Comprehensive Evaluation Method, System and Equipment for Adjustable Potential of Virtual Power Plant Considering Communication and Action Delay

Abstract

The present invention discloses a comprehensive evaluation method, system and device for the adjustable potential of a virtual power plant that takes into account both communication and action time delays. The method constructs a cloud-edge collaborative network system framework for regulating the virtual power plant, constructs a communication system delay model, an equipment action time model and a communication reliability model. Then, the k-shortest path algorithm is used to establish a virtual power plant maximum adjustable potential evaluation model based on timedelay step search. Finally, a method based on the proximal policy optimization algorithm is proposed to solve the optimization model. The present invention accurately describes the comprehensive evaluation of the adjustable potential of a virtual power plant considering the influence of the communication system and the resource action time, provides a decision-making basis for the virtual power plant to participate in primary and secondary frequency modulation and peak shaving of the power system, and is beneficial to its support for the frequency stability and power balance of the power system.

Description

Technical Field

[0001] The present invention relates to the field of comprehensive evaluation of adjustable potential of flexible loads, and particularly to a comprehensive evaluation method, system and device for adjustable potential of a virtual power plant that takes into account both communication and action time delays. Background Art

[0002] As an emerging technology and market entity, a virtual power plant is considered an effective means to alleviate the peak-valley difference of grid loads and maintain the balance between power supply and demand in the future. The quantitative relationship between the response speed and adjustable capacity of a virtual power plant can reflect its ability to respond to system regulation requirements at different time scales, and this ability is mainly affected by communication time and resource regulation action performance. Therefore, the comprehensive evaluation of the adjustable potential of a virtual power plant considering the influence of communication systems and resource action times can provide a decision-making basis for its participation in primary and secondary frequency modulation and peak shaving of the power system, and is conducive to supporting the frequency stability and power balance of the power system.

[0003] In the field of comprehensive evaluation of adjustable potential of flexible loads, many scholars have carried out research, but the existing research cannot meet the actual dispatching needs of the power grid. The actual power grid dispatching also requires the evaluation of comprehensive regulation capabilities such as response speed and duration. And the traditional centralized architecture used will bring a series of disadvantages at the communication transmission level. At the same time, the existing research has not considered the problem of deviation in response capacity caused by factors such as time delay, packet loss, and error code during information transmission. Summary of the Invention

[0004] Object of the Invention: The present invention aims to provide a comprehensive evaluation method for the adjustable potential of flexible loads of a virtual power plant that takes into account both response capacity and response time delay by comprehensively considering various influencing factors such as equipment action response time delay, communication reliability, packet loss, and error code during communication from the perspective of a cyber-physical system; another object of the present invention is to provide a comprehensive evaluation system and device for the adjustable potential of a virtual power plant that takes into account both communication and action time delays.

[0005] Technical Solution: The comprehensive evaluation method for the adjustable potential of a virtual power plant that takes into account both communication and action time delays according to the present invention includes the following steps:

[0006] Construct a communication system time delay model, an equipment action time model, and a communication reliability model respectively according to the virtual power plant cloud-edge collaborative network system framework;

[0007] According to the virtual power plant cloud-edge collaborative network system framework, the communication system time delay model, the equipment action time model, and the communication reliability model, use the KSP algorithm to obtain the optimal path and backup path between the source node m and the destination node n, and determine the total transmission time delay t between the source node m and the destination node n during data transmission d and construct a virtual power plant maximum adjustable potential evaluation model based on timedelay step size search;

[0008] Through the PPO algorithm, the maximum adjustable potential evaluation model of the virtual power plant is optimized to obtain the online evaluation of the adjustable potential of the virtual power plant.

[0009] Furthermore, the cloud-edge collaborative network system framework of the virtual power plant includes a cloud layer, an edge layer, and a response terminal; the cloud layer is used to centrally integrate the computing power resources of numerous servers, responsible for processing global data services, and the data services include comprehensive evaluation, predicting the aggregated adjustable capacity of the virtual power plant, decomposing scheduling instructions to each edge node, and overall communication optimization control of the cloud-edge-terminal system; the edge layer is used to calculate the aggregated load resource model at the edge node, and the edge nodes of the edge layer are set at positions close to the user side.

[0010] Furthermore, the communication system delay model is as follows:

[0011] The wired transmission delay between the central cloud and the edge node is

[0012]

[0013] where x i as a binary variable, x i = 1 represents that there is data packet transmission between the cloud node and the edge node i; otherwise x i = 0; L i represents the size of the downstream data packet of the i-th node; r i is the wired transmission rate between the central cloud and the edge node; b i represents the channel bandwidth between cloud nodes, P i represents the data packet transmission power, G i represents the channel gain, and G i = d i -α d i represents the distance between the i-th node and the central cloud node, -α is the signal fading factor, and s represents the transmission noise;

[0014] The wired propagation delay between the cloud node and the edge node is

[0015]

[0016] where v is the propagation speed of the optical signal in the optical fiber;

[0017] The downstream data delay t between the cloud edge and the edge node i is

[0018]

[0019] where Nh represents the routing hop count, t f represents the processing delay of a single device;

[0020] The computing delay of the j-th edge node is

[0021]

[0022] where, f j is the computing power of the j-th edge node, e j is the number of CPU cycles required for the computing task;

[0023] The wireless download delay t p is

[0024]

[0025] where, represents the queue position of user p under the j-th regional node, and T is the minimum indivisible period in wireless transmission; j

[0026] The wireless delay t p is described by a Pareto distribution, and the probability distribution function P is

[0027]

[0028] where, τ0 represents the minimum value of the inter-node delay, γ is a parameter related to the load rate, and the computing delay of the cloud server is a constant, t p is the wireless delay generated during the process of the base station transmitting to the demand response terminal.

