A virtual power plant intelligent scheduling decision system and method based on multi-source data fusion

Abstract

The application provides a virtual power plant intelligent scheduling decision system and method based on multi-source data fusion, wherein the system is provided with a data acquisition module, a communication module, a data fusion module, an intelligent decision module and an execution control module; the system method acquires data by deploying the data acquisition module at distributed energy nodes and power grid monitoring points, then transmits the data to the data fusion module by the communication module, performs feature extraction on the data after cleaning, standardization and space-time equivalence, mines potential relationships between different types of data, and performs weighted fusion; the fused data enters the intelligent decision module, selects a suitable intelligent algorithm for solving based on a preset scheduling target, generates scheduling instructions of each distributed energy resource, then issues the scheduling instructions to the execution control module, the execution control module controls and processes corresponding equipment, and real-time data feedback is collected to the intelligent decision module, so that the system can stably operate.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, specifically to a virtual power plant intelligent dispatching decision-making system and method based on multi-source data fusion. Background Technology

[0002] With the advancement of the global energy transition, distributed power sources, such as photovoltaic and wind power generation, energy storage devices, and controllable loads, are being connected to the grid in large quantities. Virtual power plants, as an important technical means to integrate decentralized resources and achieve coordinated regulation, have become a key path to improve grid flexibility and the capacity to absorb renewable energy. Virtual power plants, through information and communication technologies and energy management systems, aggregate dispersed distributed energy resources into a "virtual whole" that can participate in electricity market transactions and grid dispatch, effectively balancing supply and demand and alleviating grid pressure. However, existing virtual power plant dispatch decisions face multiple challenges. On the one hand, the randomness and intermittency of distributed energy resources significantly increase the difficulty of dispatch, such as the weather-related impacts on wind and solar power generation and the uncertainty of user electricity consumption behavior. On the other hand, traditional dispatch systems rely heavily on single types of data, such as real-time grid power and historical load data, making it difficult to comprehensively capture the dynamic characteristics of resources within the virtual power plant. At the same time, the geographical dispersion of distributed energy resources and the multi-source nature of data collection, including grid monitoring data, meteorological data, and user electricity consumption data, result in problems such as inconsistent formats, asynchronous timing, and fragmented information, leading to a lack of accuracy and timeliness in dispatch decisions and making it difficult to achieve optimal resource allocation and efficient collaborative operation within the virtual power plant. To address this, the applicant proposes a virtual power plant intelligent dispatching decision-making system and method based on multi-source data fusion, which effectively integrates heterogeneous data from multiple sources and achieves precise dispatching decisions through intelligent algorithms. This enhances the virtual power plant's ability to manage distributed energy resources and ensures stable grid operation and efficient energy utilization. Summary of the Invention

[0003] To address the aforementioned technical issues, this invention proposes a virtual power plant intelligent dispatching and decision-making system and method based on multi-source data fusion. The system collects data by deploying a data acquisition module at distributed energy nodes and grid monitoring points. The data is then transmitted to a data fusion module via a communication module. This module cleans, standardizes, and performs spatiotemporal alignment of the data before feature extraction, uncovering potential relationships between different data types, and then weighted and fused. The fused data enters the intelligent decision-making module, which selects an appropriate intelligent algorithm based on a preset dispatching objective to generate dispatching instructions for each distributed energy resource. These instructions are then sent to the execution control module, which controls the corresponding equipment and collects data in real time, feeding it back to the intelligent decision-making module to ensure stable system operation.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A virtual power plant intelligent dispatching and decision-making system and method based on multi-source data fusion is characterized by: the system comprising a data acquisition module, a communication module, a data fusion module, an intelligent decision-making module, and an execution control module; the data acquisition module is deployed at distributed energy resource nodes and power grid monitoring points, and connects to multi-source heterogeneous data terminals such as smart sensors, smart meters, meteorological monitoring equipment, energy storage status monitors, and real-time acquisition of power grid operation data, distributed energy generation data, energy storage device status data, user electricity consumption data, and meteorological data; the communication module adopts edge computing technology to achieve low-latency data transmission, supports compatibility between power-specific communication protocols and general protocols, and ensures data transmission between modules and real-time uploading to the cloud platform; the data fusion module preprocesses the acquired multi-source data through data cleaning, standardization, and spatiotemporal alignment technologies to eliminate data noise and format differences. The system utilizes feature extraction and correlation analysis algorithms to uncover potential relationships between different types of data, constructs a unified data model, and performs weighted fusion of multi-source prediction results. The weights are dynamically adjusted based on historical prediction errors to improve prediction reliability and provide structured data support for subsequent decision-making. The intelligent decision-making module includes an objective function construction module, a constraint processing module, and an algorithm scheduling center. Multiple built-in intelligent algorithms combine external information such as electricity market price signals and grid dispatch instructions to analyze and calculate the data output by the data fusion module, aiming at system economy, stability, and renewable energy absorption rate, to generate the optimal dispatch decision scheme. The execution control module sends the dispatch instructions generated by the intelligent decision-making module to the distributed energy resource control terminal, enabling precise control of the charging and discharging of power generation equipment and energy storage equipment, as well as controllable load regulation, and providing real-time feedback of execution results to form a closed-loop dispatch management system.

