Power quality adjustment resource capability dynamic evaluation method and system based on adaptive federated learning and voltage out-of-limit, electronic equipment and computer storage medium

Abstract

The invention relates to a power quality adjustment resource capability dynamic evaluation method and system based on adaptive federated learning and voltage out-of-limit, electronic equipment and a computer storage medium, and the method comprises the steps: building a power quality adjustment capability evaluation model of adjustable resources, and monitoring the voltage and current changes of the adjustable equipment in real time, providing an evaluation index of the adjustable resource electric energy quality adjustment capability according to the active power output, the reactive power output, the voltage and the current change condition of the adjustable resource, and evaluating the active / reactive power adjustment capability of the adjustable resource according to the evaluation index; the original data, the curve and the evaluation result are analyzed and trained by using adaptive federated learning and a convolutional neural network, and the trained model can dynamically and quickly analyze the active and reactive adjustment performance of the adjustable resources. According to the invention, the problem of active and reactive power regulation operation capability evaluation of the power grid adjustable device is solved, and auxiliary tools and convenient means are provided for power grid dispatching operation.

Description

Technical Field

[0001] The present invention belongs to the technical field of power quality regulation capability evaluation of adjustable resources in a distribution system, and specifically relates to a method, system, electronic device and computer storage medium for dynamic evaluation of power quality regulation resource capability based on adaptive federated learning and voltage over-limit. Background Art

[0002] With the continuous increase in the penetration rate of renewable energy and the continuous expansion of the electric vehicle industry, the operation mode and dynamic behavior of the distribution network have become increasingly complex after the access of a large number of distributed energy sources. The rapid growth of renewable energy installed capacity and the centralized distribution of renewable energy sites have had a great impact on the operation of the power grid and power consumption.

[0003] The increase in the proportion of renewable energy in the energy structure has shifted the peak-shaving and frequency-regulating pressure of the power grid to renewable energy sources, which has put forward higher requirements on the active and reactive power regulation capabilities of renewable energy sites. The anti-peaking characteristics of renewable energy such as wind power and the uncertainty of photovoltaic output make it impossible for them to generate electricity freely according to human will in the future, which poses new and greater challenges to the regulation capabilities of adjustable resources. Most of the current research content focuses on the power prediction, absorption capacity assessment, coordinated control strategy control mode and other aspects of adjustable energy, but there are not many evaluations and analyses of the actual active and reactive power regulation capabilities of adjustable resources, which makes it impossible to truly evaluate whether they can meet the peak-shaving and frequency regulation requirements under the new situation of the power grid.

[0004] Therefore, a dynamic evaluation method for the power quality regulation capability of single adjustable resources based on voltage over-limit is urgently needed. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method, system, electronic device and computer storage medium for dynamically evaluating the power quality regulation resource capacity based on adaptive federated learning and voltage over-limit.

[0006] The present invention solves the technical problem by the following technical solutions:

[0007] A method for dynamically evaluating power quality regulation resource capabilities based on adaptive federated learning and voltage over-limit, the method comprising the following steps:

[0008] S1. Based on the operation mode of the regional power grid, the voltage characteristics of the power grid are simulated and analyzed to obtain the active data of each node parameter and adjustable resources under the stable operation state, as well as the system node data and the active power data of the adjustable resources under the extreme voltage. Similarly, the reactive data of each node parameter and adjustable resources under the stable operation state, as well as the system node data and the reactive power data of the adjustable resources under the extreme voltage are obtained;

[0009] S2. Propose evaluation indicators for the power quality regulation capability of single adjustable resources, and evaluate the adjustable equipment based on the simulation data results, including: regulation direction, regulation amplitude, regulation response time, regulation duration, regulation ramp rate, etc., and analyze the influencing factors of the adjustable resource regulation capability;

[0010] According to the system node data obtained under the extreme voltage state, the original system data is stored as a matrix and a 28x28 grayscale image as a data set. The power quality regulation evaluation index of the single adjustable resource is used as the data set label. The original data and evaluation results of the active / reactive regulation capability of the adjustable resource are converted into an adjustable resource power quality regulation capability evaluation data set that is suitable for adaptive federated learning.

[0011] S3. Based on the traditional federated average learning method, CNN neural network and power system distribution network structure, a three-layer model of central server, regional server and adjustable equipment is constructed, and an adjustable resource power quality regulation capability evaluation scheme based on improved federated learning is proposed. It is trained according to the adjustable resource power quality regulation data set to obtain a quickly implemented adjustable resource power quality regulation capability evaluation network model.

