Distributed photovoltaic-energy storage-flexible load cluster collaborative control method and system

Abstract

The application discloses a kind of distributed photovoltaic-energy storage-flexible load cluster collaborative control method and system, input feature input time series convolution network, obtain the carbon potential of each key node in future;Calculate the total carbon cost of photovoltaic cluster in dispatching period T, maximum reducible power, energy storage aging cost, response probability, state space, action space;Minimize total cost using reward function;Iterative update each agent power allocation ratio under up-regulation working condition, each agent power allocation ratio under down-regulation working condition;Introduce carbon potential confidence factor to dynamically correct weight;The historical data in past preset time period and numerical weather forecast in future preset time period are input into long short-term memory network, to obtain the node carbon potential, photovoltaic output and flexible load in future preset time period. Every time interval uses objective function to carry out model predictive control optimization.The application designs electric-carbon coupled multi-objective optimization framework, embeds carbon constraint into cluster dispatching decision.

Description

Technical Field

[0001] This invention relates to the field of distributed photovoltaic-energy storage-flexible load cluster coordinated control technology, and in particular to a distributed photovoltaic-energy storage-flexible load cluster coordinated control method and system considering carbon flow tracking. Background Technology

[0002] With the accelerated construction of new power systems, distributed photovoltaic (PV) power, energy storage, and flexible loads are widely integrated into scenarios such as industrial parks, commercial and industrial distribution areas, and virtual power plants. The distribution network is transforming from a traditional passive network to an active network with a high proportion of distributed energy. Distributed power sources and loads are deployed at multiple points and connected in a decentralized manner, resulting in significant spatiotemporal fluctuations in source and load, bidirectional power flow, and flexible and varied operating modes. This places higher demands on the coordinated utilization and efficient management of energy. Green and low-carbon, safe and efficient, and economically optimal have become the core guiding principles for system operation. Coordinated management of electricity and carbon emissions, as well as integrated operation of power generation, grid, load, and storage, have become key technological directions for improving energy utilization efficiency.

[0003] For the clustered coordinated control of distributed photovoltaic, energy storage, and flexible loads, existing technologies mainly revolve around multi-agent reinforcement learning, distributed consensus protocols, and model predictive control, achieving multi-timescale coordination of power allocation, voltage regulation, and renewable energy consumption. In carbon flow analysis, node carbon potential tracking and proportional allocation methods have been applied to transmission network carbon emission accounting. Regarding electricity-carbon coordinated scheduling, existing research has explored joint decision-making on economic efficiency and low-carbon performance by introducing carbon prices as a penalty term in the optimization objective. In terms of control architecture, hierarchical distributed schemes balance overall performance and local robustness through upper-level global optimization and lower-level local rapid response. However, how to deeply integrate real-time carbon flow tracking with multi-timescale clustered control of source, storage, and load to achieve dynamic coordination of electricity-carbon coupling remains a direction for continuous exploration in current technological development. Summary of the Invention

[0004] Objectives of the Invention: The first objective of this invention is to provide a distributed photovoltaic-energy storage-flexible load cluster collaborative control method that achieves joint optimization of low carbon emissions and economic efficiency, suppresses deep charging and discharging to extend lifespan, significantly improves the lifespan friendliness of resource allocation and actual load response rate, balances global carbon-economic optimization and local robustness, and effectively addresses communication anomalies while ensuring convergence speed. The second objective of this invention is to provide an electronic device for executing the above method. The third objective of this invention is to provide a computer-readable storage medium for storing the above method. The fourth objective of this invention is to provide a distributed photovoltaic-energy storage-flexible load cluster collaborative control system that achieves joint optimization of low carbon emissions and economic efficiency, suppresses deep charging and discharging to extend lifespan, significantly improves the lifespan friendliness of resource allocation and actual load response rate, balances global carbon-economic optimization and local robustness, and effectively addresses communication anomalies while ensuring convergence speed.

[0005] Technical solution: The distributed photovoltaic-energy storage-flexible load cluster coordinated control method of the present invention includes:

[0006] (1) Compute nodes exist Carbon potential over time : Where g is the power supply serial number. For power inflow node The set of branch paths, For nodes Flow to Node power, Carbon potential at upstream nodes; For nodes A collection of distributed power sources connected at the location; Provide power to the power source; This represents the carbon emission factor corresponding to the power source.

[0007] (2) Input the input features into a temporal convolutional network to obtain the carbon potential of each key node in a future preset time period.

