CN102073766A - Method for accelerating asynchronous parallel computation of transient stability of power grid based on single-computer multiple-core mode - Google Patents

Abstract

Based on the stand-alone multi-core hardware architecture, the present invention proposes a method for rapidly improving transient stability asynchronous parallel computing, including: (1) realizing fast calculation of fault sets in batches, parallel batches, and automatic load balancing batches of transient stable faults, and for each fault The calculation results realize the stability discriminant analysis, and realize the statistical analysis report for the calculation results of all faults in the fault set; (2) realize the batch calculation and analysis of the transient stability operation; The process speed control method of batch calculation speed, proposes and realizes the technical method of mixing interactive calculation and automatic calculation of transient stability. The present invention does not change the calculation algorithm and implementation form of transient stability, realizes parallel batch calculation based on task decoupling mode, realizes parallel calculation of massive fault calculation, ensures the independence of transient stability calculation itself, and improves the efficiency of transient stability calculation .

Description

Asynchronous Parallel Computing Method for Accelerating Power Grid Transient Stability Based on Single Computer and Multi-core Mode

Technical field:

The invention relates to an asynchronous parallel computing method, in particular to an asynchronous parallel computing method for accelerating power grid transient stability based on a single machine multi-core mode.

Background technique:

Power system simulation is the most basic analysis and research method in power system research, planning, design, construction, commissioning, operation and maintenance. According to different classification methods, power system simulation has a variety of different performance types. According to the real-time analysis of simulation, the performance is offline and online simulation, and according to the means of realizing simulation, it is divided into dynamic simulation, full digital simulation, and digital-analog hybrid simulation. simulation. Differences in the fineness of simulation model description and simulation step size can be expressed in electromechanical and electromagnetic transient simulations. The invention relates to electromechanical simulation analysis by computer.

Electromechanical simulation is generally divided into steady-state calculation and analysis and transient calculation and analysis. Steady-state calculation mainly refers to the power flow calculation for the digital model grid established by the research grid. It is to construct a base state of grid operation for other calculations and analysis, which is a possible operating state of the grid. Calculation of transient stability is carried out through digital modeling of generators, excitation systems, commissioning systems, power system stabilizers, static and dynamic loads, line transformers, etc. The process of whether all generators in the grid can maintain synchronous operation.

Transient stability calculation is an indispensable analysis method in power system planning and operation analysis, and plays a pivotal role. It not only provides power system transient stability checks for planning system power supply layout and reasonable transmission, channel layout and network optimization, reactive power compensation and voltage control, protection configuration and stability control measures, but also contributes to the safety and stability of the power system. Reliable operation provides quantitative basis, and can be used to study various measures to improve transient stability, and provide basis for calculating relay protection and automatic device parameter setting.

In the domestic power system, for the main power transmission network, the transient stability calculation tools used are mainly PSD-BPA and PSASP developed by the China Electric Power Research Institute. These two calculation modes are mainly to perform a single transient stability calculation by setting a single fault with a transient stability digital model or to establish a simple fault set to realize a single fault cascading calculation. For the calculation of massive faults and combined faults of multiple operating modes, the parallel computing method of computer clusters is adopted.

The rapid development of the power grid, new energy construction, new voltage levels, various complex AC and DC transmission, micro-grid operation, and various power electronic device applications have led to a sharp increase in the number of fault sets and methods for power grid transient stability calculation and analysis. However, a large number of calculations and analyzes are often not only carried out in fixed workplaces. A large number of calculations and analyzes may be carried out at construction sites, debugging sites, conference sites, etc. It is unrealistic to establish a parallel computer cluster at each massive fault calculation site. It is also uneconomical. How to solve the problem of rapid calculation of massive faults brought about by the development of the power grid is a realistic and urgent problem to be solved.

According to the "PSD-BPA Transient Stability Program User Manual", the typical calculation steps of transient stability are as follows:

1) Collect funds for the simulated target grid, establish basic equipment parameters and network topology model data, and use the power flow program to establish the basic operation mode of the target grid.

