Virtual power plant intelligent aggregation optimization control method for multi-type flexible resources

Abstract

The invention discloses a virtual power plant intelligent aggregation optimization control method for multi-type flexible resources, and the method comprises the steps: constructing a dynamic characteristic model of distributed resources, wherein the dynamic characteristic model comprises a photovoltaic output probability prediction model, an energy storage SOC-life coupling model, an electric vehicle behavior chain model, an adjustable load constraint model, and an industrial interruptible load model; an edge agent node calculates an adjustable potential interval of a resource cluster in real time and uploads the adjustable potential interval to a cloud end, a global optimization target is solved on the cloud end based on an improved sparrow search algorithm (ISSA), after a scheduling instruction is generated, model parameters are corrected in a rolling mode according to actual output deviation calculated in real time, and a scheduling result is obtained. And triggering a resource fault emergency strategy for prediction deviation and resource fault problems occurring in the operation process of the virtual power plant. Through an edge-cloud collaborative architecture and a multi-stage optimization strategy, accurate modeling, optimization aggregation and intelligent scheduling of distributed resources are realized.

Description

Technical Field

[0001] The present invention relates to a technology in the field of virtual power plant control, specifically to an intelligent aggregation optimization control method for virtual power plants oriented to multiple types of flexible resources. Background Art

[0002] Virtual power plants (VPPs) integrate distributed energy resources (such as photovoltaics, wind power, energy storage, and controllable loads) to form a flexible virtual equivalent power generation entity, providing peak shaving, frequency regulation, and demand response services to the power system. However, existing technologies, such as Chinese patent document CN113708365A (A Method and System for Optimizing Energy Management and Control of Virtual Power Plants Based on an Edge-Cloud Architecture), often use simplified models to describe the characteristics of distributed resources. These methods fail to fully consider physical constraints (such as energy storage life loss and thermal inertia of temperature-controlled loads) and uncertainties (such as renewable energy forecast errors), resulting in a discrepancy between dispatch instructions and actual execution capabilities. When some resources fail or forecast deviations exceed thresholds, the lack of effective emergency response mechanisms and dynamic correction strategies can easily lead to system power imbalances and even affect grid stability. Summary of the Invention

[0003] In response to the above-mentioned shortcomings of the existing technology, the present invention proposes an intelligent aggregation optimization control method for virtual power plants for multi-type flexible resources. Through an edge-cloud collaborative architecture and a multi-stage optimization strategy, it realizes accurate modeling, optimized aggregation and intelligent scheduling of distributed resources.

[0004] The present invention is achieved through the following technical solutions:

[0005] The present invention relates to an intelligent aggregation optimization control method for a virtual power plant for multi-type flexible resources. After constructing a dynamic characteristic model of distributed resources including a photovoltaic output probability prediction model, an energy storage SOC-life coupling model, an electric vehicle behavior chain model, an adjustable load constraint model, and an industrial interruptible load model, an edge proxy node calculates the adjustable potential interval of a resource cluster in real time and uploads it to the cloud. In the cloud, an improved sparrow search algorithm (ISSA) is used to solve the global optimization target. After generating a dispatch instruction, the model parameters are rollingly corrected according to the actual output deviation calculated in real time. A resource failure emergency strategy is triggered for prediction deviations and resource failure problems that occur during the operation of the virtual power plant.

[0006] Technical Effects

[0007] The present invention constructs a distributed resource dynamic characteristic model that includes a photovoltaic output probability prediction model, an energy storage SOC-life coupling model, an electric vehicle behavior chain model, an adjustable load constraint model, and an industrial interruptible load model. Combining uncertainty analysis with equipment physical characteristics, the present invention realizes the calculation of the adjustable power range of each resource at a time scale of hours and a resolution of 15 minutes. Among them, the energy storage SOC-life coupling model uses an improved Arrhenius model to reflect the accelerated loss of the battery due to deep discharge, and the electric vehicle behavior chain model uses a neural network including a graph neural network (GNN) layer, a graph pooling layer, and a causal convolutional neural network (CCNN) layer to predict vehicle travel parameters. The edge agent node divides resources into clusters based on geographic location and communication topology, and uses SIMD instructions to accelerate the calculation of the overall potential boundary of the cluster, with a processing delay of <50ms. It also realizes efficient aggregation and upload of data through a TLV coding structure and a multi-level communication optimization mechanism (such as heartbeat packets, change rate transmission, and dynamic adjustment of the sending interval). The cloud uses an improved sparrow search algorithm (ISSA) based on tent chaotic mapping and dynamic weight adjustment to solve nonlinear optimization models. This algorithm generates the initial population through chaotic mapping, dynamically adjusts the proportion of discoverers, updates the positions of followers by classification, and performs projection corrections for out-of-bounds solutions. To address prediction deviations and resource failures during operation, actual output deviations are calculated in real time on a 5-minute basis. The extended Kalman filter (EKF) is used to update model parameters online, and a hierarchical emergency response mechanism is activated, including the rapid deployment of backup resources and dynamic command reallocation.

