V2G vehicle cluster ordered charging and discharging scheduling system based on multi-objective optimization

Abstract

The invention aims to provide the V2G vehicle cluster ordered charging and discharging scheduling system based on multi-objective optimization, the response rate is high, the carbon emission reduction is high, the cycle life of the battery side is prolonged, the load peak-valley difference of the power grid side is relatively low, and the annual charging and discharging cost of the user side is relatively low. The system comprises a cloud scheduling platform, a vehicle end response module, a third-party SDK aggregator and an incentive mechanism platform, the cloud scheduling platform is in communication connection with the third-party SDK aggregator, and the cloud scheduling platform is in communication connection with the vehicle end response module through a plurality of edge computing nodes. The vehicle end response module is in communication connection with the incentive mechanism platform, the cloud scheduling platform is provided with a five-dimensional target optimization engine, a GNN-MOPSO module and a DDPG real-time engine, and the incentive mechanism platform is in communication connection with a carbon exchange and an electricity market. The method is applied to the technical field of power grid dispatching.

Description

Technical Field

[0001] This invention is applied to the technical field of power grid dispatching, and in particular relates to a V2G vehicle cluster orderly charging and discharging dispatching system based on multi-objective optimization. Background Technology

[0002] Smart grid demand response and V2G (Vehicle-to-Grid) technology for new energy vehicles are important directions for the coordinated optimization of power systems and new energy vehicles. They achieve power supply and demand balance and user benefits through bidirectional charging and discharging. V2G technology allows new energy vehicles to charge and store electrical energy during off-peak hours and then transmit electrical energy back to the grid during peak or emergency periods, forming a "mobile energy storage" function.

[0003] Currently, V2G scheduling mainly adopts the following two schemes: A. Centralized optimization: The power grid dispatch center collects all vehicle SOC, electricity price, and load data, establishes a linear programming model with a single objective, and issues charging and discharging plans. B. Distributed Heuristic: Using charging piles or aggregators as units, heuristic algorithms such as genetic algorithms and particle swarm optimization are used to continuously optimize local charging and discharging power.

[0004] However, existing technologies have the following drawbacks: ① The goal is singular, and the peak-valley difference is reduced by only 8%-10%, which cannot take into account the interests of the power grid, users, batteries, carbon emissions, and response reliability. ② Battery-free cycle aging quantification model, annual capacity decay rate after scheduling ≥3%; ③ The centralized architecture has poor real-time performance. When tens of thousands of vehicles are connected, the computational dimension explodes, and the latency is >1 second. ④ Lack of user profiling and differentiated incentives, resulting in a response rate of <22%; ⑤ The dual-track mechanism of carbon credits and electricity prices leaves no room for a viable business model.

[0005] For example, Chinese patent CN119834268A discloses a method for smoothing frequency fluctuations based on V2G technology, which adopts a centralized optimization architecture. However, this method has poor real-time performance, and the computational dimension explodes when tens of thousands of vehicles are connected, resulting in a delay of >1 second. Therefore, it is necessary to provide a V2G vehicle cluster orderly charging and discharging scheduling system based on multi-objective optimization, which has a high response rate, high carbon emission reduction, extended battery cycle life, lower load peak-valley difference on the grid side, and lower annual charging and discharging cost on the user side. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a V2G vehicle cluster orderly charging and discharging scheduling system based on multi-objective optimization, which has high response rate, high carbon emission reduction, extended cycle life on the battery side, low load peak-valley difference on the grid side, and low annual charging and discharging cost on the user side.

[0007] The technical solution adopted in this invention is as follows: This invention includes a cloud scheduling platform, a vehicle-side response module, a third-party SDK aggregator, and an incentive mechanism platform. The cloud scheduling platform is communicatively connected to the third-party SDK aggregator. The cloud scheduling platform is communicatively connected to the vehicle-side response module through several edge computing nodes. The vehicle-side response module is communicatively connected to the incentive mechanism platform. The cloud scheduling platform is deployed with a five-dimensional objective optimization engine, a GNN-MOPSO module, and a DDPG real-time engine. The incentive mechanism platform is communicatively connected to the carbon exchange and the electricity market. Among them, GNN-MOPSO is a combination of graph neural networks and multi-objective particle swarm optimization algorithm. The GNN-MOPSO module is a hybrid method that combines the structured data processing capability of graph neural networks with the global search capability of multi-objective optimization algorithms. DDPG is Deep Deterministic Policy Gradient, which is an algorithm that combines deep reinforcement learning and deterministic policy gradient. SDK is a software development kit. An SDK is a collection of development tools used by software engineers to create application software for specific software packages, software frameworks, hardware platforms, operating systems, etc. It generally appears in the form of a collection of APIs, documentation, examples, and tools.

