Multi-energy complementary optimization scheduling method and system for distributed energy of virtual power plant

Abstract

The invention discloses a multi-energy complementary optimal scheduling method and system for distributed energy of a virtual power plant, belongs to the technical field of virtual power plants and energy management, and particularly relates to a virtual power plant distributed energy optimal scheduling method and system based on dynamic monitoring, multi-target optimization and multi-energy complementation. The method is suitable for a complex energy system comprising photovoltaic, wind energy and energy storage equipment and a controllable load, and is especially suitable for improving the stability and the energy efficiency of a power grid in a high-proportion renewable energy access scene.

Description

Technical Field

[0001] This invention belongs to the field of virtual power plants and energy management technology, specifically a method and system for optimizing distributed energy scheduling in virtual power plants based on dynamic monitoring, multi-objective optimization, and multi-energy complementarity. This method is applicable to complex energy systems that include photovoltaics, wind power, energy storage equipment, and controllable loads, and is particularly targeted at improving grid stability and energy efficiency in scenarios with a high proportion of renewable energy access. Background Art

[0002] With the large-scale integration of distributed energy resources (such as photovoltaics and wind power), virtual power plants (VPPs) need to coordinate the output volatility of multiple energy sources, the dynamic load demand, and the stability of the power grid. Existing technologies have the following problems: 1. Strong dependence on static models: Traditional scheduling methods are mostly based on historical static data, which makes it difficult to adapt to the dynamic changes in energy production and load demand in real time, resulting in low energy efficiency; 2. Inadequate coordination mechanisms: There is a lack of effective coordination between distributed energy resources, and energy storage systems and load-side management are not fully optimized, resulting in serious energy waste; 3. High algorithm complexity: Existing optimization models (such as mixed integer nonlinear programming) are computationally complex, difficult to meet high-frequency scheduling needs, and do not incorporate the complementary characteristics of multiple energy sources; 4. Low data security and transparency: Operational data records rely on centralized systems, which are subject to tampering risks and affect the credibility of scheduling decisions.

[0003] To address the above issues, the present invention proposes a scheduling method that integrates dynamic monitoring, multi-energy complementary prediction and intelligent optimization, and realizes the efficient, safe and adaptive operation of virtual power plants through the Internet of Things, machine learning and blockchain technology. Summary of the Invention

[0004] The core objective of this invention is to provide a multi-energy complementary optimization scheduling method and system to solve the energy coordination, economic scheduling and robust operation problems of virtual power plants in a dynamic environment. Specifically, it includes:

[0005] 1. Accurately match energy supply and demand through dynamic monitoring and multi-model forecasting;

[0006] 2. Combine multi-objective optimization algorithms with topology analysis to reduce system energy consumption and costs;

[0007] 3. Use blockchain and machine learning to improve data security and scheduling adaptability.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions and method implementation steps: Step S1: Dynamic data collection and preprocessing Data source: Deploy an IoT sensor network to collect real-time operating data from distributed power sources (photovoltaic, wind power), energy storage devices (batteries, supercapacitors), and loads (industrial, commercial), including power, temperature, wind speed, light intensity, and electricity price signals. Edge computing: Data cleaning (denoising, outlier removal) and normalization are performed at local nodes, and data is preliminarily classified through lightweight AI models (such as random forests) to reduce cloud transmission pressure. Step S2: Multi-energy output and load forecasting Photovoltaic output forecast: Using the LSTM model, we input meteorological data (radiation intensity, cloud cover, temperature) and historical output sequences, and output the predicted photovoltaic power value for the next 24 hours. The formula is as follows: Among them, G(t) is the solar radiation intensity, T(t) is the photovoltaic panel temperature, H PV Provide historical data. Wind power output prediction: Based on the Weibull distribution fitting of wind speed probability density, the output is calculated in combination with the wind turbine characteristic curve: Load forecasting: The ARMA model is used to analyze historical load time series, and ambient temperature and electricity price fluctuation factors are introduced for dynamic correction: Step S3: Multi-objective optimization scheduling modeling Objective function: in, is the overall system efficiency, λ(t) is the real-time electricity price, and the weight coefficients ω1, ω2, and ω3 are dynamically adjusted through the entropy weight method. Constraints: Power balance: P pv +P Wind +P storage +P Grid =L 总 ■Energy storage charge and discharge limit: SOC min ≤SOC(t)≤SOC max ■Voltage stability: V min ≤V i (t)≤V max Step S4: Hybrid optimization algorithm solution MOPSO-GA hybrid strategy: 1) Fast non-dominated sorting: Initialize the particle swarm, calculate the objective function value, and stratify according to the dominance relationship; 2) Elite retention: merge the current and historical non-dominated solutions and filter the diverse solution set by crowding distance; 3) Genetic operation: perform crossover (simulating binary crossover) and mutation (Gaussian mutation) on the particle swarm to avoid local optimality. Gradient descent local optimization: for energy storage charging and discharging rate γ charge / discharge , with the goal of minimizing the power matching error, dynamically adjust: Step S5: Dynamic Scheduling and Feedback Control Topology priority scheduling: Based on the electrical connection relationship, the nearest distributed energy is prioritized to reduce line losses. Demand response strategies: During low electricity price periods: Energy storage is charged and controllable loads (such as air conditioners and chillers) increase operating power. During high electricity price periods: Energy storage is discharged, and the load power drops to the minimum guaranteed value. Blockchain evidence storage: Operation data (output, electricity price, dispatch instructions) is encrypted and uploaded to the chain to ensure that it cannot be tampered with. Neural network evaluation: Build an LSTM-BP hybrid network to evaluate system energy efficiency in real time and dynamically correct prediction model parameters.