[0012] Moreover, in order to realize the simulation analysis of the grid voltage characteristics based on the regional grid operation mode, the S1 first constructs a mathematical model of a commonly used active / reactive compensation device:

[0013] (1) Synchronous generator

[0014] If the load is powered by multi-stage transformer, the generator alone cannot meet the load's voltage regulation requirements. Since the voltage of the load close to the synchronous generator cannot be adjusted too high, the voltage adjustment of the load far away must rely on other voltage regulation methods to solve the problem comprehensively. The two-order model of the traditional synchronous generator is simulated, and the rotor motion equation is:

[0015]

[0016] Where: J is the moment of inertia, D is the damping coefficient, T m , T e and T d are the mechanical torque, electromagnetic torque and damping torque of the synchronous generator respectively, ω0 is the synchronous angular velocity of the power grid, and δ is the power angle of the synchronous generator;

[0017] The relationship between the electromagnetic torque and the electromagnetic power output by the virtual synchronous generator is:

[0018]

[0019] Where: e a 、e b 、ec and i a 、i b 、i c is the output potential and output current of VSG;

[0020] The relationship between the synchronous generator output reference voltage is:

[0021]

[0022] (2) Static VAR Compensator

[0023] The dynamic equation of the static VAR compensator SVC model is:

[0024]

[0025] Where: V r is the voltage amplitude of the SVC node; V ref is the voltage reference value; K D is the feedback coefficient; G is the gain; T is the time constant; B SVC0 , ΔB SCV They are the steady-state value and variation of the SVC equivalent fundamental susceptance respectively;

[0026] (3) Photovoltaic inverter

[0027] The reactive power regulation capability of the static VAR generator SVG photovoltaic unit lies in the photovoltaic inverter therein. The photovoltaic inverter is a power electronic component, and its theoretical reactive power output limit is:

[0028]

[0029] Where: S pυ is the apparent power of the photovoltaic unit; Q pυ It is the reactive power output of the photovoltaic unit.

[0030] Moreover, based on S1, the response speed and stability of the active power compensation device are simulated. By changing its current or voltage input, the changes in the voltage amplitude of the power system nodes are observed, and the indicators such as the adjustment response time are recorded to evaluate its reactive power adjustment capability; the active power adjustment range of the active power compensation device is simulated, and the changes in active power are observed by changing its working state, and the adjustment accuracy of the active power compensation device is simulated. By changing its output power, the gap between the actual output power and the expected output power is observed; similarly, the reactive power data of each node parameter and adjustable resource under the stable operation state is recorded, and the system node data and the reactive power data of the adjustable resources are collected under the extreme voltage condition, and the response speed and stability of the reactive power compensation device are simulated.

[0031] Moreover, the S3 performs federated learning training on the adjustable resource power quality regulation capability dataset as follows:

[0032] Based on the data set, the evaluation model is obtained through adaptive federated learning training. Federated learning is aimed at decentralized multi-user scenarios. Each user, i.e., the client, has his own data set {D1, ..., D N}, the federated learning method is to train a model M by participating users FED , while user D i Data is kept locally and not transferred to a central server;

[0033] A federated learning system usually consists of a data holder and a central server. The data holder performs data training locally. The gradients generated by iterations are desensitized and uploaded to the server as interactive information instead of uploading local data. The server returns aggregated parameters for updating the model.

[0034] Consider an environment with a central server and N data holders. The goal of federated learning is to solve the model M using the local data of the data holders. FED The risk minimization problem is:

[0035]

[0036] Where: M FED For the server model, M FEDi is the local model held by the i-th data holder;

[0037] When the system starts the tth round of aggregation, the server selects the data holder client to be aggregated, and the server sets the server model M FED (t-1) is transmitted to each selected data holder client, and then the client uses the received server model M FED (t-1) Perform local training;

[0038] After training is completed, client i will use the locally updated model M FEDi (t-1) is uploaded to the server. The server aggregates the models uploaded by the client according to the following formula and uses a certain learning rate μ to aggregate the global model M FED (t) To update:

[0039]

[0040] After the training is completed, the network model of power quality regulation of adjustable resources is obtained, which can quickly evaluate and estimate the power quality regulation capability of adjustable resources without complex iterative calculations. While protecting privacy, it can realize dynamic estimation of the power quality regulation capability of adjustable resources in the power grid.