2) Fund collection target grid unit, excitation, commissioning, PSS (power system stabilizer), line transformer zero-sequence parameters, etc., establish basic stability model parameters, and conduct trial calculations through transient stability calculation programs to ensure transient stability model data reasonableness and correctness.

3) According to the parameter information in the power flow model, a fault data is established according to certain format requirements. The fault data can be bus faults, transformer faults, line faults, DC faults, or even cascading faults. The fault process may include multiple processes, including fault start, fault disappearance, switch trip, switch reset, switch re-trip, etc.

4) Perform transient stability calculation with faults, conduct stability judgment analysis on the calculation results, and analyze power grid characteristics by calculating output curves and calculating output data.

The above steps are suitable for the interactive calculation of the transient stability of the power grid, and can realize the detailed calculation of the transient stability of the manual control of a single fault form. However, it cannot adapt to the rapid automatic calculation and analysis of massive faults in multiple ways, does not utilize the computing multi-core processing capability, and lacks selective support for different application methods of engineering calculations.

Invention content:

Aiming at the deficiencies of the prior art, the purpose of the present invention is to utilize the powerful processing capability of single-machine multi-core computers, without changing the transient stability calculation program itself and the authorization judgment method, to realize the parallel batch calculation of massive fault multi-mode operations, and to improve the power grid disturbance. Calculation and analysis speed reduces the labor intensity of power grid simulation analysis. Through the pre-established multiple basic operation modes of the power grid, the power flow calculation program pre-calculates them to form multiple basic operation modes. The number of iterations of power flow calculation is limited, the calculation speed is very fast, and the transient stability calculation can only be performed after a converged power flow operation mode is obtained, so it is not necessary to implement batch or parallel calculation of power flow in multiple ways in advance, and breakthroughs in automatic power flow adjustment technology are not ruled out here Finally, in order to obtain a large number of power flow trial calculations in the base state operation mode, a parallel computing method is adopted. It is also necessary to prepare massive fault sets according to the established power flow model data according to different analysis requirements and according to the specified format. In addition, it is necessary to construct a transient stability digital model according to the power grid analysis purpose and analysis level of detail, and it is necessary to ensure that the transient stability digital model is calculated correctly and reasonably without fault disturbance data.

After the data is ready, according to the present invention, the asynchronous parallel calculation of transient stability and acceleration of massive fault sets under multiple operation modes is realized on a single machine.

The improvement of the asynchronous parallel calculation method for transient stability of power grid based on single-machine multi-core mode provided by the present invention is that the method includes the following steps:

(1) Determine the base state operation mode of the power grid simulation calculation, establish the grid transient stability model parameters that are compatible with the base state power flow level, and prepare the base state mode for the power grid that needs to be studied; establish different types of fault sets according to the differences in analysis purposes, and adjust Grid transient stability model;

(2) Determine the corresponding relationship between the ground state power flow, fault set and transient stability model, construct a transient stability calculation job task group, and form a job batch task group for all analysis targets in the form of task units, that is, job task batch queue data, job batch The queue is executed in the serial serialization mode, that is, the transient stability calculation associated data pairs are formed sequentially according to the priority level of the calculation;

(3) Judging the integrity and validity of the data, and starting the calculation of transient stability operations;

(4) Decompose the current job fault set data into a single fault data queue;

(5) Carry out the batch calculation of transient stability of the selected fault set according to the specified calculation mode;

(6) Form a comprehensive summary report of transient stability calculation of the entire fault set;

(7) After all transient stability calculation operations are completed, all operation calculations are ended.

The calculation method of the first preferred solution provided by the present invention is improved in that,

The integrity judgment includes the existence of base state data, the existence of fault set data, and the existence of transient model data;

The validity judgment includes the rationality and validity judgment of a single fault in the fault set.

The calculation method of the second preferred solution provided by the present invention is improved in that the calculation mode is divided into:

Basic cascade serial batch calculation method;

Parallel batch computing mode with specified number of cores;

Dynamic automatic load balancing calculation mode.

