CN102593830B - Parallel identification method for model parameters of electric power system - Google Patents

Abstract

The invention discloses a parallel identification method for electric system model parameters, which belongs to the technical field of electric system modeling. The present invention realizes the parameter identification of the wide-area power system model through the interactive iteration between the establishment of the optimization algorithm and the power system simulation software, and more importantly, realizes the parameter identification through the parallel processing of the power system simulation calculation task in the computer cluster environment. The parallelization of identification can effectively shorten the time of parameter identification and improve the practicability of parameter identification of wide-area power system model. In addition, the present invention can be applied to various power system simulation software currently used in my country's electric power industry, and can be applied to many modern optimization algorithms such as particle swarm algorithm, ant colony algorithm, simulated evolution algorithm, etc., and has good promotion and application prospects.

Description

A Parallel Identification Method of Power System Model Parameters

technical field

The invention relates to a parallel identification method for electric system model parameters, which belongs to the technical field of electric power system modeling.

Background technique

The results of power system simulation calculations are the basic basis for the power production department to guide the actual grid operation, and the accuracy of the simulation results depends to a large extent on the accuracy of the model parameters. In terms of parameter acquisition of power system models, the current mainstream method is parameter identification. The so-called parameter identification is to use the input and output data measured by the model to be tested in a certain dynamic process to continuously adjust the model parameters through the optimization algorithm so that the model simulation results are as close as possible to the actual measurement results.

When identifying the model parameters of the wide-area power system, on the one hand, it is necessary to simultaneously optimize multiple types of parameters, and on the other hand, it is necessary to calculate the value of the objective function that reflects the overall dynamic behavior of the power system, that is, the optimization process and the calculation process of the objective function value are interactive. . The calculation of the objective function value is based on the simulation of the dynamic process of the power system, which requires the establishment of data exchange between the optimization algorithm and the power system simulation calculation software, that is, the intermediate parameter results obtained by optimizing once can automatically replace the parameters in the simulation system. Then the dynamic response of the system is obtained by the simulation calculation software and output to the optimization program, the optimization program calculates the value of the objective function, and further optimizes to obtain new parameter values.

However, the simultaneous optimization of multiple parameters is involved in the parameter identification process of the wide-area power system, which greatly increases the calculation amount of the optimization algorithm, and the simulation of the wide-area power system also takes a long time, so identifying a set of parameters usually requires Ten hours or even several days. Therefore, improving the speed of parameter identification is of great significance for the practical application of wide-area power system parameter identification, and the use of parallel processing technology is an ideal solution.

Computer cluster technology is a parallel processing technology that organizes multiple computers to work together. This technology uses a high-speed communication network to connect a group of workstations (computing nodes) in a certain structure to form a loosely coupled parallel computing environment; and then realize unified scheduling and coordinated processing through parallel programming and the support of a visual human-computer interaction environment , so as to form an efficient parallel processing system to solve the same problem collaboratively.

Contents of the invention

The purpose of the present invention is to overcome the defects in the prior art, and propose a parallel identification method for power system model parameters, which realizes the parameter identification of wide-area power system models through the interaction between the establishment of optimization algorithms and power system simulation software , more importantly, in the computer cluster environment, the parallel processing of power system simulation computing tasks has realized the parallelization of parameter identification, effectively shortened the time of parameter identification, and improved the practicability of parameter identification of wide-area power system models . The present invention specifically adopts the following technical solutions:

The parameter identification of the wide-area power system model is realized by establishing the interaction between the optimization algorithm and the power system simulation software. The interaction between the optimization algorithm and the power system simulation software in the parameter identification process is generally controlled by the optimization algorithm, and through an interactive interface program In the computer cluster environment, the parallelization of parameter identification is realized through the parallel processing of power system simulation calculation tasks. Specifically, the computer cluster is used as the carrier of parameter parallel identification, and a parallel computing method of master-slave mode is adopted, namely One computer in the computer cluster is used as a parallel computing task management machine, which is mainly responsible for the execution of optimization algorithms and the scheduling of parallel computing tasks. The other computers are used as computing nodes, mainly responsible for power system simulation computing tasks. It is a set of possible values of the parameters to be identified. In each iteration of the optimization algorithm, many computing tasks are assigned to each computing node of the computer cluster at the same time to realize the parallelization of simulation computing tasks.

