Photovoltaic- energy storage rolling scheduling method and system based on ultra-short-term power prediction

Abstract

This application relates to the field of power dispatching technology, specifically a photovoltaic-energy storage rolling dispatching method and system based on ultra-short-term power prediction. The method includes: collecting multi-source heterogeneous data on electricity load, cloud change images, and photovoltaic power generation; constructing a cloud shading impact model through a neural network to obtain the mapping relationship between shading time, intensity, and power generation; predicting electricity load changes and photovoltaic power using historical load data and cloud images; integrating various data and calculating power differences, and balancing the difference through an energy storage system; constructing a bidirectional graph model of dynamic changes in photovoltaic and energy storage, and adjusting the dynamic connection and structure between nodes; and adopting rolling dispatching optimization to formulate a reasonable energy storage charging and discharging strategy to achieve dynamic balance and efficient operation of the photovoltaic-energy storage system.

Description

Technical Field

[0001] This application relates to the field of power dispatching technology, and specifically discloses a photovoltaic-energy storage rolling dispatching method and system based on ultra-short-term power prediction. Background Technology

[0002] Currently, distributed photovoltaic (PV) power generation and energy storage systems are increasingly widely used in power distribution networks, but they still face many challenges in actual operation. In particular, on ultra-short timescales, affected by cloud movement and complex weather conditions, PV power generation fluctuates drastically, making it difficult to achieve power balance between power generation and consumption.

[0003] Existing scheduling methods are unable to fully integrate heterogeneous data from multiple sources such as electricity load, photovoltaic power generation, and meteorology. They lack the ability to dynamically model and predict the impact of cloud cover, and cannot accurately reflect the impact of cloud changes on photovoltaic power generation, resulting in insufficient power prediction accuracy.

[0004] Furthermore, the dispatch strategies for photovoltaic-related energy storage are generally quite extensive, failing to flexibly adjust charging and discharging behavior based on real-time power differences and energy storage status, making it difficult to achieve efficient coordination and dynamic balance between photovoltaic and energy storage systems. Due to the lack of systematic methods for ultra-short-term power forecasting and rolling dispatch, the operating efficiency and power balancing capabilities of distributed photovoltaic and energy storage systems in distribution networks are significantly limited.

[0005] In view of this, this application proposes a photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction. Summary of the Invention

[0006] To achieve the above objectives, this invention provides a photovoltaic-energy storage rolling dispatch method and system based on ultra-short-term power prediction, the specific technical solution of which is as follows:

[0007] A photovoltaic-storage rolling dispatch method based on ultra-short-term power prediction includes:

[0008] Collect electricity load data, cloud change image data, and photovoltaic power generation data to form a multi-source heterogeneous dataset;

[0009] A cloud shading impact model was constructed, and a neural network was used to correlate cloud change image data with photovoltaic power generation data to establish a photovoltaic power generation prediction model based on cloud shading time, shading intensity and photovoltaic power generation.

[0010] Based on historical electricity load data, predict the time-series changes in electricity load. At the same time, based on cloud change image data, predict the timing of cloud shading on photovoltaic panels during cloud movement, and calculate the power generation of photovoltaic panels.

[0011] By integrating electricity load data, photovoltaic power generation data, and cloud impact data onto a unified time axis, the difference between photovoltaic power generation and electricity consumption is calculated, and the power difference is balanced by energy storage.

[0012] A two-way graph model of dynamic changes in photovoltaic and energy storage is constructed. The connection relationship between photovoltaic nodes and energy storage nodes is established in the two-way graph model. The graph model structure is dynamically adjusted according to the energy storage charging and discharging status. Energy storage is scheduled through the graph model.

[0013] Implement rolling scheduling optimization. Based on the current periodic graph model structure and photovoltaic power generation prediction results, formulate energy storage charging and discharging strategies. Store excess electricity and make up for power deficits through energy storage nodes to achieve dynamic balance of photovoltaic and energy storage power.

[0014] Preferably, smart meters are installed at each node of the distribution network to collect electricity load data at fixed time intervals. The electricity load data includes active power, reactive power, voltage, and current.

[0015] Cloud imaging cameras with solar tracking capabilities are deployed around the photovoltaic power station to continuously capture sky images and perform image preprocessing.

[0016] Power acquisition devices are installed in each photovoltaic array to collect photovoltaic power generation data, including photovoltaic output power, light intensity, module temperature and ambient temperature, and the geographical coordinates of each photovoltaic array are recorded.

[0017] Preferably, the collected cloud change image data is segmented to extract cloud feature images, and the sky area is divided into cloud area and clear sky area using a threshold segmentation method;

[0018] Calculate cloud coverage, cloud movement speed, and average gray value of cloud regions; construct a quantitative model of cloud shading intensity, defining cloud shading intensity as the ratio of measured light intensity under cloud shading to theoretical light intensity under clear sky conditions; and calculate the shading time of each photovoltaic array by the cloud based on cloud feature images and the geographical location of the photovoltaic arrays.

[0019] Preferably, a cloud occlusion impact model is designed, and a feedforward neural network containing an input layer, a hidden layer, and an output layer is constructed. The input layer receives the cloud feature vector and the photovoltaic power generation power at the previous moment, and the output layer outputs the predicted photovoltaic power attenuation coefficient. The constructed feedforward neural network is trained and optimized. The training set is constructed using historically synchronously collected cloud change image data and photovoltaic power generation power data, and the network weights are updated using an optimization algorithm.

[0020] A cloud shading impact model is established, which correlates cloud shading time and intensity with photovoltaic power attenuation coefficient to construct a photovoltaic power prediction model. The cloud impact data includes the timing of cloud shading on photovoltaic panels.