The calculation method of the third preferred solution provided by the present invention is improved in that the steps of the basic cascade serial batch calculation method are as follows:

(a) get one group of single fault data from the fault data queue that described step (4) decomposes and obtain, form calculation data group with mode data, transient stability model data pairing;

(b) start the transient stability calculation kernel with the calculation data set in (a);

(c) Jump to (e) after the transient stability calculation is completed;

(d) Accumulate each judgment detection cycle and compare it with the set maximum transient stability calculation time-consuming threshold. If the threshold exceeds the preset threshold, the transient stability calculation process will be forcibly terminated, and the The operation process is recorded in the fault calculation result; wherein, from step (3) to the end of all operations is a total detection time period, which is divided into several detection periods according to the amount of work.

(e) Extract the results for stability judgment and analysis, judge as a stable result, that is, a fault calculation is completed, put the relevant conclusions in the result list, and exit this fault calculation;

(f) Extract the data items of the fault data queue. After the faults in the current calculation operation have been calculated, summarize all the calculation results in this operation to form a statistical summary, and exit the calculation of this operation.

The calculation method of the fourth preferred solution provided by the present invention is improved in that the steps of the parallel batch calculation mode are as follows:

1) Multi-core exists and the set value is greater than 1, jump to 2);

2) Establish a data queue equal to the specified number of cores;

3) When there is an idle queue in the calculation data queue, a group of single fault data is filled into the idle data queue from the fault data queue decomposed in the step (4), and it is reserved in the idle data queue before it is designated to be applied by the transient stability calculation program. in the queue;

4) All queues have been filled with data, go to step 5);

5) The designated calculation core has a vacant calculation core, and a set of parameters is taken from the data queue, and the transient stability calculation process is started on the idle core, and the transient stability calculation of the fault data taken is performed;

6) In a detection cycle, when the transient stability calculation is detected, the result analysis process is started immediately, and the power grid accident calculation results are classified and analyzed. After the result analysis is completed, that is, a fault calculation is completed, and the fault calculation is exited. , jump to 8);

7) All parallel transient stability calculations take longer than the specified dead limit threshold, forcibly terminate the transient stability calculation process exceeding the threshold, and jump to 8);

8) End the calculation, and immediately release the occupied calculation core and data parameter queue;

9) In the process of filling the parallel data queue, if the current fault job still has unfilled fault data, it will be filled into the vacant data queue;

The calculation method of the fifth preferred solution provided by the present invention is improved in that the automatic load balancing calculation step is:

I. Establish a data queue equal to the specified number of cores;

II. When there is an idle queue in the calculation data queue, get a group of single fault data from the fault data queue obtained by decomposing the step (4) and fill it into the idle data queue, and keep it in the idle data queue before it is designated to be applied by the transient stability calculation program in the queue;

III. All queues have been filled with data, go to step V;

IV. The specified calculation core has a vacant calculation core, and a set of parameters is taken from the data queue, and the transient stability calculation process is started on the idle core according to the load judgment result, and the transient stability calculation of the fault data taken is performed, and the dynamic load The judgment formula is as follows:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; cpu</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; idle</span>}}}{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; period</span>}}}\\ {\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; meme</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; use</span>}}}{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; total</span>}}}\\ {\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; cpu</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; setting</span>}}\\ {\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; meme</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; setting</span>}}\end{array}\right.$

The calculation method of the sixth preferred solution provided by the present invention is improved in that: the sub-steps of the step (2) include:

① Select the ground state mode data for calculation convergence;

② Selection of fault set data for transient stability calculation;

③ Select the transient stability data matching the ground state power flow mode;

④ Integrate the data information obtained in ①-③ into an independent computing operation T _i ;

⑤Repeat the process ①-④, establish two or more independent computing jobs, sort the task priorities according to the weight factor P, form a computing priority sequence, and arrange the sequence of computing tasks according to the computing priority levels; for the jobs that have been executed, Set the weight factor P to zero, and when the new job starts to execute, the tasks will be sorted according to the weight factor P again, because the weight factor P of the job that has been calculated is 0, and it is placed at the end of the descending order;

Establish dual queue management weight factors, one of which is used to manage the weight factor set by the queue, and the other is used to manage the change of the weight factor during calculation.