[0008] As can be seen from the above scheme, this application constructs a five-dimensional refined objective function to achieve a win-win situation for the power grid, users, batteries, carbon emissions, and response reliability; establishes a quantitative model for battery aging to ensure that the annual capacity decay rate is ≤1%; adopts a "cloud-edge-device" layered architecture to ensure that the access latency for tens of thousands of vehicles is <50 ms; introduces user profiles and differentiated incentives to improve the response rate to ≥57%; and designs a dual-track mechanism of carbon credits and electricity prices to form a sustainable commercial closed loop.

[0009] A preferred embodiment is that the five-dimensional objective optimization engine includes a five-dimensional objective function, which includes formulas 1-5, where formula 1 is for grid load fluctuation: in, Let t be the total load of the power grid. The average load during the scheduling period; Formula 2 - User charging and discharging costs: in, and Let be the electricity purchase price and the electricity sales price at time t, respectively. and These represent the power purchased and the power sold at time t, respectively. Formula 3 - Battery Cycle Aging: in, This indicates the depth of SOC change in a single cycle. This indicates the total cycle life of the battery at a specific cycle depth; Formula 4 - Dynamic weighting of carbon emission factors: in, The carbon emission factor at time t is dynamically adjusted based on the real-time carbon intensity of the power grid. Formula 5 - User Response Reliability: Determined by K-Prototypes profiles, among which, Let i be the response weight. This represents the historical response rate for user i.

[0010] A preferred embodiment is that both the GNN-MOPSO module and the DDPG real-time engine are used as hybrid intelligent algorithms, with the GNN-MOPSO module applied to the day-ahead phase and the DDPG real-time engine applied to the real-time phase. The current phase includes the following steps: Step S1, collecting historical load, electricity price, weather, and vehicle travel data; Step S2: Model the vehicle-pile-network topology using GNN and output the node embeddings; Step S3: Embed and initialize MOPSO particles, and iteratively solve the Pareto front; Step S4: Output the charging and discharging plan for the next day at 96 points; The real-time phase includes the following steps: Step S5: Collect real-time deviations at edge nodes; Step S6: DDPG network performs online inference and outputs corrected power ΔP; Step S7: The edge node issues a correction command with a delay of <50 ms.

[0011] A preferred embodiment is that the vehicle-side response module includes a SOC / SOH prediction unit, a federated learning client, and a secure communication unit. The cloud-based scheduling platform updates the global reference plan every 15 minutes. The edge computing nodes complete local state estimation every 1 second. The vehicle-side response module transmits SOC, SOH, and charging demand every 100 ms. The federated learning client aggregates model gradients every 24 hours without uploading raw data. SOC is the state of charge, and SOH is the state of health. SOC and SOH are core parameters in the battery management system, used to evaluate the battery's remaining capacity and overall performance.

[0012] A preferred embodiment is that the carbon exchange corresponds to a carbon credit track, where carbon credits are calculated as: response electricity × carbon emission factor × kc CO2. The generated carbon credits are stored in a credit wallet and can be redeemed for charging vouchers, insurance discounts, and parking benefits. The electricity market corresponds to an electricity price subsidy track, which uses a Vickrey-Clarke-Groves secondary auction, with high-reliability users having priority in winning bids and receiving premium subsidies.

[0013] A preferred approach is to use K-Prototypes to cluster the raw data, including age, travel mileage, charging time period, SOC range, and historical response rate, to generate three categories of response priorities: high, medium, and low. In the MOPSO objective function, different weights are assigned to the five-dimensional objectives f1-f5 to achieve different prices for the same charging station and different weights for the same vehicle.

[0014] A preferred embodiment is that the cloud-based scheduling platform provides aggregator APIs, carbon exchange APIs, grid demand response APIs, payment APIs, and regulatory APIs. Attached Figure Description

[0015] Figure 1 This is the overall system architecture diagram; Figure 2 This is a diagram of the five-dimensional optimization objective function structure; Figure 3 This is a flowchart of the GNN-MOPSO+DDPG hybrid algorithm; Figure 4 This is a time sequence diagram of cloud-edge-device collaborative data flow; Figure 5 This is a schematic diagram of the dual-track incentive mechanism of carbon credits and electricity prices; Figure 6 It is a diagram of user profile clustering and differentiated scheduling strategies; Figure 7 This is a diagram of the multi-API access architecture of the business model platform. Detailed Implementation