[0041] A system for dynamically evaluating power quality regulation resource capabilities based on adaptive federated learning and voltage over-limit, characterized in that: based on the evaluation method, it includes:

[0042] The power grid simulation module simulates and analyzes the voltage characteristics of the power grid based on the operation mode of the regional power grid, obtains the active output of each node parameter and adjustable resources under the stable operation state, and the system node data and the active output of adjustable resources under the extreme voltage. Similarly, the reactive output of each node parameter and adjustable resources under the stable operation state, and the system node data and the reactive output of adjustable resources under the extreme voltage.

[0043] The evaluation module proposes evaluation indicators for the power quality regulation capability of single adjustable resources, and evaluates the adjustable equipment based on the above calculation results, including: regulation direction, regulation amplitude, regulation response time, regulation duration, regulation ramp rate, and analyzes the influencing factors of the power quality regulation capability of adjustable resources;

[0044] A conversion module is configured to convert load values ​​and active / reactive regulation capability data obtained by distribution network simulation into an adjustable resource power quality regulation capability data set adapted to adaptive federated learning;

[0045] The adaptive federated learning module is configured to construct an adaptive federated learning model based on federated learning, CNN neural network and adaptive method, and train it according to the adjustable resource power quality regulation capability data set to obtain an evaluation network model for quickly realizing the adjustable resource power quality regulation capability.

[0046] An electronic device, characterized in that it includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the evaluation method.

[0047] A computer-readable storage medium stores a computer program, wherein the program implements the evaluation method when executed by a processor.

[0048] The advantages and beneficial effects of the present invention are:

[0049] The present invention can realize the rapid and accurate evaluation of the power quality regulation capability of the adjustable resources in the power distribution system by means of the neural network model, which helps to make full use of the power quality regulation capability of the adjustable resources, improve the utilization rate of the adjustable resources, and reduce the investment in reactive power compensation devices, thereby improving the economic efficiency of the power distribution system. The use of the adaptive federated learning algorithm to evaluate the power quality regulation capability helps to improve the ability to control the power quality regulation capability of the adjustable resources in the power distribution system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of the present invention;

[0051] Figure 2 It is a topological diagram of a 123-node distribution network node system of the present invention;

[0052] Figure 3 is a structural block diagram of the evaluation system of the present invention;

[0053] Figure 4 It is a structural schematic diagram of the electronic device of the present invention;

[0054] Figure 5 is a probability density curve diagram of voltage distribution of the present invention;

[0055] Figure 6 This is a comparison result diagram of the evaluation value and theoretical value of the adjustable active power capability of the photovoltaic access A phase of the present invention. DETAILED DESCRIPTION

[0056] The present invention is further described in detail below through specific examples. The following examples are only illustrative and not restrictive, and the protection scope of the present invention cannot be limited thereto.

[0057] like Figure 1 As shown, a method for dynamically evaluating the power quality regulation resource capability based on adaptive federated learning and voltage over-limit is provided, wherein the steps of the method are as follows:

[0058] Step S101, according to the operating characteristics of the actual three-phase unbalanced distribution network, a three-phase unbalanced distribution system model is established, and the power quality regulation capability of the adjustable resources is simulated and tested under different loads. For the three-phase unbalanced distribution system, the active power and reactive power of the power system nodes are expressed as:

[0059]

[0060] Among them: Phase sequence Node i, j∈{1, 2, ..., N}, M is the number of nodes containing measurement elements in the three-phase unbalanced power distribution system; and are the voltage amplitude and voltage phase angle at node i respectively; The phase angle difference between the phase and the j-node ψ phase is The internode conductance is The inter-node susceptance is and i-node The injected active power and reactive power of the phase; and are the active power and reactive power of the branches associated with node i and node j respectively.

[0061] Since the output of renewable energy is random, combined with the output scenario and power flow analysis of renewable energy, it can be seen that the node voltage of the distribution network containing a high proportion of renewable energy has a probabilistic distribution characteristic. Figure 5 The probability density curve of voltage distribution is schematically described, where region I is a qualified voltage region, region II is a low voltage region, region III is a high voltage region, and region IV is an extremely high voltage region. For regions II and III, voltage regulation can be achieved by controlling reactive power compensation of adjustable resources; and for extremely high voltage region IV, active power regulation of adjustable resources can be used as an additional means of regulation. The renewable regulating energy described in this patent mainly includes photovoltaic power stations and wind turbines.