[0008] (3) Calculate the total carbon cost of the photovoltaic cluster during the scheduling period T. .

[0009] ,in, As a carbon price penalty factor, Real-time carbon potential for public connection points in the park / area. In order to purchase power from the main grid, Let t be the duration of the time period, t be the t-th time period within the scheduling cycle, and T be the total number of time periods within the scheduling cycle.

[0010] (4) Calculate the photovoltaic cluster exist Maximum power reduction during the period In the formula, Contribute to the theoretical prediction of photovoltaics.

[0011] (5) Calculate the aging cost of energy storage :

[0012] ,

[0013] In the formula, The baseline aging factor is denoted as ε; e is the natural constant, e = 2.718. As an aging acceleration factor, To accumulate the number of loops, For rated cycle life, For charging and discharging power, This is the rated capacity.

[0014] (6) Calculate the probability of response :

[0015] .

[0016] In the formula, Adjustable power for temperature-controlled flexible loads, room temperature deviation. , Set the temperature for the user. Indoor temperature; The minimum acceptable temperature deviation for the user. The maximum temperature deviation acceptable to the user. The width of the fuzzy interval.

[0017] Calculate the upper limit of adjustable power for temperature-controlled flexible loads. Adjustable power lower limit for temperature-controlled flexible loads : , ,in, The rated adjustable capacity is for temperature-controlled flexible loads. The rated adjustable capacity is for temperature-controlled flexible loads.

[0018] Preferably, after step (6), the following steps are also included:

[0019] (7) Calculate the first An intelligent agent at time state space :

[0020] .

[0021] Among them, net load , For the first An agent in time period The power of the electrical load, For the first An agent in time period The actual output power of photovoltaic power In the state of energy storage charge, The overall responsiveness probability of aggregated flexible loads. The carbon potential of the adjacent node.

[0022] Calculate the action space ,in, This is the adjustment amount for the total power of the photovoltaic cluster. The proportion borne by energy storage; minimizing the total cost using a reward function. :

[0023] ,in, for The weighting coefficients, for The weighting coefficients, for The weighting coefficients, Penalties for breach of contract regarding user comfort Cost of purchasing and selling electricity.

[0024] Preferably, after step (7), the following steps are also included:

[0025] (8) Adjust the charging margin during charging. Assign a value to the neighbor agent Increase the local adjustable margin Increase the power consumption margin when using electricity. Assign a value to the neighbor agent Increase the local adjustable margin During discharge, the discharge margin will be reduced. Assign a value to the neighbor agent Reduce local adjustable margin Reduce the power consumption margin when using electricity. Assign a value to the neighbor agent Reduce local adjustable margin Among them, the charging margin is increased. Discharge down-adjustment margin Electricity consumption margin Electricity consumption reduction margin .

[0026] Iterative updates adjust the power allocation ratio of each agent under different operating conditions. 2. Reduce the power distribution ratio of each agent under operating conditions : .

[0027] Adjust the iteration formula: .

[0028] In the formula, k is the time index. To converge the step size, for The set of neighbors; Adjacency weight; as an agent Target reference regulation power, Act as a neighbor Target reference regulation power, Act as a neighbor The local adjustable margin has been increased. Act as a neighbor The local adjustable margin is reduced.

[0029] Introducing carbon potential confidence factor Dynamically adjusted weights :

[0030] ,in, These are the initial weights based on the communication topology; This represents the carbon potential of the node where p is located. Act as a proxy for the carbon potential of the node containing node q, which is a neighbor. This represents the region's maximum carbon potential.

[0031] Preferably, after step (8), the following steps are also included:

[0032] (9) Input historical data within the past preset time period and numerical weather forecasts within the future preset time period into the long short-term memory network to obtain the node carbon potential, photovoltaic output and flexible load within the future preset time period; wherein, the historical data includes the actual power of the flexible load and the actual output of the photovoltaic.

[0033] Preferably, after step (9), the method further includes: periodically optimizing the model predictive control using an objective function; wherein the objective function is:

[0034] .

[0035] in, The first penalty coefficient, The second penalty coefficient is N, where N is the prediction time domain length. This is the energy storage capacity planned for the current period. This is the carbon potential reference value as planned recently; The starting time; For time indexing; For the first The cost of purchasing and selling electricity during a given time period; This represents the weighting coefficient of carbon costs in the overall objective. For the first Carbon costs over a given period; This represents the weighting coefficient of energy storage aging costs in the overall objective. For the first Energy storage aging costs over time; For the first The charging and discharging power of energy storage during a given time period; For the first Real-time carbon potential of the common connection point during the time period.