The beneficial effects of the present invention are: solving the problem of too long time-consuming parameter identification of the wide-area power system in the past, improving the practicability of parameter identification of the wide-area power system model, thereby improving the accuracy of power system simulation analysis and correctly formulating power grid construction planning and the way it works has a positive effect. In addition, the present invention can be applied to various power system simulation software currently used in my country's electric power industry, and can be applied to many modern optimization algorithms such as particle swarm algorithm, ant colony algorithm, simulated evolutionary algorithm, etc., so it has good promotion and application prospects.

Description of drawings:

Attached Figure 1 is a schematic diagram of the interaction between the optimization algorithm and the power system simulation software during the parameter identification process

Accompanying drawing 2 is the flow chart of the power system model parameter parallel identification method proposed by the present invention

Accompanying drawing 3 is the general execution flowchart of the optimized algorithm through parallel programming

Detailed ways:

The parallel identification method of power system model parameters of the present invention realizes the parameter identification of the wide-area power system model by establishing the interaction between the optimization algorithm and the power system simulation software, and the interaction between the optimization algorithm and the power system simulation software in the parameter identification process is generally composed of Optimize the algorithm control, and realize it through an interactive interface program (or a program with different names but similar functions); in the computer cluster environment, the parallelization of parameter identification is realized through the parallel processing of power system simulation calculation tasks, specifically with the computer As the carrier of parameter parallel identification, the cluster adopts a master-slave mode of parallel computing, that is, one computer in the computer cluster acts as a parallel computing task management machine (hereinafter referred to as "manager machine"), which is mainly responsible for the execution of optimization algorithms and parallel computing tasks. Scheduling of calculation tasks, the rest of the computers are used as calculation nodes, mainly responsible for power system simulation calculation tasks, each power system simulation calculation task actually represents a set of possible values of the parameters to be identified, in each iteration of the optimization algorithm will be Numerous computing tasks are distributed to each computing node of the computer cluster at the same time to realize the parallelization of simulation computing tasks.

"Power System Analysis Software Package (PSASP)" is one of the main tools for power system simulation in my country's electric power enterprises. Taking PSASP as an example below, the implementation manner of the present invention will be described in detail in conjunction with the accompanying drawings.

Figure 1 is a schematic diagram of the interaction between the optimization algorithm and the power system simulation software in the parameter identification process. The parameter identification process is generally controlled by the optimization algorithm, and the specific interaction process is as follows:

1. The optimization algorithm first determines a set of values for the parameters to be identified.

2. Send the set of parameter values (after format conversion) to the power system simulation software through the interactive interface program.

3. The interactive interface program calls the power system simulation software for simulation calculation.

4. After the calculation is completed, the interactive interface program calculates the optimized objective function value according to the simulation output and returns it to the optimization algorithm.

5. The optimization algorithm determines the next optimization operation according to the returned objective function value.

6. Repeat the above steps until the optimization algorithm reaches the set number of iterations or the value of the objective function reaches the expected value, and finally output the parameter optimization results.

Fig. 2 is a flow chart of the power system model parameter parallel identification method proposed by the present invention. When PSASP is used as an embodiment, it specifically includes the following steps:

1. Prepare the power system simulation data package for parameter identification on the management machine. The main content includes the format of the data of the target response curve (usually the measured dynamic response curve is used), and the matching of the output content of the power system simulation software. Depending on the power system simulation software, sometimes it is necessary to set the code for the parameters to be identified, copy the simulation calling program to the simulation data package, and so on. Set the name of the directory where the model is located as "PowerSystem", the specific operations include:

(1) Arrange the data of the target response curve (usually the measured power system dynamic response curve) required for parameter identification according to the format of the PSASP simulation output results, and name it "FN1.DAT" and store it in the "PowerSystem\" directory . The PSASP simulation output file is "PowerSystem\temp\FN1.DAT", in which each output curve is stored in columns, and each column is separated by ",".

(2) According to the actual situation of the target response curve, set the corresponding network faults and simulation output content in the PSASP interface, and require the content, quantity, and time length of the simulation output to be consistent with the target response curve.

(3) Set the "code" of the parameters to be identified in the PSASP interface.

