Regional distributed photovoltaic power prediction method and system

Abstract

The invention relates to the technical field of photovoltaic power prediction, and discloses a regional distributed photovoltaic power prediction method and system, and the photovoltaic power prediction comprises the steps: S1, obtaining meteorological data in real time, introducing a time sequence processing capability model, and capturing the nonlinear influence of the weather change on the photovoltaic power; s2, introducing an online learning and self-adaptive mechanism, enabling the prediction model to have an online learning capability, and dynamically adjusting model parameters along with the inflow of new real-time meteorological data; s3, implementing rolling prediction based on a short time window, and dynamically adjusting the power generation plan: when a sudden weather change is detected, triggering a quick response program, and adjusting the power generation plan according to the latest prediction result; and S4, integrating model prediction results, and adopting a probability prediction method to provide probability distribution of photovoltaic power output instead of a deterministic numerical value, so as to reflect an uncertainty range of future photovoltaic power change.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power prediction, and in particular to a regional distributed photovoltaic power prediction method and system. Background Art

[0002] Photovoltaic power generation is significantly affected by weather conditions (such as light intensity, temperature, cloud cover, etc.) and has strong volatility and uncertainty, which poses challenges to the stable operation and efficient dispatch of the power system.

[0003] In order to ensure the safe and stable operation of the power grid, it is necessary to accurately predict the power generation of photovoltaic power stations in order to reasonably arrange power production and consumption, balance supply and demand, and prevent abnormal grid frequency and voltage caused by sudden fluctuations in photovoltaic power generation.

[0004] After searching, the invention patent with Chinese patent number CN110414748A discloses a regional distributed photovoltaic power prediction method and system. Compared with the prior art, the invention patent with Chinese patent number CN110414748A first determines the historical meteorological data and historical output power of the photovoltaic power station, and pre-processes the determined historical meteorological data and historical output power; then builds a neural network model based on the pre-processed historical meteorological data and historical output power, and trains the neural network model; and after judging that the network parameters in the neural network model fall into the local optimum, the network parameters are heuristically searched and adaptively adjusted, so that the neural network model can jump out of the local optimum and continue to reduce the training error; at this time, based on the model obtained by the above training, the output power prediction information of the next prediction period can be obtained according to the meteorological data prediction information of the next prediction period, which improves the prediction accuracy compared with the prior art;

[0005] However, while improving the prediction accuracy by improving the model, there will be sudden weather changes in the data acquisition process. In particular, the rapid movement of clouds will significantly change the intensity of solar radiation reaching the photovoltaic panels. In particular, the rapidly changing clouds will cause drastic fluctuations in the instantaneous sunlight intensity, which will cause the model prediction to have a large prediction deviation, and directly affect the power generation of the photovoltaic system. Therefore, a regional distributed photovoltaic power prediction method and system are proposed herein. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcoming in the prior art that the prediction accuracy is easily affected by the weather, and to propose a regional distributed photovoltaic power prediction method and system.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A regional distributed photovoltaic power prediction method, the photovoltaic power prediction steps are:

[0009] S1: Acquire meteorological data in real time and introduce models with time series processing capabilities to capture the nonlinear impact of weather changes on photovoltaic power generation;

[0010] S2: Introducing online learning and adaptive mechanisms to enable the prediction model to have online learning capabilities and dynamically adjust model parameters as new real-time meteorological data flows in;

[0011] S3: Implement rolling forecasts based on shorter time windows and dynamically adjust power generation plans: When sudden weather changes are detected, a rapid response program is triggered to adjust the power generation plan based on the latest forecast results;

[0012] S4: Integrate the model prediction results and use a probabilistic prediction method to provide a probability distribution of PV power output rather than a deterministic value, reflecting the uncertainty range of future PV power changes.