[0036] The distributed photovoltaic-energy storage-flexible load cluster collaborative control system of the present invention includes:

[0037] Carbon flow tracing module, used for computing nodes exist Carbon potential over time : Where g is the power supply serial number. For power inflow node The set of branch paths, For nodes Flow to Node power, Carbon potential at upstream nodes; For nodes A collection of distributed power sources connected at the location; Provide power to the power source; The carbon emission factor corresponding to the power source is used; the input features are input into a temporal convolutional network to obtain the carbon potential of each key node within a preset future time period; the total carbon cost of the cluster within the scheduling period T is calculated. ,in, As a carbon price penalty factor, Real-time carbon potential for public connection points in the park / area. In order to purchase power from the main grid, Let t be the duration of the time period, t be the t-th time period within the scheduling cycle, and T be the total number of time periods within the scheduling cycle.

[0038] The multi-objective optimization module is used to input the input features into a temporal convolutional network to obtain the carbon potential of each key node within a preset future time period; and to calculate the total carbon cost of the photovoltaic cluster within the scheduling period T. :

[0039] .

[0040] in, As a carbon price penalty factor, Real-time carbon potential for public connection points in the park / area. In order to purchase power from the main grid, T represents the duration of the time period, where T is the total duration of the future preset time period.

[0041] Calculate photovoltaic clusters exist Maximum power reduction during the period In the formula, Contribute to photovoltaic theoretical predictions; calculate energy storage aging costs. .

[0042] In the formula, The baseline aging factor is denoted as e; e is a mathematical constant. As an aging acceleration factor, To accumulate the number of loops, For rated cycle life, For charging and discharging power, Given the rated capacity; calculate the probability of response. .

[0043] .

[0044] In the formula, Adjustable power for temperature-controlled flexible loads, room temperature deviation. , Set the temperature for the user. Indoor temperature; The minimum acceptable temperature deviation for the user. The maximum temperature deviation acceptable to the user. The width of the fuzzy interval.

[0045] Calculate the upper limit of adjustable power for temperature-controlled flexible loads. Adjustable power lower limit for temperature-controlled flexible loads : , ,in, The rated adjustable capacity is for temperature-controlled flexible loads. The rated adjustable capacity is for temperature-controlled flexible loads.

[0046] This invention innovatively establishes a carbon flow tracking model based on real-time calculation of node carbon potential and prediction using a temporal convolutional network. It embeds carbon cost as a dynamic penalty factor into the multi-objective optimization function of photovoltaic cluster scheduling. By quantifying the carbon source, transmission path, and consumption attribution of each kilowatt-hour, it changes the traditional scheduling model that focuses solely on economic efficiency, achieving joint optimization of low-carbon performance and economic efficiency, and effectively avoiding the detrimental scheduling of "low electricity price and high carbon emissions."

[0047] This invention achieves real-time quantification of carbon costs by constructing a dynamic calculation and prediction model for node carbon potential; embeds carbon constraints into cluster scheduling decisions by designing a multi-objective optimization framework that couples electricity and carbon; and establishes a hierarchical distributed collaborative control architecture to ensure global carbon-economic optimization while achieving rapid response and carbon-aware allocation of local resources, thereby solving the problems of lack of carbon flow tracking, insufficient electricity-carbon coordination, and poor cluster control adaptability in existing technologies.

[0048] An electronic device according to the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods described herein.

[0049] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described herein.

[0050] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: 1. This invention achieves low-carbon economic operation through electricity-carbon synergy. By dynamically embedding carbon costs into multi-objective optimization through real-time carbon flow tracking and node carbon potential prediction, it changes the traditional single economic scheduling mode, effectively reducing operating costs and carbon emissions, and avoiding the problem of high carbon emissions with low electricity prices. 2. It improves energy storage lifespan and flexible load response rate. By adopting an energy storage aging cost index model and a flexible load fuzzy membership model, it suppresses deep charging and discharging of energy storage to extend its lifespan. At the same time, it flexibly adjusts the load according to user comfort, significantly improving the actual load response rate and enhancing user experience. 3. It enhances robustness and multi-timescale control capabilities. The combination of upper-layer global optimization and lower-layer carbon-sensing consensus protocol maintains power allocation accuracy even in the event of communication anomalies; combined with model predictive control rolling correction, it reduces the risk of tie line overruns, achieving full-coverage precise control from day-ahead planning to second-level response. Attached Figure Description

[0051] Figure 1 This is a system architecture diagram of the present invention;

[0052] Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0053] This invention proposes a distributed photovoltaic-energy storage-flexible load cluster collaborative control method, which achieves multi-objective collaborative optimization of low carbon, economy and safety through carbon potential sensing, dynamic potential assessment, hierarchical distributed collaborative control and model prediction rolling correction.