[0029] Furthermore, the communication reliability model is as follows:

[0030] The reliability S between nodes m and n mn is

[0031]

[0032] where, N mn represents the number of links between m and n; S(e l ) represents the reliability of link e l , and information element failures, transmission packet losses, and bit errors will all affect the reliability of the link; the reliability S mn The upper limit of the constrained routing hop count is set to

[0033] Furthermore, the KSP algorithm is used to obtain the optimal path and the backup path between the source node m and the destination node n, and the deviation link weight of the transmission path between m and n is as follows:​

[0034]

[0035] Among them, W mn is the link weight between nodes m and n, and t mn is the time delay between nodes m and n, and c l represents the packet loss coefficient, represents rounding up, and t s is the delay caused by packet loss and bit error, which is a part of the total time delay and is set as a constant, and d is a random number generated between 0 and 1.

[0036] Furthermore, the total transmission time delay t between the source node m and the destination node n during the data transmission process d is

[0037]

[0038] Among them, t ij is the wired time delay generated when the edge node issues a scheduling instruction through the radio base station, t a is the action time of the device, and t s is the delay caused by packet loss and bit error.

[0039] Furthermore, the virtual power plant maximum adjustable potential evaluation model based on timedelay step search is

[0040]

[0041] Among them, F is the total adjustable capacity of the central cloud virtual power plant, represents the maximum adjustable capacity of the p j th demand response terminal within the coverage range of the jth radio base station, t allow is the maximum allowable response time delay, and x ij is the path selection result of the edge node and the BS node in the load scheduling communication link.

[0042] The virtual power plant adjustable potential comprehensive evaluation system that takes into account communication and action time delays according to the present invention includes

[0043] a model construction module, which is used to respectively construct a communication system time delay model, a device action time model, and a communication reliability model according to the virtual power plant cloud-edge collaborative network system framework; and is used to obtain the optimal path and the backup path of the source node m and the destination node n by using the KSP algorithm according to the virtual power plant cloud-edge collaborative network system framework, the communication system time delay model, the device action time model, and the communication reliability model, and determine the total transmission time delay t between the source node m and the destination node n during the data transmission process d , and construct a virtual power plant maximum adjustable potential evaluation model based on timedelay step search;

[0044] A strategy optimization evaluation module, which is used to determine the optimal strategy for the maximum adjustable potential evaluation model of the virtual power plant through the PPO algorithm, and obtain the online evaluation of the adjustable potential of the virtual power plant.

[0045] The computer device of the present invention includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the above method are implemented.

[0046] The computer-readable storage medium of the present invention stores a computer program thereon. When the computer program is executed by a processor, the steps of the above method are implemented.

[0047] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: 1. The present invention adopts a cloud-edge collaborative network system framework to solve the problems of large data transmission delay, difficult storage and utilization of massive data, easy occurrence of transmission failures and limited communication capabilities in the centralized architecture. The edge nodes provide computing resources at the network edge, which are closer to the terminal devices and users, helping to relieve the computing pressure of the cloud nodes, reduce the communication network delay, and strengthen network information security at the same time; 2. The present invention can consider the influence of cloud-edge architecture communication and device action time on the adjustable capacity response delay, quantify the regulation performance of the virtual power plant on a short time scale, and solve the problem that the current research cannot meet the actual dispatching requirements of the power grid; 3. The present invention takes into account the problem of deviation in the response capacity during the dispatching process caused by problems such as delay, packet loss, and error code in the information transmission process in the field of comprehensive evaluation of flexible load adjustable potential, making the evaluation result more reliable; 4. The present invention uses the k shortest paths KSP algorithm to iteratively calculate the optimal path and backup path from the source node to the destination node according to the delay parameters of each edge of the network, providing more flexibility and optimization space for decision-makers. When facing a complex and changeable network environment, it can more effectively cope with uncertain factors such as network congestion and failures; 5. The present invention uses the proximal policy optimization algorithm to solve the virtual power plant adjustable potential evaluation method for the maximum adjustable potential evaluation model of the virtual power plant, realizing the rapid online evaluation of the adjustable potential of the virtual power plant that takes into account both the response capacity and the response delay, significantly improving the policy performance and sample efficiency, and effectively avoiding the drastic performance fluctuations during the training process. Description of the Drawings

[0048] Figure 1 It is a flowchart of the present invention;

[0049] Figure 2 It is a cloud-edge collaborative data interaction architecture diagram;

[0050] Figure 3 It is a cloud-edge-terminal data interaction framework diagram;

[0051] Figure 4 It is a flow chart for adjustable potential evaluation and scheduling;

[0052] Figure 5 It is a response mode diagram of temperature control load;

[0053] Figure 6 It is a diagram of approximate action time models for various devices;

[0054] Figure 7 It is a flow chart for evaluating the adjustable potential of a virtual power plant;

[0055] Figure 8 It is a schematic diagram of the IEEE 33-bus system including a communication network;

[0056] Figure 9 It is a curve diagram of the adjustable potential of the load on a typical day;

[0057] Figure 10 It is a reward curve diagram of each algorithm;

[0058] Figure 11 It is a diagram of the evaluation results of the daily adjustable potential of a virtual power plant;

[0059] Figure 12 It is a diagram of the proportion of delay of each part of the BS node;

[0060] Figure 13 It is a diagram of the evaluation results of the adjustable potential of the load in each mode;

[0061] Figure 14 It is a diagram of the deviation of adjustable potential caused by packet loss;

[0062] Figure 15 It is a diagram of the deviation of wireless transmission delay caused by the load rate. Specific implementation manners

[0063] The present invention will be further described below with reference to the accompanying drawings.