[0013] In S1, we first access the satellite remote sensing data service, track the cloud dynamics in real time through satellite images, and obtain the impact of cloud cover on the photovoltaic array in real time. Based on real-time meteorological data, we extract features closely related to photovoltaic power, including the sliding window statistics of the main solar radiation intensity and the periodic characteristics of temperature. We organize the continuous meteorological data into a time series format, and sample it every minute and every hour at fixed time intervals to form a continuous time series data set. In the feature engineering stage, we introduce nonlinear transformation to express the nonlinear relationship between meteorological parameters and photovoltaic power, based on the formula:

[0014] f(x,y)=x 2 *log(y)e

[0015] x and y are independent variables, representing different physical quantities and data points in the data on the impact of cloud cover on the photovoltaic array;

[0016] x 2 Represents the square of variable x, a nonlinear transformation of variable x;

[0017] log(y) represents the logarithmic function with the natural base e as the base, reflecting the relative growth rate of the variable y relative to the exponential growth;

[0018] The two meteorological parameters are combined to capture the nonlinear effects of the interaction.

[0019] In S1, the nonlinear relationship model long short-term memory network is selected to construct a time series model. According to the characteristics of photovoltaic power output, different meteorological parameters are used as the input of the model, and photovoltaic power is used as the output. The model is trained using historical meteorological data and photovoltaic power data of the corresponding time period. Once new meteorological data comes in, the model will make real-time predictions of photovoltaic power based on the latest meteorological conditions. Based on the nonlinear activation function inside the model, the nonlinear relationship between meteorological factors and photovoltaic power output is captured, and the nonlinear interactions between different meteorological factors are learned, including the joint effects of temperature and humidity on photovoltaic efficiency.

[0020] In S2, an online learning framework is designed to allow the model to incrementally update model parameters when receiving new data. Each time a new batch of data is received, the gradient is calculated. For the loss function L(θ; x_t, y_t), where θ represents the model parameters, x_t, y_t represent the features and target values ​​of the tth sample, the model parameters are updated according to the gradient, and the update amplitude is adjusted using the learning rate α:

[0021] θ←θ-α*▽_θL(θ;x_t,y_t)

[0022] α is the learning rate, which controls the amplitude of each parameter update. Then, an adaptive learning strategy is used to dynamically adjust the learning rate. Combined with model performance monitoring, when the model prediction error exceeds the threshold, the learning rate is temporarily increased to adapt to the change.

[0023] In S3, a time window of 15 minutes, 30 minutes or 1 hour is set to make rolling forecasts for the upcoming time units. Based on the model of time series processing capability, the most recent historical meteorological data and photovoltaic power data are input, and the photovoltaic power forecast value in the future time period is output. Thresholds and rules are set to monitor the difference between the current meteorological data and the forecast results. When a sudden weather change is detected, the pre-set rapid response program is triggered to immediately execute the next stage of tasks and adjust the power generation plan based on the latest forecast results. The power dispatching center adjusts the power generation plan to adjust the load of thermal power plants and hydropower stations, dispatches energy storage equipment to release or absorb electricity, or adjusts the power transmission direction and power in the power grid.

[0024] In S4, multiple prediction models are trained simultaneously, the same meteorological parameters are input into each model to predict the PV power output, and the prediction results of all models are integrated and weighted average is adopted:

[0025] P_pred=w1*P_model1+w2*P_model2+...+wn*P_modeln

[0026] P_pred is the integrated prediction result;

[0027] P_modeli is the predicted value of the i-th model;

[0028] w1,w2,...,wn are the weights of their respective models, and satisfy w1+w2+...+wn=1;

[0029] The probability distribution of photovoltaic power output is estimated by Bayesian method, and the possibility distribution of future power output is simulated based on the posterior distribution of model parameters. For each model parameter θ, p(θ|D) is the posterior probability distribution, and D represents the historical data set:

[0030] p(P_out|D)=∫p(P_out|θ)*p(θ|D)dθp(P_out|θ) is the probability density function of the photovoltaic power output when the parameter θ is given, and p(θ|D) is the posterior probability distribution of the parameter θ.