[0021] Preferably, a load forecasting model based on a long short-term memory network is established, in which the long short-term memory network predicts the load for future periods using historical load sequences;

[0022] The cloud movement trajectory is predicted based on a cloud shading impact model. A continuous sequence of cloud change images is used to predict the cloud movement trajectory, and the start time of shading of each photovoltaic array is calculated based on the predicted cloud trajectory. The cloud impact data includes the timing of cloud shading of the photovoltaic panels.

[0023] Preferably, the photovoltaic power generation power blocked by clouds is calculated. For each photovoltaic array, the photovoltaic power generation power is calculated as the product of the conversion efficiency, effective area, solar irradiance, and shading coefficient of the photovoltaic array. The total photovoltaic power generation power is the sum of the power generation power of all photovoltaic arrays, and the photovoltaic power generation power prediction results are corrected and optimized.

[0024] Preferably, time synchronization and alignment are performed on multi-source data, including electricity load, cloud change images, and photovoltaic power generation, to establish a reference time series;

[0025] Calculate the real-time power difference between electricity consumption and photovoltaic power generation, calculate the net power difference at any time, introduce a safety margin factor to correct the power difference, and calculate the cumulative amount of the power difference.

[0026] Preferably, the charging and discharging requirements of energy storage are determined based on the sign and magnitude of the power difference. When there is a power surplus, the charging power requirement is determined, and when there is a power deficit, the discharging power requirement is determined.

[0027] The change in the state of charge of energy storage is calculated based on the state of charge, charging and discharging power, charging and discharging efficiency, and rated capacity of energy storage at the previous moment.

[0028] Formulate a power balance strategy, establish power balance constraints, and define an energy storage priority function.

[0029] Preferably, a bidirectional graph model structure of photovoltaic-energy storage is constructed, defining a set of nodes and a set of directed edges. The set of nodes includes a subset of photovoltaic nodes and a subset of energy storage nodes. The attributes of photovoltaic nodes include power generation and azimuth, and the attributes of energy storage nodes include capacity, maximum charge and discharge power, and state of charge.

[0030] In a bidirectional graph model, a bidirectional connection relationship is established between photovoltaic and energy storage nodes. The charging path from the photovoltaic node to the energy storage node and the discharging path from the energy storage node to the photovoltaic node are defined. Each edge carries a weight attribute that considers electrical distance and transmission efficiency.

[0031] Preferably, the feature is that a dynamic edge deletion mechanism is created based on a bidirectional graph model of charge and discharge states to define the operating state of the energy storage nodes;

[0032] When an energy storage node is in a charging state, delete the edges pointing from the charging energy storage node to the photovoltaic node; when an energy storage node is in a discharging state, delete all edges pointing from the photovoltaic node to the discharging energy storage node.

[0033] The photovoltaic-energy storage bidirectional graph model is periodically updated, including updating the graph's topology and node attributes, updating edge weights based on predicted changes in photovoltaic power generation, and monitoring the graph's connectivity.

[0034] Preferably, a rolling scheduling time window framework is established, the rolling optimization time domain and control time domain are defined, and the rolling time domain is divided into multiple time periods. At the beginning of each rolling cycle, a scheduling plan is formulated based on the updated graph model structure and photovoltaic power generation prediction results.

[0035] An optimization objective function is constructed that is linked to the dynamic edge deletion mechanism of the bidirectional graph model. The objective function includes the remaining power imbalance, operating cost, and energy storage degradation loss.

[0036] Formulate an energy storage charging and discharging strategy based on the graph model state, establish power flow constraints based on the set of effective edges in the graph model, and define an energy storage scheduling priority index.

[0037] A photovoltaic-energy storage rolling dispatch system based on ultra-short-term power prediction is used to implement the photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction. It includes: a data acquisition module, a cloud shading calculation module, a power calculation module, an energy storage balance module, a dispatch energy storage module, and a rolling dispatch optimization module.

[0038] The data acquisition module is used to collect electricity load data, cloud change image data, and photovoltaic power generation data to form a multi-source heterogeneous dataset.

[0039] The cloud shading calculation module is used to construct a cloud shading impact model. It uses a neural network to perform correlation analysis between cloud change image data and photovoltaic power generation data, and establishes a photovoltaic power generation prediction model between cloud shading time, shading intensity and photovoltaic power generation.

[0040] The power calculation module is used to predict the temporal changes in electricity load based on historical electricity load data, and at the same time, to predict the timing of cloud shading of photovoltaic panels when clouds move based on cloud change image data, and to calculate the power generation of photovoltaic panels.

[0041] The energy storage balancing module is used to integrate electricity load data, photovoltaic power generation data and cloud impact data onto a unified time axis, calculate the difference between photovoltaic power generation and electricity consumption, and balance the power difference through energy storage.

[0042] The energy storage scheduling module is used to construct a two-way graph model of dynamic changes in photovoltaic and energy storage, establish the connection relationship between photovoltaic nodes and energy storage nodes in the two-way graph model, dynamically adjust the graph model structure according to the energy storage charging and discharging status, and schedule energy storage through the graph model.

[0043] The rolling scheduling optimization module is used to implement rolling scheduling optimization. Based on the current periodic graph model structure and photovoltaic power generation prediction results, it formulates energy storage charging and discharging strategies, stores excess electrical energy and makes up for power deficits through energy storage nodes, and achieves dynamic balance of photovoltaic and energy storage power.

[0044] The beneficial effects of this invention are as follows: This application achieves comprehensive perception of electricity load, weather changes and photovoltaic power generation, enriches the data foundation, provides support for subsequent multi-dimensional analysis and prediction, and helps to improve the accuracy of power prediction and the timeliness of dispatch response.

[0045] This application uses deep learning methods to achieve accurate modeling of the impact of cloud cover on photovoltaic power generation, which improves the ability to predict photovoltaic power fluctuations and enhances the system's adaptability and robustness to complex weather changes.