[0006] Furthermore, the data cleaning, standardization, and spatiotemporal alignment techniques of the data fusion module in the virtual power plant intelligent dispatching decision-making system based on multi-source data fusion specifically include:

[0007] Data cleaning: False alarm data from sensors are removed using outlier detection algorithms, and missing data is filled in using spatiotemporal interpolation based on adjacent node data and time trends;

[0008] Standardization: Convert data with different units into a unified dimension, and use the Z-score standardization formula to process numerical data to ensure consistency of algorithm input;

[0009] Spatiotemporal alignment: Synchronize multi-source data based on timestamps, align different collected data to the same minute-level timeline, and associate spatially distributed device data through geographic information tags;

[0010] Furthermore, the feature-level fusion of the data fusion module of the virtual power plant intelligent dispatching decision system based on multi-source data fusion involves extracting key features, extracting spatial features of the data through convolutional neural networks based on the nonlinear relationships and daily periodic features of the data, and combining them with time features extracted by LSTM for weighted fusion. The weights are dynamically adjusted based on historical prediction errors.

[0011] Furthermore, the intelligent decision-making module of the virtual power plant intelligent dispatching decision-making system based on multi-source data fusion includes an objective function construction module, a constraint condition processing module, and an algorithm scheduling center, specifically:

[0012] Objective function construction module: Based on multi-objective optimization, including economic, technical and environmental objectives. The economic objective is to minimize the operating cost of the virtual power plant; the technical objective is to maximize the renewable energy absorption rate, maintain grid frequency stability and ensure that the energy storage SOC is within the safe range of 20%-80%; the environmental objective is to minimize carbon emissions based on the power generation coefficient of thermal power units.

[0013] The constraint processing module includes power balance constraints, equipment physical constraints, and grid security constraints. The power balance constraint ensures that distributed power generation output + energy storage charging and discharging power + purchased electricity from outside the grid equals user load. The equipment physical constraint ensures that the maximum output of photovoltaic / wind turbines is ≤ rated power, and the energy storage charging and discharging power is ≤ maximum allowable value. The grid security constraint ensures that node voltage deviation is ≤ ±5%, and line transmission power is ≤ thermal stability limit.

[0014] Algorithm Scheduling Center: The algorithms include Long Short-Term Memory (LSTM) network algorithms, Particle Swarm Optimization (PSO) algorithms, and Deep Reinforcement Learning (DRLE) algorithms. It uses an algorithm selection mechanism to make decision outputs and issue instructions, and can also perform feedback monitoring and emergency handling.

[0015] Furthermore, the algorithm scheduling center's long short-term memory network algorithm, particle swarm optimization algorithm, and deep reinforcement learning algorithm of the intelligent decision-making module of the virtual power plant intelligent dispatching decision-making system based on multi-source data fusion are specifically as follows:

[0016] Long Short-Term Memory (LSTM) network algorithm: Based on a special recurrent neural network, it excels at processing time-series data and can effectively capture long-term dependencies in data. In virtual power plant scheduling, the algorithm is used to predict the power generation of distributed energy sources and user load demand. By inputting time-series information such as historical power generation data, meteorological data, and user electricity consumption behavior data, the algorithm model can learn the changing patterns of the data and predict the power generation and consumption trends in future periods, providing accurate predictive data support for scheduling decisions.

[0017] Particle Swarm Optimization (PSO) Algorithm: A global optimization algorithm based on swarm intelligence, simulating the behavior of flocks of birds foraging or schools of fish swimming. In virtual power plant scheduling, the algorithm is used to solve multi-objective optimization problems, with the optimization objectives of minimizing system operating costs, minimizing carbon emissions, and maximizing renewable energy absorption rate. The scheduling power of each distributed energy resource is used as the optimization variable, and the objective function is constructed in combination with grid constraints. The algorithm finds the optimal scheduling scheme that satisfies multiple objectives through iterative search of particles in the solution space.

[0018] Deep reinforcement learning algorithm: Combining the perception capabilities of deep learning with the decision-making capabilities of reinforcement learning, it autonomously learns the optimal strategy in complex environments. In the virtual power plant scheduling scenario, the operating state of the virtual power plant is used as the state space, and the scheduling instructions are used as the action space. The reward function is constructed based on indicators such as system benefits and grid stability. The algorithm continuously interacts with the environment and optimizes the scheduling strategy based on reward feedback, thereby achieving adaptive and dynamic intelligent scheduling decisions.

[0019] Furthermore, the prediction of the Long Short-Term Memory Network algorithm of the intelligent decision-making module of the virtual power plant intelligent dispatch decision-making system based on multi-source data fusion is specifically as follows:

[0020] Long Short-Term Memory Network Prediction Module:

[0021] Input: Historical time series data {PPV,thist,PW,thist,Lthist,Mthist}t=1thist, where thist is the length of the historical data;

[0022] Output: Forecast values ​​for the future time period {PPV, tpred, PW, tpred, Ltpred} t=1T;

[0023] The specific steps are as follows:

[0024] 1) Data preprocessing: Normalize historical data, imput missing data, and construct input feature sequences;

[0025] 2) Construction of Long Short-Term Memory Network Algorithm Model: The input layer accepts feature sequences, the number of hidden neurons in the Long Short-Term Memory Network layer is set, and long-term dependencies are learned through forget gate, input gate, and output gate. The output layer outputs predicted values ​​through intelligent decision-making.

[0026] 3) Model training: Train the long short-term memory network using historical data, with mean squared error as the loss function, and optimize the parameters through backpropagation until the loss converges;

[0027] 4) Predictive output: Input the latest historical data, and output the historical photovoltaic power generation, historical wind power generation, and historical user load demand for future periods through a trained long short-term memory network.

