Search

CN-122000972-A - Community micro-grid scheduling method considering photovoltaic energy storage

CN122000972ACN 122000972 ACN122000972 ACN 122000972ACN-122000972-A

Abstract

The invention relates to the technical field of micro-grid scheduling, in particular to a community micro-grid scheduling method considering photovoltaic energy storage, which comprises the steps of collecting multisource heterogeneous numbers of community micro-grids, constructing a photovoltaic output prediction large model based on a Transformer framework, capturing long time sequence dependence characteristics of data by using a self-attention mechanism, outputting photovoltaic power prediction results within a preset time range, constructing a user behavior understanding large model, carrying out semantic feature extraction and clustering analysis on historical user load data, constructing a multi-objective dynamic optimization scheduling model according to the photovoltaic power prediction results and user demand response potential, solving the multi-objective dynamic optimization scheduling model by using a deep reinforcement learning algorithm, issuing real-time scheduling instructions to intelligent terminal equipment in a community, and carrying out iterative correction on scheduling strategies by using an online learning mechanism of the large model.

Inventors

  • WANG JIANMING
  • WANG JIANI

Assignees

  • 湖南誉甄文化传媒有限公司

Dates

Publication Date
20260508
Application Date
20260126

Claims (10)

  1. 1. The community micro-grid scheduling method considering photovoltaic energy storage is characterized by comprising the following steps of: The method comprises the steps of collecting multi-source heterogeneous data of a community micro-grid, wherein the multi-source heterogeneous data comprise photovoltaic power generation power data, meteorological environment data, user side load data and energy storage battery running state data; Constructing a photovoltaic output prediction large model based on a transducer architecture, inputting the multi-source heterogeneous data into the photovoltaic output prediction large model, capturing long time sequence dependence characteristics of the data by using a self-attention mechanism, and outputting a photovoltaic power generation power prediction result within a preset time range; constructing a large model for understanding user behavior, extracting semantic features and performing cluster analysis on historical user load data, identifying an implicit mode of user power consumption behavior, and predicting user demand response potential in a future period; according to the photovoltaic power generation power prediction result and the user demand response potential, constructing a multi-objective dynamic optimization scheduling model by taking the minimization of the running cost of the community micro-grid, the minimization of the power fluctuation of the grid and the minimization of the carbon emission as objective functions; Solving the multi-objective dynamic optimization scheduling model by using a deep reinforcement learning algorithm to generate a real-time scheduling instruction containing energy storage charging and discharging power, photovoltaic power and a load regulation strategy; And issuing the real-time scheduling instruction to each intelligent terminal device in the community, and carrying out iterative correction on the scheduling strategy by utilizing an online learning mechanism of the large model according to actual operation data fed back by the device.
  2. 2. The method for dispatching the community micro-grid taking photovoltaic energy storage into consideration according to claim 1, wherein after the multi-source heterogeneous data of the community micro-grid are collected, the method further comprises the steps of: Preprocessing the collected original data, wherein the preprocessing comprises filling of a missing value, elimination of an abnormal value and normalization; And calculating the correlation between meteorological environment data and photovoltaic power generation power by using a Pearson correlation coefficient analysis method, and screening out key influence factors as input feature vectors of the photovoltaic output prediction large model.
  3. 3. The community micro-grid scheduling method considering photovoltaic energy storage as set forth in claim 1, wherein the constructing of the photovoltaic output prediction large model based on a transducer architecture specifically includes: Adopting an encoder-decoder structure, introducing a position coding mechanism at an encoder end to reserve time information of a time sequence; calculating weight distribution among photovoltaic power generation data at different moments through a multi-head self-attention mechanism, and focusing on historical period data with the greatest influence on the current moment; And carrying out nonlinear transformation on the output of the attention mechanism by adopting a feedforward neural network, preventing model degradation by residual connection and layer normalization operation, and outputting a photovoltaic power generation power prediction result.
  4. 4. The community micro-grid scheduling method considering photovoltaic energy storage according to claim 1, wherein after outputting the photovoltaic power generation power prediction result within the preset time range, the method further comprises: introducing an interval prediction concept, and calculating a confidence interval of a photovoltaic power generation power prediction result by using a quantile regression algorithm; and if the actually monitored photovoltaic power exceeds the confidence interval, judging that the photovoltaic power is abnormal in prediction, and triggering a model retraining mechanism to update model parameters.
  5. 5. The community micro-grid scheduling method considering photovoltaic energy storage as set forth in claim 1, wherein the constructing the user behavior understanding large model specifically includes: converting historical electricity load data of users into time sequence diagram structure data, wherein nodes represent single users, and edges represent electricity similarity among the users; Performing aggregation operation on the time sequence diagram structure data by using a graph neural network, and extracting space-time electricity utilization characteristics of user groups in communities; And analyzing the text power consumption requirement submitted by the user by combining a natural language processing technology, converting the unstructured requirement into a quantifiable power constraint condition, and incorporating the quantitative power constraint condition into the user requirement response potential evaluation.