The calculation method of the seventh preferred solution provided by the present invention is improved in that, in the step ⑤, according to the dynamic change priority weight system, the dynamic execution of serialized cascading batch operation operations is realized, and its execution weight is determined by the double queue [p ₁ , p ₂ , ...., p _N ] ^T column vector control, static response manual setting control, dynamic response execution and dynamic execution order change of uncalculated jobs during execution and calculation, and completion of transient stability calculation during execution The dynamic coefficient pi _is set to zero, and for unexecuted jobs, when determining the next job, the bubble selection of computing jobs is realized in descending order of column vectors.

The calculation method of the eighth preferred solution provided by the present invention is improved in that: the empirical value of the step-by-step judgment and detection period of the transient stability calculation end judgment given in the step (c) is 1 second.

Compared with the prior art, the beneficial effects of the present invention are:

Three calculation modes for batch calculation of massive faults can be seamlessly switched between the three modes, and the transient stability calculation items that have been executed will not be affected after switching.

No matter what mode is used for the calculation of transient stability fault items, the serialized calculation method is adopted among multiple jobs. By introducing the calculation priority weight factor, the calculation priority of uncalculated tasks is dynamically adjusted, so that the calculation of dynamic control tasks becomes possible.

For the monitoring and judgment period of the transient stability calculation process, the engineering experience value is given, and the theoretical analysis concludes that if it is too short or too long, it will affect the calculation efficiency. The designed multi-core asynchronous start and parallel execution batch fault calculation mode can rapidly improve the efficiency of transient stability calculation.

In the calculation of transient stability, the load rate factor is introduced, and the computer load rate judgment mode is designed to suit the daily office conditions. The method of batch calculation of transient stability can make full use of the idle interval of the computer and maximize its processing capacity, and improve the temporary stability. State stability computing efficiency.

The invention does not change the transient stability calculation algorithm and implementation form, and realizes the parallel batch calculation based on the task decoupling mode, which ensures the independence of the transient stability calculation itself while realizing the parallel calculation of massive fault calculations.

The basic cascade serial batch calculation method realizes the serial serialized batch calculation of transient stability faults, which is the basic calculation form; the parallel batch calculation mode with a specified number of cores realizes the asynchronous calculation of multiple transient stability calculations under multiple computing cores Start-up and parallel computing can satisfy unattended fully automated high-efficiency computing analysis; the batch computing mode of dynamic automatic load balancing is an automatic working mode that considers coordination with other non-computing tasks and satisfies computing comfort and work selectivity. Here In this way, the automatic conversion of computing sleep, cascading batch computing and parallel batch computing can be realized. In the actual calculation process of faulty operations, the three calculation modes can be freely switched without interrupting the entire calculation process of all operations through manual intervention.

Description of drawings

Figure 1: A block diagram of the implementation steps of the asynchronous parallel computing method for the transient stability calculation efficiency of the speed-up power grid based on the single-machine multi-core proposed by the present invention;

Figure 2: Block diagram of sub-process steps of single-machine serial batch processing transient stability calculation;

Figure 3: Block diagram of sub-process steps of transient stability calculation sub-process of single-machine multi-core parallel batch processing;

Figure 4: Block diagram of sub-process steps of batch processing transient stability calculation with load balancing on a single machine;

Figure 5: Schematic diagram of the principle of three batch calculation modes for transient stability calculation, which mainly shows the calculation and processing process of calculating and checking fault operations and fault items.