[0028] Furthermore, the implementation method of the virtual power plant intelligent dispatching decision-making system based on multi-source data fusion includes the following steps:

[0029] Step 1: Data Acquisition and Preprocessing; Multi-source heterogeneous data is acquired through the data acquisition module, and cleaned and standardized in the data fusion module. Missing data is filled using interpolation or machine learning-based prediction methods to ensure data integrity and consistency.

[0030] Step 2: Data Fusion and Analysis; Key features of the data are extracted using feature engineering techniques, and temporal and spatial relationships between data are explored through spatiotemporal correlation analysis algorithms; Principal component analysis, independent component analysis, and other dimensionality reduction algorithms are used to reduce data dimensionality and improve computational efficiency.

[0031] Step 3: Intelligent scheduling decision-making; The fused data is input into the intelligent decision-making module, and based on the preset scheduling objectives, an appropriate intelligent algorithm is selected to solve the problem and generate scheduling instructions for each distributed energy resource;

[0032] Step 4: Command Execution and Feedback; The execution control module sends the scheduling command to the corresponding device and collects the device operating status data in real time and feeds it back to the intelligent decision-making module so as to dynamically adjust the scheduling plan and ensure the stable operation of the system.

[0033] The benefits of this application are:

[0034] 1. A virtual power plant intelligent dispatching decision-making system and method based on multi-source data fusion: The data acquisition module covers distributed energy, power grid operation, user load and external information. Combined with edge computing technology, it achieves low-latency data transmission of ≤100ms, solving the pain points of "incomplete data acquisition and delayed transmission" in traditional dispatching, and realizing full-dimensional data coverage and low-latency transmission.

[0035] 2. The virtual power plant intelligent dispatching decision-making system and method execution control module based on multi-source data fusion uses encrypted communication to send instructions to the equipment controller, avoiding instruction transmission leakage or interruption; at the same time, deviation monitoring is set up to solve the blindness of traditional "no feedback after instruction is issued"; thus achieving accuracy and security in instruction execution.

[0036] 3. The intelligent dispatching decision-making system and method for virtual power plants based on multi-source data fusion transforms multi-source heterogeneous data into structured features through a four-level processing of "cleaning, standardization, spatiotemporal alignment, and feature fusion," avoiding decision-making biases caused by the traditional "direct use of single data" and achieving refined data fusion.

[0037] 4. The intelligent dispatching decision-making system and method for virtual power plants based on multi-source data fusion breaks through the limitations of traditional single data dispatching by using deep multi-source data fusion technology. Through spatiotemporal feature correlation and multi-model weighted fusion, it solves the problems of large prediction errors in wind and solar power generation, reducing them to ≤8%, and difficulty in predicting load fluctuations.

[0038] 5. The intelligent dispatching decision-making system and method for virtual power plants based on multi-source data fusion utilizes an adaptive algorithm dispatching mechanism, combined with the advantages of long short-term memory network prediction, particle swarm optimization, and deep reinforcement learning dynamic decision-making, to achieve intelligent operation of the entire "prediction-optimization-execution" process, improving dispatching economy by 15%-20% compared to traditional methods;

[0039] 6. The intelligent dispatching decision-making system and method for virtual power plants based on multi-source data fusion adopts a distributed-centralized collaborative control, which supports both rapid local response of single devices and global optimization through centralized cloud optimization, taking into account both response speed and system efficiency. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0041] Figure 2 This is a schematic diagram of the data processing flow of the system method of the present invention. Detailed Implementation

[0042] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0043] like Figure 1-2The diagram illustrates a virtual power plant intelligent dispatching and decision-making system and method based on multi-source data fusion. The system comprises a data acquisition module, a communication module, a data fusion module, an intelligent decision-making module, and an execution control module. The data acquisition module is deployed at distributed energy resource nodes and power grid monitoring points, connecting to intelligent sensors, smart meters, meteorological monitoring equipment, energy storage status monitors, and real-time acquisition of multi-source heterogeneous data such as power grid operation data, distributed energy generation data, energy storage device status data, user electricity consumption data, and meteorological data. The communication module employs edge computing technology to achieve low-latency data transmission, supporting compatibility between power-specific communication protocols and general protocols, ensuring data transmission between modules and real-time uploading to the cloud platform. The data fusion module preprocesses the acquired multi-source data through data cleaning, standardization, and spatiotemporal alignment techniques to eliminate data noise and format differences. Based on feature extraction and correlation analysis algorithms, the system mines the potential relationships between different types of data, constructs a unified data model, and performs weighted fusion of multi-source prediction results. The weights are dynamically adjusted based on historical prediction errors to improve prediction reliability and provide structured data support for subsequent decision-making. The intelligent decision-making module includes an objective function construction module, a constraint processing module, and an algorithm scheduling center. The built-in intelligent algorithms can combine external information such as electricity market price signals and grid dispatch instructions to analyze and calculate the data output by the data fusion module with objectives such as system economy, stability, and renewable energy absorption rate, and generate the optimal dispatch decision scheme. The execution control module sends the dispatch instructions generated by the intelligent decision-making module to the distributed energy resource control terminal to achieve precise control of the charging and discharging of power generation equipment and energy storage equipment, and controllable load adjustment, and provides real-time feedback on the execution results to form a closed-loop dispatch management system.