[0016] like Figure 1As shown, in this embodiment, the present invention includes a cloud scheduling platform 1, a vehicle-side response module 2, a third-party SDK aggregator 3, and an incentive mechanism platform 4. The cloud scheduling platform 1 is communicatively connected to the third-party SDK aggregator 3. The cloud scheduling platform 1 is communicatively connected to the vehicle-side response module 2 through several edge computing nodes 5. The vehicle-side response module 2 is communicatively connected to the incentive mechanism platform 4. The cloud scheduling platform 1 is deployed with a five-dimensional objective optimization engine, a GNN-MOPSO module, and a DDPG real-time engine. The incentive mechanism platform 4 is communicatively connected to a carbon exchange and an electricity market. GNN-MOPSO is a combination of graph neural networks and multi-objective particle swarm optimization algorithm. The GNN-MOPSO module is a hybrid method that combines the structured data processing capability of graph neural networks with the global search capability of multi-objective optimization algorithms. DDPG is Deep Deterministic Policy Gradient, an algorithm that combines deep reinforcement learning and deterministic policy gradient. SDK is a software development kit, which is a collection of development tools used by software engineers to create application software for specific software packages, software frameworks, hardware platforms, operating systems, etc. It generally appears in the form of a collection of APIs, documentation, examples, and tools.

[0017] The cloud-based scheduling platform 1 deploys a five-dimensional target optimization engine, a GNN-MOPSO module, and a DDPG real-time engine; the edge computing node 5 uses Docker microservices, with a CPU of ≤4 cores, memory of ≤8 GB, and supports MQTT+TLS1.3; the vehicle-side response module 2 includes SOC / SOH prediction, a federated learning client, and a secure communication unit; the incentive mechanism platform 4 connects with the carbon exchange and the electricity market, providing interfaces for carbon credit clearing and electricity price subsidy settlement; and the third-party SDK aggregator 3 provides an open API to support time-of-use pricing and demand response event access.

[0018] like Figure 2 As shown, in this embodiment, the five-dimensional target optimization engine includes a five-dimensional optimization objective function, which includes formulas 1-5, where formula 1 is for grid load fluctuation: in, Let t be the total load of the power grid. The average load during the scheduling period; Formula 2 - User charging and discharging costs: in, and Let be the electricity purchase price and the electricity sales price at time t, respectively. and These represent the power purchased and the power sold at time t, respectively. Formula 3 - Battery Cycle Aging: in, This indicates the depth of SOC change in a single cycle. This indicates the total cycle life of the battery at a specific cycle depth; Formula 4 - Dynamic weighting of carbon emission factors: in, The carbon emission factor at time t is dynamically adjusted based on the real-time carbon intensity of the power grid. Formula 5 - User Response Reliability: Determined by K-Prototypes profiles, among which, Let i be the response weight. This represents the historical response rate for user i.

[0019] like Figure 1 and Figure 2 As shown, in this embodiment, the five-dimensional optimization objective function corresponds to a five-dimensional objective: f1: Minimize grid load fluctuations; f2: Minimize user charging and discharging costs; f3: Minimize battery cycle aging loss (based on Arrhenius+Cycle-life model). f4: Dynamic weighted minimization of carbon emission factors; f5: Maximize user response reliability (profile weight × response rate).

[0020] like Figure 3 As shown, in this embodiment, both the GNN-MOPSO module and the DDPG real-time engine are used as hybrid intelligent algorithms. The GNN-MOPSO module is applied to the day-ahead phase, and the DDPG real-time engine is applied to the real-time phase. The current phase includes the following steps: Step S1: Collect historical load, electricity price, weather, and vehicle travel data; Step S2: Model the vehicle-pile-network topology using GNN and output the node embeddings; Step S3: Embed and initialize MOPSO particles, and iteratively solve the Pareto front; Step S4: Output the charging and discharging plan for the next day at 96 points; The real-time phase includes the following steps: Step S5: Collect real-time deviation frequency, electricity price, and SOC at the edge nodes; Step S6: DDPG network performs online inference and outputs corrected power ΔP; Step S7: The edge node issues a correction command with a delay of <50 ms.

[0021] Daytime scheduling (core pseudocode): Input: Vehicle-pile-network adjacency matrix A, node features X; Output: Pareto frontier charge and discharge plan (taking into account five dimensions from f1 to f5); 1. GNN ← 3-layer GCN(SiLU, 64dim) → embed; 2. MOPSO ← Particle 120, Iteration 200, Embedding Initialization; 3. Non-dominated sorting → Pareto set; 4. Fuzzy entropy weight selection → 96-point plan.