[0064] The comprehensive evaluation method for the adjustable potential of a virtual power plant that takes into account communication and action delays according to the present invention includes the following steps:

[0065] (1) Construct a cloud-edge collaborative network system framework for regulating a virtual power plant;

[0066] State Grid Corporation divides the construction of the ubiquitous power Internet of Things into four levels: the application layer, the platform layer, the network layer, and the perception layer in the "Construction Outline of the Ubiquitous Power Internet of Things". This principle can represent the cyber-physical architecture of a virtual power plant. Based on the above cyber-physical architecture, as Figure 2As shown in the figure, the present invention proposes a cloud-edge collaborative architecture for a virtual power plant, which is used to evaluate and predict the adjustable capacity of a large number of flexible loads participating in the virtual power plant and to achieve optimal internal resource scheduling. Different from the centralized architecture, the edge nodes are placed close to the user side, and the central cloud relies on the virtual power plant cloud to centrally integrate the computing power resources of many servers and is responsible for processing global high-complexity data services, including comprehensively evaluating and predicting the aggregated adjustable capacity of the virtual power plant, decomposing the scheduling instructions to each edge node, and overall communication optimization control of the cloud-edge-end system. Technology service providers and load aggregators are located at the edge nodes. The edge servers they use have the characteristics of low energy consumption, low latency, and fast read-write speed, and are suitable for computing tasks facing the terminal with high reliability and low latency requirements. The edge nodes are connected to the central cloud and the distribution substation through the power backbone network, and a networking method combining optical fiber communication, dense wavelength division multiplexing technology, and synchronous digital hierarchy is mostly adopted. The distribution substation connects to the response terminal through the power access network, and usually adopts various communication methods such as LTE, 4 / 5G, Wi-Fi, and Ethernet. This architecture makes full use of the computing power resources of the cloud and edge ends, combines the three-layer regulation requirements of the virtual power plant cloud, edge end, and user terminal, and meets the requirements of regulation latency, reliability, accuracy, and computing efficiency.

[0067] As Figure 3 Shown in the figure is the cloud-edge-end data interaction framework applied to the virtual power plant. The data of each layer is collected to the cloud center. By calculating and analyzing the collected data, the cloud layer processes the scheduling control and external cooperation services of the virtual power plant at the system level. During the scheduling process at all levels, the model center of the cloud layer is responsible for storing and calculating global, aggregated, and high-complexity models of all edge layers, so as to ensure the global optimal scheduling within the virtual power plant. The model center of the edge layer is responsible for calculating the aggregated load resource model near the edge node to share the computing pressure of the cloud center, reduce the large amount of latency generated by the direct interaction between the cloud layer and the terminal data in the centralized scheduling, and ensure the data privacy and security of the virtual power plant serving users.

[0068] Cloud layer. The virtual power plant cloud server has rich storage and computing resources and can interact with platforms such as the power dispatching center, the power grid cloud platform, and the power grid data center. During the comprehensive evaluation of the aggregable adjustable capacity of the virtual power plant, the cloud server is mainly responsible for using communication system optimization methods to plan the information transmission path for network data in the system, so as to optimize the delay and reliability of data transmission in the information system, and evaluate the aggregable adjustable capacity of the virtual power plant that takes into account both communication and action time based on the uploaded aggregated terminal cluster data. During the adjustable capacity prediction process, the edge-side evaluation data is transmitted into the edge-side prediction model, and at the same time, the cloud-side evaluation data is input into the aggregable adjustable capacity prediction model, and the cloud-edge mechanism is used to achieve refined adjustable capacity prediction. According to external data such as electricity price data and unit data, the cloud platform conducts economic dispatching calculations for the virtual power plant to participate in the power system, ensuring economic dispatching on the basis of supply-demand balance. Then, according to the obtained day-ahead and intra-day dispatching plans, it guides the optimized dispatching of the adjustable resources inside the virtual power plant.

[0069] Edge layer. The edge layer plays a role in connecting the terminal and the cloud layer in cloud-edge collaboration. The edge nodes have both the uplink communication ability with the virtual power plant cloud platform and the downlink communication ability with the distribution substation, including edge controllers, edge servers, and edge gateways. Each edge node is not only responsible for the data processing and operation management of the distribution substation area where the regional data is uploaded, but also has the function of assisting the analysis of large models deployed on the cloud server in the evaluation and prediction of the aggregable adjustable capacity of the edge. In the evaluation of the aggregable adjustable capacity of the virtual power plant, the edge nodes are responsible for collecting the uploaded user operation data, evaluating the adjustable capacity of users in the region based on the differential evaluation model configured on the edge side, and uploading the cluster data, which is further evaluated by the cloud platform. In addition, the edge nodes can also be configured with small applications for users, such as short-term load baseline prediction and demand response plan filling, to help users refine the adjustment of future electricity consumption behaviors.

[0070] Terminal. The response terminal devices are deployed on the user side and are responsible for real-time collecting power data and device operation data and uploading them to the distribution substation. The user terminals and the distribution substation have a one-to-many correspondence. The terminals are mainly oriented towards intelligence, controlling multiple deployed intelligent systems, and comprehensively regulating the operation of various intelligent electrical devices according to the downloaded terminal control instructions.

[0071] The evaluation and dispatching process of the adjustable potential of the virtual power plant based on cloud-edge collaboration is as follows:

[0072] 1) The automatic demand response contract is signed by the user, and the user data is regularly sent to the base station through the regional network.

[0073] 2) The edge node EN collects the data of the customer loads within the coverage area, predicts the adjustable capacity and response time of the adjustable load based on the real-time and historical data, and uploads the results to the cloud node.

[0074] 3) The central cloud plans the information transmission path based on the network topology, and evaluates the adjustable potential of the area including reliability and delay metrics based on the uploaded data.

[0075] 4) When it senses the imbalance between power grid supply and demand, the computing cloud optimizes the result and sends it to the EN. After edge optimization, the scheduling instruction is transmitted to the terminal, and the terminal executes and feeds back the result.

