Photovoltaic power generation intelligent scheduling method and device based on dynamic supply and demand prediction

Abstract

The invention relates to a photovoltaic power generation intelligent scheduling method and device based on dynamic supply and demand prediction. The photovoltaic power generation intelligent scheduling method based on dynamic supply and demand prediction comprises the following steps: obtaining position information of a current cloud cluster, combining the position information with wind speed data, and carrying out gridding prediction by adopting Kalman filtering to obtain a cloud cluster coverage state; performing shielding calculation on the cloud cluster coverage state by adopting a shielding probability function to obtain a cloud cluster shielding probability; and performing joint calculation on the cloud cluster coverage state and the cloud cluster shielding probability by adopting an illumination intensity attenuation model and a photoelectric power linear model to obtain a photovoltaic power generation power prediction value, performing difference calculation on the photovoltaic power generation power prediction value and an electrical load prediction curve to obtain predicted supply and demand power, and performing scheduling distribution on the power storage system according to the predicted supply and demand power. And completing predictive power supply scheduling of power utilization. The photovoltaic power generation intelligent scheduling method based on dynamic supply and demand prediction has the advantage of remarkably improving the real-time utilization rate of photovoltaic electric energy in severe weather.

Description

Technical Field

[0001] The present invention relates to the field of supply and demand forecasting for power supply or distribution, and in particular to a method and device for intelligent scheduling of photovoltaic power generation based on dynamic supply and demand forecasting. Background Art

[0002] With modern society's growing interest in renewable energy, photovoltaic power generation, as a clean, efficient, and renewable energy source, has become a widely used and large-scale energy technology. Traditional photovoltaic power stations are typically built in dry, rainless, and sunny areas such as plains, deserts, and mountains. However, photovoltaic power generation systems often require large tracts of land. Therefore, coastal areas, especially offshore photovoltaic projects, are becoming a new direction for renewable energy development, aiming to maximize the utilization of solar energy resources in open areas at sea.

[0003] However, although offshore photovoltaics is a relatively new way of resource development, in actual applications, photovoltaic power generation in coastal areas is often severely affected by meteorological factors, especially during typhoon season and the influence of marine climate, which can easily cause meteorological changes such as cloud thickness and cloud movement trajectory to become extremely complex and difficult to predict, resulting in highly nonlinear attenuation of light intensity and even violent fluctuations. Such fluctuations in light intensity directly affect the power generation of photovoltaic power stations, making the prediction of power generation difficult and unstable.

[0004] In practical applications, especially in microgrids of photovoltaic power stations, the real-time power generation of photovoltaic power stations is usually used as the main energy supply source to minimize the loss in the energy conversion and transmission process. When the real-time power generation of photovoltaic power stations is not enough to meet the current power load, the storage system needs to be called for power compensation. When the storage capacity is insufficient, the dispatching system will finally enable the external or main power grid to supply power. Therefore, under extreme climatic conditions such as marine climate or typhoons, due to the influence of meteorological factors such as rapid changes in clouds and drastic fluctuations in wind speed, not only will photovoltaic power generation fluctuate greatly, but industrial production in coastal areas may also experience sudden changes in electricity demand. This dual uncertainty of supply and demand greatly increases the difficulty of dispatching the power system, thereby affecting the stability and security of power supply.

[0005] Existing technologies typically use prediction models based on relatively stable climates to calculate light intensity and power load, and combine these with scheduling battery systems to balance the power load. However, marine and typhoon climates are highly uncertain, making existing light intensity prediction models prone to large errors in such situations. Especially during the short period of a typhoon, the speed and direction of cloud movement are unpredictable, making it difficult for traditional methods to cope with such drastic changes. This can cause delays and errors in power supply scheduling, leading to unstable power supply and potential safety hazards. Therefore, existing technologies face the problem of insufficient accuracy when dealing with predictive power supply scheduling in marine and typhoon climates. Summary of the Invention

[0006] Based on this, the purpose of the present invention is to provide a photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting.

[0007] A photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting includes the following steps:

[0008] S1. Obtaining the current cloud location information; wherein the cloud location information includes the cloud center of mass position, cloud area, cloud velocity state vector, and cloud thickness;

[0009] S2. Combining the current cloud location information with wind speed data, using Kalman filtering for gridded prediction, to obtain the cloud coverage status within a preset time;

[0010] S3. Using an occlusion probability function to perform occlusion calculation on the cloud coverage state within a preset time, to obtain the cloud occlusion probability within the preset time;

[0011] S4. Using a light intensity attenuation model to perform attenuation prediction on the cloud coverage state and cloud occlusion probability within a preset time, to obtain a light intensity prediction value within the preset time;

[0012] S5. Using a photovoltaic power linear model to convert the light intensity prediction value within the preset time, to obtain the photovoltaic power prediction value within the preset time;

[0013] S6. Calculate the difference between the power load forecast curve within a preset time and the corresponding photovoltaic power generation power forecast value to obtain the forecast supply and demand power within the preset time;

[0014] S7. Dispatch and allocate the power storage system according to the predicted supply and demand power within a preset time, and complete the predictive power supply dispatch for electricity consumption.

[0015] Compared to existing technologies, the photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting described in this invention predicts the cloud movement trajectory in the future time period and converts and calculates it by combining the light intensity attenuation model with the photovoltaic power linear model. This method can then adjust the storage system in advance, improve the real-time utilization rate of photovoltaic power, and effectively increase the response speed of the scheduling strategy to avoid power shortages or waste caused by sudden weather changes.

[0016] At the same time, the Kalman filter of the present invention uses the wind speed prediction value with time foresight (LSTM prediction) as a control item to perform grid prediction on the current cloud position information, thereby significantly improving the accuracy of cloud layer change prediction under sudden climate changes.

[0017] Furthermore, step S2 includes the following sub-steps:

[0018] S21, gridding the prediction area according to the cloud area and wind speed data in the current cloud location information to obtain a gridded prediction area set;

[0019] The specific representation of the prediction area set after grid division is as follows:

[0020] G={G i,j |i,j∈[1,M]}

[0021] Where G represents the set of prediction areas after grid division, and the prediction area refers to the area where the light intensity changes due to the projection of the cloud moving in the sky; G i,j The grid at row i and column j in the current forecast area includes the location information of the current cloud cluster. The total number of grids is M×M. The specific calculation is as follows:

[0022]

[0023] Where, Area represents the total area of the prediction area; Indicates rounding up; M max Indicates the maximum number of grid constraints; Size grid It represents the actual area of a single grid, and its specific calculation is as follows:

[0024]

[0025] Where, Resolution represents the mesh size factor; Ignore the threshold for cloud effects;

[0026] λ res Represents the adaptive grid resolution adjustment factor, and its specific calculation is as follows:

[0027]

[0028] Where, ‖u‖ represents the modulus of wind speed data; u base Indicates the reference wind speed, which is obtained by the statistical weighted average of historical data; A cloud represents the cloud area in the current cloud location information; β represents the influencing factor of wind speed and area changes; is the proportion of cloud area;

[0029] S22. According to the preset time, a Kalman filter is used in combination with wind speed data to predict the set of prediction areas after gridding, and the cloud coverage status within the preset time is obtained.

[0030] Based on this, the present invention improves the accuracy and robustness of cloud trajectory prediction under sudden weather changes by adopting an adaptive grid division strategy based on wind speed data and cloud area while ensuring computational efficiency.

[0031] Furthermore, the specific iterative expression of the Kalman filter is as follows:

[0032] The cloud state of the grid in row i and column j is:

[0033]

[0034] Where x cloud,ij is the cloud state of the grid in row i and column j; x c,ij and y c,ij Represents the grid G ij The position of the local cloud center of mass in v x and v y represents the velocity state of the cloud at the i-th row and j-th column grid; Indicates the rate of change of cloud area;

[0035] The state equation for the grid in row i and column j is expressed as follows:

[0036]

[0037] Where, is the grid G ij The cloud state at the predicted time k; F ij is the grid G ij The state transfer matrix is expressed as follows:

[0038]

[0039] Where Δt is the time step; represents the attenuation term of the area change rate, and τ represents the cloud diffusion time constant;

[0040] B is the wind speed control item, which is specifically expressed as follows:

[0041]

[0042] Where, β x and β y Used to express the weight of the influence of wind speed on cloud speed; is the process noise; u (k) is the external control input, i.e., the observed variable;

[0043] Represents the continuous velocity components of the wind speed in the x and y directions. It is obtained by using LSTM combined with historical data for prediction. Its specific representation is as follows:

[0044]

[0045] Where u (init) Represents the historical wind speed data of the initial measurement; u (0) Indicates the currently measured wind speed data; u (k-1) Represents the predicted wind speed data at time k-1; LSTM 风速 In order to predict the wind speed, a long short-term memory recurrent neural network was trained and fitted using historical data as the training data set.