[0044] The data cleaning, standardization, and spatiotemporal alignment techniques of the data fusion module in the virtual power plant intelligent dispatching decision-making system based on multi-source data fusion are as follows:

[0045] Data cleaning: False alarm data from sensors are removed using outlier detection algorithms, and missing data is filled in using spatiotemporal interpolation based on adjacent node data and time trends;

[0046] Standardization: Convert data with different units into a unified dimension, and use the Z-score standardization formula to process numerical data to ensure consistency of algorithm input;

[0047] Spatiotemporal alignment: Synchronize multi-source data based on timestamps, align different collected data to the same minute-level timeline, and associate spatially distributed device data through geographic information tags;

[0048] The feature-level fusion of the data fusion module of the virtual power plant intelligent dispatch decision-making system based on multi-source data fusion is to extract key features. For the nonlinear relationship and daily periodic features of the data, the spatial features of the data are extracted by convolutional neural network, and the temporal features are extracted by LSTM. The weights are dynamically adjusted based on historical prediction errors.

[0049] The intelligent decision-making module of the virtual power plant intelligent dispatching decision-making system based on multi-source data fusion, as shown, includes an objective function construction module, a constraint processing module, and an algorithm scheduling center.

[0050] Objective function construction module: Based on multi-objective optimization, including economic, technical and environmental objectives. The economic objective shown is to minimize the operating cost of the virtual power plant; the technical objective shown is to maximize the renewable energy absorption rate, maintain grid frequency stability and ensure that the energy storage SOC is within the safe range of 20%-80%; the environmental objective shown is to minimize carbon emissions based on the power generation coefficient of thermal power units.

[0051] The constraint processing module includes power balance constraints, equipment physical constraints, and grid security constraints. The power balance constraint shown is to ensure that the output of distributed power sources + the charging and discharging power of energy storage + the power purchased from outside the grid equals the user load. The equipment physical constraint shown is that the maximum output of photovoltaic / wind turbines ≤ the rated power, and the charging and discharging power of energy storage ≤ the maximum allowable value. The grid security constraint shown is that the node voltage deviation ≤ ±5%, and the line transmission power ≤ the thermal stability limit.

[0052] Algorithm Scheduling Center: This center employs algorithms including Long Short-Term Memory (LSTM) networks, Particle Swarm Optimization (PSO) algorithms, and Deep Reinforcement Learning (DRL) algorithms. It utilizes an algorithm selection mechanism for decision-making and instruction issuance, while also providing feedback monitoring and emergency handling. The specific algorithms are as follows:

[0053] Long Short-Term Memory (LSTM) network algorithm: Based on a special recurrent neural network, it excels at processing time-series data and can effectively capture long-term dependencies in the data. In virtual power plant scheduling, the algorithm is used to predict distributed energy generation capacity and user load demand. By inputting time-series information such as historical power generation data, meteorological data, and user electricity consumption behavior data, the algorithm model can learn the data's changing patterns and predict future power generation and consumption trends, providing accurate predictive data support for scheduling decisions. The LSM prediction module is shown below.

[0054] Input: Historical time series data {PPV,thist,PW,thist,Lthist,Mthist}t=1thist, where thist is the length of the historical data;

[0055] Output: Forecast values ​​for the future time period {PPV, tpred, PW, tpred, Ltpred} t=1T;

[0056] The specific steps are as follows:

[0057] 1) Data preprocessing: Normalize historical data, imput missing data, and construct input feature sequences;

[0058] 2) Construction of Long Short-Term Memory Network Algorithm Model: The input layer accepts feature sequences, the number of hidden neurons in the Long Short-Term Memory Network layer is set, and long-term dependencies are learned through forget gate, input gate, and output gate. The output layer outputs the predicted value through a fully connected layer.

[0059] 3) Model training: Train the long short-term memory network using historical data, with mean squared error as the loss function, and optimize the parameters through backpropagation until the loss converges;

[0060] 4) Predictive Output: Input the latest historical data, and output the historical photovoltaic power generation, historical wind power generation, and historical user load demand for future periods through a trained Long Short-Term Memory network.

[0061] Particle Swarm Optimization (PSO) Algorithm: A global optimization algorithm based on swarm intelligence, simulating the behavior of flocks of birds foraging or schools of fish swimming. In virtual power plant scheduling, the algorithm is used to solve multi-objective optimization problems, with the optimization objectives of minimizing system operating costs, minimizing carbon emissions, and maximizing renewable energy absorption rate. The scheduling power of each distributed energy resource is used as the optimization variable, and the objective function is constructed in combination with grid constraints. The algorithm finds the optimal scheduling scheme that satisfies multiple objectives through iterative search of particles in the solution space.

[0062] Deep reinforcement learning algorithm: Combining the perception capabilities of deep learning with the decision-making capabilities of reinforcement learning, it autonomously learns the optimal strategy in complex environments. In the virtual power plant scheduling scenario, the operating state of the virtual power plant is used as the state space, and the scheduling instructions are used as the action space. The reward function is constructed based on indicators such as system benefits and grid stability. The algorithm continuously interacts with the environment and optimizes the scheduling strategy based on reward feedback, thereby achieving adaptive and dynamic intelligent scheduling decisions.

[0063] The implementation steps of the virtual power plant intelligent dispatching decision system based on multi-source data fusion are as follows:

[0064] Step 1: Data Acquisition and Preprocessing; Multi-source heterogeneous data is acquired through the data acquisition module, and cleaned and standardized in the data fusion module. Missing data is filled using interpolation or machine learning-based prediction methods to ensure data integrity and consistency.