[0076] As Figure 4 shown in the adjustable potential evaluation and scheduling process, the two-stage scheduling process of the negative virtual power plant under the cloud-edge collaborative architecture is divided into three stages as a whole. The first stage is the data upload stage, where each terminal node collects the real-time operation data of the load within the node range, and the regional node collects and processes the regional operation data; the second stage is the data processing stage. After the cloud node receives the scheduling instruction issued by the power dispatching center, it sends a request to the regional node. The regional node sends the regional data to the selected edge node along the path with the minimum preset upload delay. The edge node evaluates the real-time adjustable capabilities of each region and calculates the active power and reactive power of each region. The data calculation results are sent to the corresponding regional node and cloud node at the same time; the third stage is the optimization and instruction issuance stage, which is the optimization of the process of responding to the instruction and issuing it to the terminal during the actual scheduling process. The optimization process is divided into two stages: cloud optimization and edge optimization. Cloud optimization includes two parts. The first part is the response power optimization of the regional node to obtain the total sum of the demand response scheduling power within each region; the second part is the optimization of the information transmission path to select the terminal node corresponding to edge computing and plan the transmission path. The node selection information is sent to the regional node, and the regional node uploads the adjustable capability evaluation result and power data to the edge node along the planned route to prepare for edge optimization. The object of edge optimization is the response power of the terminal node, and the response instruction is issued to the terminal node for execution by the terminal.

[0077] The cloud-edge collaborative system established by the present invention is represented by a weighted undirected graph G = {V, E} for the network topology structure. Among them, is the node set of the network, and N v represents the number of nodes, is the link set of the network, and N e represents the number of links. The network nodes include a central cloud server, an edge server, a distribution substation, and a router. The set of I server nodes is denoted as When i = 0, it represents the cloud server located at the regional main substation. Assume that J distribution substations distributed in various residential communities and commercial buildings are equipped with deployed wireless base stations (BS), and the set is denoted as The jth BS wireless coverage area contains P j user demand response terminals, and its set is denoted as The base station transmits data to and from the server and server nodes via an optical fiber network, and data interaction is completed through a router.

[0078] (2) According to the cloud-edge collaborative network system framework, models for communication delay, device action time, and communication reliability were constructed respectively;

[0079] The communication system delay model is as follows:

[0080] After receiving the scheduling task, the cloud server first sends a control instruction to the edge node, resulting in a wired downlink data delay t i . In the distributed optimization process under the cloud-edge information system architecture, the cloud server optimizes the communication path and the total demand response volume within the coverage area of each BS, and the edge node optimizes the response volume of each user demand response terminal that can be responded to. The delay generated in this process includes the cloud server calculation delay and the edge node calculation delay The edge node issues a scheduling instruction, resulting in a wired delay t ij after passing through the BS, and finally a wireless delay t p is generated during the process of transmitting from the base station to the demand response terminal. Since the scale of the user response backhaul data is small, it can be ignored. The action time of the device is expressed as t a .

[0081] The end-to-end delay in the power system includes inherent delay, transmission delay, propagation delay, and queuing delay. Since the inherent delay accounts for a very small part of the total delay, it is ignored. According to Shannon's theorem, the wired transmission rate r i between the central cloud and the edge node is:

[0082]

[0083] In the formula: x i is used as a binary variable, x i = 1 represents the existence of data packet transmission between the cloud node and the edge node i, otherwise x i = 0, L i represents the size of the downlink data packet of the i-th node, b i represents the channel bandwidth between cloud nodes, P i represents the data packet transmission power, G i represents the channel gain, and G i = d i -α , d i represents the distance between the i-th node and the central cloud node, -α is the signal fading factor, and s represents the transmission noise.

[0084] The propagation delay is related to the end-to-end distance. The wired propagation delay between the cloud node and the edge node EN It can be expressed as:

[0085]

[0086] In the formula: v is the propagation speed of the optical signal in the optical fiber, generally taking 2 / 3c. The downlink data delay t between the cloud edge and the EN i can be expressed as follows:

[0087]

[0088] In the formula: N h represents the number of routing hops, and t f represents the processing delay of a single device.

[0089] For the wired delay t between the EN and the BS ij , its calculation principle is the same as that of the finite delay between the cloud node and the EN, and is the same as Equation (3).

[0090] The cloud server, according to the scheduling instructions issued by the power dispatching center, after optimization in the cloud, and according to the edge optimization requirements, allocates the optimization tasks to the EN. Define the computing power of the jth edge node as f j , the number of CPU cycles required for the computing task as e j , and its computing delay can be expressed as:

[0091]

[0092] The delay generated by wireless transmission includes the upload delay in the data upload process and the download delay generated in the instruction download process. The wireless upload delay can be calculated using the M / M / 1 queuing model. However, in this paper, since the process of uploading user data to the regional node is carried out in advance, the wireless upload delay is not considered. During the download process, the regional node issues scheduling instructions according to the load adjustable ability. The larger the adjustable ability of the terminal, the higher its ranking in the queue. Since the data packets transmitted during the download stage are smaller, the wireless download delay t p can be expressed as:

[0093]

[0094] In the formula represents the queue position of user p j under the jth regional node, and T is the smallest indivisible period in wireless transmission.

[0095] In addition, generally, the computing resources of the cloud server are very large, so the computing delay of the cloud server can be set as a constant. The delay of the information link between two nodes in the access network will show probabilistic characteristics. When the service load is certain, the wireless transmission delay can be described by the Pareto distribution, and the probability distribution function is expressed as:

[0096]

[0097] In the formula: τ0 represents the minimum value of the time delay between nodes, and γ is a parameter related to the load rate.