Since PSASP itself does not support the direct call of external programs, it does not provide the programming modification function of parameters. Various parameters of the simulation system can be set in the PSASP interface, and then the calculation job needs to be "refreshed" before the calculation is performed. Its function is to write various parameters required for the simulation into the specific In the file (the simulation system parameters are written into "PowerSystem\Lib\DATALIB.DAT", and the static load ratio in the load model is written into the last column of "PowerSystem\Temp\ST.S6"), so that the model parameters are separated from the database environment so that for reading in computing programs. The parameter identification program can modify the parameters of the simulation system by modifying the specific positions in the "DATALIB.DAT" and "ST.S6" files. However, since the content of these two files will change with the change of the simulation system, considering the versatility, it is necessary to set the "code" for the parameters to be identified in the above files to identify the location of parameter modification.

The code of the parameter to be identified can be set directly in the PSASP interface. The parameter code to be set is also a number, but it needs to be significantly different from the parameter value in per unit system. After the code is set up, you need to refresh the calculation job. You can then continue to change the parameter codes in the "DATALIB.DAT" and "ST.S6" files to alphabetic form, but this is only necessary if a large number of static load ratios need to be identified.

(4) Set the search range of the parameters to be identified. The search range can be determined based on empirical values, typical values, or node-level identification results.

(5) Copy the PSASP calculation program to the model directory.

Many simulation calculation functions of PSASP are realized by calling different executable programs in its installation directory, that is, each executable program corresponds to a simulation calculation function. Although PSASP does not directly provide call instructions for simulation calculation, it only needs to run the executable program corresponding to the required calculation function to realize the call. The present invention mainly uses three calculation modules of PSASP flow ("Wmlf.exe"), transient stability ("Wmud.exe"), and transient stability/UPI ("Wmupst.exe"). Copy the above three programs and "lforDLL.DLL" to the "PowerSystem\Temp\" directory, and then they can be directly called by external programs without the PSASP environment. This method is not to crack the PSASP, so the PSASP software dongle is still needed when calling the calculation function, and does not damage the commercial interests of the PSASP author.

2. The management machine notifies each computing node to start the parallel computing client. The client is used to receive the power system simulation data package issued by the management machine and the calculation tasks during parameter identification (and return the calculation results to the management machine).

3. The management machine sends the power system simulation data package to each computing node, which requires the cooperation between the management machine and the parallel computing client on the computing node.

4. The management machine starts the optimized algorithm scheduler programmed through parallelization.

The invention uses a computer cluster as a carrier for parameter parallel identification, and adopts a master-slave mode parallel computing method. According to this feature, the idea of parallelizing modern optimization algorithms is to parallelize computing tasks. Each computing task can be an "ant" in the ant colony algorithm, a "particle" in the particle swarm algorithm, or An "individual" in the genetic algorithm actually represents a set of possible values of the parameters to be identified. In each iteration of the optimization algorithm, there are many computing tasks at the same time, and the parallelization of the optimization algorithm is realized by distributing these computing tasks to each computing node of the computer cluster.

The optimized algorithm after parallel programming is a parallel optimized algorithm scheduler (or a program with different names but similar functions) running on the computer cluster management machine. No matter which optimization algorithm is used, its general execution flow is shown in Figure 3, and the specific steps are:

(1) The parallel optimization algorithm scheduler is started, and some parameters of the optimization algorithm are set, such as the maximum number of movements of ant colony, inertia weight of particle swarm, mutation probability of genetic algorithm, etc.

(2) Calculate and determine all computing tasks in the current round of iteration (such as the position of ants in ant colony algorithm, the position of particles in particle swarm algorithm, the genetic composition of individuals in genetic algorithm, etc.).

(3) Send computing tasks to the idle computing nodes of the computer cluster, and wait if there are no idle computing nodes.

(4) After all the calculation tasks of the current iteration are delivered, wait for all calculation nodes to return the calculation results. If a computing node fails to return the calculation result within the specified time, the computing task will be assigned to other computing nodes for recalculation.

(5) Check the limit of the number of iterations and whether the minimum error reaches the expected value to determine whether to proceed to the next iteration. If iteration is needed, go back to step (2) and continue, otherwise output the optimization result.

5. The optimization algorithm scheduler on the management machine first calculates all the computing tasks of the current iteration, and then sends the tasks to each computing node. The calculation task is actually a set of possible values of the parameters to be identified.

6. After receiving the calculation task, each calculation node calls PSASP to calculate and return the error value between the simulation output curve and the target response curve (that is, the objective function value of the optimization algorithm) to the management machine. The specific steps are:

(1) Modify the "DATALIB.DAT" and "ST.S6" files in the power system simulation data package, so that the values of the parameters to be identified are the values of the current calculation task.