[0054] Figure 1 The system architecture diagram of this invention is shown below. This invention comprises a three-layer structure: a cloud-based collaborative management and control platform, an edge-based collaborative control terminal, and a sensing and execution unit. Data interaction and collaborative control between these layers are achieved through standardized communication links. The cloud-based collaborative management and control platform adopts a cloud computing architecture and deploys a carbon flow tracking module, a multi-objective optimization module, and a data processing and storage module. The carbon flow tracking module is responsible for calculating the carbon potential of distribution network nodes, tracing carbon flow paths, and calculating carbon emissions. The multi-objective optimization module uses system operating costs, carbon emissions, renewable energy absorption rate, and network losses as objectives to conduct global optimization scheduling decisions. The data processing and storage module is responsible for storing, cleaning, and analyzing massive amounts of operational data, providing data support for scheduling strategies. The cloud-based collaborative management and control platform and the edge-based collaborative control terminal communicate via 4G / Ethernet, with a transmission latency of less than 60ms, enabling reliable issuance of global optimization commands and uploading of terminal operational data.

[0055] The edge collaborative control terminal, as a core device on the edge side, integrates a data acquisition module, an edge computing module, and an instruction execution and feedback module. The data acquisition module collects operational data from distributed photovoltaic units, energy storage system units, flexible load clusters, and grid interface units. The edge computing module performs data preprocessing, local rapid calculation, and short-term forecasting. The instruction execution and feedback module receives cloud-based optimization instructions, distributes them to each sensing and execution unit, and transmits device status and execution results back in real time. The sensing and execution units include distributed photovoltaic units, energy storage system units, flexible load clusters, and grid interface units. Distributed photovoltaic units consist of photovoltaic arrays, string inverters, and environmental sensors, converting solar energy into electrical energy and uploading output data. Energy storage system units consist of lithium battery packs, energy storage converters (PCS), and battery management systems (BMS), regulating charging and discharging according to instructions. Flexible load clusters include adjustable loads such as smart air conditioners, electric water heaters, and charging piles, adjusting power demand according to control instructions. The grid interface unit includes smart meters, transformers, and a communication unit, enabling power and data interaction between the system and the external power grid.

[0056] The core of the method in this invention is a multi-timescale collaborative optimization model of electricity-carbon coupling. Its goal is to minimize the operating cost and total carbon emissions of the cluster while satisfying system power balance, equipment operating constraints, and user comfort. The method includes: real-time carbon flow tracking; dynamic potential assessment of source-storage-load; hierarchical cluster collaborative optimization of electricity-carbon coupling; and closed-loop rolling correction based on carbon potential prediction. The specific steps are as follows: Figure 2 As shown.

[0057] (a) Carbon flow tracking module, used for real-time carbon flow tracking and nodal carbon potential calculation.

[0058] This step aims to provide spatiotemporally resolved carbon signals for cluster collaborative control, realizing the transformation from "ubiquitous carbon emission statistics" to "real-time carbon flow sensing".

[0059] (1) Carbon flow tracking module, used for real-time calculation of carbon potential at distribution network nodes.

[0060] Based on the injected power and power flow distribution at each node obtained from the distribution network state estimation, an improved proportional allocation method considering the zero-carbon characteristics of distributed photovoltaic power is used to calculate the real-time carbon potential node by node. Nodes are defined. exist Carbon potential over time : , Where g is the power supply serial number. For power inflow node The set of branch paths, For nodes Flow to Node power, Carbon potential at upstream nodes; For nodes A collection of distributed power sources connected to the site, including photovoltaics and energy storage (which are considered power sources when discharging); Provide power to the power source; This represents the carbon emission factor corresponding to the power source (0 for photovoltaics, and the carbon potential stored during charging is used during energy storage discharge). The calculation period is 1 minute, covering all distributed photovoltaic access points and load nodes. Carbon potential The unit is / kWh.

[0061] (2) Carbon flow tracking module, used for future carbon potential prediction.