[0065] Step 2: Data Fusion and Analysis; Key features of the data are extracted using feature engineering techniques, and temporal and spatial relationships between data are explored through spatiotemporal correlation analysis algorithms; Principal component analysis, independent component analysis, and other dimensionality reduction algorithms are used to reduce data dimensionality and improve computational efficiency.

[0066] Step 3: Intelligent scheduling decision-making; The fused data is input into the intelligent decision-making module, and based on the preset scheduling objectives, an appropriate intelligent algorithm is selected to solve the problem and generate scheduling instructions for each distributed energy resource;

[0067] Step 4: Command Execution and Feedback; The execution control module sends the scheduling command to the corresponding device and collects the device operating status data in real time and feeds it back to the intelligent decision-making module so as to dynamically adjust the scheduling plan and ensure the stable operation of the system.

[0068] The implementation method of the intelligent dispatching and decision-making system for virtual power plants based on multi-source data fusion is described in detail below, taking a virtual power plant in an industrial park as an example. This virtual power plant covers the following distributed energy resources and loads:

[0069] Distributed power sources: two 100kW rooftop photovoltaic power stations, one 50kW small wind farm, and one 30kW emergency diesel generator;

[0070] Energy storage equipment: 1 set of 100kWh lithium battery energy storage system, with a charge / discharge efficiency of 90% and a safe SOC range of 20%-80%;

[0071] User load: 3 industrial enterprises (30% of which are adjustable loads, such as production line motors), 1 commercial office building (the adjustable load is the air conditioning system), and 50 residential users (mainly basic loads); External connections: connected to the regional distribution network (allowing power exchange), real-time electricity market price (0.3 yuan / kWh from 0:00 to 8:00, 0.6 yuan / kWh from 8:00 to 22:00, and 0.4 yuan / kWh from 22:00 to 24:00), and a weather station (providing forecast data for sunlight, wind speed, and temperature);

[0072] Detailed implementation of each module of the system:

[0073] Data acquisition module:

[0074] Multi-source data real-time acquisition and deployment equipment:

[0075] Photovoltaic power station: Install temperature sensors (to monitor photovoltaic panel temperature, sampling frequency 1 minute / time) and power sensors (to collect real-time output, accuracy ±1%).

[0076] Wind farm: Deploy wind speed sensors (range 0-20m / s) and speed sensors (to monitor wind turbine rotor speed);

[0077] Energy storage system: Equipped with a SOC monitor (sampling frequency 30 seconds / time) and a charge / discharge power meter (records real-time charge / discharge status);

[0078] On the user side: Industrial enterprises install smart meters (to collect three-phase current and active power), commercial office buildings install load controllers (to record air conditioner power), and residential users are equipped with smart meters (to count total electricity consumption).

[0079] Power grid monitoring points: Install voltage / current transformers (monitoring access point voltage, range 0-400V, sampling frequency 50Hz) and line loss monitors;

[0080] External data interfaces: connecting to meteorological platforms (obtaining data on solar irradiance ≥500W / m² for the next 24 hours from 8:00 to 16:00, and wind speed ≥3m / s for the next 10:00 to 15:00) and electricity market trading platforms (real-time electricity price updates).

[0081] Communication module: Employs an edge computing terminal (deployed in the park's substation), supports power-specific protocols (for power grid data transmission) and general protocols (for meteorological and user data), with data transmission latency ≤100ms, ensuring real-time upload to the cloud platform;

[0082] Data fusion module:

[0083] Multi-source data preprocessing and fusion data cleaning:

[0084] Outlier detection: The 3σ criterion is used to remove abnormal values ​​of photovoltaic power output (e.g., if the photovoltaic power output suddenly increases to 150kW at 9:00 on a certain day, far exceeding the rated 100kW, it is judged as a false alarm by the sensor and removed).

[0085] Missing value completion: Wind power data is missing at a certain time (e.g., 14:30). Based on the wind speed trend (3-5m / s) of the previous 10 minutes (14:20-14:30) and the data of adjacent wind farms at the same time, it is completed to 32kW through linear interpolation.

[0086] Standardization: Numerical data is standardized using Z-score. For example, photovoltaic power output (unit: kW) and irradiance (unit: W / m²) are converted into dimensionless data in the range [-1, 1].

[0087] Spatiotemporal alignment:

[0088] Time alignment: Meteorological data (original sampling frequency 5 minutes / time) and photovoltaic power output (1 minute / time) are uniformly interpolated to a 1-minute time axis to ensure timestamp synchronization;

[0089] Spatial alignment: By associating data through geographic tags (such as the photovoltaic power station being located on the roof of Building A in the park, with coordinates X1, Y1), spatial matching of "sunlight intensity (Building A area) - photovoltaic output (Building A power station)" is achieved.

[0090] Feature-level fusion:

[0091] Spatial features were extracted using CNN: Convolution operation was performed on the temperature distribution data (20×20 pixel matrix) of three photovoltaic panel arrays to extract the features of temperature hotspot regions (which affect power generation efficiency).

[0092] Combine LSTM to extract time features: input user load data from the past 7 days (1440 minutes / day) to learn the daily cycle pattern of load (e.g., industrial load peaks from 9:00 to 18:00, and residential load peaks from 18:00 to 22:00).

[0093] Decision-level fusion: The output results of three models (LSTM, ARIMA, and Random Forest) are weighted and fused for photovoltaic power output prediction.