[0009] As a preferred solution of the multi-energy complementary optimization scheduling method and system of the virtual power plant distributed energy described in the present invention, the system is composed of the following modules (see Appendix Figure 1 ):

[0010] Data acquisition module: including photovoltaic / wind energy sensors, smart meters, and edge computing gateways.

[0011] Predictive Analysis Module: Runs LSTM (energy forecasting) and ARMA (load forecasting) models in parallel.

[0012] Optimization solution module: MOPSO-GA algorithm engine, outputs Pareto optimal solution set.

[0013] Dynamic scheduling module: combines topology analysis, real-time electricity prices and energy storage status to generate scheduling instructions.

[0014] Feedback control module: blockchain-based data storage and neural network dynamic optimization.

[0015] The beneficial effects of the present invention are: Improved efficiency of multi-energy complementarity: The prediction error of the LSTM and ARMA dual models is reduced to less than 5%, and the photovoltaic-energy storage synergy efficiency is increased by 15%. Economic optimization: Through time-of-use electricity price response, the electricity purchase cost is reduced by 20% to 30%. Computational efficiency: The MOPSO-GA hybrid algorithm is 40% faster than the traditional MINLP algorithm, meeting 5-minute real-time scheduling requirements. Security and reliability: Blockchain technology ensures data transparency, neural networks dynamically correct model deviations, and system failure rates are reduced by 50%. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 : System architecture diagram, showing data flow and module interaction.

[0017] Figure 2 :MOPSO-GA algorithm flow chart, including initialization, non-dominated sorting, genetic operation and other steps.

[0018] Figure 3 :Example of multi-energy complementary scheduling scenario (photovoltaic + energy storage + grid coordination). DETAILED DESCRIPTION

[0019] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] Example 1: A virtual power plant in an industrial park includes the following equipment: 1) 500kW photovoltaic system, installed on the factory roof; 2) 300kW wind energy is distributed outside the park; 3) 1MWh lithium battery energy storage, with a charge and discharge efficiency of 92%; 4) Controllable load: chiller (maximum power 200kW) and central air conditioning system (total power 150kW); 5) Grid connection: 10kV distribution network access, time-of-use electricity price difference between peak and valley is 0.8 yuan / kWh (peak period: 18:00-22:00, valley period: 0:00-6:00)