[0046] This application achieves time-series alignment and dynamic fusion of multi-source data, which can reflect the power supply and demand balance in real time, providing a basis for precise adjustment of energy storage systems and effectively reducing the impact of power fluctuations on the distribution network.

[0047] This application uses a graphical model to dynamically reflect the operating status and interaction relationship of photovoltaic and energy storage devices, thereby enabling intelligent and flexible scheduling of the energy storage system and enhancing the overall collaborative optimization capability of the system.

[0048] This application, based on predictive rolling optimization scheduling, effectively improves energy storage utilization efficiency and system dynamic balance capability, reduces grid operation risk, and increases the photovoltaic consumption ratio and the economics of distributed energy. Attached Figure Description

[0049] Figure 1 A flowchart of the photovoltaic-energy storage rolling scheduling method based on ultra-short-term power prediction provided by the present invention;

[0050] Figure 2Flowchart for preparing distributed photovoltaic-energy storage scheduling data provided by this invention;

[0051] Figure 3 Flowchart for constructing the cloud occlusion impact model provided by this invention;

[0052] Figure 4 A flowchart for accurate power prediction provided by this invention;

[0053] Figure 5 This invention provides a flowchart for the scheduling of a photovoltaic-energy storage dynamic graph model.

[0054] Figure 6 The structural diagram of the photovoltaic-energy storage rolling dispatch system based on ultra-short-term power prediction provided by the present invention. Detailed Implementation

[0055] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0056] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention can also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0057] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0058] Example 1

[0059] Reference Figures 1 to 5 The accompanying drawings are for the first embodiment of the present invention, as shown below. Figure 1 As shown, a photovoltaic-energy storage rolling scheduling method based on ultra-short-term power prediction is provided.

[0060] Step 1: Collect electricity load data, cloud change image data, and photovoltaic power generation data to form a multi-source heterogeneous dataset; see [link / reference] Figure 2 This is a flowchart for preparing distributed photovoltaic-energy storage scheduling data for this step.

[0061] Electricity load data is collected by installing smart meters at various nodes of the distribution network at fixed time intervals. Collect electricity load data, including active power. reactive power ,Voltage and current ,in express Active power at any given time express Reactive power at any given moment express Voltage value at time, express The current value at a given moment. For example, It can be set to 1 minute. By collecting electricity load data at high frequency, it is possible to accurately capture the dynamic changes in load in the short term, providing a refined data foundation for subsequent load forecasting.

[0062] Cloud layer change image data is collected, and cloud imaging cameras with solar tracking capabilities are deployed around the photovoltaic power station. These cameras track the sun via solar trackers and are controlled to ensure they are always aligned with the direction of sunlight incidence. The cloud imaging cameras then display images at time intervals. Continuous capture of sky images, for example, The timer can be set to 30 seconds. Preprocessing is performed on the acquired raw image, including image denoising, contrast enhancement, and size normalization. The RGB color image is converted to a grayscale image. ,in Represents grayscale value, and These represent the horizontal and vertical coordinates of the image, respectively. Indicates the acquisition time. The difference between images at adjacent time points is calculated. ,in express The image grayscale difference values ​​at different times are used to extract the motion features of the clouds. By acquiring cloud images and using the difference processing method, the movement trajectory and speed of the clouds can be accurately identified, providing reliable image data support for predicting cloud shading of photovoltaic panels.

[0063] Collect photovoltaic power generation data by installing power acquisition devices in each photovoltaic array at time intervals. Collect photovoltaic power generation data. For example, The time can be set to 30 seconds. The collected data includes photovoltaic output power. Light intensity Component temperature and ambient temperature ,in express Photovoltaic output power at any given time express Light intensity at any given time express The temperature of the photovoltaic module at any given time. express The ambient temperature at any given time. The geographical coordinates of each photovoltaic array are also recorded. ,in Indicates the first The longitude of each photovoltaic array Indicates the first The latitude of each photovoltaic array. By comprehensively collecting power data and environmental parameters of photovoltaic power generation, the actual operating status of photovoltaic power generation can be fully reflected.

[0064] Construct a multi-source heterogeneous dataset of electricity load, cloud images, and photovoltaic power generation. ,in This represents a dataset of electricity loads. This represents a dataset of cloud images. This represents a dataset of photovoltaic power generation.

[0065] This step, through the collection and integration of the aforementioned multi-source heterogeneous data, forms a comprehensive dataset containing multi-dimensional information on electricity load, cloud changes, and photovoltaic power generation. This provides a comprehensive and accurate data foundation for subsequent power prediction and scheduling optimization, and improves the accuracy and reliability of distributed photovoltaic-energy storage system scheduling.

[0066] Step 2: Construct a cloud shading impact model, using a neural network to correlate cloud change data with photovoltaic power generation data, establishing a mapping relationship between cloud shading time, shading intensity, and photovoltaic power generation; see [link / reference] Figure 3 This is a flowchart for constructing the cloud occlusion impact model in this step.

[0067] The collected cloud image data was segmented to extract cloud feature images; a threshold segmentation method was used to divide the sky area into cloud areas and clear sky areas. Cloud coverage was defined. ,in express Cloud coverage at any given time express The pixel area covered by clouds in a time-lapse image. This represents the total pixel area of ​​the image. The cloud movement speed is calculated using optical flow. ,in express The speed of cloud movement at all times and These represent the horizontal and vertical velocity components of the cloud layer's movement in the image, respectively. The average grayscale value of the cloud region is calculated as the grayscale feature of the cloud layer. ,in express The average gray value of the cloud region at any given time. This indicates the number of pixels in the cloud area. Indicates the first The grayscale value of each pixel. Multi-dimensional feature extraction methods for clouds can comprehensively characterize the morphology and motion characteristics of clouds, providing accurate feature information for subsequent occlusion impact analysis.