[0008] Compared with the existing technology, the present invention achieves accurate modeling of distributed resources. The photovoltaic output probability prediction model generates an output range with a 95.4% confidence level through the 3σ principle, and can be updated daily to adapt to seasonal changes and equipment attenuation characteristics. The energy storage model takes into account both the state of charge dynamics and cycle life loss to ensure that the SOC does not exceed the limit during scheduling. The electric vehicle model calculates real-time adjustable potential based on the user's historical travel data without affecting the user's travel needs. The adjustable load and industrial interruptible load models are based on the heat balance equation and the user's declared time period, respectively, to balance the comfortable temperature and production continuity. The cluster grouping and computational optimization of the edge proxy nodes achieves real-time aggregation and efficient transmission of the resource adjustable potential range. The communication optimization mechanism reduces the amount of data transmission, ensuring that information is uploaded to the cloud in a timely and accurate manner. The improved sparrow search algorithm improves the efficiency and accuracy of solving the global optimization goal, and can generate the optimal scheduling plan under complex constraints to ensure the interactive balance of power between the virtual power plant and the main grid, while taking into account the minimization of operating costs and the accurate tracking of grid instructions. The rolling correction and graded emergency mechanism enable the system to respond to prediction deviations and resource failures in real time, improve the accuracy of subsequent scheduling predictions through online updates of model parameters, and quickly restore the system power balance through standby resource call and instruction reallocation strategies, ensuring the stable operation of the virtual power plant under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 Flowchart of the present invention;

[0010] Figure 2 Cloud-edge-end architecture for virtual power plants;

[0011] Figure 3 Graph of the neural network prediction method. DETAILED DESCRIPTION

[0012] like Figure 1 As shown, this embodiment relates to a virtual power plant intelligent aggregation optimization control method for multiple types of flexibility resources, including:

[0013] Step 1) Construct a dynamic characteristic model of distributed resources that includes a photovoltaic output probability prediction model, an energy storage SOC-life coupling model, an electric vehicle behavior chain model, an adjustable load constraint model, and an industrial interruptible load model. Combined with uncertainty analysis and equipment physical characteristics, the adjustable power range of each resource is realized on a time scale of 15 minutes with a unit of hour and a resolution of 15 minutes. Calculations, including:

[0014] 1.1 Establish a photovoltaic output probability prediction model and determine its uncertainty range: In view of the intermittent characteristics of photovoltaic and wind power, firstly, based on the irradiance and wind speed The PV output probability prediction model including the photovoltaic conversion model and the wind power prediction model is constructed based on the weather forecast data; to quantify the forecast uncertainty, the standard deviation is calculated based on the historical 720-hour forecast error data. ,use The principle generates output intervals with a 95.4% confidence level.

[0015] The photoelectric conversion model is ,in: is the component efficiency, is the total area of ​​PV panels (m²);

[0016] The wind power prediction model describes the wind speed-power relationship through a segmented power curve model. ,in: is the cut-in wind speed, is the rated wind speed, To cut out the wind speed, is the rated power of the fan (kW).

[0017] The output range This interval updates the historical data window daily to ensure that it adapts to seasonal changes and equipment attenuation characteristics, and outputs photovoltaic and wind power in Adjustable range of time ( ).