[0046] A ij Indicates the neighboring grid to the current grid G ij The influence weight matrix of Represents the influence matrix of the neighboring grid on the current grid, which is specifically expressed as follows:

[0047]

[0048] Where, represents the state vector of the neighboring grid at time k+1; v x,(i+m,j+n) and v y,(i+m,j+n) Used to represent the grid G (i+m,j+n) The corresponding cloud velocity state; R represents the impact range, and its specific calculation is as follows:

[0049]

[0050] In the formula, base represents the baseline influence range; represents the cloud area change rate of the current grid at time k-1; γ is the sensitivity coefficient of the area change rate;

[0051] α mn The grid influence weight representing the horizontal and vertical offset is calculated as follows:

[0052]

[0053] Where, distance mn is the Gaussian attenuation weight value, which is specifically expressed as follows:

[0054]

[0055] Where [m,n] is the grid offset; Δx and Δy represent the actual side lengths of individual grids, i.e.

[0056] [m·Δx,n·Δt] is used to represent the actual position offset of the domain grid [i+m,i+n] relative to the current grid;

[0057] represents the cloud displacement vector affected by wind speed; σ mn is the standard deviation, and its specific calculation is as follows:

[0058]

[0059] Where, σ base is the benchmark standard deviation; δ is the standard deviation sensitivity coefficient;

[0060] For the k+1th moment grid G ij The Kalman filter prediction of the cloud state is specifically expressed as follows:

[0061]

[0062] Where, Represents the grid G ij The predicted cloud state at time k+1; is the grid G ij The prediction error covariance matrix of ; Represents the grid G ij The posterior error covariance matrix at time k; Q ij Represents the grid G ij The process noise covariance matrix of ;

[0063] When the prediction is obtained at the k+1th moment, the grid G ij After the cloud state is determined, the Kalman filter is updated, which is specifically expressed as follows:

[0064]

[0065] Where, Represents the grid G ij The Kalman filter matrix H ij Represents the grid G ij The observation matrix; R represents the covariance matrix of the observation noise; It represents the pseudo observation value at the k+1th moment obtained by LSTM prediction, which is used to replace the real observation data. Its specific representation is as follows:

[0066]

[0067] Where, represents the initial measured cloud state; Indicates the actual cloud state currently measured; LSTM 云团 Long short-term memory recurrent neural network for predicting cloud state;

[0068] When the Kalman filter update within the preset time is completed, the specific calculation of the global cloud coverage state within the preset time at the kth moment is expressed as follows:

[0069]

[0070] In the formula, Cover (k) It is used to indicate the cloud coverage status corresponding to the k-th moment, that is, the cloud area within the prediction area at the k-th moment; Represents the grid G ij The initial cloud area; Represents the grid G ij The area change rate predicted by the Kalman filter at time r.

[0071] Based on this, the present invention adopts LSTM to predict wind speed to enhance the long-term prediction accuracy of Kalman filter for cloud motion trajectory, and introduces a dynamic weight coupling mechanism of adjacent grids to enhance the dynamic correlation between spatial distance, wind speed direction and area change rate, thereby ensuring the local consistency of the predicted cloud motion trajectory.

[0072] In addition, pseudo observations are generated through LSTM to further improve the error compensation of long-term prediction and enhance the long-term prediction ability of Kalman filter.

[0073] Furthermore, the occlusion probability function is specifically expressed as follows to calculate the cloud occlusion probability at the kth moment:

[0074]

[0075] Where, represents the cloud occlusion probability at the kth moment; represents the historical occlusion probability term; is the historical occlusion weight at the kth moment, and its specific calculation is as follows:

[0076]

[0077] Where, represents the rate of change of the cloud coverage area at the kth moment; β cover represents the historical impact adjustment coefficient;

[0078] Indicates the coverage degree of the current cloud group to the forecast area; represents the cloud diffusion probability gain at the kth moment, and its specific calculation is as follows:

[0079]

[0080] Where, γ diffusion represents the diffusion influence adjustment coefficient; It indicates the degree of influence of the saturation rate of change, where θ is the saturation threshold; Represents the remaining space of the predicted area, that is, the space that can be diffused.

[0081] The present invention uses an occlusion probability function to smooth the continuity of cloud changes, and dynamically adjusts the calculation method of occlusion probability in combination with historical occlusion weights, so that it can better adapt to cloud states changing at different speeds.

[0082] In addition, the cloud coverage rate is calculated to ensure that the occlusion probability is consistent with the actual cloud area, and the diffusion probability gain term is introduced to optimize the gradual change process of the cloud boundary, thereby significantly improving the prediction accuracy of the degree of influence of cloud movement on sunlight, especially improving the adaptability and accuracy of the prediction of the degree of influence on sunlight in drastic climate changes such as typhoons and thunderstorms.

[0083] Furthermore, the specific calculation of the light intensity prediction value at the kth moment obtained by the light intensity attenuation model is expressed as follows:

[0084]

[0085] Where, represents the final predicted light intensity value at the kth moment; Represents the historical compensation item, where w hist Represents the historical light intensity compensation weight coefficient, and its specific calculation is as follows:

[0086]

[0087] Where λ hist Indicates the attenuation rate of the impact of historical data; represents the rate of change of occlusion probability at the kth moment; represents the diffusion compensation term, and its specific calculation is as follows:

[0088]

[0089] Where θ vIndicates the cloud coverage area change rate threshold; I (k) The predicted value of light intensity at time k without historical compensation is expressed as follows:

[0090]

[0091] Where I0 represents the current clear sky radiation intensity; represents the effective optical path term, where w area is the cloud area attenuation coefficient, and its specific calculation is expressed as:

[0092]

[0093] Where, α tyoe Indicates the cloud type coefficient; h cloud Indicates the thickness of the cloud layer; h ref Indicates the reference height of the cloud;

[0094] represents the transmittance correction term, and μ represents the weight coefficient of cloud thickness. The specific calculation is as follows:

[0095] μ=μ0·(1-exp(-δ μ ·h cloud ))

[0096] Where μ0 represents the basic optical thickness coefficient; δ μ represents the thickness sensitivity coefficient, which is obtained by Mie scattering calculation;

[0097] d eff It represents the equivalent path length correction factor, and its specific calculation is as follows:

[0098]

[0099] Where, represents the altitude angle of the sun at the kth moment.

[0100] Based on this, the present invention combines the light intensity prediction value with the historical compensation term to dynamically smooth the light intensity changes, thereby improving the continuity and accuracy of the light intensity prediction; at the same time, the boundary effect of fast-moving clouds is corrected through the diffusion compensation term to avoid abnormal fluctuations in light intensity caused by sudden changes in clouds, thereby improving the accuracy of the light intensity prediction.

[0101] In addition, the light intensity prediction value is dynamically adjusted through the transmittance correction term based on cloud thickness, solar altitude angle and optical scattering characteristics, so that the prediction value is more consistent with the attenuation of the actual physical process.

[0102] Furthermore, the photovoltaic power linear model is used for conversion to obtain the specific expression of the photovoltaic power prediction value at the kth moment as follows:

[0103]

[0104] Where, represents the final photovoltaic power generation power forecast value at the kth moment; It represents the photovoltaic power generation forecast value at the kth moment without historical smoothing, and its specific expression is as follows:

[0105]

[0106] Where η pv Indicates the comprehensive efficiency coefficient of photovoltaic modules; A pv is the total effective photovoltaic units of the photovoltaic power station; ∈ loss represents the photovoltaic power loss term; represents the historical smoothing compensation term; w pv is the dynamic smoothing weight of power generation, and its specific calculation is as follows:

[0107]

[0108] Where λ pv is the historical smoothing adjustment coefficient of photovoltaic power generation; is the rate of change of generated power.

[0109] Based on this, the present invention introduces a dynamic smoothing mechanism based on historical data in the photovoltaic power conversion process, that is, the current light intensity prediction value is integrated with the historical power generation power state through the historical smoothing compensation term, so as to achieve effective modeling of the hysteresis characteristics of photovoltaic power generation, thereby making the photovoltaic power generation power prediction value closer to the actual response curve and improving the physical consistency of the prediction.

[0110] Furthermore, the specific method of obtaining the power load forecast curve within the preset time includes the following steps:

[0111] S61. Sampling and normalizing historical power load data through a sliding window to obtain a power load data sequence;

[0112] S62A, performing statistical calculation on the power load data sequence to obtain a statistical characteristic vector of the power load data;

[0113] S62B, using one-dimensional Haar wavelet to perform wavelet decomposition on the power load data sequence to obtain high- and low-frequency feature vectors of the power load data;

[0114] S63, concatenating and fusing the statistical feature vector and the high-frequency and low-frequency feature vectors of the power load data to obtain a fused feature vector;

[0115] S64. Use the power load regression model to predict the fused feature vector to obtain a power load prediction curve within a preset time;

[0116] After obtaining the power load forecast curve within the preset time, the specific expression of the difference calculation at the kth moment is as follows:

[0117]

[0118] Where, represents the predicted supply and demand power at the kth moment; represents the power load forecast value at the kth moment in the power load forecast curve; η gird Indicates the power transmission loss coefficient; δ safe is the safety margin, and its specific calculation is as follows:

[0119] δ safe =λ safe μ(wave high )+γ safe ·σ(wave high )

[0120] Where λ safe represents the fluctuation intensity coefficient; γ safe represents the uncertainty coefficient; μ(wave high ) represents the mean of the high-frequency components in the high- and low-frequency feature vectors of the power load data; σ(wave high ) represents the standard deviation of the high-frequency components in the high- and low-frequency feature vectors of the electricity load data.