  6. 6. The community micro-grid scheduling method considering photovoltaic energy storage as set forth in claim 1, wherein the constructing a multi-objective dynamic optimization scheduling model specifically includes: Establishing an objective function: ; Wherein the method comprises the steps of The running cost comprises electricity purchasing cost, equipment maintenance cost and energy storage depreciation cost; power fluctuation variance of the power grid; Alpha, beta and gamma are weight coefficients of various indexes respectively; the constraint conditions comprise upper and lower limit constraint of the state of charge of the energy storage system, charge and discharge power constraint, photovoltaic absorption constraint and power balance constraint, and the constraint conditions are dynamically updated along with time step.
  7. 7. The community micro-grid scheduling method considering photovoltaic energy storage according to claim 1, wherein the method for solving the multi-objective dynamic optimization scheduling model by using a deep reinforcement learning algorithm specifically comprises the following steps: Constructing an intelligent body, and defining the current state space of the community micro-grid as a vector containing photovoltaic predicted output, current load, SOC state and real-time electricity price; defining a charging and discharging action and a load regulation action of an energy storage system as an action space; setting a reward function, and giving positive rewards when the objective function J value is reduced by scheduling actions and the constraint condition is met, and giving negative penalties on the contrary; training the intelligent agent through a near-end strategy optimization algorithm or a deep Q network algorithm to enable the intelligent agent to output an optimal real-time scheduling instruction; according to the actual operation data fed back by the equipment, the scheduling strategy is iteratively corrected by using an online learning mechanism of a large model, and the method specifically comprises the following steps: A rolling prediction and feedback correction mechanism is established, and the actually collected photovoltaic output and load data are input into a model at intervals of set time; Calculating an error between a predicted value and an actual value, and adjusting network weights of the photovoltaic output prediction large model and the user behavior understanding large model by using a back propagation algorithm; New samples are continuously input, so that the model has the self-adaptive evolution capability along with seasonal changes and custom transition of users; The method also comprises the following steps of: Monitoring voltage and frequency parameters of a connection point of the micro power grid and the large power grid; when a large power grid is detected to be faulty or the parameters are limited, automatically switching to an off-grid operation mode; and under the off-grid operation mode, rapidly generating an island operation scheduling strategy taking the key load power supply priority as a core by using the large model, and adjusting the voltage and the frequency of the energy storage system to control and support the community micro-grid in a voltage source mode.
  8. 8. A community micro-grid dispatching system taking into account photovoltaic energy storage as defined in any one of claims 1 to 7, comprising: The multi-dimensional data sensing module is used for collecting multi-source heterogeneous data in the community micro-grid in real time, wherein the multi-source heterogeneous data comprise real-time output data of a photovoltaic power generation unit, meteorological environment monitoring data, battery state data of an energy storage system, intelligent load data of a user side and power data interacted with an external large power grid; The AI fusion prediction module is connected with the multidimensional data perception module and is configured with a deep learning prediction model based on space-time feature fusion, and is used for receiving the multi-source heterogeneous data, extracting space-time correlation features of photovoltaic output and behavior features of user load, and outputting a photovoltaic power prediction curve and a community load demand prediction curve in a future preset time window; the dynamic optimization decision module is connected with the AI fusion prediction module and is used for constructing an objective function which takes the lowest system operation cost, the lowest power fluctuation and the optimal energy storage battery health loss as multiple objectives, and generating an optimal scheduling strategy at the current moment by adopting an improved multi-objective particle swarm optimization algorithm in combination with the prediction curve and preset constraint conditions; The edge calculation control module is connected with the dynamic optimization decision module and is used for receiving the optimal scheduling strategy, analyzing the optimal scheduling strategy into specific control instructions aiming at the photovoltaic inverter, the energy storage converter and the intelligent load terminal, simultaneously executing feedback by the real-time monitoring equipment, and carrying out local closed loop correction if deviation occurs; and the man-machine interaction and alarm module is used for visually displaying the running state of the micro-grid, the prediction data and the scheduling strategy and triggering the hierarchical alarm when the abnormal working condition or the prediction error is detected to exceed the threshold value.
  9. 9. An electronic device, comprising: At least one memory non-transitory storing computer-executable instructions; At least one processor configured to execute the computer-executable instructions, Wherein the computer executable instructions when executed by the processor implement a community micro grid scheduling method taking into account photovoltaic energy storage according to any of claims 1-7.
  10. 10. A computer readable storage medium storing computer executable instructions which when executed by at least one processor implement a community micro grid scheduling method taking into account photovoltaic energy storage according to any of claims 1-7.