(2) Call "Wmud.exe" (transient stability) or "Wmupst.exe" (transient stability/UPI) in the "PowerSystem\Temp\" directory to calculate

(3) Wait for the calculation to end, then read the data in the PSASP output file "PowerSystem\Temp\FN1.DAT", compare it with the data of the target response curve (stored in "PowerSystem\FN1.DAT"), and calculate the error value. In this step, the calculation may be terminated abnormally due to the unreasonable value of the parameters in the calculation task. This needs to be judged by checking the modification time of the "PowerSystem\Temp\FN1.DAT" file. If the calculation is abnormally terminated , it should return an agreed numerical code to the management machine to indicate that the current parameter combination is unreasonable.

(4) Return the error value between the simulation output curve and the target response curve to the management machine, and then wait for the next calculation task.

7. The optimization algorithm scheduler on the management machine judges whether to perform the next round of iterative optimization based on the results returned by each computing node. If the optimization continues, repeat steps 5 and 6, otherwise output the parameter identification results.

Claims (2)

1. a parallel identification method for model parameters of electric power system, it is characterized in that: by setting up the mutual parameter identification of realizing the wide area power system model between optimized algorithm and PSASP software, being controlled by optimized algorithm generally alternately of optimized algorithm and PSASP software in the parameter identification process, and realize by an interactive interface program, realized the parallelization of parameter identification by the parallel processing to the electric system simulation calculation task under the computer cluster environment, specifically using the carrier of computer cluster as the Parameter Parallel identification, adopt a kind of parallel computation mode of master slave mode, be that a computer in computer cluster is as parallel computation task management machine, mainly bear the execution of optimized algorithm and the scheduling of parallel computation task, all the other computers are as computing node, mainly bear the electric system simulation calculation task, the actual representative of each electric system simulation calculation task be one group of parameter to be identified may value, in each of optimized algorithm is taken turns iteration, numerous calculation tasks are assigned to each computing node of computer cluster to realize the parallelization of simulation calculation task simultaneously, the concrete reciprocal process of optimized algorithm and PSASP software is: at first (1) optimized algorithm determines one group of value of parameter to be identified, (2) will after this group parameter value format transformation, be handed down to PSASP software by the interactive interface program, (3) interactive interface routine call PSASP software carries out simulation calculation, (4), after having calculated, the interactive interface program is according to simulation data calculation optimization target function value and return to optimized algorithm, (5) optimized algorithm is determined next step Optimum Operation according to the target function value returned, (6) repeat above step, until optimized algorithm reaches iterations or the target function value of setting, reach expection, last output parameter optimum results.

2. a kind of parallel identification method for model parameters of electric power system according to claim 1, is characterized in that the method specifically comprises the steps:

(1) prepare the electric system simulation packet for parameter identification on parallel computation task management machine;

(2) parallel computation task management machine notifies each computing node to start the parallel computation client;

(3) parallel computation task management machine is handed down to each computing node by the electric system simulation packet;

(4) parallel computation task management machine starts the optimized algorithm scheduler program through parallel programming;

(5) the optimized algorithm scheduler program on parallel computation task management machine is assigned calculation task to each computing node;

(6) each computing node calls PSASP software according to calculation task and is calculated, and returns to the error numerical value between simulation data curve and target response curve to parallel computation task management machine;

(7) result that the optimized algorithm scheduler program on parallel computation task management machine returns according to each computing node, judge whether to carry out the next round optimizing, if continue optimizing, and repeating step (5) and step (6), otherwise output parameter identification result.

3. a kind of parallel identification method for model parameters of electric power system according to claim 1 and 2, it is characterized in that: described optimized algorithm is a process parallel programming, operate in the parallel optimization algorithm scheduler program on parallel computation task management machine, no matter specifically adopt which kind of optimized algorithm, its general execution step is:

1 parallel optimization algorithm scheduler program starts, and some parameters of optimized algorithm itself are arranged;

The all calculation tasks of 2 calculative determinations when the previous round iteration;

The 3 idle computing nodes to computer cluster issue calculation task, if there is no idle computing node wait for;

After the 4 all calculation tasks when the epicycle iteration all issue, wait for that all computing nodes return to result of calculation, if there is computing node to fail to return at the appointed time result of calculation, its distribution of computation tasks is recalculated to other computing nodes;

5 check whether iterations restriction and minimal error reach desired value, thereby determine whether to carry out the next round iteration, if also need iteration, get back to step 2 and continue to carry out, otherwise the output optimum results.

Images