[0094] Historical errors: LSTM error 5%, ARIMA error 8%, Random Forest error 7%;

[0095] Weight calculation: the smaller the error, the higher the weight;

[0096] Data storage: Time-series database stores high-frequency data, such as 1-minute photovoltaic output and load power (supports 100,000 writes per second);

[0097] Relational databases store static data such as photovoltaic panel rated power (100kW), energy storage battery type (lithium iron phosphate), and user type (industrial / commercial).

[0098] Intelligent decision-making module:

[0099] Multi-objective optimization and construction of algorithmic scheduling objective functions (taking 24-hour short-term scheduling as an example):

[0100] Economic objective: Minimize operating costs = Cost of electricity purchased from outside the grid (electricity price × purchased volume) + Fuel cost of diesel generator (3 yuan / kWh × generated volume) - Revenue from selling electricity to the grid (electricity price × sold volume);

[0101] Technical targets: Renewable energy consumption rate ≥95%, energy storage SOC maintained at 20%-80%.

[0102] Constraints:

[0103] Power balance constraint: To ensure real-time matching of power supply and demand within the virtual power plant, the balance relationship must be satisfied that "the sum of the power output of all power sources and the power interacting with the grid equals the total load demand," as detailed below:

[0104] The total power supply capacity comprises the actual output of distributed photovoltaic power stations within the virtual power plant, the actual output of small wind farms, the power generation of emergency diesel generators, the power output of energy storage systems (positive values ​​indicate energy storage discharge and power supply to the system, while negative values ​​indicate energy storage charging and power extraction from the system), and the power purchased from the regional distribution network. Simultaneously, if the virtual power plant generates excess power, it can sell power to the distribution network, and this sold power must be deducted from the total power supply capacity. Ultimately, the difference between the aforementioned power supply capacity and the deducted power sales must equal the total electricity demand of all users within the coverage area of ​​the virtual power plant to ensure no power shortage or excess waste.

[0105] Equipment physical constraints: To prevent distributed energy equipment from malfunctioning due to operation exceeding its performance limits, physical boundaries need to be set for the actual output of each device, as follows:

[0106] Photovoltaic power plants: Due to the rated power limitation of photovoltaic panels, their actual output must be between 0 (no sunlight) and 100kW (rated full power generation), and must not exceed the maximum power generation capacity of the equipment;

[0107] Wind farms: Due to the limitations of wind turbine rated power and wind speed, their actual output must be between 0 (no wind) and 50kW (rated full power) to prevent wind turbines from operating under overload conditions;

[0108] Energy storage system: Due to the limitations of battery charge / discharge rate and safety performance, its power needs to be between -50kW (maximum charging power, drawing power from the system) and 50kW (maximum discharging power, supplying power to the system) to avoid overcharging and discharging damage to the battery.

[0109] Power grid security constraints: To ensure the secure interconnection between the virtual power plant and the regional distribution network, the core security indicators for power grid operation must be met, as follows:

[0110] Node voltage constraints: The node voltage of the virtual power plant connected to the distribution network must be based on the distribution network reference voltage of 380V, with an allowable deviation range of no more than ±5%. That is, the actual voltage must be between 361V (380V×95%) and 399V (380V×105%) to prevent the equipment from burning out due to excessive voltage or the equipment from failing to operate normally due to excessively low voltage.

[0111] Line transmission power constraint: The transmission power of the connecting lines between the virtual power plant and the distribution network must not exceed the thermal stability limit of 200kW to avoid overheating of the lines due to power overload, which could lead to safety accidents such as tripping or line damage.

[0112] Algorithm Scheduling Center:

[0113] Short-term dispatch (24 hours):

[0114] The PSO algorithm is used to solve the multi-objective optimization problem: Particle swarm size 30, number of iterations 50;

[0115] Optimization variables: photovoltaic power output, energy storage charging and discharging power, and off-grid electricity purchases at different time periods;

[0116] Output: 100kW full photovoltaic power generation from 8:00 to 16:00, energy storage charging to 80kWh during off-peak hours (0:00-8:00), and discharging to supplement the load during peak hours (8:00-22:00).

[0117] Real-time scheduling:

[0118] Dynamic adjustment using the DRL algorithm:

[0119] State space: current SOC (e.g., 30kWh), real-time load (e.g., 120kW), actual photovoltaic output (e.g., 80kW), current electricity price (0.6 yuan / kWh);

[0120] Operational range: Energy storage discharge of 10kW / 20kW / 30kW, or no operation;

[0121] Reward function: The lower the cost, the higher the absorption rate, and the more stable the voltage, the higher the reward;

[0122] Decision result: When the photovoltaic output suddenly drops by 20kW (actual 80kW → predicted 00kW), the DRL triggers the energy storage to discharge an additional 20kW to maintain power balance.

[0123] Execution control module:

[0124] Command issuance and closed-loop control command issuance:

[0125] The intelligent decision-making module generates scheduling instructions and sends them to the device controllers via an encrypted protocol: Photovoltaic inverter: 100kW output from 8:00 to 16:00; Energy storage PCS: 30kW charging power (up to 80kWh) from 0:00 to 8:00, 40kW discharging power from 12:00 to 14:00; Commercial office building air conditioning: 10kW load reduction from 14:00 to 16:00 (from 50kW to 40kW).