[0121] Based on this, the present invention dynamically samples historical electricity load data through a sliding window and integrates statistical feature extraction with Haar wavelet decomposition methods to simultaneously capture long-term trends and short-term mutations in the load curve, achieving highly sensitive recognition and accurate prediction of load surges or dips, significantly improving the model's prediction stability and accuracy.

[0122] In addition, in the process of calculating the difference between supply and demand power, the present invention introduces a safety margin compensation mechanism based on high-frequency fluctuation characteristics, quantitatively models the load fluctuation intensity by using the mean and standard deviation of the high-frequency component, and dynamically adjusts the safety margin accordingly to enhance the system's adaptability to sudden power disturbances, thereby effectively suppressing supply and demand imbalances and power dispatch fluctuations caused by prediction errors.

[0123] A photovoltaic power generation intelligent scheduling device based on dynamic supply and demand forecasting includes a cloud location information acquisition unit, a cloud location prediction unit, a cloud shading probability calculation unit, a light intensity attenuation prediction unit, a light intensity-power conversion unit, a supply and demand power calculation unit, and a predictive power supply scheduling unit;

[0124] The cloud position information acquisition unit is used to acquire the current cloud position information; wherein the cloud position information includes the cloud center of mass position, cloud area, cloud velocity state vector and cloud thickness;

[0125] The cloud position prediction unit is used to combine the current cloud position information with the wind speed data, use Kalman filtering to perform grid prediction, and obtain the cloud coverage status within a preset time;

[0126] The cloud occlusion probability calculation unit is used to perform occlusion calculation on the cloud coverage state within a preset time using an occlusion probability function to obtain the cloud occlusion probability within the preset time;

[0127] The light intensity attenuation prediction unit is used to use a light intensity attenuation model to perform attenuation prediction on the cloud coverage state and cloud occlusion probability within a preset time, and obtain a light intensity prediction value within the preset time;

[0128] The light intensity-power conversion unit is used to convert the light intensity prediction value within a preset time using a photovoltaic power linear model to obtain the photovoltaic power prediction value within the preset time;

[0129] The power supply and demand calculation unit is used to calculate the difference between the power load prediction curve within a preset time and the corresponding photovoltaic power generation power prediction value to obtain the predicted power supply and demand within the preset time;

[0130] The predictive power supply scheduling unit is used to schedule and allocate the power storage system according to the predicted supply and demand power within a preset time, thereby completing the predictive power supply scheduling for electricity consumption.

[0131] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0132] Figure 1 This is a simplified structural diagram of the photovoltaic power generation intelligent scheduling device based on dynamic supply and demand forecasting according to the present invention;

[0133] Figure 2 This is a simplified flow chart of the photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting according to the present invention;

[0134] Figure 3 A simplified schematic diagram of the prediction area. DETAILED DESCRIPTION

[0135] In order to solve the problem of insufficient accuracy faced by the existing technology when coping with predictive power supply scheduling in marine climate and typhoon climate, the present invention obtains the current cloud position information, and combines it with wind speed data, and uses the Kalman filter algorithm to perform grid prediction on the cloud, so as to obtain the cloud coverage status within a preset time, and uses the occlusion probability function to perform occlusion calculation on the cloud status to generate the cloud occlusion probability within the preset time; then, based on the cloud occlusion probability and the cloud status, the light intensity attenuation model is used to perform attenuation prediction, so as to obtain the light intensity prediction value within the preset time period, and the light intensity prediction value is converted into the photovoltaic power generation power prediction value through the photovoltaic power linear model; then, the photovoltaic power generation power prediction value and the power load prediction value are calculated to obtain the predicted supply and demand power, and the storage system is dispatched and allocated according to the predicted supply and demand power to complete the predictive power supply scheduling.

[0136] Accordingly, the present invention calculates the photovoltaic power generation power forecast value based on weather forecasts and calculates it with the power load forecast value to accurately identify future power gaps or surpluses, and conducts storage and power supply scheduling planning in advance, thereby improving the intelligent scheduling capabilities of local energy or microgrids under extreme climatic conditions and avoiding energy waste or power shortages.

[0137] Based on the above design, the present invention proposes a photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting, and based on this method proposes a photovoltaic power generation intelligent scheduling device based on dynamic supply and demand forecasting.

[0138] Please also see Figure 1 and Figure 2 , a simplified structural diagram of the photovoltaic power generation intelligent scheduling device based on dynamic supply and demand forecasting according to the present invention, Figure 2 This is a simplified flow chart of the photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting described in the present invention.

[0139] The photovoltaic power generation intelligent scheduling device based on dynamic supply and demand prediction includes a cloud position information acquisition unit 1, a cloud position prediction unit 2, a cloud shading probability calculation unit 3, a light intensity attenuation prediction unit 4, a light intensity-power conversion unit 5, a supply and demand power calculation unit 6 and a predictive power supply scheduling unit 7.

[0140] The cloud location information acquiring unit 1 is used to execute step S1: acquiring the location information of the current cloud.

[0141] Specifically, the cloud position information includes the cloud center of mass position, cloud area, cloud velocity state vector and cloud thickness;

[0142] The cloud cluster centroid position is used to locate the core position of the current cloud cluster. The cloud cluster reflection wave data is obtained through the weather radar, and the cloud cluster outline in the reflection wave data is extracted. The center of mass position is calculated based on the cloud cluster outline.

[0143] The cloud area is converted into the cloud area according to the reflected wave intensity of the cloud region in the weather radar image;

[0144] The cloud velocity vector is obtained by obtaining the radial velocity of the cloud through weather radar and calculating the horizontal and vertical movement speeds of the cloud in combination with multi-view data.

[0145] The cloud thickness is obtained by measuring the intensity and propagation time of the echo signal by the weather radar. It should be noted that the weather radar has its own cloud positioning and tracking system, and the current position information parameters of the cloud can be directly obtained through the system. However, this is not the invention of the present invention and will not be elaborated on here.

[0146] The cloud position prediction unit 2 is used to execute step S2: combining the current cloud position information with the wind speed data, using Kalman filtering to perform grid prediction, and obtaining the cloud coverage status within a preset time.

[0147] Specifically, the prediction area is gridded according to the cloud area and wind speed data in the current cloud location information to obtain a set of prediction areas after gridding, which is specifically expressed as follows:

[0148] G={G i,j |i,j∈[1,M]}

[0149] Where G represents the set of prediction areas after gridding. The prediction area refers to the area where the light intensity changes due to the shadow cast by the movement of clouds in the sky. The change of the sky area will directly affect the photovoltaic power generation. Usually, the area to be monitored is delineated based on the image of the weather radar. For details, please refer to Figure 3 , Figure 3 A simple schematic diagram of the prediction area; G i,j Represents the grid at row i and column j in the current forecast area, which contains the location information of the current cloud cluster. The total number of grids is M×M, and the specific calculation is as follows:

[0150]

[0151] Where Area represents the total area of the prediction area, which is a known parameter; Indicates rounding up; M max Indicates the maximum number of grid constraints, set the corresponding value according to the corresponding computing power; Size grid It represents the actual area of a single grid, and its specific calculation is as follows:

[0152]

[0153] Where Resolution represents the grid size factor, which is used to control the grid size. The default setting in this invention is 10km. 2 , represents the basic resolution when there is no wind speed and cloud influence; The cloud impact ignoring threshold is set by default to 0.0001. When the cloud clusters have an influence that can be ignored, the as the current grid resolution factor;

[0154] λ res Represents the adaptive grid resolution adjustment factor, which is used to dynamically adjust the grid size according to the changes in cloud speed and area to adapt to different cloud characteristics. Its specific calculation is as follows:

[0155]

[0156] Where, ‖u‖ represents the modulus of wind speed data, which is used to reflect the moving speed of clouds; u base Indicates the reference wind speed, which is obtained by the statistical weighted average of historical data; A cloud Represents the cloud area in the current cloud location information; β represents the influencing factor of wind speed and area change, which defaults to 0.5 and is used to balance the relative importance of wind speed and cloud coverage; It is the cloud area ratio item, which is used to reflect the coverage of the cloud cluster to the forecast area.

[0157] Then, according to the preset time, the Kalman filter is used in combination with the wind speed data to predict the set of prediction areas after the grid is divided to obtain the cloud coverage status within the preset time. The specific iterative expression of the Kalman filter is as follows:

[0158] Assume that the cloud state of the grid in row i and column j is:

[0159]

[0160] Where x cloud,ij is the cloud state of the grid in row i and column j, which is used to represent the state vector of the current grid; x c,ij and y c,ij Represents the grid G ij The position of the local cloud centroid in is used to characterize the shape distribution of the cloud in the grid;

[0161] v x and v y represents the velocity state of the cloud at the i-th row and j-th column grid; Represents the rate of change of cloud area, i.e. partial derivative, which is used to capture the deformation of the cloud;

[0162] Based on this, the state equation for the grid in row i and column j is expressed as:

[0163]

[0164] Where, is the grid G ij The cloud state at the predicted time k; F ij is the grid G ij The state transition matrix is used to determine the state change from the previous moment to the current moment, and its specific expression is as follows:

[0165]

[0166] Where Δt is the time step, which defaults to 10 minutes and is used to update the cloud state; The attenuation term representing the area change rate is used to reflect the inertia of cloud diffusion. τ represents the cloud diffusion time constant, which is a known parameter obtained by statistical averaging of historical data.