[0062] (1) Wind power regulation capability

[0063] The adjustment range of the wind turbine reactive power is affected by its output active power P (or wind speed υ and terminal voltage U s ), so the upper and lower limits of the wind turbine’s reactive power adjustable capacity are:

[0064]

[0065] Where Q max and Q min are the upper and lower limits of the reactive power adjustable capacity of the doubly fed wind turbine, respectively. When determining the reactive power adjustable range of a wind farm, the influence of wind speed variability needs to be considered. Take the midpoint υ of the wind speed interval (i-1, i], i = 1, 2, ..., k i =i-0.5 as the wind speed in this interval. Therefore, the wind speed v i =i-0.5, the upper and lower limits of the reactive power adjustable capacity of the doubly-fed wind turbine are:

[0066]

[0067] in and are the upper and lower bounds of the reactive power adjustable capacity of the i-th wind turbine respectively.

[0068] When the natural wind speed is υ, the upper and lower limits of the reactive power adjustable capacity of the wind farm are:

[0069]

[0070] in and are the upper and lower bounds of the reactive power regulation capability of the wind farm, respectively. i is the loss coefficient, p i is the proportion of wind turbines in the wind speed interval (i-1, i] under the natural wind speed V0. Therefore, we only need to know the proportion of wind turbines in each wind speed interval (i-1, i] under a certain natural wind speed V0 to the total number of wind turbines in the wind farm. i The total reactive power adjustable capacity range of the wind turbines in each wind speed interval (i-1, i] is calculated based on the reactive power adjustable capacity range of a single wind turbine.

[0071] (2) PV inverter regulation capability

[0072] The regulation capability is analyzed according to the main constraints of the inverter operation, where S N , U N , I N are the rated power, rated voltage and rated current of the inverter respectively; L is the filter reactance; U dc is the DC bus voltage; U pcc is the access point voltage; f is the switching frequency. Assume that the active power injected by the front end of the inverter is P, and the reactive power injected by the inverter into the system is Q. Under the voltage-oriented control of the grid-connected point, the active / reactive power injected by the inverter into the system through the grid-connected point can be expressed as:

[0073]

[0074] The maximum current constraints are:

[0075]

[0076] Combining the above formula with the following formula gives:

[0077]

[0078] The maximum reversible voltage constraints of the inverter are:

[0079]

[0080] Combining the above formulas, we can get:

[0081]

[0082] Set the lower limit of the inverter reactive power adjustable capability to Q min , the upper limit of reactive power adjustable capacity is Q max According to the above formula, the upper and lower limits of reactive power adjustable capacity are:

[0083]

[0084] Voltage U at the access point of the PV inverter pcc Under rated conditions, U N ,In actual scenarios, the voltage at the access point will change with the ,system operating conditions, and generally fluctuates within the rated voltage of 0.9pu to 1.1pu.

[0085] Photovoltaic and wind turbines are connected to a three-phase unbalanced power system and simulated under different load conditions to test the active / reactive regulation range of adjustable resources without exceeding the voltage limit of the power system nodes.

[0086] Step S102, proposes evaluation indicators for the power quality regulation capability of a single adjustable resource, and evaluates the adjustable device according to the simulation data results, specifically including: regulation direction, regulation amplitude, regulation response time, regulation duration, regulation ramp rate, and analyzes the influencing factors of the adjustable resource regulation capability.

[0087] Step S103, adaptive federated learning training is performed on the generated power quality regulation resource capability data set. Based on the data set, adaptive federated learning training is used to obtain a regulation capability evaluation model, so as to achieve rapid evaluation of the power quality regulation capability of the adjustable equipment in the distribution network and solve the problem of low resource utilization of the distribution network.

[0088] Based on the distribution network structure of the power system, in order to achieve adaptive functions, on the basis of the traditional two-layer model of federated learning, the distribution network for distributed renewable energy access is divided into three levels: central server, sub-regional server and measurement equipment. For the central cloud server, there are model optimization problems as follows:

[0089]

[0090] Therefore, the allocation method can be used to solve the optimization problem. Since the sub-datasets and the equipment models on both sides are allocated one-to-one, the above formula can be written as:

[0091]

[0092] in

[0093] in, is the decision variable. From the above formula, we can see that each subset must be assigned to each device one-to-one. If we write the above formula in linear programming form, we have:

[0094]

[0095] Where Ap = b

[0096] in, is the loss vector is the decision variable, is the constraint matrix, is a vector of 1s.