[0098] The device action time model is as follows:

[0099] In the present invention, the user equipment in potential evaluation is divided into four types: electric vehicles, temperature control loads, non-industrial interruptible loads, and industrial loads.

[0100] 1) Electric vehicles

[0101] The operator can arrange the electric vehicles parked at the charging station for equivalent energy storage and communicate with the BS by accessing the access point. It is assumed that the electric vehicles can only be parked in the electric vehicle cluster and the parking state of the electric vehicles remains unchanged during the scheduling period. After receiving the scheduling signal, the charging station sets the target charging capacity of the electric vehicle and broadcasts data packets to all electric vehicles, and the corresponding electric vehicle receives the data packets and responds.

[0102] 2) Temperature control loads

[0103] The temperature control load can provide operating reserves by reducing power consumption. Typical temperature control loads are air conditioning loads and electric water heaters. When working normally, the temperature control load has two states: operating mode and standby mode, and the two operating powers alternate. As Figure 5 shown, taking the cooling mode as an example, when the room temperature is lower than the set maximum allowable temperature, by increasing the set temperature, the load operating power drops from P set1 to P set2 to provide adjustable capacity. When the temperature control load receives the control signal to cut the load at time t sig , its state has two cases: (a) cooling mode and (b) standby mode. When in the cooling mode, its operating power P se immediately becomes the standby power P std . Due to the hysteresis effect of the indoor temperature change, at time t re the room temperature reaches the set temperature, and the power of the temperature control load changes back to the new operating power P set2 . In this case, the air conditioner is equivalent to executing the response plan almost immediately. When in the standby mode at time t sig , the temperature control load maintains the standby power P std . After the room temperature rises, it restarts at time t re . If the response plan is not executed, it will enter the cooling mode at the original set temperature at time t o . At this time, the equivalent action time t ref of the temperature control load can be expressed as:

[0104] t ref = t o -t sig (8)

[0105] 3) Interruptible load

[0106] The types of interruptible loads are diverse and relatively scattered, including lighting equipment, kitchen equipment, washing machines, etc. Such loads are allowed to have conditional power outages under the incentive mechanism, and the loads during the shifted operation period are also included in this category. Its control strategy adopts switch control, with a fast response speed, usually in the millisecond to second level. However, due to the large differences in types, adjustment accuracy, and communication environments among different loads, the adjustable capacity of the aggregation group is a function that increases with time.

[0107] 4) Industrial load

[0108] Industrial loads account for the largest proportion of adjustable capacity in demand response. Due to the high precision of their signal receiving and control devices, the stability during the response process is also the highest. Some industrial loads can immediately respond to the response signal and have the function of continuous adjustment, such as electrolytic aluminum, the steel industry, and pumping stations. Their ramp rates after response are small, and the action times range from 30 seconds to several minutes; another part of industrial loads need to maintain on or off for a long time, including hydraulic presses, forging presses, grinding machines, etc. Such loads are not suitable for participating in system frequency modulation, so they are not considered in this invention.

[0109] In summary, the approximate action time models of various devices are as Figure 6 shown. The figure describes the relationship between the action time and the adjustable capacity. The adjustable capacity of the load increases with time after the action time of the load and finally reaches its target adjustable capacity. Since the target adjustable capacity, action time, and ramp rate of different types of loads are different in practice, the starting points, steepness, and highest points of each curve in the figure are also different. The specific values should be obtained through industry research. This figure is only a schematic diagram. For the convenience of calculation, this invention represents the action time of the device as t a .

[0110] The communication reliability is modeled as follows:

[0111] The reliability of the communication system represents the probability of error-free transmission in communication. The reliability S mn between nodes m and n can be expressed as:

[0112]

[0113] M In the formula: N mn represents the number of links between m and n, and S(e l ) represents link e lReliability. Faults in information components, packet loss during transmission, and bit errors can all affect the reliability of the link. Since there are usually more than one communication link from the BS to the cloud node and between cloud nodes, when a node or link fails, the continuity will not be affected by the switchover to the backup path, but additional switchover latency will be generated. When all information transmission paths between nodes are interrupted, information transmission interruption will occur. When packets are lost during transmission, due to the influence of the retransmission mechanism, the sending end will retransmit the packets after a period of time, and this process will generate retransmission latency t s . In terms of transmission bit errors, since the probability of bit errors occurring in the physical information system is extremely low, generally not greater than 10 5 , therefore, the errors in information transmission caused by bit errors can be ignored. In addition, as the number of routing hops increases, the routing reliability will decrease, so an upper limit on the number of routing hops is set as a constraint condition.

[0114] (3) According to the cloud-edge collaborative network system framework and the models of communication latency, device action time, and communication reliability, a virtual power plant maximum adjustable potential evaluation model based on timedelay step search is established;

[0115] Different from centralized scheduling, the cloud-edge information system reduces the latency generated by uploading and downloading power data by adding edge nodes near the user side, scheduling user terminals within the management scope of the edge nodes, and interacting with the cloud server. In order to accurately quantify the demand response scheduling latency under the cloud-edge communication system, the power grid demand-side management platform needs to evaluate the scheduling potential of load resources affected by the communication system in real time, and present the relationship curve between the response time and the adjustable capacity, providing support for the precise scheduling of demand response services. When the power grid is power unbalanced, the operator can schedule the virtual power plant according to precise latency requirements.