Detailed ways

The block diagram of the specific embodiment of the present invention is shown in Figure 2, and the asynchronous parallel computing method based on the single-machine multi-core speed-up power grid transient stability computing efficiency includes the following steps:

(1) According to the requirements of power grid analysis and calculation, from the power flow topology model data of the power grid network, according to the composition of power grid equipment including generators, busbars, AC lines, transformers, transient stabilizers, DC equipment, etc., form the transient stability analysis requirements of the power grid failure set. Realize the power flow calculation based on the steady state model of the power grid, and form the base state operation mode of the power grid simulation calculation. The parameters of the transient stability model of the power grid established based on the level of the ground state power flow, mainly include the prime mover model of the generator in the power grid, the wind turbine and its control system model, the excitation regulator model, the governor model, the DC control system model, and the power system stability model. Transformer model, static and dynamic load model, SVC/STATCOM/FLC and other power electronic equipment models, power grid zero sequence including transformer zero sequence and line zero sequence models, etc. And prepare for various ground state methods for the power grid that needs to be studied, including large dry, small dry, large large, small full, etc., to ensure the convergence of the ground state power flow mode; according to the analysis purpose such as N-1 fault scanning, N-2 fault screening, etc. The difference sets up different types of fault sets and adjusts the transient stability model of the power grid.

(2) Determine the corresponding relationship between the ground state power flow, fault set and transient stability model, and build a task group for transient stability calculation. In the form of task units, the job batch task group of all analysis targets is formed, that is, the job task batch queue data. The job batch queue is executed in serial serialization mode, so the associated data pairs for transient stability calculations should be formed in sequence according to the priority of calculations. The sub-items of this step include:

① Select the ground state mode data for calculation convergence;

② Selection of fault set data for transient stability calculation;

③ Select the transient stability data matching the ground state power flow mode;

④Integrate the data information obtained from ①-③ into an independent calculation operation T _i , where T represents the operation, and i represents the sequence number generated by this operation, which must be accumulated sequentially.

N代表设置的全部计算作业数。在单一作业中，设置有作业优先权重因子即P＝[p₁，p₂，....，p_N]^T，权重因子P_i表明了作业优先级，为大于零的自然数。当作业序列执行时，按权重因子对任务优先级进行排序，形成计算优先序列。对于已经执行完成的作业，将权重因子置零，在新的作业开始执行时，将重新按权重因子对任务排序，因已经计算的作业权重因子为0，被放置到降序末端，新的作业权重因子如果为0，表明所有计算作业已经结束，否则就启动该作业进行计算。为了保存权重因子供下次应用，要建立双队列管理权重因子，其中一个队列管理设置的权重因子，另一个用于管理计算时的权重因子的变化。⑤ Repeat the process ①-④ to create multiple independent computing jobs, arrange the sequence of computing tasks according to the computing priority, and establish a complete set of batch job tasks

N represents the number of all calculation jobs set. In a single job, a job priority weight factor is set, that is, P=[p ₁ , p ₂ , ..., p _N ] ^T , and the weight factor P _i indicates the job priority, which is a natural number greater than zero. When the job sequence is executed, the task priorities are sorted by weight factors to form a computing priority sequence. For the jobs that have been executed, the weight factor is set to zero. When the new job starts to execute, the tasks will be sorted according to the weight factor again. Because the weight factor of the job that has been calculated is 0, it is placed at the end of the descending order, and the new job weight If the factor is 0, it indicates that all calculation jobs have ended, otherwise, the job is started for calculation. In order to save the weight factor for the next application, it is necessary to establish a dual queue management weight factor, one of which manages the weight factor set by the queue, and the other is used to manage the change of the weight factor during calculation.

In step ⑤, according to the dynamic change priority weight system, the dynamic execution of serialized and cascaded batch computing jobs is realized. Its execution weight is controlled by double-queue [p ₁ , p ₂ , ..., p _N ] ^T column vectors, static response is controlled by manual setting, and dynamic response is the change of dynamic execution order of uncalculated jobs during execution and calculation. The dynamic coefficient p _i of the completed transient stability calculation is set to zero, and for the unexecuted job, when the next job is determined, the bubbling selection of the calculation job is realized according to the descending order of the column vector.