[0097] Based on the traditional federated average algorithm, this paper proposes an improved adaptive federated learning method, which effectively improves the accuracy. This improved federated learning method is applied to the problem of evaluating the intelligent connection and regulation capability of electric energy in power systems, and a dynamic evaluation method for the active and reactive regulation capability of adjustable resources based on adaptive federated learning is proposed.

[0098] Traditional federated learning algorithms usually need to transmit models frequently during the training process, and distributed nodes are often limited by resources such as communication and bandwidth, making it difficult to achieve frequent communication with the server. With the increasing complexity of the learning model structure, the amount of data for model updates has increased significantly, further exacerbating the communication overhead of federated learning and placing a serious burden on nodes and networks. The present invention proposes an adaptive federated learning method, which introduces an intermediate server layer based on the two-layer model of traditional federated learning in order to avoid large amounts of data transmission and reduce communication rounds. The distribution network is divided into three levels: central server, sub-regional server, and measurement equipment.

[0099] The adaptive federated learning method is different from the traditional federated learning. During the training process, the model of the measurement device is not directly transmitted to the central server, but first transmitted to the sub-region server. The sub-region server selects the best device model based on the performance of the model on the entire sub-region data set, and then transmits the best model to the central server. Compared with the traditional two-layer federated learning, the adaptive federated learning algorithm greatly reduces the number of communications and reduces the communication overhead.

[0100] In addition, the sub-regional servers only transmit the best model to the central server for aggregation, while traditional federated learning transmits all models (regardless of their effectiveness) to the central server. Therefore, compared with traditional federated learning, adaptive federated learning not only significantly reduces the communication burden, but also greatly improves the accuracy of the trained model. The specific training process is as follows. Before the initial global training begins, the central server initializes the models of all its sub-regional servers and measurement devices, and sends the initial model w 0 It is then sent to the sub-region server, which then sends w 0The data is then transmitted to each measurement device. In each round of global training t, the measurement device performs L local trainings according to the stochastic gradient descent algorithm, and the central server performs central aggregation based on the updated local model. In the first local training of each round of global aggregation, the sub-region server divides the sub-region data set into multiple sub-data sets equal to the number of measurement devices, and assigns the sub-data sets to the measurement devices one by one. Taking the tth global training as an example, the central server aggregates the model w t-1 Transmitted to the sub-region server and measurement equipment, the measurement equipment in the model w t-1 In the first local training, the sub-region server obtains the sub-dataset allocation solution α that minimizes the sum of the loss function according to the distributed simplex algorithm, and allocates the sub-datasets to the measurement devices one by one according to the solution α. ​​The measurement devices allocate the sub-datasets {x jk ,y jk} Local training using stochastic gradient descent algorithm:

[0101]

[0102] Where η is the learning rate, and are the local network models of the lth and l+1th times respectively, and the loss function f(·) is the root mean square error (RMSE):

[0103]

[0104] After L local trainings are completed, each measurement device accesses its own sub-region dataset and converts the device model Run it on the remaining sub-datasets to get the model The loss value f on each sub-dataset jk , calculate its entire sub-region data set D i The average loss value on

[0105]

[0106] Among them, N i is the number of measuring devices in sub-region i. The sub-region server calculates the number of measuring devices under it. Values ​​are selected in sub-area dataset D i The best performance is The smallest measuring device as the best device model Will As a sub-region server model:

[0107]

[0108] After local training is completed, the sub-region servers will use their respective best models Transmit to the central server for central aggregation:

[0109]

[0110] The aggregated model w t+1 The next round of global training is performed as the global model of the t+1th round. After the Tth round of global training, the final model w is obtained. T ,This model can achieve accurate and rapid evaluation of the power quality regulation capability of adjustable resources in a three-phase unbalanced system while protecting the privacy of sub-regional data.

[0111] In the method of the present invention, each level of the distribution network performs different stages of federated learning and undertakes different adaptive functions, namely monitoring, analysis, planning and execution. The sub-regional measurement equipment is monitored and controlled through local training and distributed simplex algorithm, and each sub-regional server is based on the average loss value. Analyze the on-device model and plan the next round of local training.