[0167] B is the wind speed control term, which is used to describe the influence of wind speed and other observed variables on cloud clusters. Its specific expression is as follows:

[0168]

[0169] Where, β x and β y The weight used to represent the influence of wind speed on cloud speed is obtained through actual data calibration, and the present invention does not specifically limit the influence weight;

[0170] is process noise, which conforms to Gaussian distribution and is used to reflect the randomness in the evolution of cloud state; u (k) is the external control input, and is the observed variable, that is, Represents the continuous velocity components of the wind speed in the x and y directions. This invention uses LSTM combined with historical data for prediction, which is specifically expressed as follows:

[0171]

[0172] Where u (init) Represents the historical wind speed data of the initial measurement, which is used to characterize the past state of wind speed and is obtained by measuring the echo of the meteorological radar ultrasonic wave; u (0) Indicates the currently measured wind speed data, which is used to characterize the current state of the wind speed and is obtained by measuring the echo of the weather radar ultrasonic wave; u (k-1)Represents the predicted wind speed data at time k-1, which is used to characterize the future state of wind speed and is obtained by combining LSTM with historical data for prediction; LSTM 风速 A long short-term memory recurrent neural network for predicting wind speed is trained and fitted using historical data as a training dataset. It is used to obtain the predicted values of the continuous velocity components of the wind speed in the x and y directions at the kth moment by inputting a sequence of historical wind speed components.

[0173] A ij Indicates the neighboring grid to the current grid G ij The influence weight matrix is used to indicate the degree to which the state of the current grid is affected by the adjacent grids. It represents the influence matrix of the neighboring grid on the current grid, which is specifically expressed as follows:

[0174]

[0175] Where, represents the state vector of the neighboring grid at time k+1, which only contains the velocity state of the cloud; v x,(i+m,j+n) and v y,(i+m,j+n) Used to represent the grid G (i+m,j+n) The corresponding cloud velocity state; R represents the influence range, which is used to indicate the range of adjacent grids that affect the current grid. The specific calculation is as follows:

[0176]

[0177] In the formula, base represents the baseline influence range, which is used to represent the basic influence range, and its default value is 1; Indicates the cloud area change rate of the current grid at time k-1; γ is the area change rate sensitivity coefficient, which is used to indicate the degree of influence of the cloud area change rate of the current grid on the affected range, and its default value is 0.2;

[0178] α mn The grid influence weights representing the horizontal and vertical offsets are used to indicate the degree of influence of the velocity state of the cloud cluster in the grid with a horizontal offset of m and a vertical offset of n on the velocity state of the cloud cluster in the current grid. The specific calculation is:

[0179]

[0180] Where, distance mn is the Gaussian attenuation weight value, which is used to indicate the attenuation degree of the influence of the grid with a horizontal offset of m and a vertical offset of n on the current grid. It is specifically expressed as follows:

[0181]

[0182] Where [m,n] is the grid offset, which is used to indicate the position offset relative to the current grid; Δx and Δy represent the actual side lengths of individual grids, that is, And [m·Δx,n·Δy] is used to represent the actual position offset of the domain grid [i+m,i+n] relative to the current grid; represents the cloud displacement vector affected by wind speed; σ mn is the standard deviation, which is used to control the magnitude of the grid effect when the horizontal axis is offset by m and the vertical axis is offset by n. The specific calculation is as follows:

[0183]

[0184] Where, σ base is the benchmark standard deviation, with a default value of 0.5 km, obtained through experiments and historical data. δ is the standard deviation sensitivity coefficient, calibrated based on experiments, and used to control the degree of influence of the area change of the cloud cluster in the grid with a horizontal axis offset of m and a vertical axis offset of n on the current grid.

[0185] Based on this, the neighborhood influence weight depends on the degree of match between the actual displacement of the cloud cluster and the spatial position of the grid. When the wind speed direction is consistent with the neighborhood direction, the weight is higher. By weighting the information of the domain grid, the subsequent Kalman filter can make a smoother and more consistent prediction of the current grid under the influence of multiple adjacent grids, thereby reducing the prediction error caused by insufficient local information.

[0186] Among them, for the k+1th moment grid G ij The Kalman filter prediction of the cloud state is specifically expressed as follows:

[0187]

[0188] Where, Represents the grid G ij The predicted cloud state at time k+1 is obtained by predicting the posterior state estimate based on the kth time; is the grid G ij The prediction error covariance matrix is used to represent the uncertainty of the prediction state at the k+1th moment; Represents the grid G ij The posterior error covariance matrix at time k is the prediction error covariance matrix after observation update; Q ij Represents the grid G ij The process noise covariance matrix is used to represent the external uncertainty factors that may affect the cloud state during the prediction process, thereby avoiding excessive reliance on historical data in the prediction process;

[0189] When the prediction is obtained at the k+1th moment, the grid Gij After the cloud state is determined, the Kalman filter is updated, which is specifically expressed as follows:

[0190]

[0191] Where, Represents the grid G ij The Kalman filter matrix is used to adjust the correction amount of the predicted state; H ij Represents the grid G ij The observation matrix is used to map the state variables to the observation space, thereby ensuring that the observation error can be calculated correctly; R represents the covariance matrix of the observation noise, which is used to represent the uncertainty of the observation data; It represents the pseudo observation value at the k+1th moment obtained by LSTM prediction, which is used to replace the real observation data, so that the Kalman filter can perform more accurate state updates in the long-term prediction process. Its specific representation is as follows:

[0192]

[0193] Where, Indicates the initial measured cloud state, which is used to characterize the historical data of the cloud state; Indicates the actual cloud state of the current measurement, used to characterize the actual cloud state of the current measurement; LSTM 云团 In order to predict the cloud state, a long short-term memory recurrent neural network is trained and fitted using historical data as a training dataset. It is used to obtain the pseudo-observation value of the cloud state at the k+1th moment by inputting the sequence of measured cloud state and predicted cloud state.

[0194] Finally, when the Kalman filter update within the preset time is completed, the specific calculation of the global cloud coverage state within the preset time at the kth moment is as follows:

[0195]

[0196] In the formula, Cover (k) It is used to indicate the cloud coverage status corresponding to the k-th moment, that is, the cloud area within the prediction area at the k-th moment; Represents the grid G ij The initial cloud area is obtained by clipping the corresponding cloud area in the current cloud location information; Represents the grid G ij The area change rate predicted by the Kalman filter at time r.

[0197] Based on this, the present invention uses LSTM wind speed prediction as a nonlinear but time-related input control variable to improve the long-term prediction of cloud motion trajectory by Kalman filtering under extreme weather conditions, and reduce the accumulated deviation of Kalman filtering in the long-term prediction process, thereby significantly improving the accuracy of cloud coverage status prediction.

[0198] In addition, by adopting the weight influence term of the neighboring grids, the state equation of the Kalman filter is enhanced to model the cloud state changes in the local area, avoiding the accumulation of single grid prediction errors affecting the prediction of the entire cloud, thereby improving the prediction accuracy of the cloud coverage state; at the same time, by dynamically adjusting the weight influence term of the neighboring grids, it is ensured that the Kalman filter prediction process can adapt to cloud movements of different scales, thereby improving the prediction adaptability.

[0199] Since the Kalman filter traditionally relies on real measurement values as observation quantities for updating, but the real measurement values in the long-term prediction process are limited, the present invention predicts the cloud state through LSTM to generate pseudo observation values to replace the real observation data, thereby significantly enhancing the long-term prediction capability of the Kalman filter. Moreover, since LSTM introduces nonlinear and time-series related characteristics, the prediction accuracy of the cloud coverage state is significantly improved.

[0200] The cloud occlusion probability calculation unit 3 is used to execute step S3: using an occlusion probability function to perform occlusion calculation on the cloud coverage state within a preset time to obtain the cloud occlusion probability within the preset time.

[0201] Specifically, the specific expression for calculating the cloud occlusion probability at the kth moment by using the occlusion probability function is as follows:

[0202]

[0203] Where, represents the cloud occlusion probability at the kth moment; represents the historical occlusion probability term, which is used to represent the impact of the historical occlusion probability on the current time k; is the historical occlusion weight at the kth moment. The lower its value, the more suitable it is for predicting rapidly changing clouds. Its specific calculation is as follows:

[0204]

[0205] Where, Indicates the rate of change of the cloud coverage area at the kth moment, which is used to indicate the cloud diffusion rate; β cover Represents the historical impact adjustment coefficient, which is used to control the impact of the change rate on the weight. Its default value is 0.1;

[0206] Indicates the coverage degree of the current cloud group to the forecast area; It represents the cloud diffusion probability gain at the kth moment, which is used to correct the cloud diffusion effect when the cloud changes drastically. Its specific calculation is as follows:

[0207]

[0208] Where, γ diffusion Indicates the diffusion influence adjustment coefficient, and its default value is 0.5; Indicates the degree of influence of the saturation rate of change. The value range is limited by tanh to prevent the correction term from growing infinitely. Its θ is the saturation threshold, which is used to adjust the amplitude effect when the area change rate exceeds the threshold. Its default value is 0.5; Represents the remaining space of the predicted area, that is, the diffusible space, which is used to calculate the corrected occlusion probability when there is diffusible space.