[0068] Construct a quantitative model of cloud occlusion intensity and define cloud occlusion intensity. ,in express The cloud cover intensity at any given time, with values ​​ranging from [value range missing]. , This represents the measured light intensity when cloud cover is present. This represents the light intensity under theoretical clear-sky conditions. Based on cloud images and the geographical location of the photovoltaic arrays, the duration of cloud shading on each photovoltaic array is calculated. ,in, Indicates from From the moment the clouds completely obscure the view... The time required for each photovoltaic array express From the edge of the cloud layer to the first The distance between the photovoltaic arrays was determined. By quantifying the intensity and duration of cloud obstruction, the impact of clouds was accurately characterized.

[0069] Design a neural network model to address the impact of cloud occlusion, constructing a feedforward neural network comprising an input layer, hidden layers, and an output layer; the input layer receives feature vectors. ,in, express The input feature vector at time t, express The photovoltaic power generation capacity at the previous moment. Two hidden layers are set; the first hidden layer contains... There are 10 neurons, and the activation function is the ReLU function. ,in, The second hidden layer contains the input values ​​of the neurons; There are 10 neurons, and the activation function is the tanh function. The output layer outputs the predicted photovoltaic power attenuation coefficient. ,in express The photovoltaic power attenuation coefficient caused by cloud cover at any given time, with a value range of [value missing]. The constructed multi-layer neural network structure can effectively capture the nonlinear relationship between cloud features and photovoltaic power.

[0070] The constructed multi-layer neural network is trained and optimized using historically collected cloud image data and photovoltaic power generation data to build the training set. A loss function is defined. ,in, This represents the mean squared error loss. Indicates the number of training samples. Indicates the first The predicted attenuation coefficient for each sample. Indicates the first The actual attenuation coefficient of each sample. The Adam optimization algorithm is used to update the network weights, and the learning rate is set to adaptive adjustment. Iterative training is performed using the backpropagation algorithm until the loss function converges. After training, the mapping relationship between cloud features and photovoltaic power attenuation is obtained. ,in This represents the mapping function of the trained neural network.

[0071] A comprehensive impact model of cloud shading is established, correlating cloud shading time, shading intensity, and photovoltaic power attenuation coefficient to construct a photovoltaic power prediction model. ,in, express Photovoltaic power forecasts must take into account the impact of cloud cover. express The maximum photovoltaic power under ideal conditions at any given time. For cloud shading predictions for future times, the shading start time for each photovoltaic array is calculated based on the cloud movement trajectory and speed. and duration of occlusion ,in, Indicates the first The moment when the photovoltaic array begins to be shaded Indicates the first The duration of time a photovoltaic array is shaded This indicates the length of the cloud layer in the direction of movement.

[0072] By constructing a cloud shading impact model, we achieved a precise correlation analysis between cloud changes and photovoltaic power generation, established a dynamic mapping relationship between cloud shading time, shading intensity and photovoltaic power, significantly improved the accuracy of ultra-short-term photovoltaic power prediction, and provided a reliable model foundation for subsequent power prediction and scheduling optimization.

[0073] Step 3: Predict temporal changes in electricity load based on historical electricity load data, and simultaneously predict the timing of cloud shading of photovoltaic panels based on cloud image data, then calculate the photovoltaic panel power generation; see [link to relevant documentation]. Figure 4 This is a flowchart of the power accuracy prediction process for this step.

[0074] A load forecasting model based on Long Short-Term Memory (LSTM) networks is established, and an LSTM network is constructed. The LSTM network uses historical load data to predict future loads. The input layer of the LSTM network receives historical load sequences. ,in Represents the historical load input vector. Indicates at time The former Load values ​​at each time step This represents the time step of historical data. Through the gating mechanism of the LSTM network, long-term dependencies in load data are captured to obtain load forecasts for future times. ,in, express Forecasted load at any given time This indicates the prediction time domain.

[0075] To predict cloud movement trajectories, based on the cloud occlusion impact model constructed in step 2, a series of continuous cloud image sequences are used. The centroid coordinates of the clouds are then calculated. ,in and They represent The horizontal and vertical coordinates of the cloud centroid at any given time. and Indicates the cloud region number 1 The horizontal and vertical coordinates of each pixel are used to predict the movement trajectory of the cloud centroid using the Kalman filter algorithm. The state equation is: ,in express The state vector at time t, Represents the state transition matrix. This indicates process noise.

[0076] Based on the predicted cloud trajectory, calculate the start time when each photovoltaic array is shaded. ,in Indicates the first The moment when the photovoltaic array begins to be shaded express From the edge of the cloud layer to the first The distance between each photovoltaic array. The trajectory prediction method based on image sequences can accurately predict the timing of cloud shading of each photovoltaic array.

[0077] Calculate the photovoltaic power generation considering cloud cover, for the first... The formula for calculating the power generation of a photovoltaic array is as follows: ,in Indicates the first A photovoltaic array in Power generation at any given moment Indicates the first The conversion efficiency of a photovoltaic array Indicates the first The effective area of ​​each photovoltaic array express Solar irradiance at any given time Indicates the first A photovoltaic array in The occlusion coefficient at any given time. When hour, ;when hour, ,in Indicates the first The duration of shading for each photovoltaic array and These are the shading intensity and power attenuation coefficient obtained in step 2, respectively. The total photovoltaic power generation is... ,in express The total power generation of all photovoltaic arrays at any given time. This indicates the total number of photovoltaic arrays.

[0078] By establishing a time-series electricity load forecasting model and a photovoltaic power calculation method under the influence of cloud cover, accurate forecasting of load demand and photovoltaic power generation was achieved. The dynamic impact of cloud movement on photovoltaic power generation was fully considered, providing accurate power forecasting data for subsequent power balance and energy storage scheduling, and improving the scheduling accuracy and operating efficiency of distributed photovoltaic-energy storage systems.