[0018] 1.2 Constructing a coupled energy storage SOC-life model: The scheduling of the energy storage system needs to take into account both the state of charge (SOC) dynamics and cycle life loss. Its state update equation is: ,in: is the scheduling time step, Indicates the charging efficiency, represents the discharge efficiency, Expressed as the battery rated capacity (kWh). The life loss cost function uses the improved Arrhenius model to reflect the accelerated loss of the battery due to deep discharge. ,in: is the loss coefficient, is the SOC sensitivity coefficient. The energy storage system must meet safety constraints , and power constraints , .

[0019] The adjustable potential of the energy storage SOC-lifetime coupling model is , the negative sign indicates the discharge direction, ensuring that the SOC does not exceed the limit during the scheduling process.

[0020] 1.3 Establish an electric vehicle behavior chain model and calculate its adjustable load: Based on the user's historical travel data including charging start and end time (SOC change), use Figure 3 The neural network structure shown predicts vehicle travel parameters The network takes the adjacency matrix and feature matrix as input, and performs feature extraction and pattern recognition through the GNN layer, graph pooling layer and CCNN layer in sequence, and finally outputs the key parameters of the trip. The real-time adjustable potential is dynamically calculated according to the current SOC. , ,in: The minimum SOC set by the user when leaving the vehicle. When charging power is capped , discharge power upper limit , ensuring that it does not affect users' travel needs.

[0021] 1.4 Constructing an adjustable load constraint model: dynamically following the heat balance equation according to the indoor temperature , in the comfortable temperature range set by the user The adjustable cooling load is linearly related to the temperature deviation ,in: is the wall heat exchange coefficient, For cooling efficiency, is the outdoor temperature, is the real-time indoor temperature, The midpoint of comfortable temperature, is the temperature-power conversion coefficient. The negative sign indicates that the cooling power can be reduced when the temperature rises.

[0022] 1.5 Constructing an industrial interruptible load model: Based on the key period set reported by users (accurate to 15 minutes), full base load reduction is allowed during non-critical periods ,in: The real-time operating power (kW) of the equipment is collected from the PLC in real time via the Modbus protocol to ensure a balance between production continuity and regulation feasibility.

[0023] Step 2) The edge proxy node calculates the adjustable potential interval information of all resources and aggregates it, then uploads it to the cloud via a lightweight protocol. Specifically, it includes:

[0024] 2.1 Cluster Grouping: As an edge proxy node connecting the underlying physical devices and the cloud control center, all resources are divided into non-overlapping clusters based on geographic location and communication topology. Specifically: Based on the principle of physical proximity, resources under the same substation feeder are grouped into the same cluster, such as distributed power sources and controllable loads in industrial parks; complementary resources are clustered, such as charging stations and supporting energy storage systems; the coverage range of edge gateway devices (such as PMC-1606) is used as the boundary to ensure that the number of devices in a single cluster does not exceed 200. Each cluster is equipped with an independent edge agent node, which is responsible for local data collection and preprocessing, forming a three-level architecture of "device-edge-cloud".

[0025] 2.2 Based on the device-level adjustable power range output in step 1), the edge agent node calculates the aggregate adjustable power range of all devices in the same cluster obtained by grouping in step 2.1, and calculates the cluster-level potential boundary. Specifically, ,in: and For equipment exist Adjustable power upper and lower limits at all times.

[0026] The aggregation is preferably accelerated by SIMD instructions to implement parallel computing of 200 devices in a single cluster on an ARM Cortex-A53 processor, with a processing delay of <50ms, which meets the actual requirements of power grid dispatching.

[0027] 2.3 The edge proxy node encapsulates the results of step 2.2 into a binary data packet, using a TLV (Type-Length-Value) encoding structure. The data packet is published to the cloud via the MQTT protocol, using QoS 1 to ensure reliable message transmission.

[0028] The release adopts a multi-level communication optimization mechanism, sending a heartbeat packet (18 bytes) containing only a timestamp and a status code every 5 minutes for link monitoring. When the change rate is less than the threshold, only the difference with the previous moment is transmitted, that is, the change rate indicator , sort the clusters by the rate of change, high priority data Enjoy bandwidth priority and dynamically adjust the sending interval based on RTT (Round Trip Time) , where: dynamic threshold , if and only if When a message is received, data upload is triggered, realizing a hybrid transmission mechanism of periodic upload of static data and real-time response of dynamic data.

