, which are inconsistent with the actual engineering situation, and makes the prediction results closer to the actual dispatch and operation boundary.

[0050] It is understandable that the reliability of the three types of information—environment, power plant operation, and power grid—exhibits objective differences under different prediction scenarios. When a certain type of factor is in a state of drastic fluctuation, the reliability of the initial prediction results derived from that type of feature decreases. At this time, it is necessary to reduce the influence weight of the corresponding information and rely more on stable and reliable information sources to carry out predictions, so that the final results are more in line with the dominant constraints of the current scenario.

[0051] The initial photovoltaic power plant power prediction results are derived by integrating environmental characteristics, power plant operation characteristics, and grid demand characteristics. However, in actual operation, the reliability of different characteristics varies significantly under different scenarios. For example, when a cloud cluster passes rapidly, the predicted power output extrapolated from the current irradiance may be invalid, requiring a reduction in the contribution of environmental characteristics. When the risk of grid curtailment is high, the changing trend of grid demand characteristics becomes more critical for predicting the actual grid-connected power. Therefore, it is necessary to perform a correction based on the reliability of the characteristics before outputting the prediction results.

[0052] The stability of each feature is determined by the degree of change of its corresponding change sequence within the current prediction window.

[0053] The stability of environmental characteristics is determined based on the solar irradiance change sequence and cloud displacement. If the difference between adjacent time steps in the irradiance change sequence fluctuates little and the cloud displacement is lower than the preset threshold, it indicates that the environmental conditions are stable at the current moment and the stability of environmental characteristics is high. The predicted power generation obtained in this way has high reliability. Conversely, it indicates that the environment is in a period of drastic change and the stability is relatively low.

[0054] The stability of the power plant's operating characteristics is determined based on the inverter state sequence and the power plant's available capacity change sequence. If the inverter state does not change within the time window and the difference between each time step in the available capacity change sequence is small, it indicates that the equipment is operating smoothly and the power plant's operating characteristics are highly stable. Conversely, if the inverter state changes or the available capacity fluctuates significantly, the stability is low.

[0055] The stability of the power grid demand characteristics is determined based on the predicted trend of electricity demand and the historical sequence of power rationing events: if the predicted electricity demand shows a monotonic change within the time window and the frequency of power rationing events in the historical sequence of power rationing events is lower than a preset threshold, it indicates that the power grid absorption status is stable and the stability of the power grid demand characteristics is high; conversely, if the electricity demand fluctuates drastically or the frequency of power rationing is high, the stability is low.

[0056] In practice, each level of stability is quantified by a stability coefficient, which ranges from 0 to 1. A higher value indicates a more reliable feature. The stability coefficient is calculated online based on the change sequences of each feature acquired before the current target prediction time.

[0057] The environmental characteristic stability coefficient is determined based on the standard deviation of the irradiance variation sequence and cloud displacement: the standard deviation is close to 1 when it is less than the calibrated threshold and close to 0 when it is greater than the threshold, using linear interpolation. The irradiance threshold is taken as 25% and 75% of the historical window standard deviation, which are the boundaries between stable and severe conditions, respectively; the cloud displacement threshold is taken as the average historical displacement plus one standard deviation.

[0058] The power plant's operational stability coefficient is determined based on the number of inverter state transitions and the magnitude of available capacity changes: it approaches 1 when there are fewer than 3 state transitions per 24 hours and the capacity fluctuation is less than 5% of the rated capacity; otherwise, it approaches 0, using linear interpolation. The state transition threshold of 3 times per 24 hours is derived from the inverter equipment specification, and the capacity fluctuation threshold is taken as 3 times the standard deviation of the available capacity change during normal operation, approximately 5% of the rated capacity. The stability coefficient of power grid demand characteristics is determined based on the absolute value of the first-order difference in electricity demand and the proportion of power rationing events: it is close to 1 when the absolute value of the difference is less than the median and the proportion of power rationing is less than 10%, otherwise it is close to 0, using linear interpolation. The electricity demand threshold is taken as the median of the historical absolute values ​​of the first-order difference; the power rationing proportion threshold is taken as the 70th percentile of the historical window proportion, approximately 10%.

[0059] The stability coefficient preset threshold is determined by grid search with a range of 0.3 to 0.9 and a step size of 0.05, selecting the one with the smallest validation set error. The above thresholds are updated quarterly.

[0060] It is understandable that environmental characteristics and power plant operating characteristics jointly determine the generating capacity, and environmental characteristics and grid demand characteristics jointly determine the absorption constraint parameters. The generating capacity and absorption constraint parameters are multiplied to obtain the grid-connected power. Therefore, in this embodiment, reducing the contribution ratio means reducing the degree of influence of the calculation link corresponding to this feature on the final prediction result. Specifically, this is achieved through the following sub-link convergence correction method.