[0079] Step 4: Integrate the electricity load data, photovoltaic power generation data, and cloud impact data onto a unified time axis, calculate the difference between power generation and power consumption, and balance the power difference through energy storage.

[0080] Time synchronization and alignment of multi-source data, including electricity load, cloud imagery, and photovoltaic power generation, will be performed on the electricity load forecast data obtained in step 3. Photovoltaic power generation forecast data In addition, cloud impact data is unified along a timeline, including the timing of cloud shading of photovoltaic panels. A baseline time series is established. ,in Representing a unified time series, Indicates the start time of scheduling. Indicates the scheduling time step. Indicates the number of scheduling periods. For example, Set to 1 minute. For data with inconsistent sampling intervals, use cubic spline interpolation for time resampling to ensure all data have corresponding values ​​at the same time. Define a time alignment function. ,in, Indicates the first This function maps data from different sampling frequencies onto a unified time series, similar to the original data. Time synchronization processing eliminates time discrepancies between multi-source data, laying the foundation for accurate calculation of power differences.

[0081] Calculate the real-time power difference between power consumption and power generation at any given time. Calculate the net power difference ,in express The net power difference at any given time is represented by a positive value indicating a photovoltaic power surplus and a negative value indicating a power deficit. A safety margin factor is introduced to account for the impact of prediction errors. The corrected power difference is ,in This represents the power difference after considering the safety margin. Represents a symbolic function. For example, Set to 0.05. Simultaneously, calculate the cumulative amount of the power difference. ,in Indicates from the start time of scheduling to The cumulative energy difference at time intervals, For scheduling time The power difference, This represents the time index from the start of the scheduling to time t. The real-time power difference calculation method can accurately reflect the power balance between supply and demand.

[0082] Analyze the energy storage balance demand and determine the charging and discharging requirements of energy storage based on the sign and magnitude of the power difference. When When this occurs, it indicates a power surplus, and energy storage needs to absorb the excess electrical energy, requiring a charging power of [value missing]. ,in express The charging power requirement at any time, This indicates the maximum charging power of the energy storage system. When... When this occurs, it indicates a power deficit, and the energy storage system needs to release electrical energy, with a discharge power requirement of [value missing]. ,in express Discharge power requirements at any given time This represents the maximum discharge power of the energy storage system. Calculate the change in the state of charge of the energy storage system. ,in express State of charge at time t, Indicates charging efficiency. Indicates discharge efficiency. This indicates the rated capacity of the energy storage system.

[0083] Develop a power balance strategy. Establish power balance constraints. ,in express The actual discharge power of the stored energy at all times. Indicates the actual charging power. This represents the power exchanged with the grid. When the energy storage system cannot completely balance the power difference, power regulation is performed through the grid. Define the energy storage priority function. ,in express Energy storage scheduling priority at any given time. Indicates the system's rated power. When... Priority should be given to using energy storage for balancing, among which This indicates the priority threshold; the hierarchical power balancing strategy can maximize the utilization of energy storage resources and reduce dependence on the power grid.

[0084] Controlling the charging and discharging process of energy storage and defining power balance indices. ,in express The power self-balance at time t, with a value range of . A higher value indicates a stronger self-balancing capability. The average self-balancing degree within the scheduling cycle is calculated. ,in, This represents the average self-balancing degree. Monitoring the state boundary of the energy storage system ensures... ,in and These represent the lower and upper limits of the state of charge, respectively. When approaching the limits, the energy storage power allocation strategy is adjusted to prevent overcharging or over-discharging.

[0085] This step integrates multi-source data into a unified time axis and calculates the power difference, achieving precise matching analysis between photovoltaic power generation and electricity demand. By regulating the charging and discharging of energy storage, the power difference is effectively balanced, improving the autonomous operation capability of distributed photovoltaics, reducing dependence on the external power grid, and providing a key power balance mechanism for building an efficient photovoltaic-energy storage coordinated dispatch system.

[0086] Step 5: Construct a dynamic two-way graph model of photovoltaic (PV) and energy storage, establish connections between PV nodes and energy storage nodes within the model, dynamically adjust the graph model structure based on the energy storage's charging and discharging status, and schedule energy storage through the graph model; see [link / reference]. Figure 5 This is the scheduling flowchart for the photovoltaic-energy storage dynamic graph model in this step.

[0087] Construct the basic structure of the photovoltaic-energy storage bidirectional graph model; define the bidirectional graph model. ,in Represents a set of nodes. This represents the set of directed edges. The node set contains a subset of photovoltaic nodes. and energy storage node subset ,in Indicates the first One photovoltaic node, Indicates the first One photovoltaic node, , Indicates the total number of photovoltaic nodes; Indicates the first One energy storage node, Indicates the first One energy storage node, Indicates the total number of energy storage nodes. Each photovoltaic node Carrying attributes ,in Indicates the first Each photovoltaic node is in Power generation at any given moment This indicates the azimuth angle of the photovoltaic array. Each energy storage node... Carrying attributes ,in Indicates the first The capacity of each energy storage unit Indicates the maximum charging power. Indicates the maximum discharge power. Indicates the first Each energy storage unit in The state of charge at any given time. The parameters of the actual physical device are represented by a bidirectional graph model through the definition of node attributes.

[0088] In a bidirectional graph model, a bidirectional connection is established between photovoltaic (PV) and energy storage nodes. For any PV node... and energy storage nodes Establish bidirectional edges and ,in Indicates from photovoltaic node To energy storage nodes The directed edges represent charging paths. Indicates from energy storage node To photovoltaic nodes The directed edges represent discharge paths. Each edge carries a weight attribute. ,in express Time Side The weight, Indicates photovoltaic node With energy storage nodes Electrical distance between them Indicates photovoltaic node With energy storage nodes Transmission efficiency between them Indicates photovoltaic node With energy storage nodes Electrical distance between them Indicates photovoltaic node With energy storage nodes The transmission efficiency between them. Similarly, define... The weights represent the discharge path weights. The weights are designed considering electrical distance and transmission efficiency, ensuring that power preferentially flows to energy storage units that are closer and more efficient.