[0116] In order to determine the total latency t d during the data transmission process, the primary task is to determine the data transmission path, and the path selection depends on the magnitudes of the latencies of different paths. The present invention uses the KSP algorithm to iteratively obtain the optimal path and backup path from the source node to the destination node according to the latency parameters of each edge in the network. The KSP algorithm is a form of path search algorithm that uses the deviation path method to obtain the set of shortest paths between two nodes in the cloud-edge collaborative information network, so that the optimal path that meets the requirements can be selected. The present invention designs a KSP algorithm aiming at minimizing data transmission deviation, considering the latency, packet loss, and bit errors in information nodes and transmission links, using the KSP algorithm to iteratively obtain the optimal path and backup path from the source node to the destination node according to the latency parameters of each edge in the network, and at the same time, the latency t s generated by packet loss and bit errors is also considered as part of the transmission deviation. The transmission deviation link weight between nodes m and n is defined as follows: the latency t generated by packet losss is also considered as part of the total time delay. The link weight between nodes m and n is defined as follows:

[0117]

[0118] where: c l represents the packet loss coefficient, which is determined by network traffic, and t mn is the time delay between m and n. represents rounding up. d is a random number generated between 0 and 1.

[0119] In the KSP calculation process, the network topology matrix is first input. After receiving a request containing the source node m and the destination node n, the KSP algorithm searches for k shortest paths according to the link weights and searches for backup paths after removing the corresponding nodes.

[0120] When the KSP algorithm iterates out the optimal path and the backup path from the source node to the destination node according to the time delay parameters of each edge of the network, the data transmission path is determined accordingly, that is, the path with the lowest node occupancy rate in the shortest path iterated by the KSP algorithm is preferentially selected as the data transmission path, and if there is a transmission failure, it is switched to the backup path. When the transmission path is determined, the transmission time delay between two nodes during data transmission can be determined, that is, the total transmission time delay t d between the source node m and the destination node n can be expressed as:

[0121]

[0122] To evaluate the adjustable capacity of the response terminal and the scheduling time delay in each period, the optimization objective function is the total adjustable capacity F of the central cloud virtual power plant under the allowed time delay, and the optimization variables are the path selection results x i and x ij in the load scheduling communication link between the edge node and the BS node. In each evaluation, the maximum allowed response time delay t allow increases with the same step size to obtain the total maximum adjustable capacity under the discontinuous time delay sequence. The optimization process is shown as follows:

[0123]

[0124] where: F jpj represents the maximum adjustable capacity of the p j th demand response terminal within the coverage of the jth BS. is the upper limit of the routing hop count, which is a constraint condition set according to the communication reliability described above.

[0125] (4) Based on the cloud-edge collaborative network system framework in step (1), the models of communication delay, device action time, and communication reliability in step (2), and the virtual power plant maximum adjustable potential evaluation model based on timedelay step search in step (3), an evaluation method for the maximum adjustable potential based on the proximal policy optimization (PPO) algorithm is constructed to solve the total maximum adjustable capacity of the model in step (3);

[0126] Due to the complexity of the communication environment in virtual power plants and large-scale links, the present invention proposes an adjustable potential evaluation method based on the proximal policy optimization algorithm on the basis of the virtual power plant adjustable potential evaluation model with a fixed step size to solve the total maximum adjustable capacity of the virtual power plant and realize the real-time evaluation of the adjustable potential of the virtual power plant. As a deep reinforcement learning method, PPO can be regarded as a system in which an agent interacts with the environment. The adjustable potential evaluation problem can be described as the agent finding a set of actions in a determined maximum allowable response delay environment to maximize the total adjustable load, that is, the reward value obtained by the agent is the largest at this time. Through continuous interaction with the environment during the training process, the agent finally reaches an optimal policy π in each environment θ to maximize the current reward.

[0127] The virtual power plant evaluation process is as Figure 7 shown. The evaluation process is mainly divided into a training process and a testing process. The training process uses the terminal historical adjustable capacity and network status after edge cloud computing as the training set to train the neural network parameters to maximize the total adjustable capacity of the flexible load at each t allow and save the output neural network parameters. This process usually takes a long time, and a rolling training method is used to continuously update the historical data set. The testing process is the real-time evaluation process, and its input test set is the current terminal adjustable capacity and historical status, which can instantaneously evaluate the maximum adjustable capacity in the current environment.

[0128] Specifically, the adjustable potential evaluation and optimization process can be regarded as a Markov decision process. In an interaction between the agent and the environment, at time t, the environment is in state s t , and according to the current policy π, the agent executes action a t and interacts with the environment to obtain reward r t , which represents the quality of the current action. The environment moves to the next state s t+1 and repeats the above process. The key elements of the Markov decision process in the present invention are defined as follows.

[0129] 1) State space:

[0130]

[0131] The state space represents the environmental state of the system, which includes the available response volume of demand response terminals within the coverage of regional distribution substations, the maximum allowable response delay, and the distribution network status uploaded by the collection module.

[0132] 2) Action space:

[0133] A t ={x i (t), x ij (t)|i = 1:I, j = 1:J} (14)

[0134] The action space represents the variables in the optimization process, including the edge node selection x i and the BS node selection x ij , which is a discrete action space. This action space is the result of inputting the evaluation system state space into the policy function learned by the agent.

[0135] 3) Reward:

[0136]

[0137] The total adjustable capacity F is used as the reward value obtained by the agent in one learning, and the average value of Q - times Monte Carlo simulation is sampled as the output to mitigate the fluctuations in the communication environment.

[0138] During the training process, the cloud node uses historical data within a certain sliding time window to train the model and obtains the cloud optimization decision model. During the real - time decision - making process, the cloud node inputs the real - time virtual power plant state space parameters into the model, uses the trained decision model to obtain real - time optimization variables, and sends the results to the edge nodes. The edge node piece - wise linearizes the objective function, transforms the problem into a mixed - integer linear programming problem, and then uses a solver to solve it.