[0062] A temporal convolutional network (TCN) is used to predict the carbon potential of key nodes (grid connection point, photovoltaic cluster outlet, and energy storage access point) in the next 4 hours. Input features include: historical carbon potential sequence of the past 24 hours (including 15-minute resolution), numerical weather forecast (including irradiance, temperature, humidity, and 1-hour resolution), and ultra-short-term photovoltaic power output forecast (including 15-minute resolution). The TCN network consists of 4 layers of dilated convolutional blocks, and the output layer is a fully connected layer, which is updated online every 30 minutes.

[0063] (3) Carbon flow tracking module, used for carbon cost modeling of photovoltaic clusters.

[0064] The node obtained in step (1) exist Carbon potential over time Step (2) embeds the carbon potential of each key node for the next 4 hours into the cost function of photovoltaic cluster scheduling, defining the total carbon cost of the photovoltaic cluster within the scheduling period T: ,in, This is the carbon price penalty factor, in yuan / , Real-time carbon potential for the Public Connection Point (PCC) of the Industrial Park / Transformer Area. To purchase power (kW) from the main grid. Duration of the time period (in hours). Based on nodes. exist Carbon potential over time This will allow us to obtain the real-time carbon potential of the common connection point (PCC) in the photovoltaic cluster / area. This carbon cost will serve as one of the core objectives for subsequent collaborative optimization of the photovoltaic cluster.

[0065] (ii) Multi-objective optimization module, used to assess the dynamic adjustability potential of distributed photovoltaic-energy storage-flexible load.

[0066] We conduct refined modeling of three key resources within photovoltaic clusters: distributed photovoltaic (output adjustability is limited by curtailment), energy storage (charge and discharge power and lifespan constraints), and flexible loads (user comfort constraints), quantifying their actual adjustable range and adjustment costs under different operating conditions.

[0067] (1) Real-time calculation of carbon potential at distribution network nodes.

[0068] Photovoltaic power output is inherently highly volatile, and its adjustability primarily lies in "curtailable power" (i.e., curtailed power). Defining a photovoltaic cluster. exist The maximum power that can be reduced during the time period is (Unless participating in demand response, light will not be discarded by default; it will only be enabled when disposal is difficult.) This invention contributes to the theoretical prediction of photovoltaic power. It treats photovoltaic power as a priority zero-carbon power source for energy consumption, giving it the highest scheduling priority.

[0069] (2) Energy storage dynamic adjustable potential and aging cost model.

[0070] The dynamic adjustability potential of energy storage is constrained by the state of charge (SOC), charge / discharge power limits, and cycle life aging costs. In addition to conventional SOC boundaries... In addition to power limits, an aging cost factor is introduced to quantify the aging loss of each charge and discharge cycle into the energy storage aging cost. : In the formula, The baseline aging factor is (yuan / kWh); e is a mathematical constant, approximately equal to 2.71828. This is an aging acceleration factor, with a value ranging from 0.5 to 1.0. To accumulate the number of loops, For rated cycle life, This represents the charging and discharging power (positive for charging). This is the rated capacity. This cost will directly participate in the collaborative optimization of the photovoltaic cluster, avoiding excessive use of energy storage.

[0071] (3) Flexible load fuzzy membership degree adjustable potential model.

[0072] For temperature-controlled flexible loads such as variable frequency air conditioners and electric water heaters, a responsive probability interval based on a fuzzy membership function of user comfort is established. Room temperature deviation is defined. ,in Set the temperature for the user. Calculate the probability of responsiveness. .

[0073] .

[0074] In the formula, Adjustable power for temperature-controlled flexible loads. The minimum acceptable temperature deviation for users is 0.5℃ by default; The maximum acceptable temperature deviation for the user is 2.0℃ by default. The default temperature range width is 0.5℃. Users can dynamically adjust this range via the cloud by selecting either "Green & Efficient" or "Comfort First" mode. .

[0075] The adjustable power limit of this temperature-controlled flexible load Adjustable power lower limit for temperature-controlled flexible loads They are respectively: , ,in For temperature-controlled flexible loads, the rated adjustable capacity is... The rated adjustable capacity for temperature-controlled flexible loads is specified, such as an air conditioner which can be increased by 0-1.5kW or decreased by 0-1.5kW. The final output shows the adjustable power range for each temperature-controlled flexible load over a 5-30 minute timescale, along with the corresponding user comfort penalty. ,when > Penalties for breach of user comfort It is positive.

[0076] (iii) Multi-objective optimization module for hierarchical photovoltaic cluster collaborative optimization of electric-carbon coupling.