[0031] In S4, a random variable P is set to be affected by multiple factors such as light intensity, temperature, and component efficiency. out Based on the time series prediction model, the upper and lower bounds of power at different probability levels, 90% confidence intervals, and 95% confidence intervals are calculated to reflect the possible range of power generation of photovoltaic power stations in the future under the confidence level. The rolling prediction strategy based on the S3 short time window continuously updates the probability distribution prediction according to the newly flowing real-time data. When a sudden weather change or a change in the predicted distribution is detected, the rapid response program is triggered to adjust the power generation plan according to the latest probability prediction results. The energy dispatching center will dispatch other energy sources or adjust the load in advance to ensure power supply when the upper limit of the predicted power generation is low based on the probability prediction results and the upper and lower bounds of power. When the lower limit of the predicted power generation is high, the reserve capacity is appropriately reduced or the consumption ratio of photovoltaic power is increased.

[0032] A system used in a regional distributed photovoltaic power prediction method includes a meteorological data acquisition module, and the photovoltaic power prediction module includes:

[0033] Learning and Adaptation Module: used to develop online update algorithms for model parameters, so that the model can be dynamically adjusted based on real-time feedback;

[0034] Rolling forecast engine module: used to periodically perform short-term rolling forecasts and generate new forecast results at set time intervals;

[0035] Forecast result analysis module: interpret rolling forecast results, compare historical data with current forecasts, and identify sudden weather changes;

[0036] Risk assessment module: Analyze the uncertainty of prediction results, quantify prediction risks, and provide a reference for decision-making.

[0037] The learning and adaptation module includes a model building unit. The model building unit extracts features closely related to photovoltaic power based on real-time meteorological data to build a time series prediction model. The prediction result analysis module includes a power generation plan adjustment unit. The power generation plan adjustment unit formulates a power dispatching strategy that adapts to uncertainty based on the probabilistic prediction results and coordinates operations with other energy systems.

[0038] The present invention has the following beneficial effects:

[0039] 1. In the present invention, by implementing rolling forecasts based on a shorter time window, refined power generation plan adjustments can be made. When a sudden weather change is detected, the response program can be quickly triggered to adjust the power generation plan in a timely manner based on the latest forecast results to ensure stable operation and efficient dispatch of the power system.

[0040] 2. In the present invention, the charging and discharging strategies of the energy storage system are optimized by utilizing the uncertainty information of probability prediction, thereby reducing energy waste and helping the energy dispatching center to better cope with uncertain factors such as sudden weather changes, thereby ensuring the stable and efficient operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A method step diagram of a regional distributed photovoltaic power prediction method proposed by the present invention;

[0042] Figure 2 This is a system block diagram of a regional distributed photovoltaic power prediction system proposed by the present invention.

[0043] In the figure: S1: Real-time acquisition of meteorological data, introduction of a model with time series processing capabilities, and capturing the nonlinear impact of weather changes on photovoltaic power generation;

[0044] S2: Introducing online learning and adaptive mechanisms to enable the prediction model to have online learning capabilities and dynamically adjust model parameters as new real-time meteorological data flows in;

[0045] S3: Implement rolling forecasts based on shorter time windows and dynamically adjust power generation plans: When sudden weather changes are detected, a rapid response program is triggered to adjust the power generation plan based on the latest forecast results;

[0046] S4: Integrate the model prediction results and use a probabilistic prediction method to provide a probability distribution of PV power output rather than a deterministic value, reflecting the uncertainty range of future PV power changes. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Embodiment 1

[0049] like Figure 1-2 As shown, the present invention proposes a regional distributed photovoltaic power prediction method, the photovoltaic power prediction steps are:

[0050] S1: Acquire meteorological data in real time and introduce models with time series processing capabilities to capture the nonlinear impact of weather changes on photovoltaic power generation;