[0021] Implementation steps: 1. Data Collection and Preprocessing ○Sensor deployment: The photovoltaic panels are equipped with temperature sensors (DS18B20) and irradiance sensors (LI-COR LI-200R), with a sampling frequency of 1 minute. The wind turbine is equipped with an anemometer (NRG Symphonie) and a wind direction sensor, with a sampling frequency of 30 seconds; ■The energy storage system is connected to the BMS (battery management system) to monitor SOC (state of charge) and charging and discharging power in real time. ○Edge computing nodes: Use RT200W as the local gateway and run the data cleaning algorithm (median filtering + 3σ outlier removal); Data normalization formula: 2. Multi-energy output and load forecasting Photovoltaic output prediction (LSTM model): ■Input data: historical 72-hour irradiance, temperature, and output data, with a time step of 15 minutes; ■Model structure: 2-layer LSTM (128 units) + Dropout (0.2) + Dense layer (outputs 24-hour forecast); ■ Training results: Test set RMSE = 3.2%, the next day’s predicted peak value is 480kW (actual value 496kW). ○Load forecasting (ARMA model): ■Input data: cooling load time series for the past 7 days, ambient temperature, and real-time electricity price; ■Model parameters: AR order p = 3, MA order q = 2; ■Dynamic correction: For every 1°C increase in temperature, the load increases by 5%; for every 0.1 yuan / kWh increase in electricity price, the load decreases by 2%. 3. Multi-objective optimization scheduling solution ○Objective function weight setting: ■Energy consumption weight w1 = 0.5, cost weight w2 = 0.3, efficiency weight w3 = 0.2; ○MOPSO-GA algorithm parameters: parameter value Number of particles 50 Number of iterations 100 Crossover probability 0.8 (simulated binary crossover) Mutation probability 0.1 (Gaussian variation) ○Optimization result: The Pareto frontier solution is concentrated to select the solution with the lowest total cost ( Figure 2 ), the scheduling strategy is as follows: ■ Off-peak period (0:00-6:00): Power is purchased from the grid (price 0.3 yuan / kWh) and charged to SOC = 90%; ■ Flat period (6:00-18:00): PV power supply is prioritized, with any shortfalls supplemented by energy storage; Peak period (18:00-22:00): Energy storage discharges 300kW and chiller power is reduced to 120kW. 4. Dynamic Scheduling and Execution ○Topology priority rules: Nearest power supply: Photovoltaic power supply will be given priority to plant No. 1 (closest), and wind power will be used to supply peripheral equipment; Line loss screening: Eliminate nodes with voltage deviation rates greater than 5% (verified by power flow calculation). ○Energy storage charging and discharging control: Charging phase: constant current charging (0.5C), power limit is 200kW; ■Discharge phase: Dynamically adjust power based on gradient descent algorithm, with matching error <2%. 5. Feedback and Adaptive Adjustment ○Blockchain evidence storage: ■Hyperledger Fabric chaincode records hourly output, electricity prices, and SOC data, generating an unalterable log; ○Neural Network Evaluation: ■Build an LSTM-BP hybrid model, input real-time operation data, and output energy efficiency score (0-100); ■If the score is < 80 for three consecutive times, trigger model retraining (update LSTM hyperparameters).

[0022] The above implementation methods verify the feasibility and superiority of the present invention in complex scenarios through specific parameters, algorithm configuration and measured data, and provide a complete technical path for the multi-energy complementary optimization of virtual power plants.

Claims

1. A multi-energy complementary optimization scheduling method and system for distributed energy in a virtual power plant, characterized in that: include: Dynamic data collection, LSTM-ARMA dual model prediction, MOPSO-GA hybrid optimization, energy storage dynamic peak regulation and blockchain feedback control.

2. The method according to claim 1, characterized in that The multi-objective optimization objective function includes energy consumption, cost and efficiency weight coefficients. in, is the overall efficiency of the system, λ(t) is the real-time electricity price, and the weight coefficients ω1, ω2, and ω3 are dynamically adjusted by the entropy weight method. The constraints are: Power balance: P pv +P Wind +P storage +P Grid =L 总 ■Energy storage charge and discharge limit: SOC min ≤SOC(t)≤SOC max ■Voltage stability: V min ≤V i (t)≤V max 3. The method according to claim 1, characterized in that A gradient descent algorithm is used to match the energy storage charge and discharge rates with load requirements. For energy storage charge and discharge rate γ charge / discharge , with the goal of minimizing the power matching error, dynamically adjust:

4. A system for implementing the method according to claims 1-3, characterized in that: include Step S1: Data collection, deploying an IoT sensor network to collect real-time operating data of distributed power sources (photovoltaic, wind power), energy storage devices (batteries, supercapacitors), and loads (industrial, commercial), including power, temperature, wind speed, light intensity, and electricity price signals. Step S2: Predictive analysis, accurately matching energy supply and demand based on distributed energy model analysis and load forecasting. Step 3: Optimization solution, through multi-objective optimization scheduling modeling analysis, MOSO-GA hybrid algorithm and neural network evaluation algorithm, multi-dimensional, multi-model calculation to obtain the optimal solution set for dynamic scheduling. Step S4: The dynamic scheduling and feedback control module dynamically adjusts and controls the execution of the optimal solution set obtained by the hybrid algorithm to achieve the optimal effect.

Images