[0022] DDPG Real-Time Correction (Core Pseudocode): Inputs: Day-ahead plan deviation ΔP, real-time SOC, electricity price, SOH; Output: Millisecond-level power correction ΔP, delay < 50 ms; 1. State s = [ΔP, ΔSOC, price, SOH]; 2. Action a = [Pcharge, Psell]; 3. Rewards

[0023] (Where, the weight λ is set by expert experience and is associated with the Pareto weights of MOPSO). 4. Network with 2×128 ReLU, learning rate 3e-4, experience pool ; 5. Delay < 50 ms.

[0024] Closed-loop update: The DDPG experience pool → federated learning gradients → GNN embedding fine-tuning form a "day-to-day - real-time - closed-loop" rolling optimization. Specifically, federated learning aggregates the gradients from each edge node or vehicle (without uploading the original data) to fine-tune the GNN model in the cloud, improving the accuracy of day-to-day planning.

[0025] like Figure 4 As shown, in this embodiment, the vehicle-side response module 2 includes a SOC / SOH prediction unit, a federated learning client, and a secure communication unit. The cloud scheduling platform 1 updates the global reference plan every 15 minutes. The edge computing node 5 completes local state estimation every 1 second. The vehicle-side response module 2 transmits SOC, SOH, and charging demand every 100 ms. The federated learning client aggregates model gradients every 24 hours without uploading raw data. SOC is the state of charge, and SOH is the state of health. SOC and SOH are core parameters in the battery management system, used to evaluate the battery's remaining capacity and overall performance.

[0026] like Figure 5As shown, in this embodiment, the carbon exchange corresponds to the carbon credit track, where carbon credits are calculated as: response electricity × carbon emission factor × kc0.8 kgCO2 / credit. The generated carbon credits are stored in a credit wallet and can be redeemed for charging vouchers, insurance discounts, and parking benefits. The electricity market corresponds to the electricity price subsidy track, which uses a Vickrey-Clarke-Groves VCG secondary auction, with high-reliability users having priority in winning bids and receiving premium subsidies.

[0027] like Figure 6 As shown, in this embodiment, the original data includes age, travel mileage, charging period, SOC range, and historical response rate. K-Prototypes is used to cluster the original data to generate three categories of response priorities: high, medium, and low. In the MOPSO objective function, different weights are assigned to the five-dimensional objectives f1-f5 to achieve different prices for the same charging station and different weights for the same vehicle.

[0028] like Figure 7 As shown in this embodiment, the cloud-based scheduling platform 1 provides aggregator APIs, carbon exchange APIs, grid demand response APIs, payment APIs, and regulatory APIs. It supports the simultaneous injection of three price signals: time-of-use pricing, demand response events, and carbon prices; aggregators can call the SDK to create virtual power plants (VPPs) for bidding.

[0029] In this embodiment, the key implementation parameters are as follows: scheduling period ΔT = 15 min; particle swarm size 120, iterations 200 generations; DDPG network 2×128 nodes, learning rate 3e-4, experience pool Federal LSTM-SOC predicts MAPE = 1.8%; Real-world testing on 1260 vehicles shows: peak-to-valley difference decreased by 27.4%, annual user revenue increased by 418 yuan, battery cycle life increased by 9.6%, and carbon emissions were reduced by 2.3 kg / vehicle / month.

[0030] In this embodiment, the present application has the following technical effects: On the grid side: the load peak-to-valley difference decreased by ≥25%, and the number of main transformer overloads decreased by ≥30%; On the user side: Annual charging and discharging costs decrease by ≥15%, and additional carbon credits yield ≥180 yuan / vehicle; Battery side: Cycle life extended by ≥9%, capacity decay rate reduced by ≥6%; Carbon emission side: Average monthly emission reduction per vehicle ≥ 2 kg 10,000 vehicles reduce emissions by approximately 240 tons per year; On the operations side: response rate improved by ≥35%, maintenance labor costs decreased by ≥40%, forming a sustainable commercial closed loop.

[0031] Table 1 shows the 6-month operating data of 1260 vehicles (comparative experiment, p < 0.01):

[0032] Although the embodiments of the present invention are described with reference to actual solutions, they do not constitute a limitation on the meaning of the present invention. Modifications to the embodiments and combinations with other solutions based on this specification will be obvious to those skilled in the art.