[0209] Based on this, the present invention establishes a nonlinear probability equation driven by the coverage area change rate to integrate the historical occlusion probability term with the restricted cloud diffusion probability gain term to achieve accurate dynamic modeling of cloud occlusion probability. At the same time, by adaptively adjusting the historical occlusion weight, the dependence on historical data in scenarios with rapidly changing clouds, such as typhoons and marine climate, is reduced, thereby enhancing the ability to predict cloud occlusion with short-term and drastic changes, thereby effectively reducing error accumulation.

[0210] In addition, the present invention also introduces a cloud diffusion probability gain to adaptively correct the occlusion probability, ensuring that under conditions of rapidly spreading clouds, the prediction can be dynamically adjusted as the cloud diffuses, reducing the lag of the occlusion prediction. At the same time, it ensures that the cloud is close to complete coverage, and the cloud diffusion probability gain can automatically decay to avoid the accumulation of calculation errors, thereby improving the accuracy of cloud occlusion probability calculation under severe weather conditions such as strong convection or typhoons.

[0211] The light intensity attenuation prediction unit 4 is used to execute step S4: using a light intensity attenuation model to perform attenuation prediction on the cloud coverage state and cloud shading probability within a preset time, and obtain a light intensity prediction value within the preset time.

[0212] Specifically, the specific calculation of the light intensity prediction value at the kth moment obtained by the light intensity attenuation model is expressed as follows:

[0213]

[0214] Where, I (k) Represents the predicted value of light intensity at the kth moment; I0 represents the current clear sky radiation intensity, which is used to represent the maximum solar radiation in a cloudless state and is a constant obtained based on historical measurement data of the corresponding area; represents the effective optical path term, which is used to characterize the direct shielding effect of cloud area coverage on the solar radiation path. area is the cloud area attenuation coefficient, and its specific calculation is expressed as:

[0215]

[0216] Where, α type represents the cloud type coefficient, which is obtained by statistical calculation of historical measurement data. Its value is related to the current predicted cloud type. The present invention does not specifically limit the value of the cloud type coefficient. cloud Indicates the cloud thickness, obtained by weather radar measurement; h ref Indicates the reference height of the cloud, which is a constant;

[0217] It represents the transmittance correction term, which is used to describe the attenuation effect of cloud thickness and occlusion probability on light intensity. Its μ represents the weight coefficient of cloud thickness. Its specific calculation is as follows:

[0218] μ=μ0·(1-exp(-δ μ ·h cloud ))

[0219] Where μ0 represents the basic optical thickness coefficient, which is obtained by inverting the meteorological radar reflectivity factor and the empirical formula and is a constant; δ μ represents the thickness sensitivity coefficient, which is obtained by Mie scattering calculation;

[0220] d eff It represents the equivalent path length correction factor, which is used to correct the geometric influence of the sun's altitude angle on the cloud penetration path to express the equivalent path length of light when passing through the cloud. The specific calculation is as follows:

[0221]

[0222] Where, represents the altitude angle of the sun at the kth moment, which is a known parameter.

[0223] Accordingly, the light intensity attenuation model of the present invention separates the cloud coverage area term from the occlusion probability term, thereby distinguishing the different effects of the geometric occlusion effect of clouds and the optical transmission effect of clouds on light intensity, thereby improving the adaptability to light attenuation under different weather conditions; in addition, the present invention introduces an equivalent path length correction coefficient to ensure the accuracy of light intensity prediction in different time periods, so as to reduce the error accumulation in the calculation process and thus improve the accuracy of light intensity prediction.

[0224] Furthermore, the illumination intensity attenuation model also includes diffusion compensation and historical smoothing calculation, which are specifically expressed as follows:

[0225]

[0226] Where, represents the final predicted light intensity value at the kth moment; Represents the historical compensation term, which is used to smooth the illumination prediction and avoid short-term fluctuations affecting the overall trend. hist Represents the historical light intensity compensation weight coefficient, and its specific calculation is as follows:

[0227]

[0228] Where λ hist Indicates the attenuation rate of historical data impact, which is used to control the impact of historical data on prediction. The default value is 0.2; represents the rate of change of occlusion probability at the kth moment; represents the diffusion compensation term, which is used to compensate for the local light leakage caused by the rapid movement of clouds. Its specific calculation is as follows:

[0229]

[0230] Where θ v The cloud coverage area change rate threshold is used to measure whether the current cloud area change rate requires diffusion compensation. Its value is obtained through statistical calculations based on historical data. When the cloud area change rate is greater than the cloud coverage area change rate threshold, it means that the cloud is moving rapidly or the edge is thinning out. This may cause sunlight to penetrate the cloud gaps for a short period of time, temporarily increasing the actual light intensity. Therefore, diffusion compensation is required.

[0231] Based on this, the present invention further supplements the light intensity prediction model with a historical compensation term to reduce the impact of short-term fluctuations on the light intensity prediction; and, by combining the diffusion compensation term, improves the prediction accuracy of fast-moving clouds and reduces the impact of short-term light leakage, thereby significantly improving the accuracy of light intensity prediction.

[0232] The light intensity-power conversion unit 5 is used to execute step S5: using a photovoltaic power linear model to convert the light intensity prediction value within the preset time to obtain the photovoltaic power prediction value within the preset time.

[0233] Specifically, the photovoltaic power linear model is used for conversion to obtain the photovoltaic power prediction value at the kth moment. The specific expression is as follows:

[0234]

[0235] Where, represents the predicted value of photovoltaic power generation at the kth moment; η pvIndicates the comprehensive efficiency coefficient of photovoltaic modules, which is used to express the ratio of convertible electrical energy under unit light intensity and is a known parameter; A pv is the total effective photovoltaic unit of the photovoltaic power station, which corresponds to the prediction area and is a known parameter; loss Represents the photovoltaic power loss term, which is used to correct power generation losses caused by environmental factors. Generally, the linear regression method is used to fit the historical data of photovoltaic power stations to construct the photovoltaic power loss term. However, since the power loss terms of different photovoltaic power stations may be affected by factors such as the environment, hardware aging, or climate, the present invention does not specifically limit the specific method for obtaining the photovoltaic power loss term.

[0236] Furthermore, the illumination intensity attenuation model also includes a historical smoothing calculation, which is specifically expressed as follows:

[0237]

[0238] Where, represents the final photovoltaic power generation power forecast value at the kth moment; represents the historical smoothing compensation term, which is used to smooth the photovoltaic power generation power forecast value; w pv is the dynamic smoothing weight of power generation, which is used to determine the degree of influence of the historical photovoltaic power generation forecast value on the current photovoltaic power generation forecast value. Its specific calculation is as follows:

[0239]

[0240] Where λ pv The historical smoothing adjustment coefficient of photovoltaic power generation is used to control the impact of drastic changes in a short period of time. Its default value is 0.4. is the rate of change of generated power.

[0241] Based on this, the present invention combines historical smoothing items to ensure that the predicted value of photovoltaic power generation presents a certain transition effect, that is, gradually decreases or gradually increases. This is because the photovoltaic material itself has a certain lag effect on changes in light. Therefore, the present invention uses historical data to smooth the predicted value of photovoltaic power generation to ensure that the predicted value is more in line with the gradual change trend of actual power generation, thereby avoiding scheduling errors caused by prediction errors.

[0242] The supply and demand power calculation unit 6 is used to execute step S6: performing difference calculation between the power load prediction curve within a preset time and the corresponding photovoltaic power generation power prediction value to obtain the predicted supply and demand power within the preset time.

[0243] Specifically, the method for obtaining the power load forecast curve within the preset time includes the following steps:

[0244] S61. Sampling and normalizing historical power load data through a sliding window to obtain a power load data sequence;

[0245] Specifically, the i-th normalized power load data in the load data sequence is The specific calculation is as follows:

[0246]

[0247] Where, L i Represents the i-th historical electricity load data in the window; Represents a historical electricity load data set, which is specifically expressed as follows:

[0248]

[0249] Where i base Represents the index of the current time point, that is, the most recently acquired power load data; range represents the size of the sliding window; wherein, the step size of the power load data is Δt, which is consistent with the time step size of the aforementioned photovoltaic power generation power.

[0250] S62A, performing statistical calculation on the power load data sequence to obtain a statistical characteristic vector of the power load data;

[0251] Specifically, the statistical characteristic vector of the power load data is specifically expressed as follows:

[0252]

[0253] Where μ win Represents the mean value of electricity load data, which is used to reflect the overall level of electricity load; ρ peak Represents the peak-to-valley difference rate, which is used to measure the peak-to-valley difference in electricity load; σ win Indicates the standard deviation of electricity load data, which is used to reflect the volatility of electricity load.

[0254] S62B, using one-dimensional Haar wavelet to perform wavelet decomposition on the power load data sequence to obtain high- and low-frequency feature vectors of the power load data;

[0255] Specifically, the high-frequency and low-frequency feature vectors f of the power load data are wave The specific representation is as follows:

[0256]

[0257] Where, μ(wave high ) is the mean of the high-frequency components of the power load data, which is used to quantify sudden fluctuations to reflect sudden changes in power load; μ(wave low) represents the mean of the low-frequency component of the electricity load data and is used to quantify the long-term trend to represent the periodic characteristics of the electricity load.