[0077] A hierarchical architecture of "upper-layer global electricity-carbon coupling optimization + lower-layer carbon sensing and consistent allocation" is adopted to achieve coordinated control of distributed photovoltaic-energy storage-flexible load clusters.

[0078] (1) Global optimization at the upper level: deep reinforcement learning with multi-agent electro-carbon coupling.

[0079] Each distributed photovoltaic-energy storage-flexible load aggregate (such as a transformer substation or a virtual power plant unit) is considered as an intelligent agent. The definition of the first... An intelligent agent at time state space : ,in Net load, For the first An intelligent agent at time The electrical load power, For the first An intelligent agent at time The actual output power of photovoltaic power In the state of energy storage charge, The overall responsiveness probability of aggregated flexible loads. The carbon potential of the adjacent node.

[0080] Action space ,in, This represents the total power adjustment of the photovoltaic cluster, expressed in kW. Positive values ​​indicate load reduction / charging, while negative values ​​indicate load increase / discharging. The remaining power is shared by flexible loads, with energy storage accounting for a certain proportion.

[0081] Minimize total cost using reward function To maximize overall benefits, costs that need to be considered include electricity purchase and sales costs. Total carbon cost Energy storage aging costs Penalties for violations related to user comfort .

[0082] ,in They are respectively The weighting coefficients are determined offline via Bayesian optimization. A Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm is employed, with centralized training and distributed execution. Offline training is conducted using one year's worth of historical photovoltaic, load, electricity price, and carbon potential data, with online fine-tuning every 24 hours.

[0083] (2) Lower-layer rapid response: Carbon perception consensus protocol.

[0084] The upper layer outputs the total power adjustment reference value for each agent every 5 minutes. The lower-level local rapid response unit needs to allocate the total adjustment amount between the energy storage terminals and flexible load terminals within the photovoltaic cluster. Assume there are M adjustable resource agents within the photovoltaic cluster (each energy storage PCS or load aggregator is a resource agent), and define the charging up-adjustment margin, discharging down-adjustment margin, power consumption up-adjustment margin, and power consumption down-adjustment margin respectively.

[0085] Energy storage agent: Charging margin increased Discharge down-adjustment margin Both types of margins are constrained by the energy storage SOC operating limits. Load proxy: Electricity consumption upscaling margin. Electricity consumption reduction margin .

[0086] Increase charging margin during charging Assign a value to the neighbor agent Increase the local adjustable margin Increase the power consumption margin when using electricity. Assign a value to the neighbor agent Increase the local adjustable margin During discharge, the discharge margin will be reduced. Assign a value to the neighbor agent Reduce local adjustable margin Reduce the power consumption margin when using electricity. Assign a value to the neighbor agent Reduce local adjustable margin In both the upward and downward adjustment scenarios, the corresponding consensus protocols are used to iteratively update the power allocation ratio of each agent under the converged upward adjustment condition. 2. Reduce the power distribution ratio of each agent under operating conditions .

[0087] Adjust the iteration formula:

[0088] .

[0089] In the formula, To achieve convergence, a value of 0.5 is used; for The set of neighbors; Adjacency weight; as an agent Target reference regulation power, Act as a neighbor Target reference regulation power, , Neighbor agents Increase the local adjustable margin and decrease the local adjustable margin.

[0090] This invention introduces a carbon potential confidence factor. Dynamically adjusted weights : ,in, These are the initial weights based on the communication topology; The carbon potential confidence factor ranges from 0.2 to 0.5. This represents the carbon potential of the node where p is located. Act as a proxy for the carbon potential of the node containing node q, which is a neighbor. This design maximizes the carbon potential in the region, granting nodes with low carbon potential a higher power scheduling weight, prioritizing the absorption of clean electricity or the reduction of carbon emissions. The protocol converges in no more than 10 iterations. If proxy communication is interrupted, it automatically switches to a compensatory consensus based on an autoregressive moving average model.

[0091] (iv) Multi-objective optimization module, used for closed-loop rolling correction based on carbon potential prediction.

[0092] To address the random fluctuations in photovoltaic output and load, as well as carbon potential prediction errors, this invention introduces model predictive control (MPC) to achieve closed-loop rolling correction.

[0093] (1) Carbon potential and power ultra-short-term prediction.

[0094] A long short-term memory network is used to predict nodal carbon potential, photovoltaic output, and load for the next two hours. The input window contains historical data from the past two hours and numerical weather forecasts for the next hour. The system outputs the prediction error correction for the next eight time periods every 15 minutes. The historical data includes the actual power of flexible loads and the actual output of photovoltaic power.