Claims

1. A multi-objective optimization-based V2G vehicle cluster orderly charging and discharging scheduling system, characterized in that: It includes a cloud scheduling platform (1), a vehicle end response module (2), a third party SDK aggregator (3) and an incentive mechanism platform (4), the cloud scheduling platform (1) is in communication connection with the third party SDK aggregator (3), the cloud scheduling platform (1) is in communication connection with the vehicle end response module (2) through a plurality of edge computing nodes (5), the vehicle end response module (2) is in communication connection with the incentive mechanism platform (4), the cloud scheduling platform (1) is deployed with a five-dimensional target optimization engine, a GNN-MOPSO module and a DDPG real-time engine, the incentive mechanism platform (4) is in communication connection with a carbon exchange and a power market. 2.The multi-objective optimization based V2G vehicle cluster orderly charging and discharging scheduling system according to claim 1, characterized in that, The five-dimensional objective optimization engine includes a five-dimensional optimization objective function including Equations 1-5, Equation 1 - grid load fluctuation: wherein, is the total grid load at time t, is the average load over the dispatch period; Formula 2 - User charge and discharge cost: Where, and are the electricity purchase price and electricity sale price at time t, respectively, and are the electricity purchase power and electricity sale power at time t, respectively. Equation 3 - Battery cycle aging: wherein, represents the SOC change depth for a single cycle, represents the total cycle life of the battery at a specific cycle depth; Formula 4 - Carbon emission factor dynamic weighting: Where, is the carbon emission factor at time t, dynamically adjusted according to the real-time carbon intensity of the grid. Formula 5 - User Response Reliability: is determined by the K-Prototypes image, where, is the response weight for user i, is the historical response rate for user i. 3.The multi-objective optimization based V2G vehicle cluster orderly charging and discharging scheduling system according to claim 2, characterized in that: The five-dimensional optimization target function corresponds to five-dimensional targets: f1: minimize grid load fluctuation; f2: minimize user charging and discharging cost; f3: minimize battery cycle aging loss (based on Arrhenius+Cycle-life model); f4: dynamically weighted minimization of carbon emission factor; f5: maximize user response reliability (image weight x response rate). 4.The multi-objective optimization based V2G vehicle cluster orderly charging and discharging scheduling system according to claim 1, characterized in that: The GNN-MOPSO module and the DDPG real-time engine are both hybrid intelligent algorithms, the GNN-MOPSO module is applied in the day-ahead stage, and the DDPG real-time engine is applied in the real-time stage; The day-ahead stage includes the following steps: Step S1, collect historical load, price, weather and vehicle travel data; Step S2, GNN modeling of vehicle-pile-grid topology, output node embedding; Step S3, embed MOPSO particles, and iteratively solve the Pareto frontier; Step S4, output the next day's 96-point charging and discharging plan; The real-time stage includes the following steps: Step S5, the edge node collects real-time deviations (frequency, price, SOC); Step S6, DDPG network online inference, output correction power ΔP; Step S7, the edge node issues a correction instruction with a delay of <50 ms. 5.The multi-objective optimization based V2G vehicle cluster orderly charging and discharging scheduling system according to claim 3, characterized in that: The vehicle end response module (2) includes an SOC / SOH prediction unit, a federated learning client and a secure communication unit, the cloud scheduling platform (1) updates the global reference plan every 15 minutes; the edge computing node (5) completes local state estimation every 1 second; the vehicle end response module (2) returns SOC, SOH and charging demand every 100 ms; the federated learning client aggregates model gradients once every 24 hours without uploading original data. 6.The multi-objective optimization based V2G vehicle cluster orderly charging and discharging scheduling system according to claim 1, characterized in that: The carbon exchange corresponds to a carbon credit track, carbon credits: response power x carbon emission factor x kc (0.8 kgCO2 / credit), the generated carbon credits are stored in a credit wallet, and the carbon credits can be exchanged for charging coupons, insurance discounts and parking discounts; the power market corresponds to a price subsidy track, and adopts Vickrey-Clarke-Groves (VCG) second auction, high reliability users are given priority in bidding and obtain premium subsidies. 7.The multi-objective optimization based V2G vehicle cluster orderly charging and discharging scheduling system according to claim 5, characterized in that: The original data includes age, travel mileage, charging period, SOC interval and historical response rate. K-Prototypes is used to cluster the original data to generate high, medium and low three types of response priorities. Different weights are given to five-dimensional targets f1-f5 in the MOPSO objective function to realize the same stake different price and the same car different value. 8.The multi-objective optimization based V2G vehicle cluster orderly charging and discharging scheduling system according to claim 1, characterized in that: The cloud scheduling platform (1) provides aggregator API, carbon exchange API, power grid demand response API, payment API and regulatory API.

Images