[0029] Step 3) The cloud uses the Improved Sparrow Search Algorithm (ISSA) to solve the received aggregated results. After generating dispatch instructions based on the obtained global optimization objectives, they are sent to the edge agent nodes to ensure the optimized operation of the virtual power plant in a complex power grid environment. Specifically, the following steps are performed:

[0030] 3.1 The cloud side sets the adjustable power range of the cluster based on the data uploaded by the edge proxy node in step 2 , a nonlinear optimization model is constructed with the goal of minimizing operating costs and accurately tracking grid instructions. Specifically: ,in: is the total scheduling time, is the time step, for Real-time electricity prices are obtained from the electricity market trading platform; The power that the virtual power plant interacts with the main grid; For cluster The equipment loss cost is calculated based on the energy storage life loss model in step 1; Power reference instructions issued by the power grid dispatching center; is the weight coefficient, which is used to balance the power purchase cost and instruction tracking accuracy; the constraints of the optimization model include: power balance constraint, which ensures the interactive balance of power between the virtual power plant and the main grid, that is, ; Resource adjustable range constraints ensure that the actual scheduling power of each cluster is within the adjustable power range determined in step 2, that is, ; Ramp rate constraint, which limits the power change rate of each cluster to avoid frequent start and stop of equipment and grid impact ,in: Set according to the resource type.

[0031] 3.2 Solve the nonlinear optimization model generated in step 3.1 by using an improved sparrow search algorithm based on tent chaos mapping and dynamic weight adjustment, specifically including:

[0032] i) The initial population is generated by using Tent chaotic mapping, and the diversity of the initial solution is improved through the ergodicity and randomness of the chaotic system. Specifically: , the chaotic sequence Map to the solution space and get the initial population position ,in: 、 are the lower and upper bounds of the decision variables, respectively.

[0033] ii) Dynamically adjust the proportion of discoverers, and the discoverer updates its position according to the following rules: When performing a global search , otherwise perform a local search , where: PD is the probability of alertness, is the iteration coefficient, is the maximum number of iterations, is the dynamic weight, is a random number that follows a standard normal distribution.

[0034] iii) Followers are treated differently based on their fitness. Specifically, followers with lower fitness are treated by enhancing their local development capabilities through the following formula: ; Followers with higher fitness move closer to the global optimal solution through the following formula: ,in: For Levy flight strategy, is the flight coefficient.

[0035] The differentiation processing mentioned above means comparing the fitness of the follower with the average fitness of the population. If the individual fitness is lower than the average fitness of the population, it is classified as having lower fitness, otherwise it is classified as having higher fitness, thereby ensuring that the algorithm preferentially guides high-fitness individuals to search for the global optimal solution during the iteration process, while encouraging low-fitness individuals to enhance their local development capabilities to avoid falling into local optimality.

[0036] iv) If individual fitness If the fitness is greater than the population average, it indicates that the individual is trapped in a local optimum and needs to move closer to the global optimal solution. Specifically: Otherwise, a small exploration is performed near the current location. ,in: is the dynamic exploration factor ( ), is a random number in [0,1].

[0037] v) Use the projection method to correct the out-of-bounds solution to ensure that the solution meets the resource adjustable range constraints, specifically: .

[0038] 3.3 The optimal scheduling solution is obtained by improving the sparrow search algorithm After that, the cloud generates standardized scheduling instructions and sends them to the corresponding edge agent nodes through the MQTT protocol.

[0039] The aforementioned sending preferably adopts a two-way sending confirmation mechanism, that is, the edge agent node needs to return an ACK confirmation within 1 second after receiving the instruction. If the confirmation is not received, the cloud will resend it every 500ms, and retry up to 3 times; at the same time, the edge agent node reports the actual execution power every 15 seconds, and the cloud calculates the tracking error in real time. ,when An alarm is triggered and rolling optimization is started.

[0040] Step 4) To address forecast deviations and resource failures that may arise during virtual power plant operation, the model parameters are adjusted based on actual output deviations, and multi-strategy emergency responses and rapid command updates are triggered to ensure stable operation of the system under complex operating conditions. Specifically, the following steps are performed:

[0041] 4.1 The edge agent node calculates the actual output deviation of each cluster and the actual output deviation of the optimized scheduling instruction in real time with a cycle of 5 minutes And collect and upload to the cloud in real time, when The rolling correction process is triggered, that is, step 4.2 is executed, and deviation alarm information is generated and uploaded to the operation and maintenance platform, where: is the optimal scheduling solution obtained in step 3, For cluster exist Actual output power at the moment, dynamic deviation threshold , is the cluster rated power, The total length of a day.