[0139] To verify the effectiveness and rationality of the virtual power plant adjustable potential comprehensive evaluation method considering both communication and action delays based on the cloud - edge collaborative architecture proposed in the present invention, the present invention uses the IEEE33 - node distribution network system as shown in Figure 8 for demand response evaluation, and the voltage level is 12.66 kV. One cloud node, three edge nodes, and 33 distribution substations including BS are set up. The nodes in the power backbone network are connected through routers. Each BS covers 100 user demand response terminals. The BSs are located in three different regions. Among them, the node numbers 0 - 11 are residential areas, including interruptible loads, temperature - controlled loads, and electric vehicles; the node numbers 12 - 24 are commercial areas, mainly with temperature - controlled loads; the node numbers 25 - 32 are industrial areas, mainly with industrial loads.

[0140] The typical daily adjustable potential curves of the four types of loads at the demand response terminals are as shown in Figure 9As shown in the figure. The theoretical maximum adjustable capacity and the network state parameter matrix calculated from the loads of each response terminal in a certain place within ten days are selected as the training set of the algorithm. For the evaluation of the adjustable potential of the virtual power plant, we set five scenario modes. Mode 1 is the situation considering the communication system delay, equipment action time, and communication reliability. Mode 2 is the situation only considering the equipment action time. Mode 3 is the situation where the whole system is residential areas. Mode 4 is the situation where the whole system is commercial areas. Mode 5 is the situation where the whole system is industrial areas. The relevant simulation parameter settings are shown in the following table.

[0141]

[0142] Optimization results:

[0143] In this invention, the PPO algorithm and the DDPG algorithm are used for example simulation, and the comparison of the reward curves in the training process is as Figure 10 shown. The reward value in the PPO training process gradually increases and tends to be stable, and converges at about 11,000 steps, indicating that the agent has found a strategy to maximize the total adjustable capacity of the load during the training process. As a traditional DRL algorithm, the DDPG algorithm converges at about 12,000 steps, with large fluctuations in the convergence process, and the reward convergence performance in the training process is weaker than that of PPO. This shows that in the simulation environment set in this invention, the stability and convergence of PPO are better than those of traditional DRL algorithms. Using the adjustable capacity of the response terminal on the eleventh day as the input of the test set, the RMSE value of the deviation between the online evaluation result of the anti-normalized PPO and the true value is 0.129, and the evaluation result is relatively accurate.

[0144] The evaluation results of the daily adjustable potential of the central cloud virtual power plant considering time delay are as Figure 11As shown, the total adjustable load in the area within the maximum allowable response delay range of [0, 500 ms] with a step of 5 ms at any moment of a day is shown. In Mode 1, the communication system delay, device action time, and communication reliability are considered. In Mode 2, only the device action time is considered. The total adjustable capacity of Mode 1 is 20.3% less than that of Mode 2. This is because Mode 2 lacks the part of the communication delay in the allowable response delay, resulting in an error where the entire allowable response delay is provided by the device action delay. When the allowable response delay is small, due to the existence of communication delay, in Mode 1, only the terminals close to the server node can be scheduled, and the number of terminals that can respond is small, with little adjustable potential, and it is in the slow growth period. Mode 2 has the potential to schedule most terminals in advance, mainly including temperature control loads and interruptible loads with fast action speeds, resulting in the largest error. As the allowable response delay increases, more demand response terminals can be added. After 100 ms, due to the existence of electric vehicle loads, their high ramp rate leads to a large increase in the adjustable capacity, and the error of Mode 2 is also relatively large. After 200 ms, the growth of the adjustable capacity is mainly provided by temperature control industrial loads and industrial loads, which is determined by their long ramp-up time. It can be predicted that several seconds after the dispatch instruction is issued, the adjustable capacity can still increase due to the growth of the response capacity of industrial loads.

[0145] As Figure 12 Shown is the proportion of each part in the communication delay of each BS node at 12:00 with an allowable response delay of 100 ms. At this time, all response terminals participate in the scheduling. The wired delay is related to the distance between the BS node and the edge node. The closer the node is, the smaller the proportion of the wired delay in the total communication delay, and the greater its schedulable action time. Since all terminals under each BS node are scheduled, the wireless delay and the computing delay approach equality.

[0146] Modes 3, 4, and 5 are respectively set to the cases where the node distribution network system is all residential areas, commercial areas, and industrial areas. Figure 13 It shows the adjustable potential of the virtual power plant in each mode at 12:00. The shaded part represents the range of the adjustable potential under Monte Carlo sampling with a packet loss rate of 1% and a lightly loaded access network. The volatility of the adjustable potential is slightly higher when the delay is low. By comparing each mode, it can be obtained that the adjustable load potential in general within 300 ms of the allowable response delay is 23.4% and 12.7% higher than that of Modes 3 and 5 respectively, and 22.3% lower than that of Mode 4.

[0147] Sensitivity analysis:

[0148] To analyze the deviation of the load adjustable potential evaluation caused by the loss of data packets during the transmission process, Figure 14 shows the allowable response delay t allowThe variation of the maximum adjustable potential of the load with the packet loss rate at 50 ms, 100 ms, 150 ms, and 200 ms respectively. The shaded part is the distribution range of the adjustable potential after 100 Monte Carlo samplings, and the sample average is represented by a line. t allow When t allow = 50 ms, the fluctuation range of the sampling results is relatively large. This is because at this time, the adjustable potential mainly depends on the communication delay. The increase in the packet loss rate causes fluctuations in the communication delay. When there are fewer schedulable terminals, even a small fluctuation will increase the evaluation deviation. As the allowed response delay increases, the proportion of the communication delay in the total allowed delay decreases, and the evaluation fluctuations caused by it also gradually decrease. Due to the existence of the retransmission delay, when the packet loss rate increases, the adjustable capacity shows a slow decreasing trend.

[0149] As Figure 15 shown in the variation relationship between the wireless transmission delay in the access network and the load rate, the Monte Carlo sampling results are represented by scatter points. The change in the access network load rate has little impact on the overall delay. When the load rate increases, an extreme scenario of doubled delay may occur, affecting the accuracy of the evaluation results.