[0095] (2) Electric-carbon synergistic MPC rolling optimization.

[0096] MPC optimization is triggered every 15 minutes. (Based on the current moment...) Starting with actual measured values ​​(including photovoltaic output, load, energy storage SOC, and nodal carbon potential), we construct the future. Time period ( The optimization problem is given by the equation (4~8, 15 minutes per time period), with the objective function being: ,in , The first and second penalty coefficients are used to suppress excessive deviations from the current plan; and These are the planned energy storage capacity and carbon potential reference values ​​for the current day; This is the starting point for scrolling optimization; For time indexing; For the first Electricity purchase and sale costs for a given period (in yuan); This represents the weighting coefficient of carbon costs in the overall objective. For the first Carbon cost per period (in yuan); This represents the weighting coefficient of energy storage aging costs in the overall objective. For the first Energy storage aging cost over time (RMB); For the first The charging and discharging power of energy storage during a given time period; For the first Real-time carbon potential at the point of common coupling during the time period. Constraints include power balance, energy storage SOC and power limits, grid interaction power limits, and adjustable upper and lower limits for flexible loads.

[0097] (3) Closed-loop execution.

[0098] MPC executes calculation instructions only for the first period: energy storage charging and discharging power setpoints, flexible load power plans, and photovoltaic cluster interconnection power reference values. The next control cycle (15 minutes later) re-optimizes using new measured data, forming a closed loop, thereby continuously correcting the impact of carbon flow tracking and prediction errors on photovoltaic cluster control.

Claims

1. A distributed photovoltaic-energy storage-flexible load cluster coordinated control method, characterized in that, include: (1) Compute nodes exist Carbon potential over time : ， Where g is the power supply serial number. For power inflow node The set of branch paths, For nodes Flow to Node power, Carbon potential at upstream nodes; For nodes A collection of distributed power sources connected at the location; Provide power to the power source; The carbon emission factor corresponding to the power source; (2) Input the input features into a temporal convolutional network to obtain the carbon potential of each key node in a future preset time period; (3) Calculate the total carbon cost of the photovoltaic cluster during the scheduling period T. : ， in, As a carbon price penalty factor, Real-time carbon potential for public connection points in the park / area. In order to purchase power from the main grid, Let t be the duration of the time period, t be the t-th time period within the scheduling cycle, and T be the total number of time periods within the scheduling cycle. (4) Calculate the photovoltaic cluster exist Maximum power reduction during the period In the formula, Contribute to theoretical predictions of photovoltaics; (5) Calculate the aging cost of energy storage : ， In the formula, The baseline aging factor is denoted as e; e is the natural constant. As an aging acceleration factor, To accumulate the number of loops, For rated cycle life, For charging and discharging power, Rated capacity; (6) Calculate the probability of response : ， In the formula, Adjustable power for temperature-controlled flexible loads, room temperature deviation. , Set the temperature for the user. Indoor temperature; The minimum acceptable temperature deviation for the user. The maximum temperature deviation acceptable to the user. The width of the fuzzy interval; Calculate the upper limit of adjustable power for temperature-controlled flexible loads. Adjustable power lower limit for temperature-controlled flexible loads : , , in, The rated adjustable capacity is for temperature-controlled flexible loads. The rated adjustable capacity is for temperature-controlled flexible loads.

2. The distributed photovoltaic-energy storage-flexible load cluster coordinated control method according to claim 1, characterized in that, Following step (6), the following is also included: (7) Calculate the first An intelligent agent at time state space : ， Among them, net load , For the first An agent in time period The power of the electrical load, For the first An agent in time period The actual output power of photovoltaic power In the state of energy storage charge, The overall responsiveness probability of aggregated flexible loads. The carbon potential of adjacent nodes; Calculate the action space ,in, This is the adjustment amount for the total power of the photovoltaic cluster. The proportion to be borne by energy storage; Minimize total cost using reward function : ， in, for The weighting coefficients, for The weighting coefficients, for The weighting coefficients, Penalties for breach of contract regarding user comfort Cost of purchasing and selling electricity.