[0051] S2: Introducing online learning and adaptive mechanisms to enable the prediction model to have online learning capabilities and dynamically adjust model parameters as new real-time meteorological data flows in;

[0052] S3: Implement rolling forecasts based on shorter time windows and dynamically adjust power generation plans: When sudden weather changes are detected, a rapid response program is triggered to adjust the power generation plan based on the latest forecast results;

[0053] S4: Integrate the model prediction results and use a probabilistic prediction method to provide a probability distribution of PV power output rather than a deterministic value, reflecting the uncertainty range of future PV power changes.

[0054] In S1, we first access the satellite remote sensing data service, track the cloud dynamics in real time through satellite images, and obtain the impact of cloud cover on the photovoltaic array in real time. Based on real-time meteorological data, we extract features closely related to photovoltaic power, including the sliding window statistics of the main solar radiation intensity and the periodic characteristics of temperature. We organize the continuous meteorological data into a time series format, and sample it every minute and every hour at fixed time intervals to form a continuous time series data set. In the feature engineering stage, we introduce nonlinear transformation to express the nonlinear relationship between meteorological parameters and photovoltaic power, based on the formula:

[0055] f(x,y)=x 2 *log(y)e

[0056] x and y are independent variables, representing different physical quantities and data points in the data on the impact of cloud cover on the photovoltaic array;

[0057] x 2 Represents the square of variable x, a nonlinear transformation of variable x;

[0058] log(y) represents the logarithmic function with the natural base e as the base, reflecting the relative growth rate of the variable y relative to the exponential growth;

[0059] The two meteorological parameters are combined to capture the nonlinear effects of the interaction.

[0060] In S1, the nonlinear relationship model long short-term memory network is selected to construct a time series model. According to the characteristics of photovoltaic power output, different meteorological parameters are used as the input of the model, and photovoltaic power is used as the output. The model is trained using historical meteorological data and photovoltaic power data of the corresponding time period. Once new meteorological data comes in, the model will make real-time predictions of photovoltaic power based on the latest meteorological conditions. Based on the nonlinear activation function inside the model, the nonlinear relationship between meteorological factors and photovoltaic power output is captured, and the nonlinear interactions between different meteorological factors are learned, including the joint effects of temperature and humidity on photovoltaic efficiency.

[0061] In S2, an online learning framework is designed to allow the model to incrementally update model parameters when receiving new data. Each time a new batch of data is received, the gradient is calculated. For the loss function L(θ; x_t, y_t), where θ represents the model parameters, x_t, y_t represent the features and target values ​​of the tth sample, the model parameters are updated according to the gradient, and the update amplitude is adjusted using the learning rate α:

[0062]

[0063] α is the learning rate, which controls the amplitude of each parameter update. Then, an adaptive learning strategy is used to dynamically adjust the learning rate. Combined with model performance monitoring, when the model prediction error exceeds the threshold, the learning rate is temporarily increased to adapt to the change.

[0064] In S3, a time window of 15 minutes, 30 minutes or 1 hour is set to make rolling forecasts for the upcoming time units. Based on the model of time series processing capability, the most recent historical meteorological data and photovoltaic power data are input, and the photovoltaic power forecast value in the future time period is output. Thresholds and rules are set to monitor the difference between the current meteorological data and the forecast results. When a sudden weather change is detected, the pre-set rapid response program is triggered to immediately execute the next stage of tasks and adjust the power generation plan based on the latest forecast results. The power dispatching center adjusts the power generation plan to adjust the load of thermal power plants and hydropower stations, dispatches energy storage equipment to release or absorb electricity, or adjusts the power transmission direction and power in the power grid.