[0089] A dynamic edge-removal mechanism is created based on a bidirectional graph model of charge and discharge states, and the operating state function of the energy storage node is defined. ,in Indicates the first One energy storage node in The running status at any given moment, This represents the power of the node; a positive value indicates charging, a negative value indicates discharging, and zero indicates standby. When... When charging, perform the edge deletion operation. ,in This represents the set of edges after removing all nodes from the energy storage node. Edges pointing to photovoltaic nodes. When During the discharge state, perform the edge deletion operation. This means deleting all connections between photovoltaic nodes and energy storage nodes. The edges. For example, when energy storage node 1 is in a charging state, all edges are deleted. Edges of this type indicate that the energy storage node does not output power during the charging process. The dynamic edge deletion mechanism avoids power flow conflicts in the bidirectional graph model, ensuring the physical feasibility of the photovoltaic-energy storage structure and guaranteeing the safety of energy storage during operation.

[0090] The energy storage scheduling algorithm is implemented based on a graph model. At each scheduling time, the power difference calculated in step 4 is used. Energy storage power is allocated through a graphical model; when At that time, calculate the rechargeable power of each energy storage node. ,in Indicates the first One energy storage node in Available charging power at any given time Indicates the upper limit of the state of charge. This represents the scheduling time step. Based on the graph-based maximum flow algorithm, it calculates the power flow allocation from the photovoltaic node to the energy storage node. ,in express From photovoltaic nodes Flow to energy storage nodes The power was calculated; through power flow calculation using a graphical model, coordinated scheduling of multiple photovoltaic and energy storage systems was achieved.

[0091] The bidirectional graph model of photovoltaic-energy storage is updated periodically. At the end of each scheduling cycle, the topology and node attributes of the graph are updated, and the state of charge of the energy storage nodes is updated. ,in Indicates the first The charging efficiency of each energy storage unit. The edge weights are updated based on the predicted changes in photovoltaic power. Monitor the connectivity of the graph and define connectivity metrics. ,in express Graph connectivity at time step Indicates the current number of valid edges. Indicates the initial total number of edges. When At that time, among them, This represents the minimum connectivity threshold, which triggers graph structure optimization and adjustment.

[0092] This step constructs a dynamic bidirectional graph model of photovoltaic-energy storage, realizing a structured representation of the complex structure of photovoltaic-energy storage. The dynamic edge deletion mechanism effectively avoids charging and discharging conflicts, and the graph-based power flow algorithm realizes optimized scheduling among multiple nodes, improving the utilization efficiency of energy storage resources and enhancing the flexibility and reliability of the system. It provides an innovative graph theory solution for the intelligent scheduling of large-scale distributed photovoltaic-energy storage systems.

[0093] Step 6: Implement rolling scheduling optimization. Based on the current periodic graph model structure and power prediction results, formulate energy storage charging and discharging strategies to store excess energy and make up for power deficits through energy storage nodes, achieving dynamic balance of photovoltaic-energy storage power.

[0094] Establish a time window framework for rolling scheduling and define the rolling optimization time domain. and control time domain ,in Indicates the time length for prediction and optimization. Indicates the actual duration of control execution, and satisfies Divide the rolling time domain into A period of time, namely ,in This indicates the time interval for the rolling schedule. For example, Set to 2 hours. Set to 30 minutes. Set to 5 minutes. At the start of each rolling cycle. Based on the updated graph model structure in step 5 Based on the power prediction results obtained in step 3, formulate future... Scheduling plan within the time period. Execution control time domain. After a scheduling instruction is issued, the time window rolls forward one control time domain to enter the next optimization cycle. The rolling optimization mechanism can dynamically adjust the scheduling strategy based on the latest forecast information and the status of photovoltaic-energy storage, thereby improving the adaptability of scheduling.

[0095] Construct an optimization objective function that is linked to the dynamic edge-removal mechanism of the bidirectional graph model, and define the objective function for rolling optimization. ,in This represents the overall optimization objective value. Indicates the time period index for the rolling time domain division. Indicates the starting time point for rolling optimization. express The remaining power imbalance at time . Indicates operating costs, Indicates energy storage degradation loss, , , These represent the weight coefficients for each item.

[0096] Residual power imbalance The calculation formula is ,in Indicates the total photovoltaic power. Indicates load power. Indicates the first The charging and discharging power of each energy storage unit. Operating costs include charging and discharging losses. ,in and These represent the unit cost coefficients for charging and discharging, respectively. and These represent the actual charging and discharging power, respectively. Through multi-objective optimization, a comprehensive consideration of power balance, economy, and equipment lifespan is achieved.

[0097] Formulate an energy storage charging and discharging strategy based on the state of the graph model, according to the set of valid edges in the current graph model. Establish power flow constraints and ,in and These represent the charging and discharging power flows, respectively. This indicates the maximum charging power limit. This indicates the maximum discharge power limit. and The indicator functions represent the states when the energy storage node... The value is 1 when the device is charging or discharging, and 0 otherwise. Define the energy storage scheduling priority index. ,in, Indicates the first Each energy storage unit in The scheduling priority at any given moment Indicates the reference state of charge. This represents the health status coefficient of the energy storage unit. Higher-priority energy storage units participate in charge / discharge scheduling first. The strategy based on the graph model state ensures the consistency of scheduling decisions with the photovoltaic-energy storage topology.