[0061] When the stability of environmental characteristics falls below a preset threshold, the reliability of the power generation calculation based on meteorological data is deemed insufficient. In this case, the power generation calculation is no longer based solely on the extrapolation of current irradiance. Instead, a convergence correction is applied to the fluctuation range of the initial predicted power generation. The power output benchmark during a period of relatively stable environmental conditions in the recent past is used as a reference. For example, if the irradiance change rate was lower than the set threshold in the previous 24 hours, the steady-state average output during that period is used as the correction benchmark. This constrains the current predicted power generation to converge towards this benchmark, with the correction magnitude determined by the degree to which the environmental characteristic stability coefficient deviates from the threshold. This avoids drastic jumps in the predicted power generation due to sudden changes in irradiance or rapid cloud movement.

[0062] When the stability of the power plant's operating characteristics falls below a preset threshold, the reliability of the output conversion process corresponding to the equipment's operating condition is deemed insufficient. In this case, a convergence correction is performed on the capacity conversion portion of the initial predicted power output. Using the change rate of available capacity under the recent normal operating conditions of the equipment as a reference, the available capacity coefficient at the current prediction time is brought closer to this reference, thus limiting the interference of equipment operating condition jumps on the predicted power output to a reasonable range.

[0063] When the stability of grid demand characteristics falls below a preset threshold, the reliability of the calculation process for the grid absorption constraint parameters is deemed insufficient. In this case, the initial grid absorption constraint parameters are converged and corrected. Using the steady-state absorption level of the power plant in the same historical period (e.g., the same working day type and time period within the past 30 days) as a reference benchmark, the currently predicted absorption constraint parameters are aligned with this benchmark, thus constraining their adjustment range and preventing significant deviations in grid-connected power prediction due to fluctuations in electricity demand forecasts or abnormal determinations of power generation limitation status.

[0064] In each of the above correction steps, the value at the convergence point can be achieved through conventional weighted interpolation according to the principle. The correction magnitude is related to the stability coefficient of the corresponding feature. The lower the stability coefficient, the greater the correction magnitude.

[0065] When the stability of two or more features is simultaneously below a preset threshold, the multiple factors are deemed unreliable, no correction is made, and the initial prediction result is used directly as the final output.

[0066] After corrections at the corresponding stages mentioned above, the predicted power generation capacity, the predicted actual grid-connected power capacity, and the predicted restricted power capacity are recalculated to form the final power prediction result for the photovoltaic power plant.

[0067] This implementation predicts the theoretical upper limit of grid-connected power of photovoltaic power plants under given grid-connected capacity and electricity demand boundaries, rather than simulating grid dispatch strategies. Energy storage charging and discharging, and adjustable load response belong to the grid dispatch execution layer, and their ultimate impact on grid-connected power is already reflected in the statistical characteristics of historical power curtailment states. If energy storage absorption or transmission channels are restricted during a certain period, the corresponding period in historical data will be represented as a power curtailment event. This information is input into the model through the historical power curtailment state sequence. Therefore, this scheme does not separately model the intermediate links of energy storage and dispatch, but directly describes the boundary conditions using grid absorption constraint parameters.

[0068] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A photovoltaic power plant power prediction method based on multimodal data, characterized in that, include: Acquire multimodal data of the area where the photovoltaic power station is located, including environmental factor data, photovoltaic power station operation data, and grid-side demand data; The multimodal data are subjected to feature extraction processing to obtain the environmental features, power plant operation features, and power grid demand features at the corresponding target prediction time. The predicted power generation capacity of the photovoltaic power station under conditions unconstrained by the power grid is determined based on the environmental characteristics and the operating characteristics of the photovoltaic power station. Based on the environmental characteristics and the power grid demand characteristics, the predicted power demand on the power grid side is determined, and the power grid absorption constraint parameters are determined based on the predicted power demand, grid-connected capacity, and historical power generation restriction status. The initial photovoltaic power generation prediction result is determined based on the photovoltaic power generation prediction result and the grid absorption constraint parameter. Based on the environmental change status, power plant operation status, and grid absorption status corresponding to the target prediction time, the stability of the corresponding characteristics on the initial photovoltaic power plant power prediction result is determined, and the initial photovoltaic power plant power prediction result is corrected according to the stability to obtain the final photovoltaic power plant power prediction result. Output the final predicted power of the photovoltaic power plant within the target prediction time, including the predicted value of the photovoltaic power generation capacity, the predicted value of the actual grid-connected power, and the predicted value of the confined power.

2. The photovoltaic power plant power prediction method based on multimodal data according to claim 1, characterized in that, The environmental factor data includes solar irradiance, ambient temperature, wind speed, and satellite cloud imagery; The photovoltaic power plant operation data includes historical output power, inverter operating status, and available power plant capacity; The grid-side demand data includes historical projected electricity demand, grid-connected capacity, and historical power rationing status.