[0065] In S4, multiple prediction models are trained simultaneously, the same meteorological parameters are input into each model to predict the PV power output, and the prediction results of all models are integrated and weighted average is adopted:

[0066] P_pred=w1*P_model1+w2*P_model2+...+wn*P_modeln

[0067] P_pred is the integrated prediction result;

[0068] P_modeli is the predicted value of the i-th model;

[0069] w1,w2,...,wn are the weights of their respective models, and satisfy w1+w2+...+wn=1;

[0070] The probability distribution of photovoltaic power output is estimated by Bayesian method, and the possibility distribution of future power output is simulated based on the posterior distribution of model parameters. For each model parameter θ, p(θ|D) is the posterior probability distribution, and D represents the historical data set:

[0071] p(P_out|D)=∫p(P_out|θ)*p(θ|D)dθp(P_out|θ) is the probability density function of the photovoltaic power output when the parameter θ is given, and p(θ|D) is the posterior probability distribution of the parameter θ.

[0072] In S4, a random variable P is set to be affected by multiple factors such as light intensity, temperature, and component efficiency. out Based on the time series prediction model, the upper and lower bounds of power at different probability levels, 90% confidence intervals, and 95% confidence intervals are calculated to reflect the possible range of power generation of photovoltaic power stations in the future under the confidence level. The rolling prediction strategy based on the S3 short time window continuously updates the probability distribution prediction according to the newly flowing real-time data. When a sudden weather change or a change in the predicted distribution is detected, the rapid response program is triggered to adjust the power generation plan according to the latest probability prediction results. The energy dispatching center will dispatch other energy sources or adjust the load in advance to ensure power supply when the upper limit of the predicted power generation is low based on the probability prediction results and the upper and lower bounds of power. When the lower limit of the predicted power generation is high, the reserve capacity is appropriately reduced or the consumption ratio of photovoltaic power is increased.

[0073] In this embodiment, when a sudden weather change is detected, the rapid response program is triggered. The current prediction model finds that due to the sudden strong storm, the sunlight at the location of the photovoltaic power station will be greatly weakened in the next two hours, resulting in the predicted photovoltaic output being significantly lower than the original plan. The system's built-in weather change detection algorithm captures this change and triggers a pre-set "rapid response program" to start the rapid response program. After the rapid response program is triggered, the original power generation plan is immediately stopped and a new power generation plan is calculated based on the latest prediction results.

[0074] For example: It was originally estimated that photovoltaic power generation could meet the 10MW electricity demand in the region within these two hours, but now it is predicted that it can only provide 5MW of electricity. At the same time, a new power generation plan is implemented to notify nearby fossil fuel power plants to increase output power by 5MW to make up for the reduction in photovoltaic output; or to dispatch existing energy storage equipment to release stored electricity during this period; or to request the grid dispatch center to allocate power resources from other regions. At the same time, the inverters and other equipment of the photovoltaic power station will also be adjusted according to the new power generation plan to maximize the use of currently available solar energy resources.

[0075] Embodiment 2

[0076] like Figure 1-2 As shown, based on the first embodiment, a system used in a regional distributed photovoltaic power prediction method includes a meteorological data acquisition module, and the photovoltaic power prediction module includes:

[0077] Learning and Adaptation Module: used to develop online update algorithms for model parameters, so that the model can be dynamically adjusted based on real-time feedback;

[0078] Rolling forecast engine module: used to periodically perform short-term rolling forecasts and generate new forecast results at set time intervals;

[0079] Forecast result analysis module: interpret rolling forecast results, compare historical data with current forecasts, and identify sudden weather changes;

[0080] Risk assessment module: Analyze the uncertainty of prediction results, quantify prediction risks, and provide a reference for decision-making.

[0081] The learning and adaptation module includes a model building unit. The model building unit extracts features closely related to photovoltaic power based on real-time meteorological data to build a time series prediction model. The prediction result analysis module includes a power generation plan adjustment unit. The power generation plan adjustment unit formulates a power dispatching strategy that adapts to uncertainty based on the probabilistic prediction results and coordinates operations with other energy systems.