[0042] 4.2 To address the issue of time-varying characteristics of resources such as photovoltaics and energy storage, an extended Kalman filter (EKF) is used to update the parameters of the dynamic characteristic model of distributed resources in step 1 online, and then step 4.3 is performed, specifically including:

[0043] i) Taking the SOC-lifetime coupling model of the energy storage system as an example, the state space model is constructed as follows: ,in: is the state of charge, is the parameter vector to be updated; is the identity matrix, is the output prediction function based on the current parameters, is the actual measured value. The measurement equation is ,in: To measure noise, it follows a Gaussian distribution , by iteratively calculating the Kalman gain , state estimation covariance matrix , realize dynamic correction of model parameters and improve the accuracy of subsequent scheduling predictions.

[0044] ii) Taking the photovoltaic output probability prediction model as an example, the state space model is constructed as follows: ,in: is the efficiency of photovoltaic modules, is the standard deviation of the prediction error; is the parameter vector to be updated; output prediction function Corresponding to the photovoltaic output forecast value, is the actual photovoltaic output measurement value. The measurement equation is: . Updated through iteration and , adapt to seasonal changes and equipment attenuation characteristics.

[0045] iii) Taking the electric vehicle behavior chain model as an example, the state space model is constructed as follows: in: is the weight and bias parameter vector of the neural network (GNN layer, graph pooling layer, CCNN layer); output prediction function is the predicted vehicle travel parameter , is the actual user travel data (charging start / end time, SOC change). The measurement equation is: The EKF method is used to dynamically modify the neural network parameters to improve the accuracy of travel behavior prediction.

[0046] iv) Taking the adjustable load constraint model as an example, the state space model is constructed as follows: ,in: is the indoor temperature, is the wall heat exchange coefficient, For cooling efficiency, is the parameter vector to be updated; output prediction function Corresponding to the predicted indoor temperature value at the next moment, is the actual measured indoor temperature. The measurement equation is: Dynamically update thermal characteristic parameters to optimize adjustable cooling load prediction.

[0047] v) Taking the industrial interruptible load model as an example, the state space model is constructed as follows: ,in: is the equipment base load power, is the parameter to be updated; output prediction function (Power can be interrupted during non-critical periods), is the actual base load power collected through the Modbus protocol. The measurement equation is: . Correct basic load parameters in real time to ensure the feasibility of interruptible load scheduling.

[0048] 4.3 The edge agent node monitors the device heartbeat packet in real time (sent every 5 minutes). If it fails to receive the heartbeat packet for 3 consecutive times or receives an abnormal device status signal (such as energy storage SOC exceeding the limit, photovoltaic output dropping beyond the threshold), it is determined to be a resource failure and the fault information is uploaded to the cloud. hour( is the power shortage of the fault resource, is the cluster rated power), start strategy 1; otherwise, start strategy 2.

[0049] Strategy 1: Rapidly call backup resources from a pre-set backup resource pool (including reserved energy storage capacity, interruptible load, etc.), the goal is to solve the problem with the minimum call cost: ,in: For backup resources The call cost coefficient, Enable state variables for resources, Fault resources The power shortage.

[0050] Strategy 2: Dynamic reallocation of instructions to eliminate faulty resources Then, the improved sparrow search algorithm (ISSA) in step 3 is called again and optimized under the modified constraints: ,by dynamically adjusting the scheduling instructions of non-faulty resources, the system power balance can be quickly restored.

[0051] 4.4 New dispatch instructions generated by the cloud after deviation correction or fault processing , and is sent to the edge proxy node via a low-latency communication channel (such as a 5G private network or industrial-grade WiFi). This channel uses a hard real-time transmission protocol with an end-to-end latency of ≤10ms, making it suitable for rapid command updates in fault emergency scenarios. Unlike the MQTT protocol used for conventional dispatch instructions in step 3.3, this mechanism automatically switches to a low-latency channel when a prediction deviation exceeds a threshold or a resource failure is detected, ensuring that emergency commands are transmitted first and meeting the requirements for safe and stable power grid operation.