[0258] S63, concatenating and fusing the statistical feature vector and the high-frequency and low-frequency feature vectors of the power load data to obtain a fused feature vector;

[0259] Specifically, the fused feature vector f fusion The specific representation is as follows:

[0260]

[0261] S64. Use the power load regression model to predict the fused feature vector to obtain a power load prediction curve within a preset time;

[0262] Specifically, the power load regression model is expressed as follows:

[0263]

[0264] Where, It represents the power load forecast curve within the preset time, which is specifically expressed as follows:

[0265]

[0266] Where, represents the predicted value of the power load at the kth moment; bias represents the bias matrix of the power load regression model, which is obtained by training the power load regression model with historical data. The specific expression of bias is as follows:

[0267]

[0268] Where b k represents the bias term at the kth moment, which is used to adjust the overall bias of the power load forecast value at the kth moment, thereby correcting the systematic error in the historical data; β represents the weight matrix of the power load regression model, which is used to control the impact of different features on the power load forecast. It is obtained by training the power load regression model with historical data. The specific expression of β is as follows:

[0269]

[0270] Where, β k,j Represents the weight value of the jth dimension at the kth moment, and the feature dimension j defaults to 5.

[0271] Next, after obtaining the power load forecast curve within the preset time, the specific expression of the difference calculation at the kth moment is as follows:

[0272]

[0273] Where, represents the predicted supply and demand power at the kth moment, which is used to predict the relationship between the demand and the predicted value of photovoltaic power generation; η gird Indicates the power transmission loss coefficient, which is used to represent the conversion loss when photovoltaic power is transmitted to the microgrid, local energy system or power grid. It is an empirical coefficient with a default value of 0.95; safe is the safety margin, a buffer value used to ensure system safety. Its specific calculation is as follows:

[0274] δ safe =λ safe μ(wave high )+γ safe ·σ(wave high )

[0275] Where λ safe represents the fluctuation intensity coefficient, which is used to reserve a certain high-frequency fluctuation mean as a buffer coefficient. It is obtained based on the statistics of historical accident risks and can be further calculated using time step exponential decay. The present invention does not specifically limit the selection of its fluctuation intensity coefficient; safe represents the uncertainty coefficient, which is used to evaluate the uncertainty of electricity consumption; σ(wave high ) represents the standard deviation of the high-frequency component of the electricity load data and is used for safety redundancy of abnormal load changes to prevent sudden changes in the system from causing an imbalance between supply and demand.

[0276] Accordingly, the present invention introduces wavelet features on the basis of global statistical features to enhance the ability to identify sudden drops or mutations in electricity consumption, thereby improving the accuracy of predicting sudden changes in electricity load; at the same time, safety margin compensation is combined with supply and demand power forecasting to ensure that in the case of drastic short-term fluctuations in electricity load, the system still has sufficient power buffering, avoiding power supply fluctuations caused by excessive differences in supply and demand, and reducing the risk of unstable power supply.

[0277] The predictive power supply scheduling unit 7 is used to execute step S7: scheduling and allocating the power storage system according to the predicted supply and demand power within a preset time, thereby completing the predictive power supply scheduling for electricity consumption.

[0278] Specifically, the scheduling allocation is expressed as follows:

[0279]

[0280] Where ER(k) represents the surplus ratio at the kth moment, which is used to characterize the current adequacy of photovoltaic power generation. Its specific calculation is as follows:

[0281]

[0282] Among them, -θ store represents the surplus threshold, which is set based on experience. When the surplus ratio exceeds this threshold, it is considered that the current photovoltaic power generation is sufficient, and part of the generated electricity is directly stored in the power storage system; SR(k) represents the supply-demand gap ratio, which is specifically calculated as follows:

[0283]

[0284] Among them, θ critical Indicates the gap danger threshold, which is used to determine whether the current power gap ratio is too low. If so, the reserve energy of the power storage system is called upon first.

[0285] Based on this, the present invention constructs a predictive supply and demand power curve as the decision-making basis for preventive scheduling strategies, thereby achieving early identification of possible power gaps or surpluses in future periods, thereby guiding the storage system to perform proactive scheduling and energy storage management, significantly improving the response efficiency and control accuracy of the scheduling strategy, and reducing the risk of power system fluctuations caused by supply and demand imbalances.

[0286] It should be noted that the predicted supply and demand power within the preset time period on which the scheduling allocation is based mainly selects several short time periods before the current moment (for example, 1 to 2 prediction time steps) as the effective scheduling basis to ensure that the storage system completes energy allocation before the load gap occurs; and the prediction results of more distant time steps are usually not directly involved in the scheduling calculation due to their high uncertainty. They can only be used as a reference for the scheduling system in some scenarios to make auxiliary energy storage or power purchase decisions. In the case of worsening weather, the prediction period can be further increased to obtain the corresponding predicted supply and demand power to achieve rolling prediction.

[0287] At the same time, because different energy storage systems have their own scheduling strategies and control logic, and their scale, energy storage capacity, and charge-discharge efficiency vary significantly, this invention does not limit the specific details of energy storage system scheduling. In particular, the core of this invention is to provide a predictive supply and demand power calculation method and provide reasonable power scheduling recommendations based on the prediction results to enable the energy storage system to make corresponding adjustments. Therefore, the specific implementation method can be adapted and optimized based on the actual energy storage system capacity, scheduling algorithm, discharge strategy, and grid feedback capabilities.

[0288] Compared with the existing technology, the present invention uses Kalman filtering combined with a light intensity attenuation model to accurately predict future photovoltaic power generation power in the case of drastic climate change, and combines historical data to perform intelligent regression prediction of power load, thereby calculating the predicted supply and demand power, and performing predictive power supply scheduling based on the predicted supply and demand power, effectively responding to power supply fluctuations under sudden weather conditions, and reducing power shortages or power waste, thereby significantly improving the stability and safety of power supply.

[0289] Based on the same inventive concept, the present application also provides an electronic device, which can be a terminal device such as a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.). The device includes one or more processors and a memory, wherein the processor is used to execute a program to implement the photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting according to an embodiment of the present invention; and the memory is used to store a computer program executable by the processor.

[0290] Based on the same inventive concept, the present application also provides a computer-readable storage medium, corresponding to the embodiment of the aforementioned photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting, wherein the computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting recorded in any of the above embodiments.

[0291] The present application may take the form of a computer program product implemented on one or more storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-usable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. The information can be computer-readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include but are not limited to: phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0292] The above-described embodiments merely represent several implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous modifications and improvements without departing from the spirit of the present invention, and the present invention is intended to encompass such modifications and variations.

Claims

1. A photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting, characterized in that: The following steps are involved: S1. Obtaining the current cloud location information; wherein the cloud location information includes the cloud center of mass position, cloud area, cloud velocity state vector, and cloud thickness; S2. Combining the current cloud location information with wind speed data, using Kalman filtering for gridded prediction, to obtain the cloud coverage status within a preset time; S3. Using an occlusion probability function to perform occlusion calculation on the cloud coverage state within a preset time, to obtain the cloud occlusion probability within the preset time; S4. Using a light intensity attenuation model to perform attenuation prediction on the cloud coverage state and cloud occlusion probability within a preset time, to obtain a light intensity prediction value within the preset time; S5. Using a photovoltaic power linear model to convert the light intensity prediction value within the preset time, to obtain the photovoltaic power prediction value within the preset time; S6. Calculate the difference between the power load forecast curve within a preset time and the corresponding photovoltaic power generation power forecast value to obtain the forecast supply and demand power within the preset time; S7. Dispatch and allocate the power storage system according to the predicted supply and demand power within a preset time, and complete the predictive power supply dispatch for electricity consumption.

2. The photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting according to claim 1 is characterized in that: The step S2 includes the following sub-steps: S21, gridding the prediction area according to the cloud area and wind speed data in the current cloud location information to obtain a gridded prediction area set; The specific representation of the prediction area set after grid division is as follows: G={G i,j |i,j∈[1,M]} Where G represents the set of prediction areas after grid division, and the prediction area refers to the area where the light intensity changes due to the projection of the cloud moving in the sky; G i,j The grid at row i and column j in the current forecast area includes the location information of the current cloud cluster. The total number of grids is M×M. The specific calculation is as follows: Where, Area represents the total area of the prediction area; Indicates rounding up; M max Indicates the maximum number of grid constraints; Size grid It represents the actual area of a single grid, and its specific calculation is as follows: Where, Resolution represents the mesh size factor; Ignore the threshold for cloud effects; λ res Represents the adaptive grid resolution adjustment factor, and its specific calculation is as follows: Where, ‖u‖ represents the modulus of wind speed data; u base Indicates the reference wind speed, which is obtained by the statistical weighted average of historical data; A cloud represents the cloud area in the current cloud location information; β represents the influencing factor of wind speed and area changes; is the proportion of cloud area; S22. According to the preset time, a Kalman filter is used in combination with wind speed data to predict the set of prediction areas after gridding, and the cloud coverage status within the preset time is obtained.