[0126] Feedback monitoring: At 12:30, the actual output of photovoltaic power was 94kW (command 100kW), with a deviation of 6% (exceeding the 5% threshold), triggering an alarm; feedback was sent to the intelligent decision-making module, and particle swarm optimization was re-optimized: the energy storage discharge power was adjusted from 40kW to 46kW to compensate for the 6kW deviation;

[0127] Emergency Response: At 15:00 on a certain day, the voltage at the access point was monitored to drop sharply to 350V (below the safe value of 361V), which was determined to be a power grid fault; the emergency module immediately cut off the commercial air conditioner (40kW) and non-critical industrial load (20kW), started the energy storage to discharge 50kW (maximum output), and maintained the power supply to the hospital's backup load (30kW) and the residential basic load (50kW) until the voltage was restored;

[0128] The specific execution flow of the method steps is as follows: Data acquisition and preprocessing: 0:00-24:00 Data such as photovoltaic, wind power, and load are collected through sensors. After cleaning (removing outliers) and standardization (Z-score processing), the data is stored in the time series database.

[0129] Data fusion analysis: Extract the nonlinear characteristics of photovoltaic-irradiance and the daily cycle characteristics of load, and weight and fuse the results of the three photovoltaic prediction models (weight 0.4 / 0.25 / 0.35) to output the predicted value for the next 24 hours (e.g., photovoltaic output of 100kW from 8:00 to 16:00).

[0130] Intelligent scheduling decision-making: Input fused data, short-term scheduling uses particle swarm optimization to solve multi-objective optimization (minimize cost and carbon emissions) to generate a 24-hour scheduling plan; real-time scheduling uses DRL for dynamic adjustment (such as compensating for photovoltaic output deviation).

[0131] Command execution and feedback: Distribute scheduling commands to equipment, monitor execution status in real time, and trigger secondary optimization when deviation exceeds limits, forming a closed loop of "collection-fusion-decision-execution-feedback";

[0132] Through this embodiment, the virtual power plant in the industrial park has achieved efficient integration and intelligent scheduling of multi-source data, increased the renewable energy consumption rate to 98%, reduced operating costs by 15%, and reduced carbon emissions by 20%, thus verifying the effectiveness of the system.

[0133] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A virtual power plant intelligent dispatching and decision-making system based on multi-source data fusion, characterized in that: The system comprises a data acquisition module, a communication module, a data fusion module, an intelligent decision-making module, and an execution control module. The data acquisition module is deployed at distributed energy resource nodes and power grid monitoring points, connecting to intelligent sensors, smart meters, meteorological monitoring equipment, energy storage status monitors, and other multi-source heterogeneous data sources, including real-time power grid operation data, distributed energy generation data, energy storage device status data, user electricity consumption data, and meteorological data. The communication module employs edge computing technology to achieve low-latency data transmission, supporting compatibility between power-specific communication protocols and general protocols, ensuring data transmission between modules and real-time uploading to the cloud platform. The data fusion module uses data cleaning, standardization, and spatiotemporal alignment technologies to process the acquired data. The system preprocesses multi-source data to eliminate data noise and format differences. Based on feature extraction and correlation analysis algorithms, it mines potential relationships between different types of data, constructs a unified data model, and performs weighted fusion of multi-source prediction results. The weights are dynamically adjusted based on historical prediction errors to improve prediction reliability and provide structured data support for subsequent decision-making. The intelligent decision-making module includes an objective function construction module, a constraint processing module, and an algorithm scheduling center. The built-in intelligent algorithms can combine external information such as electricity market price signals and power grid dispatch instructions to analyze and calculate the data output by the data fusion module with system economy, stability, and renewable energy absorption rate as objectives, and generate the optimal dispatch decision scheme. The execution control module sends the scheduling instructions generated by the intelligent decision-making module to the distributed energy resource control terminal, realizing precise control of the charging and discharging of power generation equipment and energy storage equipment, and the adjustment of controllable load, and provides real-time feedback on the execution results, forming a closed-loop scheduling management.

2. The intelligent dispatching and decision-making system for virtual power plants based on multi-source data fusion according to claim 1, characterized in that: The data cleaning, standardization, and spatiotemporal alignment techniques of the data fusion module in the virtual power plant intelligent dispatching decision-making system based on multi-source data fusion are as follows: Data cleaning: False alarm data from sensors are removed using outlier detection algorithms, and missing data is filled in using spatiotemporal interpolation based on adjacent node data and time trends; Standardization: Convert data with different units into a unified dimension, and use the Z-score standardization formula to process numerical data to ensure consistency of algorithm input; Spatiotemporal alignment: Synchronize multi-source data based on timestamps, align different collected data to the same minute-level timeline, and associate spatially distributed device data through geographic information tags.

3. The intelligent dispatching and decision-making system for virtual power plants based on multi-source data fusion according to claim 1, characterized in that: The feature-level fusion of the data fusion module of the virtual power plant intelligent dispatching decision system based on multi-source data fusion is to extract key features, extract spatial features of the data through convolutional neural networks based on the nonlinear relationship and daily periodic features of the data, and perform weighted fusion by combining the time features extracted by LSTM. The weights are dynamically adjusted based on historical prediction errors.