[0052] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principles and purpose of the present invention. The scope of protection of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. All implementation schemes within its scope shall be subject to the constraints of the present invention.

Claims

1. A virtual power plant intelligent aggregation optimization control method for multi-type flexibility resources, characterized by: By constructing a dynamic characteristic model of distributed resources that includes a photovoltaic output probability prediction model, an energy storage SOC-life coupling model, an electric vehicle behavior chain model, an adjustable load constraint model, and an industrial interruptible load model, the edge agent node calculates the adjustable potential range of the resource cluster in real time and uploads it to the cloud. In the cloud, the global optimization objective is solved based on the improved sparrow search algorithm (ISSA). After generating the dispatch instruction, the model parameters are rolled-corrected according to the actual output deviation calculated in real time, triggering a resource failure emergency strategy for the prediction deviation and resource failure problems that occur during the operation of the virtual power plant.

2. The method for intelligent aggregation and optimization control of virtual power plants for multi-type flexibility resources according to claim 1 is characterized in that: The photovoltaic output probability prediction model is constructed in the following way: first, based on the intermittent characteristics of photovoltaic and wind power, and wind speed The PV output probability prediction model including the photovoltaic conversion model and the wind power prediction model is constructed based on the weather forecast data; to quantify the forecast uncertainty, the standard deviation is calculated based on the historical 720-hour forecast error data. ,use The principle generates an output interval with a 95.4% confidence level; The photoelectric conversion model is ,in: is the component efficiency, is the total area of ​​PV panels (m²); The wind power prediction model describes the wind speed-power relationship through a segmented power curve model. ,in: is the cut-in wind speed, is the rated wind speed, To cut out the wind speed, is the rated power of the fan (kW); The output range This interval updates the historical data window daily to ensure that it adapts to seasonal changes and equipment attenuation characteristics, and outputs photovoltaic and wind power in Adjustable range of time ( ).

3. The method for intelligent aggregation and optimization control of virtual power plants for multi-type flexibility resources according to claim 1 is characterized in that: The energy storage SOC-life coupling model is constructed in the following way: the scheduling of the energy storage system needs to take into account the state of charge (SOC) dynamics and cycle life loss, and its state update equation is: ,in: is the scheduling time step, Indicates the charging efficiency, represents the discharge efficiency, Expressed as the battery rated capacity (kWh), the life loss cost function uses the improved Arrhenius model to reflect the accelerated loss of the battery due to deep discharge. ,in: is the loss coefficient, is the SOC sensitivity coefficient, and the energy storage system needs to meet the safety constraints , and power constraints , ; The adjustable potential of the energy storage SOC-lifetime coupling model is , the negative sign indicates the discharge direction, ensuring that the SOC does not exceed the limit during the scheduling process.

4. The method for intelligent aggregation and optimization control of virtual power plants for multi-type flexibility resources according to claim 1 is characterized in that: The electric vehicle behavior chain model is constructed by using a neural network structure to predict vehicle travel parameters based on user historical travel data including charging start and end time and SOC change. The network takes the adjacency matrix and feature matrix as input, and performs feature extraction and pattern recognition through the graph neural network (GNN) layer, graph pooling layer and CCN layer in sequence, and finally outputs the key parameters of the trip, and dynamically calculates the real-time adjustable potential according to the current SOC. , ,in: The minimum SOC set by the user when leaving the vehicle, when the vehicle is in the charging period When charging power is capped , discharge power upper limit , ensuring that it does not affect users' travel needs.

5. The method for intelligent aggregation and optimization control of virtual power plants for multi-type flexibility resources according to claim 1 is characterized in that: The adjustable load constraint model is constructed in the following way: according to the indoor temperature dynamics follow the heat balance equation , in the comfortable temperature range set by the user The adjustable cooling load is linearly related to the temperature deviation ,in: is the wall heat exchange coefficient, For cooling efficiency, is the outdoor temperature, is the real-time indoor temperature, The midpoint of comfortable temperature, is the temperature-power conversion coefficient. The negative sign indicates that the cooling power can be reduced when the temperature rises.