3. The photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting according to claim 2 is characterized in that: The specific iterative expression of the Kalman filter is as follows: The cloud state of the grid in row i and column j is: Where x cloud,ij is the cloud state of the grid in row i and column j; x c,ij and y c,ij Represents the grid G ij The position of the local cloud center of mass in v x and v y represents the velocity state of the cloud at the i-th row and j-th column grid; Indicates the rate of change of cloud area; The state equation for the grid in row i and column j is expressed as follows: Where, is the grid G ij The cloud state at the predicted time k; F ij is the grid G ij The state transfer matrix is expressed as follows: Where Δt is the time step; represents the attenuation term of the area change rate, and τ represents the cloud diffusion time constant; B is the wind speed control item, which is specifically expressed as follows: Where, β x and β y Used to express the weight of the influence of wind speed on cloud speed; is the process noise; u (k) is the external control input, i.e., the observed variable; Represents the continuous velocity components of the wind speed in the x and y directions. It is obtained by using LSTM combined with historical data for prediction. Its specific representation is as follows: Where u (init) Represents the historical wind speed data of the initial measurement; u (0) Indicates the currently measured wind speed data; u (k-1) Represents the predicted wind speed data at time k-1; LSTM 风速 In order to predict the wind speed, a long short-term memory recurrent neural network was trained and fitted using historical data as the training data set. A ij Indicates the neighboring grid to the current grid G ij The influence weight matrix of Represents the influence matrix of the neighboring grid on the current grid, which is specifically expressed as follows: Where, represents the state vector of the neighboring grid at time k+1; v x,(i+m,n+n) and v y,(i+m,j+n) Used to represent the grid G (i+m,j+n) The corresponding cloud velocity state; R represents the impact range, and its specific calculation is as follows: In the formula, base represents the baseline influence range; represents the cloud area change rate of the current grid at time k-1; γ is the sensitivity coefficient of the area change rate; α mn The grid influence weight representing the horizontal and vertical offset is calculated as follows: Where, distance mn is the Gaussian attenuation weight value, which is specifically expressed as follows: Where [m,n] is the grid offset; Δx and Δy represent the actual side lengths of individual grids, i.e. [m·Δx,n·Δt] is used to represent the actual position offset of the domain grid [i+m,i+n] relative to the current grid; represents the cloud displacement vector affected by wind speed; σ mn is the standard deviation, and its specific calculation is as follows: Where, σ base is the benchmark standard deviation; δ is the standard deviation sensitivity coefficient; For the k+1th moment grid G ij The Kalman filter prediction of the cloud state is specifically expressed as follows: Where, Represents the grid G ij The predicted cloud state at time k+1; is the grid G ij The prediction error covariance matrix of ; Represents the grid G ij The posterior error covariance matrix at time k; Q ij Represents the grid G ij The process noise covariance matrix of ; When the prediction is obtained at the k+1th moment, the grid G ij After the cloud state is determined, the Kalman filter is updated, which is specifically expressed as follows: Where, Represents the grid G ij The Kalman filter matrix H ij Represents the grid G ij The observation matrix; R represents the covariance matrix of the observation noise; It represents the pseudo observation value at the k+1th moment obtained by LSTM prediction, which is used to replace the real observation data. Its specific representation is as follows: Where, represents the initial measured cloud state; Indicates the actual cloud state currently measured; LSTM 云团 Long short-term memory recurrent neural network for predicting cloud state; When the Kalman filter update within the preset time is completed, the specific calculation of the global cloud coverage state within the preset time at the kth moment is expressed as follows: In the formula, Cover (k) It is used to indicate the cloud coverage status corresponding to the k-th moment, that is, the cloud area within the prediction area at the k-th moment; Represents the grid G ij The initial cloud area; Represents the grid G ij The area change rate predicted by the Kalman filter at time r.

4. The photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting according to claim 3 is characterized in that: The specific expression of the occlusion probability function for calculating the cloud occlusion probability at the kth moment is as follows: Where, represents the cloud occlusion probability at the kth moment; represents the historical occlusion probability term; is the historical occlusion weight at the kth moment, and its specific calculation is as follows: Where, represents the rate of change of the cloud coverage area at the kth moment; β cover represents the historical impact adjustment coefficient; Indicates the coverage degree of the current cloud group to the forecast area; represents the cloud diffusion probability gain at the kth moment, and its specific calculation is as follows: Where, γ diffusion represents the diffusion influence adjustment coefficient; It indicates the degree of influence of the saturation rate of change, where θ is the saturation threshold; Represents the remaining space of the predicted area, that is, the space that can be diffused.

5. The photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting according to claim 4 is characterized in that: The specific calculation of the light intensity prediction value at the kth moment obtained by the light intensity attenuation model is as follows: Where, represents the final predicted light intensity value at the kth moment; Represents the historical compensation item, where w hist Represents the historical light intensity compensation weight coefficient, and its specific calculation is as follows: Where λ hist Indicates the attenuation rate of the impact of historical data; represents the rate of change of occlusion probability at the kth moment; represents the diffusion compensation term, and its specific calculation is as follows: Where θ v Indicates the cloud coverage area change rate threshold; I (k) The predicted value of light intensity at time k without historical compensation is expressed as follows: Where I0 represents the current clear sky radiation intensity; represents the effective optical path term, where w area is the cloud area attenuation coefficient, and its specific calculation is expressed as: Where, α type Indicates the cloud type coefficient; h cloud Indicates the thickness of the cloud layer; h ref Indicates the reference height of the cloud; represents the transmittance correction term, and μ represents the weight coefficient of cloud thickness. The specific calculation is as follows: μ=μ0·(1-exp(-δ μ ·h cloud )) Where μ0 represents the basic optical thickness coefficient; δ μ represents the thickness sensitivity coefficient, which is obtained by Mie scattering calculation; d eff It represents the equivalent path length correction factor, and its specific calculation is as follows: Where, represents the altitude angle of the sun at the kth moment.

6. The photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting according to claim 5 is characterized in that: The photovoltaic power linear model is used for conversion to obtain the photovoltaic power prediction value at the kth moment as follows: Where, represents the final photovoltaic power generation power forecast value at the kth moment; It represents the photovoltaic power generation forecast value at the kth moment without historical smoothing, and its specific expression is as follows: Where η pv Indicates the comprehensive efficiency coefficient of photovoltaic modules; A pv is the total effective photovoltaic units of the photovoltaic power station; ∈ loss represents the photovoltaic power loss term; represents the historical smoothing compensation term; w pv is the dynamic smoothing weight of power generation, and its specific calculation is as follows: Where λ pv is the historical smoothing adjustment coefficient of photovoltaic power generation; is the rate of change of generated power.

7. The photovoltaic power generation intelligent scheduling method based on dynamic supply and demand forecasting according to claim 6 is characterized in that: The specific method for obtaining the power load forecast curve within the preset time includes the following steps: S61. Sampling and normalizing historical power load data through a sliding window to obtain a power load data sequence; S62A, performing statistical calculation on the power load data sequence to obtain a statistical characteristic vector of the power load data; S62B, using one-dimensional Haar wavelet to perform wavelet decomposition on the power load data sequence to obtain high- and low-frequency feature vectors of the power load data; S63, concatenating and fusing the statistical feature vector and the high-frequency and low-frequency feature vectors of the power load data to obtain a fused feature vector; S64. Use the power load regression model to predict the fused feature vector to obtain a power load prediction curve within a preset time; After obtaining the power load forecast curve within the preset time, the specific expression of the difference calculation at the kth moment is as follows: Where, represents the predicted supply and demand power at the kth moment; represents the power load forecast value at the kth moment in the power load forecast curve; η gird Indicates the power transmission loss coefficient; δ safe is the safety margin, and its specific calculation is as follows: d safe =λ safe ·μ(wave high )+c safe ·σ(wave high ) Where λ safe represents the fluctuation intensity coefficient; γ saee represents the uncertainty coefficient; μ(wave high ) represents the mean of the high-frequency components in the high- and low-frequency feature vectors of the power load data; σ(wave high ) represents the standard deviation of the high-frequency components in the high- and low-frequency feature vectors of the electricity load data.

8. A photovoltaic power generation intelligent scheduling device based on dynamic supply and demand forecasting, characterized in that: It includes a cloud position information acquisition unit, a cloud position prediction unit, a cloud shading probability calculation unit, a light intensity attenuation prediction unit, a light intensity-power conversion unit, a supply and demand power calculation unit and a predictive power supply scheduling unit; The cloud position information acquisition unit is used to acquire the current cloud position information; wherein the cloud position information includes the cloud center of mass position, cloud area, cloud velocity state vector and cloud thickness; The cloud position prediction unit is used to combine the current cloud position information with the wind speed data, use Kalman filtering to perform grid prediction, and obtain the cloud coverage status within a preset time; The cloud occlusion probability calculation unit is used to perform occlusion calculation on the cloud coverage state within a preset time using an occlusion probability function to obtain the cloud occlusion probability within the preset time; The light intensity attenuation prediction unit is used to use a light intensity attenuation model to perform attenuation prediction on the cloud coverage state and cloud occlusion probability within a preset time, and obtain a light intensity prediction value within the preset time; The light intensity-power conversion unit is used to convert the light intensity prediction value within a preset time using a photovoltaic power linear model to obtain the photovoltaic power prediction value within the preset time; The power supply and demand calculation unit is used to calculate the difference between the power load prediction curve within a preset time and the corresponding photovoltaic power generation power prediction value to obtain the predicted power supply and demand within the preset time; The predictive power supply scheduling unit is used to schedule and allocate the power storage system according to the predicted supply and demand power within a preset time, thereby completing the predictive power supply scheduling for electricity consumption.