[0098] Implement a dynamic allocation algorithm for energy storage power, and at each scheduling time, allocate power based on the predicted power difference sequence. The energy storage power allocation is solved using a quadratic programming method. Constraints are established: state of charge constraints. Power constraints and connectivity constraints based on graph models (While charging) or (During discharge), where and The adjacency matrix elements of the graph are represented. The optimal energy storage power sequence is obtained by solving the problem using the Lagrange multiplier method. The optimization results are mapped to a graphical model to update the power flow distribution. , Let be the optimal energy storage power at time t. In the graphical model The edge weights connecting node a to node b are defined. The dynamic power allocation algorithm ensures the efficient utilization of energy storage resources.

[0099] Feedback correction and balance verification are performed on the rolling schedule in the control time domain. After execution, actual power data is collected, and prediction error is calculated. and ,in, , They represent Actual and predicted values ​​of photovoltaic power generation at any given time. , They represent The actual and predicted load power values ​​at each time point are calculated. Based on error information, a Kalman filter is used to correct the prediction model parameters for the next cycle. A power balance metric is defined. ,in express The power balance at any given time; a value closer to 1 indicates a better balance. The average balance during the rolling cycle is monitored. .when At that time, among them This represents the balance threshold, which triggers adaptive adjustments to the scheduling strategy.

[0100] This step implements rolling scheduling optimization, dynamically formulating energy storage charging and discharging strategies based on real-time updated graphical model structures and power prediction results. It achieves closed-loop control of prediction, optimization, execution, and feedback, enabling the energy storage system to flexibly respond to fluctuations in photovoltaic power and load, effectively store excess energy, and promptly compensate for power deficits. This significantly improves the power balance capability and operational reliability of distributed photovoltaic-energy storage systems, providing an advanced scheduling method for achieving stable consumption of a high proportion of renewable energy.

[0101] Example 2

[0102] Reference Figure 6 This is the second embodiment of the present invention, which provides a photovoltaic-energy storage rolling dispatch system based on ultra-short-term power prediction.

[0103] The system includes: a data acquisition module, a cloud cover calculation module, a power calculation module, an energy storage balance module, a scheduling energy storage module, and a rolling scheduling optimization module.

[0104] The data acquisition module is used to collect electricity load data, cloud change image data, and photovoltaic power generation data to form a multi-source heterogeneous dataset.

[0105] The cloud shading calculation module is used to construct a cloud shading impact model. It uses a neural network to perform correlation analysis between cloud change data and photovoltaic power generation data, and establishes a mapping relationship between cloud shading time, shading intensity and photovoltaic power generation.

[0106] The power calculation module is used to predict temporal changes in electricity load based on historical electricity load data, and at the same time, to predict the timing of cloud shading of photovoltaic panels when clouds move based on cloud image data, and to calculate the power generation of photovoltaic panels.

[0107] The energy storage balancing module is used to integrate electricity load data, photovoltaic power generation data, and cloud impact data onto a unified time axis, calculate the difference between power generation and power consumption, and balance the power difference through energy storage.

[0108] The energy storage scheduling module is used to construct a two-way graph model of dynamic changes in photovoltaic and energy storage, establish the connection relationship between photovoltaic nodes and energy storage nodes in the two-way graph model, dynamically adjust the graph model structure according to the energy storage charging and discharging status, and schedule energy storage through the graph model.

[0109] The rolling scheduling optimization module is used to implement rolling scheduling optimization. Based on the current periodic graph model structure and power prediction results, it formulates energy storage charging and discharging strategies, stores excess electrical energy and makes up for power deficits through energy storage nodes, and achieves dynamic balance of photovoltaic-energy storage power.

[0110] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0111] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

Claims

1. A photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction, characterized in that, include: Collect electricity load data, cloud change image data, and photovoltaic power generation data to form a multi-source heterogeneous dataset; A cloud shading impact model was constructed, and a neural network was used to correlate cloud change image data with photovoltaic power generation data to establish a photovoltaic power generation prediction model based on cloud shading time, shading intensity and photovoltaic power generation. Based on historical electricity load data, predict the time-series changes in electricity load. At the same time, based on cloud change image data, predict the timing of cloud shading on photovoltaic panels during cloud movement, and calculate the power generation of photovoltaic panels. By integrating electricity load data, photovoltaic power generation data, and cloud impact data onto a unified time axis, the difference between photovoltaic power generation and electricity consumption is calculated, and the power difference is balanced by energy storage. A two-way graph model of dynamic changes in photovoltaic and energy storage is constructed. The connection relationship between photovoltaic nodes and energy storage nodes is established in the two-way graph model. The graph model structure is dynamically adjusted according to the energy storage charging and discharging status. Energy storage is scheduled through the graph model. Implement rolling scheduling optimization, and formulate energy storage charging and discharging strategies based on the current periodic graph model structure and photovoltaic power generation prediction results, so as to store excess energy and make up for power deficits through energy storage nodes; The collected cloud change image data is segmented to extract cloud feature images, and the sky area is divided into cloud area and clear sky area using the threshold segmentation method. Calculate cloud coverage, cloud movement speed, and average grayscale value of cloud regions; construct a quantitative model of cloud shading intensity, defining cloud shading intensity as the ratio of measured illumination intensity under cloud shading to theoretical illumination intensity under clear sky conditions; and calculate the shading time of each photovoltaic array based on cloud feature images and the geographical location of the photovoltaic arrays. A model of cloud occlusion impact is designed, and a feedforward neural network containing an input layer, a hidden layer, and an output layer is constructed. The input layer receives the cloud feature vector and the photovoltaic power generation power at the previous moment, and the output layer outputs the predicted photovoltaic power attenuation coefficient. The constructed feedforward neural network is trained and optimized. Historically synchronously collected cloud change image data and photovoltaic power generation power data are used to construct the training set, and the network weights are updated using an optimization algorithm. A cloud shading impact model is established, which correlates cloud shading time, shading intensity, and photovoltaic power generation attenuation coefficient to construct a photovoltaic power generation prediction model. The cloud impact data includes the timing of cloud shading on photovoltaic panels. Establish a load forecasting model based on long short-term memory networks, which use historical load sequences to predict loads in future periods; The cloud movement trajectory is predicted based on a cloud shading impact model. Continuous cloud change image sequences are used to predict the cloud movement trajectory, and the start time of shading for each photovoltaic array is calculated based on the predicted cloud trajectory.