[0112] In summary, the method of this example proposes a dynamic evaluation method for power quality regulation capability based on adaptive federated learning and voltage over-limit. First, for the three-phase unbalanced distribution network system with new energy access, the active / reactive regulation capability of adjustable resources such as photovoltaic inverters and wind turbines at node voltages between 0.95-1.05 under different carrying capacities is simulated and tested; the adjustable equipment is evaluated according to the simulation data results, and the influencing factors of the power quality regulation capability of adjustable resources are analyzed; based on the above two steps, the power quality regulation capability dataset of adjustable resources is obtained, and an adaptive federated learning model is constructed based on federated learning, CNN neural network and adaptive framework, and the power quality regulation capability dataset of adjustable resources of the distribution system is substituted for training to obtain a network model that quickly realizes the power quality regulation capability of adjustable resources.

[0113] In a specific embodiment, the method for dynamic evaluation of active and reactive regulation capability based on adaptive federated learning and voltage over-limit of the present application is verified on the IEEE-123 node distribution network system. The topology diagram of the distribution system is as follows: Figure 2 As shown in Figure 1, the IEEE-123 node distribution network system has single-phase and two-phase loads, and therefore has significant load imbalance. The network has different voltage levels and four voltage regulators.

[0114] like Figure 3As shown, a dynamic evaluation system for power quality regulation resource capability based on adaptive federated learning and voltage over-limit is based on the evaluation method, including:

[0115] The power grid simulation module 210 performs power grid voltage characteristic simulation analysis based on the operation mode of the regional power grid, obtains the active output of each node parameter and adjustable resources under a stable operation state, and the system node data and the active output of adjustable resources under a voltage limit condition, and similarly obtains the reactive output of each node parameter and adjustable resources under a stable operation state, and the system node data and the reactive output of adjustable resources under a voltage limit condition;

[0116] The evaluation module 220 proposes an evaluation index for the power quality regulation capability of a single adjustable resource, and evaluates the adjustable device according to the above calculation results, specifically including: regulation direction, regulation amplitude, regulation response time, regulation duration, regulation ramp rate, and analyzes the influencing factors of the power quality regulation capability of the adjustable resource;

[0117] A conversion module is configured to convert load values ​​and active / reactive regulation capability data obtained by distribution network simulation into an adjustable resource power quality regulation capability data set adapted to adaptive federated learning;

[0118] The adaptive federated learning module 230 is configured to construct an adaptive federated learning model based on federated learning, CNN neural network and adaptive method, and train it according to the adjustable resource power quality regulation capability data set to obtain an evaluation network model that quickly realizes the adjustable resource power quality regulation capability.

[0119] like Figure 4 As shown, an electronic device is characterized in that it includes: at least one processor 310, and a memory 320 communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the evaluation method, for a three-phase unbalanced distribution network system with new energy access, a simulation test is performed on the active / reactive regulation capability of adjustable resources such as photovoltaic inverters and wind turbines at a node voltage between 0.95-1.05 under different carrying capacities; the adjustable equipment is evaluated according to the simulation data results, and the influencing factors of the power quality regulation capability of the adjustable resources are analyzed; based on the above two steps, an adjustable resource power quality regulation capability data set is obtained, and an adaptive federated learning model is constructed based on federated learning, CNN neural network and adaptive framework, and the distribution system adjustable resource power quality regulation capability data set is substituted into the training to obtain a network model that quickly realizes the power quality regulation capability of the adjustable resources.

[0120] The electronic device further includes: an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330 and the output device 340 may be connected via a bus or other means. Figure 4 In the example, the connection through the bus is taken as an example. The memory 320 is the above-mentioned computer-readable storage medium. The processor 310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 320, that is, the method for evaluating the power quality regulation capability of adjustable resources in the active distribution network of the above-mentioned method embodiment is implemented. The input device 330 can receive input digital or character information, and generate key signal input related to user settings and function control of the active distribution network adjustable resource power quality regulation capability evaluation system. The output device 340 may include display devices such as display screens.

[0121] A computer-readable storage medium having a computer program stored thereon, wherein the program implements the evaluation method when executed by a processor, and the computer executable instructions are set to:

[0122] For the three-phase unbalanced distribution network system with new energy access, simulation tests are conducted on the active / reactive regulation capabilities of adjustable resources such as photovoltaic inverters and wind turbines at node voltages between 0.95-1.05 under different carrying capacities;

[0123] Evaluate the adjustable equipment based on the simulation data results and analyze the factors affecting the power quality regulation capability of the adjustable resources;

[0124] Based on the above two steps, the data set of power quality regulation capability of adjustable resources is obtained. Based on federated learning, CNN neural network and adaptive framework, an adaptive federated learning model is constructed. The data set of power quality regulation capability of adjustable resources of the distribution system is substituted into the model for training, and a network model that can quickly realize the power quality regulation capability of adjustable resources is obtained.