3. The distributed photovoltaic-energy storage-flexible load cluster coordinated control method according to claim 2, characterized in that, Following step (7), the following is also included: (8) Adjust the charging margin during charging. Assign a value to the neighbor agent Increase the local adjustable margin Increase the power consumption margin when using electricity. Assign a value to the neighbor agent Increase the local adjustable margin During discharge, the discharge margin will be reduced. Assign a value to the neighbor agent Reduce local adjustable margin Reduce the power consumption margin when using electricity. Assign a value to the neighbor agent Reduce local adjustable margin Among them, the charging margin is increased. Discharge down-adjustment margin Electricity consumption margin Electricity consumption reduction margin ; Iterative updates adjust the power allocation ratio of each agent under different operating conditions. Adjusting the power distribution ratio of each agent under different operating conditions : ， Adjust the iteration formula: , In the formula, k is the time index. To converge the step size, for The set of neighbors; Adjacency weight; as an agent Target reference regulation power, Act as a neighbor Target reference regulation power, Act as a neighbor The local adjustable margin has been increased. Act as a neighbor The local adjustable margin is reduced; Introducing carbon potential confidence factor Dynamically adjusted weights : ， in, These are the initial weights based on the communication topology; This represents the carbon potential of the node where p is located. Act as a proxy for the carbon potential of the node containing node q, which is a neighbor. This represents the region's maximum carbon potential.

4. The distributed photovoltaic-energy storage-flexible load cluster coordinated control method according to claim 3, characterized in that, Following step (8), the following is also included: (9) Input historical data within the past preset time period and numerical weather forecasts within the future preset time period into the long short-term memory network to obtain the node carbon potential, photovoltaic output and flexible load within the future preset time period; wherein, the historical data includes the actual power of the flexible load and the actual output of the photovoltaic.

5. The distributed photovoltaic-energy storage-flexible load cluster coordinated control method according to claim 4, characterized in that, Following step (9), the following is also included: At regular intervals, the objective function is used to optimize the model predictive control. The objective function is: ， in, The first penalty coefficient, The second penalty coefficient is N, where N is the prediction time domain length. This is the energy storage capacity planned for the current period. This is the carbon potential reference value as planned recently; The starting time; For time indexing; For the first The cost of purchasing and selling electricity during a given time period; This represents the weighting coefficient of carbon costs in the overall objective. For the first Carbon costs over a given period; This represents the weighting coefficient of energy storage aging costs in the overall objective. For the first Energy storage aging costs over time; For the first The charging and discharging power of energy storage during a given time period; For the first Real-time carbon potential of the common connection point during the time period.

6. A distributed photovoltaic-energy storage-flexible load cluster collaborative control system, characterized in that, include: Carbon flow tracing module, used for computing nodes exist Carbon potential over time : Where g is the power supply serial number. For power inflow node The set of branch paths, For nodes Flow to Node power, Carbon potential at upstream nodes; For nodes A collection of distributed power sources connected at the location; Provide power to the power source; The carbon emission factor corresponding to the power source is used; the input features are input into a temporal convolutional network to obtain the carbon potential of each key node within a preset future time period; the total carbon cost of the cluster within the scheduling period T is calculated. ,in, As a carbon price penalty factor, Real-time carbon potential for public connection points in the park / area. In order to purchase power from the main grid, Let t be the duration of the time period, t be the t-th time period within the scheduling cycle, and T be the total number of time periods within the scheduling cycle. The multi-objective optimization module is used to input the input features into a temporal convolutional network to obtain the carbon potential of each key node within a preset future time period; and to calculate the total carbon cost of the photovoltaic cluster within the scheduling period T. : ， in, As a carbon price penalty factor, Real-time carbon potential for public connection points in the park / area. In order to purchase power from the main grid, Where T is the duration of the time period, and T is the total duration of the future preset time period; Calculate photovoltaic clusters exist Maximum power reduction during the period In the formula, Contribute to photovoltaic theoretical predictions; calculate energy storage aging costs. : ， In the formula, The baseline aging factor is denoted as e; e is a mathematical constant. As an aging acceleration factor, To accumulate the number of loops, For rated cycle life, For charging and discharging power, Rated capacity; Calculate the probability of response : ， In the formula, Adjustable power for temperature-controlled flexible loads, room temperature deviation. , Set the temperature for the user. Indoor temperature; The minimum acceptable temperature deviation for the user. The maximum temperature deviation acceptable to the user. The width of the fuzzy interval; Calculate the upper limit of adjustable power for temperature-controlled flexible loads. Adjustable power lower limit for temperature-controlled flexible loads : , , in, The rated adjustable capacity is for temperature-controlled flexible loads. The rated adjustable capacity is for temperature-controlled flexible loads.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 5.

Images