2. The photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction according to claim 1, characterized in that, Smart meters are installed at each node of the power distribution network to collect electricity load data at fixed time intervals. The electricity load data includes active power, reactive power, voltage and current. Cloud imaging cameras with solar tracking capabilities are deployed around the photovoltaic power station to continuously capture sky images and perform image preprocessing. Power acquisition devices are installed in each photovoltaic array to collect photovoltaic power generation data, including photovoltaic output power, light intensity, module temperature and ambient temperature, and the geographical coordinates of each photovoltaic array are recorded.

3. The photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction according to claim 2, characterized in that, The photovoltaic power generation power blocked by clouds is calculated. For each photovoltaic array, the photovoltaic power generation power is calculated as the product of the photovoltaic array's conversion efficiency, effective area, solar irradiance, and shading coefficient. The total photovoltaic power generation power is the sum of the power generation power of all photovoltaic arrays. The photovoltaic power generation power prediction results are then corrected and optimized.

4. The photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction according to claim 3, characterized in that, Time synchronization and alignment of multi-source data, including electricity load, cloud change images, and photovoltaic power generation, are performed to establish a baseline time series. Calculate the real-time power difference between electricity consumption and photovoltaic power generation, calculate the net power difference at any time, introduce a safety margin factor to correct the power difference, and calculate the cumulative amount of the power difference.

5. The photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction according to claim 4, characterized in that, The charging and discharging requirements of energy storage are determined based on the sign and magnitude of the power difference. When there is a power surplus, the charging power requirement is determined, and when there is a power deficit, the discharging power requirement is determined. The change in the state of charge of energy storage is calculated based on the state of charge, charging and discharging power, charging and discharging efficiency, and rated capacity of energy storage at the previous moment. Formulate a power balance strategy, establish power balance constraints, and define an energy storage priority function.

6. The photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction according to claim 5, characterized in that, A bidirectional graph model structure for photovoltaic-energy storage is constructed, defining a set of nodes and a set of directed edges. The set of nodes contains a subset of photovoltaic nodes and a subset of energy storage nodes. The attributes of photovoltaic nodes include power generation and azimuth, while the attributes of energy storage nodes include capacity, maximum charge and discharge power, and state of charge. In a bidirectional graph model, a bidirectional connection relationship is established between photovoltaic and energy storage nodes. The charging path from the photovoltaic node to the energy storage node and the discharging path from the energy storage node to the photovoltaic node are defined. Each edge carries a weight attribute that considers electrical distance and transmission efficiency.

7. The photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction according to claim 6, characterized in that, A dynamic edge deletion mechanism is created based on a bidirectional graph model of charge and discharge states to define the operating states of energy storage nodes; When an energy storage node is in a charging state, delete the edges pointing from the charging energy storage node to the photovoltaic node; when an energy storage node is in a discharging state, delete all edges pointing from the photovoltaic node to the discharging energy storage node. The photovoltaic-energy storage bidirectional graph model is periodically updated, including updating the graph's topology and node attributes, updating edge weights based on predicted changes in photovoltaic power generation, and monitoring the graph's connectivity.

8. The photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction according to claim 7, characterized in that, Establish a time window framework for rolling scheduling, define the rolling optimization time domain and control time domain, and divide the rolling time domain into multiple time periods. At the beginning of each rolling cycle, formulate a scheduling plan based on the updated graph model structure and photovoltaic power generation prediction results. An optimization objective function is constructed that is linked to the dynamic edge deletion mechanism of the bidirectional graph model. The objective function includes the remaining power imbalance, operating cost, and energy storage degradation loss. Formulate energy storage charging and discharging strategies based on graph model states, establish power flow constraints based on the set of effective edges in the graph model, and define an energy storage scheduling priority index.

9. A photovoltaic-energy storage rolling dispatch system based on ultra-short-term power prediction, used to implement the photovoltaic-energy storage rolling dispatch method based on ultra-short-term power prediction as described in any one of claims 1 to 8, characterized in that, include: The system includes a data acquisition module, a cloud cover calculation module, a power calculation module, an energy storage balance module, a dispatch energy storage module, and a rolling dispatch optimization module. The data acquisition module is used to collect electricity load data, cloud change image data, and photovoltaic power generation data to form a multi-source heterogeneous dataset. The cloud shading calculation module is used to construct a cloud shading impact model. It uses a neural network to perform correlation analysis between cloud change image data and photovoltaic power generation data, and establishes a photovoltaic power generation prediction model between cloud shading time, shading intensity and photovoltaic power generation. The power calculation module is used to predict the temporal changes in electricity load based on historical electricity load data, and at the same time, to predict the timing of cloud shading of photovoltaic panels when clouds move based on cloud change image data, and to calculate the power generation of photovoltaic panels. The energy storage balancing module is used to integrate electricity load data, photovoltaic power generation data and cloud impact data onto a unified time axis, calculate the difference between photovoltaic power generation and electricity consumption, and balance the power difference through energy storage. The energy storage scheduling module is used to construct a two-way graph model of dynamic changes in photovoltaic and energy storage, establish the connection relationship between photovoltaic nodes and energy storage nodes in the two-way graph model, dynamically adjust the graph model structure according to the energy storage charging and discharging status, and schedule energy storage through the graph model. The rolling scheduling optimization module is used to implement rolling scheduling optimization. Based on the current periodic graph model structure and photovoltaic power generation prediction results, it formulates energy storage charging and discharging strategies, and stores excess electrical energy and makes up for power deficits through energy storage nodes.

Images