4. The intelligent dispatching and decision-making system for virtual power plants based on multi-source data fusion according to claim 1, characterized in that: The intelligent decision-making module of the virtual power plant intelligent dispatching and decision-making system based on multi-source data fusion includes an objective function construction module, a constraint condition processing module, and an algorithm dispatching center, specifically: Objective function construction module: Based on multi-objective optimization, including economic, technical and environmental objectives. The economic objective is to minimize the operating cost of the virtual power plant; the technical objective is to maximize the renewable energy absorption rate, maintain grid frequency stability and ensure that the energy storage SOC is within the safe range of 20%-80%; the environmental objective is to minimize carbon emissions based on the power generation coefficient of thermal power units. The constraint processing module includes power balance constraints, equipment physical constraints, and grid security constraints. The power balance constraint ensures that distributed power generation output + energy storage charging and discharging power + purchased electricity from outside the grid equals user load. The equipment physical constraint ensures that the maximum output of photovoltaic / wind turbines is ≤ rated power, and the energy storage charging and discharging power is ≤ maximum allowable value. The grid security constraint ensures that node voltage deviation is ≤ ±5%, and line transmission power is ≤ thermal stability limit. Algorithm Scheduling Center: The algorithms include Long Short-Term Memory (LSTM) network algorithms, Particle Swarm Optimization (PSO) algorithms, and Deep Reinforcement Learning (DRLE) algorithms. It uses an algorithm selection mechanism to make decision outputs and issue instructions, and can also perform feedback monitoring and emergency handling.

5. The intelligent dispatching and decision-making system for virtual power plants based on multi-source data fusion according to claim 4, characterized in that: The algorithms of the intelligent decision-making module of the virtual power plant intelligent dispatch decision-making system based on multi-source data fusion, namely the long short-term memory network algorithm, particle swarm optimization algorithm, and deep reinforcement learning algorithm of the dispatch center, are as follows: Long Short-Term Memory (LSTM) network algorithm: Based on a special recurrent neural network, it excels at processing time-series data and can effectively capture long-term dependencies in data. In virtual power plant scheduling, the algorithm is used to predict the power generation of distributed energy sources and user load demand. By inputting time-series information such as historical power generation data, meteorological data, and user electricity consumption behavior data, the algorithm model can learn the changing patterns of the data and predict the power generation and consumption trends in future periods, providing accurate predictive data support for scheduling decisions. Particle Swarm Optimization Algorithm: A global optimization algorithm based on swarm intelligence, simulating the behavior of flocks of birds foraging or schools of fish swimming. In virtual power plant scheduling, the algorithm is used to solve multi-objective optimization problems. The optimization objectives are to minimize system operating costs, minimize carbon emissions, and maximize renewable energy absorption rate. The scheduling power of each distributed energy resource is used as the optimization variable. The objective function is constructed by combining the grid constraints. The algorithm finds the optimal scheduling scheme that satisfies multiple objectives through iterative search of particles in the solution space. Deep reinforcement learning algorithm: Combining the perception capabilities of deep learning with the decision-making capabilities of reinforcement learning, it autonomously learns the optimal strategy in complex environments. In the virtual power plant scheduling scenario, the operating state of the virtual power plant is used as the state space, and the scheduling instructions are used as the action space. The reward function is constructed based on indicators such as system benefits and grid stability. The algorithm continuously interacts with the environment and optimizes the scheduling strategy based on reward feedback, thereby achieving adaptive and dynamic intelligent scheduling decisions.

6. The intelligent dispatching and decision-making system for virtual power plants based on multi-source data fusion according to claim 5, characterized in that: The prediction of the Long Short-Term Memory (LSTM) network algorithm of the intelligent decision-making module of the virtual power plant intelligent dispatch decision-making system based on multi-source data fusion is as follows: Long Short-Term Memory Network Prediction Module: Input: Historical time series data {PPV,thist,PW,thist,Lthist,Mthist}t=1thist, where thist is the length of the historical data; Output: Forecast values ​​for the future time period {PPV, tpred, PW, tpred, Ltpred} t=1T; The specific steps are as follows: 1) Data preprocessing: Normalize historical data, imput missing data, and construct input feature sequences; 2) Construction of Long Short-Term Memory Network Algorithm Model: The input layer accepts feature sequences, the number of hidden neurons in the Long Short-Term Memory Network layer is set, and long-term dependencies are learned through forget gate, input gate, and output gate. The output layer outputs the predicted value through a fully connected layer. 3) Model training: Train the long short-term memory network using historical data, with mean squared error as the loss function, and optimize the parameters through backpropagation until the loss converges; 4) Predictive output: Input the latest historical data, and output the historical photovoltaic power generation, historical wind power generation, and historical user load demand for future periods through a trained long short-term memory network.

7. The method for a virtual power plant intelligent dispatching decision-making system based on multi-source data fusion according to claims 1-5, characterized in that: The implementation steps of the virtual power plant intelligent dispatching and decision-making system based on multi-source data fusion are as follows: Step 1: Data Acquisition and Preprocessing; Multi-source heterogeneous data is acquired through the data acquisition module, and cleaned and standardized in the data fusion module. Missing data is filled using interpolation or machine learning-based prediction methods to ensure data integrity and consistency. Step 2: Data Fusion and Analysis; Key features of the data are extracted using feature engineering techniques, and temporal and spatial relationships between data are explored through spatiotemporal correlation analysis algorithms; Principal component analysis, independent component analysis, and other dimensionality reduction algorithms are used to reduce data dimensionality and improve computational efficiency. Step 3: Intelligent scheduling decision-making; The fused data is input into the intelligent decision-making module, and based on the preset scheduling objectives, an appropriate intelligent algorithm is selected to solve the problem and generate scheduling instructions for each distributed energy resource; Step 4: Command Execution and Feedback; The execution control module sends the scheduling command to the corresponding device and collects the device operating status data in real time and feeds it back to the intelligent decision-making module so as to dynamically adjust the scheduling plan and ensure the stable operation of the system.

Images