6. The method for intelligent aggregation and optimization control of virtual power plants for multi-type flexibility resources according to claim 1 is characterized in that: The industrial interruptible load model is constructed in the following way: according to the key period set reported by the user (accurate to 15 minutes), full base load reduction is allowed during non-critical periods ,in: The real-time operating power (kW) of the equipment is collected from the PLC in real time via the Modbus protocol to ensure a balance between production continuity and regulation feasibility.

7. The method for intelligent aggregation and optimization control of virtual power plants for multi-type flexibility resources according to claim 1 is characterized in that: The adjustable potential range of the real-time computing resource cluster specifically includes: 2.1 Cluster Grouping: As an edge proxy node connecting the underlying physical devices and the cloud control center, all resources are divided into non-overlapping clusters based on geographic location and communication topology. Specifically, based on the principle of physical proximity, resources under the same substation feeder are grouped into the same cluster, such as distributed power sources and controllable loads within an industrial park. Complementary resources are clustered, such as charging stations and supporting energy storage systems. The coverage area of ​​edge gateway devices (such as the PMC-1606) is used as the boundary to ensure that the number of devices in a single cluster does not exceed 200. Each cluster is equipped with an independent edge agent node responsible for local data collection and preprocessing, forming a three-level "device-edge-cloud" architecture. 2.2 Based on the device-level adjustable power range output in step 1), the edge agent node calculates the aggregate adjustable power range of all devices in the same cluster obtained by grouping in step 2.1, and calculates the cluster-level potential boundary. Specifically, ,in: and For equipment exist Adjustable power upper and lower limits at all times; 2.3 The edge proxy node encapsulates the results of step 2.2 aggregation into a binary data packet, using a TLV (Type-Length-Value) encoding structure. The data packet is published to the cloud through the MQTT protocol, using QoS 1 to ensure reliable message transmission.

8. The method for intelligent aggregation and optimization control of virtual power plants for multi-type flexibility resources according to claim 7 is characterized in that: The release adopts a multi-level communication optimization mechanism, sending a heartbeat packet (18 bytes) containing only a timestamp and a status code every 5 minutes for link monitoring. When the change rate is less than the threshold, only the difference with the previous moment is transmitted, that is, the change rate indicator , sort the clusters by the rate of change, high priority data Enjoy bandwidth priority and dynamically adjust the sending interval based on RTT (Round Trip Time) , where: dynamic threshold , if and only if When a message is received, data upload is triggered, realizing a hybrid transmission mechanism of periodic upload of static data and real-time response of dynamic data.

9. The method for intelligent aggregation and optimization control of virtual power plants for multi-type flexibility resources according to claim 1 is characterized in that: The scheduling instruction is obtained in the following way: 3.1 Cluster adjustable power range based on edge agent node upload in the cloud , a nonlinear optimization model is constructed with the goal of minimizing operating costs and accurately tracking grid instructions. Specifically: ,in: is the total scheduling time, is the time step, for Real-time electricity prices are obtained from the electricity market trading platform; The power that the virtual power plant interacts with the main grid; For cluster The equipment loss cost is calculated based on the energy storage life loss model in step 1; Power reference instructions issued by the power grid dispatching center; is the weight coefficient, which is used to balance the power purchase cost and instruction tracking accuracy; the constraints of the optimization model include: power balance constraint, which ensures the interactive balance of power between the virtual power plant and the main grid, that is, ; Resource adjustable range constraints ensure that the actual scheduling power of each cluster is within the adjustable power range determined in step 2, that is, ; Ramp rate constraint, which limits the power change rate of each cluster to avoid frequent start and stop of equipment and grid impact ,in: Set according to resource type; 3.2 Solve the nonlinear optimization model generated in step 3.1 by using an improved sparrow search algorithm based on tent chaos mapping and dynamic weight adjustment, specifically including: i) The initial population is generated by using Tent chaotic mapping, and the diversity of the initial solution is improved through the ergodicity and randomness of the chaotic system. Specifically: , the chaotic sequence Map to the solution space and get the initial population position ,in: 、 are the lower and upper bounds of the decision variables respectively; ii) Dynamically adjust the proportion of discoverers, and the discoverer updates its position according to the following rules: When performing a global search , otherwise perform a local search , where: PD is the probability of alertness, is the iteration coefficient, is the maximum number of iterations, is the dynamic weight, is a random number that obeys the standard normal distribution; iii) Followers are treated differently based on their fitness. Specifically, followers with lower fitness are treated by enhancing their local development capabilities through the following formula: ; Followers with higher fitness move closer to the global optimal solution through the following formula: ,in: For Levy flight strategy, is the flight coefficient; iv) If individual fitness If the fitness is greater than the population average, it indicates that the individual is trapped in a local optimum and needs to move closer to the global optimal solution. Specifically: Otherwise, a small exploration is performed near the current location. ,in: is the dynamic exploration factor ( ), is a random number in [0,1]; v) Use the projection method to correct the out-of-bounds solution to ensure that the solution meets the resource adjustable range constraints, specifically: ; 3.3 The optimal scheduling solution is obtained by improving the sparrow search algorithm After that, the cloud generates standardized scheduling instructions and sends them to the corresponding edge agent nodes through the MQTT protocol.