[0082] In this embodiment, in the process of constructing the time series model, not only conventional meteorological data are integrated, but also multiple environmental factors such as satellite remote sensing images, terrain data, and air quality are integrated. Spatiotemporal interpolation and data fusion techniques are used to improve data quality and coverage. At the same time, data enhancement technology is used to simulate the data distribution under extreme climate conditions and improve the generalization ability of the model. Not only conventional meteorological data are integrated, but also multiple environmental factors such as satellite remote sensing images, terrain data, and air quality are integrated. Spatiotemporal interpolation and data fusion techniques are used to improve data quality and coverage. Finally, data enhancement technology is used to simulate the data distribution under extreme climate conditions and improve the generalization ability of the model.

[0083] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A regional distributed photovoltaic power prediction method, characterized in that: The photovoltaic power prediction steps are: S1: Acquire meteorological data in real time and introduce models with time series processing capabilities to capture the nonlinear impact of weather changes on photovoltaic power generation; S2: Introducing online learning and adaptive mechanisms to enable the prediction model to have online learning capabilities and dynamically adjust model parameters as new real-time meteorological data flows in; S3: Implement rolling forecasts based on shorter time windows and dynamically adjust power generation plans: When sudden weather changes are detected, a rapid response program is triggered to adjust the power generation plan based on the latest forecast results; S4: Integrate the model prediction results and use a probabilistic prediction method to provide a probability distribution of PV power output rather than a deterministic value, reflecting the uncertainty range of future PV power changes.

2. A regional distributed photovoltaic power prediction method according to claim 1, characterized in that: In S1, we first access the satellite remote sensing data service, track the cloud dynamics in real time through satellite images, and obtain the impact of cloud cover on the photovoltaic array in real time. Based on real-time meteorological data, we extract features closely related to photovoltaic power, including sliding window statistics of the main solar radiation intensity and periodic characteristics of temperature. We organize continuous meteorological data into a time series format and sample it every minute and every hour at fixed time intervals to form a continuous time series data set. In the feature engineering stage, we introduce nonlinear transformation to express the nonlinear relationship between meteorological parameters and photovoltaic power, based on the formula: f(x,y)=x 2 *log(y)e x and y are independent variables, representing different physical quantities and data points in the data on the impact of cloud cover on the photovoltaic array; x 2 Represents the square of variable x, a nonlinear transformation of variable x; log(y) represents the logarithmic function with the natural base e as the base, reflecting the relative growth rate of the variable y relative to the exponential growth; The two meteorological parameters are combined to capture the nonlinear effects of the interaction.

3. A regional distributed photovoltaic power prediction method according to claim 1, characterized in that: In S1, the nonlinear relationship model long short-term memory network is selected to construct a time series model. According to the characteristics of photovoltaic power output, different meteorological parameters are used as the input of the model, and photovoltaic power is used as the output. The model is trained using historical meteorological data and photovoltaic power data of the corresponding time period. Once new meteorological data comes in, the model will make real-time predictions of photovoltaic power based on the latest meteorological conditions. Based on the nonlinear activation function inside the model, the nonlinear relationship between meteorological factors and photovoltaic power output is captured, and the nonlinear interactions between different meteorological factors are learned, including the joint effects of temperature and humidity on photovoltaic efficiency.

4. A regional distributed photovoltaic power prediction method according to claim 1, characterized in that: In S2, an online learning framework is designed to allow the model to incrementally update model parameters when receiving new data. Each time a new batch of data is received, the gradient is calculated. For the loss function L(θ; x_t, y_t), where θ represents the model parameters, x_t, y_t represent the features and target values ​​of the tth sample, the model parameters are updated according to the gradient, and the update amplitude is adjusted using the learning rate α: θ←θ-α*▽_θL(θ;x_t,y_t) α is the learning rate, which controls the amplitude of each parameter update. Then, an adaptive learning strategy is used to dynamically adjust the learning rate. Combined with model performance monitoring, when the model prediction error exceeds the threshold, the learning rate is temporarily increased to adapt to the change.