[0125] The computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the adjustable resource power quality regulation and evaluation system of the active power distribution network, etc. In addition, the computer-readable storage medium may include a high-speed random access memory, and may also include a memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the adjustable resource power quality regulation capability system of the active power distribution network via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0126] like Figure 6 The comparison result of the photovoltaic access A phase photovoltaic active adjustable capacity evaluation value and theoretical value of the present invention is shown. The distributed photovoltaic access node 7 is connected to phase A, and 1000 groups of user loads varying within a certain range are randomly generated. The photovoltaic active output under the limit voltage is obtained by Newton-Raphson algorithm and dichotomy, and then the adaptive federated learning method is used for training to obtain the photovoltaic active output evaluation model under the limit voltage; Table 1 shows the evaluation results and evaluation effects of the present invention under 10 groups of sampling numbers (100, 200, 300, 400, 500, 600, 700, 800, 900, 1000); Table 2 shows the evaluation effect of 1000 groups of photovoltaic active adjustable capacity evaluation of the present invention.

[0127] Table 1 Evaluation results and evaluation effects of 10 groups of sampling numbers

[0128]

[0129]

[0130] Table 2 Evaluation indicators of 1000 sets of data

[0131]

[0132] Although the embodiments and drawings of the present invention are disclosed for illustrative purposes, those skilled in the art will appreciate that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A method for dynamically evaluating power quality regulation resource capabilities based on adaptive federated learning and voltage over-limit, characterized by: The steps of the method are: S1. Based on the operation mode of the regional power grid, the voltage characteristics of the power grid are simulated and analyzed to obtain the active data of each node parameter and adjustable resources under the stable operation state, as well as the system node data and the active power data of the adjustable resources under the extreme voltage. Similarly, the reactive data of each node parameter and adjustable resources under the stable operation state, as well as the system node data and the reactive power data of the adjustable resources under the extreme voltage are obtained; S2. Propose evaluation indicators for the power quality regulation capability of single adjustable resources, and evaluate the adjustable equipment based on the simulation data results, including: regulation direction, regulation amplitude, regulation response time, regulation duration, regulation ramp rate, etc., and analyze the influencing factors of the adjustable resource regulation capability; According to the system node data obtained under the extreme voltage state, the original system data is stored as a matrix and a 28x28 grayscale image as a data set. The power quality regulation evaluation index of the single adjustable resource is used as the data set label. The original data and evaluation results of the active / reactive regulation capability of the adjustable resource are converted into an adjustable resource power quality regulation capability evaluation data set that is suitable for adaptive federated learning. S3. Based on the traditional federated average learning method, CNN neural network and power system distribution network structure, a three-layer model of central server, regional server and adjustable equipment is constructed, and an adjustable resource power quality regulation capability evaluation scheme based on improved federated learning is proposed. It is trained according to the adjustable resource power quality regulation data set to obtain a quickly implemented adjustable resource power quality regulation capability evaluation network model.

2. The method for dynamically evaluating power quality regulation resource capabilities based on adaptive federated learning and voltage over-limit according to claim 1 is characterized in that: S1 is to realize the simulation analysis of the grid voltage characteristics based on the regional grid operation mode. First, a mathematical model of a commonly used active / reactive compensation device is constructed: (1) Synchronous generator If the load is powered by multi-stage transformer, the generator alone cannot meet the load's voltage regulation requirements. Since the voltage of the load close to the synchronous generator cannot be adjusted too high, the voltage adjustment of the load far away must rely on other voltage regulation methods to solve the problem comprehensively. The two-order model of the traditional synchronous generator is simulated, and the rotor motion equation is: Where: J is the moment of inertia, D is the damping coefficient, T m , T e and T d are the mechanical torque, electromagnetic torque and damping torque of the synchronous generator respectively, ω0 is the synchronous angular velocity of the power grid, and δ is the power angle of the synchronous generator; The relationship between the electromagnetic torque and the electromagnetic power output by the virtual synchronous generator is: Where: e a 、e b 、e c and i a 、i b 、i c is the output potential and output current of VSG; The relationship between the synchronous generator output reference voltage is: (2) Static VAR Compensator The dynamic equation of the static VAR compensator SVC model is: Where: V r is the voltage amplitude of the SVC node; V ref is the voltage reference value; K D is the feedback coefficient; G is the gain; T is the time constant; B SVC0 , ΔB SCV They are the steady-state value and variation of the SVC equivalent fundamental susceptance respectively; (3) Photovoltaic inverter The reactive power regulation capability of the static VAR generator SVG photovoltaic unit lies in the photovoltaic inverter therein. The photovoltaic inverter is a power electronic component, and its theoretical reactive power output limit is: Where: S pv is the apparent power of the photovoltaic unit; Q pv It is the reactive power output of the photovoltaic unit.