10. The method for intelligent aggregation and optimization control of virtual power plants for multi-type flexibility resources according to claim 1 is characterized in that: To address forecast deviations and resource failures that may arise during virtual power plant operation, model parameters are adjusted based on actual output deviations, triggering multi-strategy emergency responses and rapid command updates to ensure stable system operation under complex operating conditions. Specifically, these include: 4.1 The edge agent node calculates the actual output deviation of each cluster and the actual output deviation of the optimized scheduling instruction in real time with a cycle of 5 minutes And collect and upload to the cloud in real time, when The rolling correction process is triggered, that is, step 4.2 is executed, and deviation alarm information is generated and uploaded to the operation and maintenance platform, where: is the optimal scheduling solution obtained in step 3, For cluster exist Actual output power at the moment, dynamic deviation threshold , is the cluster rated power, is the total length of a day; 4.2 To address the issue of time-varying characteristics of resources such as photovoltaics and energy storage, an extended Kalman filter (EKF) is used to update the parameters of the dynamic characteristic model of distributed resources in step 1 online, and then step 4.3 is performed, specifically including: i) Taking the SOC-lifetime coupling model of the energy storage system as an example, the state space model is constructed as follows: ,in: is the state of charge, is the parameter vector to be updated; is the identity matrix, is the output prediction function based on the current parameters, is the actual measured value, and the measurement equation is ,in: To measure noise, it follows a Gaussian distribution , by iteratively calculating the Kalman gain , state estimation covariance matrix , realize dynamic correction of model parameters and improve the accuracy of subsequent scheduling prediction; ii) Taking the photovoltaic output probability prediction model as an example, the state space model is constructed as follows: ,in: is the efficiency of photovoltaic modules, is the standard deviation of the prediction error; is the parameter vector to be updated; output prediction function Corresponding to the photovoltaic output forecast value, is the actual photovoltaic output measurement value, and the measurement equation is: , through iterative updates and , adapt to seasonal changes and equipment attenuation characteristics; iii) Taking the electric vehicle behavior chain model as an example, the state space model is constructed as follows: in: is the weight and bias parameter vector of the neural network; output prediction function is the predicted vehicle travel parameter , is the actual user travel data (charging start / end time, SOC change), and the measurement equation is: ,dynamically modify the neural network parameters through EKF to improve the ,precision of travel behavior prediction; iv) Taking the adjustable load constraint model as an example, the state space model is constructed as follows: ,in: is the indoor temperature, is the wall heat exchange coefficient, For cooling efficiency, is the parameter vector to be updated; output prediction function Corresponding to the predicted indoor temperature value at the next moment, is the actual measured indoor temperature, and the measurement equation is: ,optimize the adjustable cooling load prediction by dynamically updating the,thermal characteristic parameters; v) Taking the industrial interruptible load model as an example, the state space model is constructed as follows: ,in: is the equipment base load power, is the parameter to be updated; output prediction function (Power can be interrupted during non-critical periods), is the actual base load power collected through the Modbus protocol. The measurement equation is: , by correcting the basic load parameters in real time, the feasibility of interruptible load scheduling is guaranteed; 4.3 The edge proxy node monitors the device heartbeat packets in real time. If it fails to receive a heartbeat packet for three consecutive times or receives an abnormal device status signal, it is determined to be a resource failure and the failure information is uploaded to the cloud. 4.4 New dispatch instructions generated by the cloud after deviation correction or fault processing , and sent to the edge proxy node through a low-latency communication channel.

Images