5. A regional distributed photovoltaic power prediction method according to claim 1, characterized in that: In S3, a time window of 15 minutes, 30 minutes or 1 hour is set to make rolling forecasts for the upcoming time units. Based on the model of time series processing capability, the most recent historical meteorological data and photovoltaic power data are input, and the photovoltaic power forecast value in the future time period is output. Thresholds and rules are set to monitor the difference between the current meteorological data and the forecast results. When a sudden weather change is detected, the pre-set rapid response program is triggered to immediately execute the next stage of tasks and adjust the power generation plan based on the latest forecast results. The power dispatching center adjusts the power generation plan to adjust the load of thermal power plants and hydropower stations, dispatches energy storage equipment to release or absorb electricity, or adjusts the power transmission direction and power in the power grid.

6. A regional distributed photovoltaic power prediction method according to claim 1, characterized in that: In S4, multiple prediction models are trained simultaneously, the same meteorological parameters are input into each model to predict the PV power output, and the prediction results of all models are integrated and weighted average is adopted: P_pred=w1*P_model1+w2*P_model2+...+wn*P_modeln P_pred is the integrated prediction result; P_modeli is the predicted value of the i-th model; w1,w2,...,wn are the weights of their respective models, and satisfy w1+w2+...+wn=1; The probability distribution of photovoltaic power output is estimated by Bayesian method, and the possibility distribution of future power output is simulated based on the posterior distribution of model parameters. For each model parameter θ, p(θ|D) is the posterior probability distribution, and D represents the historical data set: p(P_out|D)=∫p(P_out|θ)*p(θ|D)dθ p(P_out|θ) is the probability density function of the PV power output given the parameter θ, and p(θ|D) is the posterior probability distribution of the parameter θ.

7. A regional distributed photovoltaic power prediction method according to claim 1, characterized in that: In S4, a random variable P is set to be affected by multiple factors such as light intensity, temperature, and component efficiency. out Based on the time series prediction model, the upper and lower bounds of power at different probability levels, 90% confidence intervals, and 95% confidence intervals are calculated to reflect the possible range of power generation of photovoltaic power stations in the future under the confidence level. The rolling prediction strategy based on the S3 short time window continuously updates the probability distribution prediction according to the newly flowing real-time data. When a sudden weather change or a change in the predicted distribution is detected, the rapid response program is triggered to adjust the power generation plan according to the latest probability prediction results. The energy dispatching center will dispatch other energy sources or adjust the load in advance to ensure power supply when the upper limit of the predicted power generation is low based on the probability prediction results and the upper and lower bounds of power. When the lower limit of the predicted power generation is high, the reserve capacity is appropriately reduced or the consumption ratio of photovoltaic power is increased.

8. A system used in a regional distributed photovoltaic power prediction method according to any one of claims 1 to 7, comprising a meteorological data acquisition module, characterized in that: The photovoltaic power prediction module includes: Learning and Adaptation Module: used to develop online update algorithms for model parameters, so that the model can be dynamically adjusted based on real-time feedback; Rolling forecast engine module: used to periodically perform short-term rolling forecasts and generate new forecast results at set time intervals; Forecast result analysis module: interpret rolling forecast results, compare historical data with current forecasts, and identify sudden weather changes; Risk assessment module: Analyze the uncertainty of prediction results, quantify prediction risks, and provide reference for decision-making; The learning and adaptation module includes a model building unit. The model building unit extracts features closely related to photovoltaic power based on real-time meteorological data to build a time series prediction model. The prediction result analysis module includes a power generation plan adjustment unit. The power generation plan adjustment unit formulates a power dispatching strategy that adapts to uncertainty based on the probabilistic prediction results and coordinates operations with other energy systems.

Images