3. The method for dynamically evaluating power quality regulation resource capabilities based on adaptive federated learning and voltage over-limit according to claim 1 is characterized in that: Based on S1, the response speed and stability of the active power compensation device are simulated. By changing its current or voltage input, the changes in the voltage amplitude of the power system node are observed, and indicators such as the adjustment response time are recorded to evaluate its reactive power regulation capability. The active power adjustment range of the active power compensation device is simulated, and the change of active power is observed by changing its working state. The adjustment accuracy of the active power compensation device is simulated, and the gap between the actual output power and the expected output power is observed by changing its output power. Similarly, the reactive power data of each node parameter and adjustable resources are recorded under stable operation state, and the system node data and reactive power data of adjustable resources are collected under extreme voltage conditions, and the response speed and stability of the reactive compensation device are simulated.

4. The method for dynamically evaluating power quality regulation resource capabilities based on adaptive federated learning and voltage over-limit according to claim 1 is characterized in that: The S3 performs federated learning training on the adjustable resource power quality regulation capability dataset as follows: Based on the data set, the evaluation model is obtained through adaptive federated learning training. Federated learning is aimed at decentralized multi-user scenarios. Each user, i.e., the client, has his own data set {D1, ..., D N }, the federated learning method is to train a model M by participating users FED , while user D i Data is kept locally and not transferred to a central server; A federated learning system usually consists of a data holder and a central server. The data holder performs data training locally. The gradients generated by iterations are desensitized and uploaded to the server as interactive information instead of uploading local data. The server returns aggregated parameters for updating the model. Consider an environment with a central server and N data holders. The goal of federated learning is to solve the model M using the local data of the data holders. FED The risk minimization problem is: Where: M FED For the server model, M FEDi is the local model held by the i-th data holder; When the system starts the tth round of aggregation, the server selects the data holder client to be aggregated, and the server sets the server model M FED (t-1) is transmitted to each selected data holder client, and then the client uses the received server model M FED (t-1) Perform local training; After training is completed, client i will use the locally updated model M FEDi (t-1) is uploaded to the server. The server aggregates the models uploaded by the client according to the following formula and uses a certain learning rate μ to aggregate the global model M FED (t) To update: After the training is completed, the network model of power quality regulation of adjustable resources is obtained, which can quickly evaluate and estimate the power quality regulation capability of adjustable resources without complex iterative calculations. While protecting privacy, it can realize dynamic estimation of the power quality regulation capability of adjustable resources in the power grid.

5. A dynamic evaluation system for power quality regulation resource capability based on adaptive federated learning and voltage over-limit, characterized by: Based on the evaluation method according to any one of claims 1 to 4, comprising: The power grid simulation module simulates and analyzes the voltage characteristics of the power grid based on the operation mode of the regional power grid, obtains the active output of each node parameter and adjustable resources under the stable operation state, and the system node data and the active output of adjustable resources under the extreme voltage. Similarly, the reactive output of each node parameter and adjustable resources under the stable operation state, and the system node data and the reactive output of adjustable resources under the extreme voltage. The evaluation module proposes evaluation indicators for the power quality regulation capability of single adjustable resources, and evaluates the adjustable equipment based on the above calculation results, including: regulation direction, regulation amplitude, regulation response time, regulation duration, regulation ramp rate, and analyzes the influencing factors of the power quality regulation capability of adjustable resources; A conversion module is configured to convert load values ​​and active / reactive regulation capability data obtained by distribution network simulation into an adjustable resource power quality regulation capability data set adapted to adaptive federated learning; The adaptive federated learning module is configured to construct an adaptive federated learning model based on federated learning, CNN neural network and adaptive method, and train it according to the adjustable resource power quality regulation capability data set to obtain an evaluation network model for quickly realizing the adjustable resource power quality regulation capability.

6. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the evaluation method described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the evaluation method according to any one of claims 1 to 4 is implemented.

Images