9. The photovoltaic power generation intelligent scheduling device based on dynamic supply and demand forecasting according to claim 8, characterized in that: The cloud position prediction unit is further configured to perform the following sub-steps: S21, gridding the prediction area according to the cloud area and wind speed data in the current cloud location information to obtain a gridded prediction area set; The specific representation of the prediction area set after grid division is as follows: G={G i,j |i,j∈[1,M]} Where G represents the set of prediction areas after grid division, and the prediction area refers to the area where the light intensity changes due to the projection of the cloud moving in the sky; G i,j The grid at row i and column j in the current forecast area includes the location information of the current cloud cluster. The total number of grids is M×M. The specific calculation is as follows: Where, Area represents the total area of the prediction area; Indicates rounding up; M max Indicates the maximum number of grid constraints; Size grid It represents the actual area of a single grid, and its specific calculation is as follows: Where, Resolution represents the mesh size factor; Ignore the threshold for cloud effects; λ res Represents the adaptive grid resolution adjustment factor, and its specific calculation is as follows: Where, ‖u‖ represents the modulus of wind speed data; u base Indicates the reference wind speed, which is obtained by the statistical weighted average of historical data; A cloud represents the cloud area in the current cloud location information; β represents the influencing factor of wind speed and area changes; is the proportion of cloud area; S22. Based on the preset time, a Kalman filter is used in combination with wind speed data to predict the set of prediction areas after the grid is divided, and the cloud coverage status within the preset time is obtained; The specific iterative expression of the Kalman filter is as follows: The cloud state of the grid in row i and column j is: Where x cloud,sj is the cloud state of the grid in row i and column j; x c,ij and y c,ij Represents the grid G ij The position of the local cloud center of mass in v x and v y represents the velocity state of the cloud at the i-th row and j-th column grid; Indicates the rate of change of cloud area; The state equation for the grid in row i and column j is expressed as follows: Where, is the grid G ij The cloud state at the predicted time k; F ij is the grid G ij The state transfer matrix is expressed as follows: Where Δt is the time step; represents the attenuation term of the area change rate, and τ represents the cloud diffusion time constant; B is the wind speed control item, which is specifically expressed as follows: Where, β x and β y Used to express the weight of the influence of wind speed on cloud speed; is the process noise; u (k) is the external control input, i.e., the observed variable; Represents the continuous velocity components of the wind speed in the x and y directions. It is obtained by using LSTM combined with historical data for prediction. Its specific representation is as follows: Where u (init) Represents the historical wind speed data of the initial measurement; u (0) Indicates the currently measured wind speed data; u (k-1) Represents the predicted wind speed data at time k-1; LSTM 风速 In order to predict the wind speed, a long short-term memory recurrent neural network was trained and fitted using historical data as the training data set. A ij Indicates the neighboring grid to the current grid G ij The influence weight matrix of Represents the influence matrix of the neighboring grid on the current grid, which is specifically expressed as follows: Where, represents the state vector of the neighboring grid at time k+1; v x,(i+m,j+n) and v y,(i+m,j+n) Used to represent the grid G (i+m，j+n) The corresponding cloud velocity state; R represents the impact range, and its specific calculation is as follows: In the formula, base represents the baseline influence range; represents the cloud area change rate of the current grid at time k-1; γ is the sensitivity coefficient of the area change rate; α mn The grid influence weight representing the horizontal and vertical offset is calculated as follows: Where, distance mn is the Gaussian attenuation weight value, which is specifically expressed as follows: Where [m,n] is the grid offset; Δx and Δy represent the actual side lengths of individual grids, i.e. [m·Δx,n·Δy] is used to represent the actual position offset of the domain grid [i+m,i+n] relative to the current grid; represents the cloud displacement vector affected by wind speed; σ mn is the standard deviation, and its specific calculation is as follows: Where, σ base is the benchmark standard deviation; δ is the standard deviation sensitivity coefficient; For the k+1th moment grid G ij The Kalman filter prediction of the cloud state is specifically expressed as follows: Where, Represents the grid G ij The predicted cloud state at time k+1; is the grid G ij The prediction error covariance matrix of ; Represents the grid G ij The posterior error covariance matrix at time k; Q ij Represents the grid G ij The process noise covariance matrix of ; When the prediction is obtained at the k+1th moment, the grid G ij After the cloud state is determined, the Kalman filter is updated, which is specifically expressed as follows: Where, Represents the grid G ij The Kalman filter matrix H ij Represents the grid G ij The observation matrix; R represents the covariance matrix of the observation noise; It represents the pseudo observation value at the k+1th moment obtained by LSTM prediction, which is used to replace the real observation data. Its specific representation is as follows: Where, represents the initial measured cloud state; Indicates the actual cloud state currently measured; LSTM 云团 Long short-term memory recurrent neural network for predicting cloud state; When the Kalman filter update within the preset time is completed, the specific calculation of the global cloud coverage state within the preset time at the kth moment is expressed as follows: In the formula, Cover (k) It is used to indicate the cloud coverage status corresponding to the k-th moment, that is, the cloud area within the prediction area at the k-th moment; Represents the grid G ij The initial cloud area; Represents the grid G ij The area change rate predicted by the Kalman filter at time r; The specific expression of the occlusion probability function for calculating the cloud occlusion probability at the kth moment is as follows: Where, represents the cloud occlusion probability at the kth moment; represents the historical occlusion probability term; is the historical occlusion weight at the kth moment, and its specific calculation is as follows: Where, represents the rate of change of the cloud coverage area at the kth moment; β cover represents the historical impact adjustment coefficient; Indicates the coverage degree of the current cloud group to the forecast area; represents the cloud diffusion probability gain at the kth moment, and its specific calculation is as follows: Where, γ diffusion represents the diffusion influence adjustment coefficient; It indicates the degree of influence of the saturation rate of change, where θ is the saturation threshold; Represents the remaining space of the predicted area, i.e., the space for diffusion; The specific calculation of the light intensity prediction value at the kth moment obtained by the light intensity attenuation model is as follows: Where, represents the final predicted light intensity value at the kth moment; Represents the historical compensation item, where w hist Represents the historical light intensity compensation weight coefficient, and its specific calculation is as follows: Where λ hist Indicates the attenuation rate of the impact of historical data; represents the rate of change of occlusion probability at the kth moment; represents the diffusion compensation term, and its specific calculation is as follows: Where θ v Indicates the cloud coverage area change rate threshold; I (k) The predicted value of light intensity at time k without historical compensation is expressed as follows: Where I0 represents the current clear sky radiation intensity; represents the effective optical path term, where w area is the cloud area attenuation coefficient, and its specific calculation is expressed as: Where, α type Indicates the cloud type coefficient; h cloud Indicates the thickness of the cloud layer; h ref Indicates the reference height of the cloud; represents the transmittance correction term, and μ represents the weight coefficient of cloud thickness. The specific calculation is as follows: μ=μ0·(1-exp(-δ μ ·h cloud )) Where μ0 represents the basic optical thickness coefficient; δ μ represents the thickness sensitivity coefficient, which is obtained by Mie scattering calculation; d eff It represents the equivalent path length correction factor, and its specific calculation is as follows: Where, represents the altitude angle of the sun at the kth moment; The photovoltaic power linear model is used for conversion to obtain the photovoltaic power prediction value at the kth moment as follows: Where, represents the final photovoltaic power generation power forecast value at the kth moment; It represents the photovoltaic power generation forecast value at the kth moment without historical smoothing, and its specific expression is as follows: Where η pv Indicates the comprehensive efficiency coefficient of photovoltaic modules; A pv is the total effective photovoltaic units of the photovoltaic power station; ∈ loss represents the photovoltaic power loss term; represents the historical smoothing compensation term; w pv is the dynamic smoothing weight of power generation, and its specific calculation is as follows: Where λ pv is the historical smoothing adjustment coefficient of photovoltaic power generation; is the rate of change of generated power.

10. The photovoltaic power generation intelligent scheduling device based on dynamic supply and demand forecasting according to claim 8, characterized in that: The specific method for obtaining the power load forecast curve within the preset time includes the following steps: S61. Sampling and normalizing historical power load data through a sliding window to obtain a power load data sequence; S62A, performing statistical calculation on the power load data sequence to obtain a statistical characteristic vector of the power load data; S62B, using one-dimensional Haar wavelet to perform wavelet decomposition on the power load data sequence to obtain high- and low-frequency feature vectors of the power load data; S63, concatenating and fusing the statistical feature vector and the high-frequency and low-frequency feature vectors of the power load data to obtain a fused feature vector; S64. Use the power load regression model to predict the fused feature vector to obtain a power load prediction curve within a preset time; After obtaining the power load forecast curve within the preset time, the specific expression of the difference calculation at the kth moment is as follows: Where, represents the predicted supply and demand power at the kth moment; represents the power load forecast value at the kth moment in the power load forecast curve; η gird Indicates the power transmission loss coefficient; δ safe is the safety margin, and its specific calculation is as follows: d safe =λ safe ·μ(wave high )+c safe ·σ(wave high ) Where λ safe represents the fluctuation intensity coefficient; γ safe represents the uncertainty coefficient; μ(wave high ) represents the mean of the high-frequency components in the high- and low-frequency feature vectors of the power load data; σ(wave high ) represents the standard deviation of the high-frequency components in the high- and low-frequency feature vectors of the electricity load data.

Images