[0150] In summary, the comprehensive evaluation method of the adjustable potential of the virtual power plant considering both communication and action delays based on the cloud-edge collaboration architecture proposed by the present invention can accurately consider the influence of the communication system and resource action time, thereby comprehensively evaluating the adjustable potential of the virtual power plant. It can provide a decision-making basis for the virtual power plant to participate in primary and secondary frequency regulation and peak shaving of the power system, and is beneficial to supporting the frequency stability and power balance of the power system.

[0151] The computer device of the present invention includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the above method are implemented.

[0152] The computer-readable storage medium of the present invention stores a computer program thereon. When the computer program is executed by a processor, the steps of the above method are implemented.

Claims

1. A comprehensive evaluation method for the adjustable potential of a virtual power plant that takes into account both communication and action time delays, characterized in that It includes the following steps: According to the virtual power plant cloud-edge collaborative network system framework, construct a communication system delay model, a device action time model, and a communication reliability model respectively; According to the virtual power plant cloud-edge collaborative network system framework, communication system delay model, device action time model, and communication reliability model, the KSP algorithm is used to obtain the optimal path and backup path between the source node m and the destination node n, and the total transmission delay t between the source node m and the destination node n during the data transmission process is determined. d A virtual power plant maximum adjustable potential evaluation model based on timedelay step search is constructed. Through the PPO algorithm, optimize the virtual power plant maximum adjustable potential evaluation model to obtain the online evaluation of the virtual power plant adjustable potential; The communication system delay model is as follows: Wired transmission delay between the central cloud and edge nodes is Among them, x i As a binary variable, x i = 1 represents the existence of data packet transmission between the cloud node and the edge node i; otherwise x i = 0; L i represents the size of the downlink data packet of the i-th node; r i is the wired transmission rate between the central cloud and the edge node; b i represents the channel bandwidth between cloud nodes, P i represents the data packet transmission power, G i represents the channel gain, and G i = d i -α where d i represents the distance between the i-th node and the central cloud node, -α is the signal fading factor, and s represents the transmission noise; Wired propagation delay between cloud node and edge node is Among them, v is the propagation speed of the optical signal in the optical fiber; The downlink data latency t between the cloud edge and the edge node i is Among them, N h represents the number of routing hops, and t f represents the processing delay of a single device; The calculation delay of the j-th edge node For where, f j is the computing power of the j-th edge node, and e j is the number of CPU cycles required for the computing task; Wireless download delay t p is Among them, represents the queue position where user p is located under the j-th regional node j and T is the smallest indivisible period in wireless transmission; Wireless time delay t p It is described by the Pareto distribution, and the probability distribution function P is Among them, τ0 represents the minimum value of the time delay between nodes, γ is a parameter related to the load rate, and the computing delay of the cloud server is a constant, and t p is the wireless time delay generated during the process of the base station transmitting to the demand response terminal; The communication reliability model is as follows: The reliability S between nodes m and n mn is Among them, N mn represents the number of links between m and n; S(e l ) represents the reliability of link e l . The failures of information components, packet loss in transmission, and error codes will all affect the reliability of the link; the reliability S mn constraint on the upper limit of the routing hop count is set to Use the KSP algorithm to obtain the optimal path and the backup path between the source node m and the destination node n. The deviation link weight of the transmission path between m and n is as follows: Among them, W mn is the link weight between nodes m and n, and t mn is the time delay between nodes m and n, and c l represents the packet loss coefficient, represents rounding up, and t s is the delay caused by packet loss and bit error, set as a constant, and d is a random number generated between 0 and 1; The total transmission delay t between the source node m and the destination node n during the data transmission d is Among them, t ij is the wired delay generated by the scheduling instruction sent by the edge node passing through the wireless base station, t a is the action time of the device, t s is the delay caused by packet loss and error code; The virtual power plant maximum adjustable potential evaluation model based on timedelay step search is Among them, F is the total adjustable capacity of the central cloud virtual power plant, represents the maximum adjustable capacity of the p j th demand response terminal within the coverage area of the jth wireless base station, t allow is the maximum allowable response delay, x ij is the path selection result between the edge node and the BS node in the load scheduling communication link.

2. The comprehensive evaluation method for the adjustable potential of a virtual power plant that takes into account both communication and action time delays according to claim 1, wherein The virtual power plant cloud-edge collaborative network system framework includes a cloud layer, an edge layer, and a response terminal; the cloud layer is used to centrally integrate the computing power resources of many servers, and is responsible for processing global data services. The data services include comprehensive evaluation, predicting the aggregated adjustable capacity of the virtual power plant, decomposing the scheduling instructions to each edge node, and the overall communication optimization control of the cloud-edge-terminal system; the edge layer is used to calculate the aggregated load resource model at the edge node, and the edge nodes of the edge layer are set at positions close to the user side.

3. A system for the comprehensive evaluation method of the adjustable potential of a virtual power plant that takes into account both communication and action time delays according to any one of claims 1-2, characterized in that, It includes A model construction module is used to respectively construct a communication system delay model, a device action time model, and a communication reliability model according to the virtual power plant cloud-edge collaborative network system framework; and is used to obtain the optimal path and standby path between the source node m and the destination node n by using the KSP algorithm according to the virtual power plant cloud-edge collaborative network system framework, the communication system delay model, the device action time model, and the communication reliability model, and determine the total transmission delay t between the source node m and the destination node n during the data transmission process d , and construct a virtual power plant maximum adjustable potential evaluation model based on timedelay step search; A policy optimization evaluation module, which is used to determine the optimal policy for the virtual power plant maximum adjustable potential evaluation model through the PPO algorithm to obtain the online evaluation of the virtual power plant adjustable potential.

4. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 2.

Images