Description

Community micro-grid scheduling method considering photovoltaic energy storage Technical Field The invention relates to the technical field of micro-grid dispatching, in particular to a community micro-grid dispatching method considering photovoltaic energy storage. Background With the transformation of global energy structures and the promotion of 'double carbon' targets, a community micro-grid is used as a key unit for connecting distributed renewable energy sources (such as photovoltaics), energy storage systems and diversified user loads, and the stable operation and efficient management of the community micro-grid are particularly important. The method comprises the following steps of 1, acquiring charging data of an electric vehicle, determining a charging behavior rule of the electric vehicle according to the charging data, 2, simulating disordered charging and discharging behaviors of the electric vehicle according to a charging mode, power battery characteristics and the charging behavior rule to obtain a total load curve of a community, 3, dividing a plurality of time periods according to the total load curve for user demands, wherein different time-sharing electricity prices are adopted for different time periods, and the electric vehicle is discharged when the electricity price is higher and is charged when the electricity price is lower; In the prior art, china patent, application number is CN202411126407.9, discloses a community micro-grid energy storage scheduling method, relates to the technical field of micro-grid energy storage, and effectively realizes effective management and control of different power utilization demands in a community through whole-flow management of demand investigation, data acquisition, energy storage system and power utilization system butt joint and equipment remote control and micro-grid scheduling. By means of peak-to-peak scheduling, emergency scheduling, peak-to-valley clipping scheduling, other intelligent scheduling and other methods, different electricity requirements can be met in a targeted manner, integral electricity fluctuation of a community power grid can be reduced, the risk of the community power grid is solved, certain economic benefits are obtained by utilizing peak-to-valley electricity difference, stability and reliability of the power grid can be improved, occurrence of voltage fluctuation and power grid faults is reduced, renewable energy sources can be integrated better through management and scheduling of energy storage facilities, development and application of the renewable energy sources are promoted, and energy conversion and sustainable development are promoted; In the prior art III, china patent, application number CN202510439197.7, discloses a community micro-grid scheduling method and system considering photovoltaic energy storage, and relates to the technical field of data processing; according to the photovoltaic energy storage data, the trend and periodicity in the data are identified through time sequence analysis to obtain a time sequence analysis result, a neural network architecture is constructed according to the time sequence analysis result, future photovoltaic energy storage data are predicted according to the neural network architecture to obtain a prediction result, and the scheduling strategy of the power equipment is determined according to the prediction result and the running state of the current power distribution network. The invention improves the utilization efficiency of the photovoltaic energy storage system, reduces the electric power cost and improves the stability of the electric power system by comprehensively utilizing the technical means of time sequence analysis, neural network prediction, linear programming and a planar cutting method; The first, second and third existing technologies fail to fully utilize the advantages of the large AI model in depth feature extraction, long time sequence dependency capture and extensive learning capability, so that problems of low prediction precision, stiff scheduling strategy and slow system response still exist when facing complex scenes such as massive distributed resource access, rapid user habit change and extreme weather frequent. Therefore, the invention provides a community micro-grid dispatching method considering photovoltaic energy storage. Disclosure of Invention In order to achieve the above purpose, the invention adopts the following technical scheme: In one aspect of the invention, a community micro-grid scheduling method considering photovoltaic energy storage is provided, which comprises the following steps: The method comprises the steps of collecting multi-source heterogeneous data of a community micro-grid, wherein the multi-source heterogeneous data comprise photovoltaic power generation power data, meteorological environment data, user side load data and energy storage battery running state data; Constructing a photovoltaic output prediction la