3. The photovoltaic power plant power prediction method based on multimodal data according to claim 2, characterized in that, Feature extraction processing is performed on the multimodal data, including: The multimodal data are sorted according to time identifiers to construct environmental numerical data sequences, satellite cloud image sequences, photovoltaic power plant operation data sequences, and grid-side demand data sequences. Environmental features are extracted from the environmental numerical data sequence and satellite cloud image sequence. The environmental features include solar irradiance variation sequence, ambient temperature variation sequence, cloud coverage ratio, cloud displacement direction, and cloud displacement amount. The operation characteristics of the photovoltaic power plant are extracted from the operation data sequence of the photovoltaic power plant, including the inverter status sequence and the change sequence of the available capacity of the power plant. The grid demand features are extracted from the grid-side demand data sequence, which include historical projected electricity demand sequence, grid-connected capacity, and historical power restriction status sequence.

4. The photovoltaic power plant power prediction method based on multimodal data according to claim 3, characterized in that, Determining the predicted power generation capacity of the photovoltaic power station under grid-free conditions based on the environmental characteristics and the operating characteristics of the photovoltaic power station includes: Based on the cloud coverage ratio, cloud displacement direction, and cloud displacement amount, determine the predicted cloud coverage ratio corresponding to the predicted time when the cloud reaches the area where the photovoltaic power station is located and the target prediction time. The predicted solar irradiance corresponding to the target prediction time is determined based on the solar irradiance change sequence, and the predicted solar irradiance is corrected based on the predicted cloud cover ratio to obtain the effective irradiance at the target prediction time. The available status of the equipment is determined based on the inverter state sequence, and the available capacity coefficient corresponding to the target prediction time is determined based on the power station available capacity change sequence. The predicted power generation capacity of the photovoltaic power station under grid-free conditions is determined based on the effective irradiance and the available capacity factor.

5. The photovoltaic power plant power prediction method based on multimodal data according to claim 3, characterized in that, Determining the predicted electricity demand on the grid side based on the environmental characteristics and the grid demand characteristics includes: Determine the predicted ambient temperature corresponding to the target prediction time based on the ambient temperature change sequence; The predicted electricity demand at the target prediction time is determined based on the historical predicted electricity demand sequence and the predicted ambient temperature.

6. The photovoltaic power plant power prediction method based on multimodal data according to claim 5, characterized in that, The step of determining the grid absorption constraint parameters based on the predicted electricity demand, grid-connected capacity, and historical power restriction status includes: The basic absorption coefficient is determined based on the comparison between the predicted electricity demand and the grid-connected capacity; The risk level of issuance restriction at the target prediction time is determined based on the historical issuance restriction status sequence. The basic absorption coefficient is corrected according to the aforementioned power generation restriction risk level to obtain the power grid absorption constraint parameters.

7. The photovoltaic power plant power prediction method based on multimodal data according to claim 6, characterized in that, The step of determining the initial photovoltaic power plant power prediction result based on the photovoltaic power generation prediction result and the grid absorption constraint parameters includes: The predicted generating power is constrained by the grid absorption constraint parameters to obtain the initial photovoltaic power plant power prediction results; The grid absorption constraint parameter is the proportion of the predicted power that can be connected to the grid, and the initial photovoltaic power plant power prediction result is equal to the product of the predicted power and the grid absorption constraint parameter. Wherein, when the predicted power generation is less than or equal to the allowable grid-connected power, the predicted power generation is determined as the actual grid-connected power prediction value; Conversely, the predicted generating power is limited according to the grid absorption constraint parameters, and the limited power value is determined as the actual grid-connected power prediction value.

8. The photovoltaic power plant power prediction method based on multimodal data according to claim 7, characterized in that, Determine the stability of the corresponding features with respect to the initial photovoltaic power plant power prediction results, including: The stability of environmental characteristics is determined based on the solar irradiance variation sequence and cloud displacement. The stability of the power plant's operating characteristics is determined based on the inverter state sequence and the power plant's available capacity change sequence. The stability of the power grid demand characteristics is determined based on the predicted electricity demand and the historical power rationing sequence. Each stability level is used to characterize the reliability of the corresponding feature for the initial photovoltaic power plant power prediction results under the current prediction scenario.

9. The photovoltaic power plant power prediction method based on multimodal data according to claim 8, characterized in that, The initial photovoltaic power plant power prediction result is corrected based on the stability level to obtain the final photovoltaic power plant power prediction result, including: When the stability of any feature is lower than a preset threshold, the contribution ratio of that feature to the initial photovoltaic power plant power prediction result is reduced, and the remaining features are used for correction to obtain the final photovoltaic power plant power prediction result.

Images