(3) Start the single-group transient stability calculation. Before calculation, it is necessary to consider re-judging the completeness and validity of power flow base state mode, fault set file and transient model data, because in the process of power grid simulation calculation, data modeling and modification, mode calculation and dynamic model parameter adjustment, specific The settings of the fault data may be changed manually. Only by considering the modification and change of the data can the latest transient stability calculation and analysis requirements be met. Therefore, at the moment before the calculation is started, the judgment of the integrity and validity of the data is beneficial to the entire parallel batch reliability and efficiency. Integrity judgment includes three judgments: the existence of ground state data, the existence of fault set data, and the existence of transient model data; the validity judgment includes the rationality and validity judgment of a single fault in the fault set; in addition, it is necessary to check the ground state power flow Whether it is converged, whether the parameter setting of the stable model is valid, and whether the stability judgment of the calculation result exists and is valid automatically. If there is no manual setting of the judgment stability value, the default judgment stability value is taken in combination with expert knowledge.

(4) For the fault set data of the current job, decompose it into a single fault data queue. When decomposing, operate according to the format row. In the fault set data, each fault row starts with an interval row, and the interval row includes three data fields: fault interval identifier, fault calculation name, and fault result data name. Among them, fault name and fault result The data name field is allowed to be absent, and this method can be automatically generated according to the name formation rules.

(5) Carry out the batch calculation of transient stability of the selected fault set according to the specified calculation mode. This method divides the calculation mode into three types: the basic cascade serial batch calculation mode, the parallel batch calculation mode with a specified number of cores, and the batch calculation mode with dynamic automatic load balancing. The three calculation modes can be switched at any time during the calculation process. According to the specified mode, the single fault set transient stability calculation is divided into three branches, and the steps are as follows:

Basic cascading serial batch calculation method. Cascade mode is a serial automatic calculation of transient stability failure modes. It disassembles the faults one by one in the fault set, continuously matches the fault data, transient stability model data, and mode data, and transmits them to the transient stability calculation core, which completes the transient stability calculation. The sub-process steps of this method are as follows:

(a) Take a group of single fault data from the fault data queue decomposed in step (4), and pair them with mode data and transient stability model data to form a calculation data group.

(b) Start the transient stability calculation core with the calculation data set in (a), and start the transient stability calculation in the form of a single fault.

(c) By continuously monitoring the progress of the transient stability calculation, it is judged whether the calculation is over. If it ends, jump to (e). If it is not over, just wait for a judgment cycle. The judging period is expressed as a delay T _d , and the default value T _d is 1 second. The specified delay time directly affects the efficiency of cascaded calculations. If the setting is too long, the theoretical limit of the waiting time for the last judgment of the end of the calculation is close to T _d , because if the calculation just ends when it enters the wait at the beginning, come back to judge again The time is just delayed by a T _d . If the worst case is considered, with an average of N faults in one job and a total of M jobs, then the limit time wasted due to round robin inspection is: N×M×T _d , the unit is second. Therefore, every time the delay T _d increases by 1 second, the entire cascaded calculation delay waste will be enlarged by N×M seconds. The delay setting value cannot be too short, because too short will lead to an increase in the frequency of judgment, and may also lead to a decrease in calculation efficiency.

It is agreed that each transient stability calculation takes the same amount of time. The following expressions hold:

(N ₀ +1)×T ₀ =(T ₀ +ΔT _ε )×N _ε

T ₀ is the default initial delay value, N ₀ is the number of times needed for uniform judgment of single transient stability calculation (assuming that the judgment interval is uniform and not subject to changes in machine conditions), then consider the time spent on judgment itself, The minimum number of times for judging the end of a single transient stability calculation is at least (N ₀ +1). If the variation of the delay setting value is represented by ΔT _ε , and the total number of judgments due to the change of the delay setting value is N _ε , then the change of the number of judgments due to the change of the delay setting value is:

$\mathrm{<span\; class="notranslate">\; \&Delta;N</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}}{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Considering the conversion time cost ΔT _ω of the entry and exit process, then, when the delay time is shortened, that is, when ΔT _ε is less than 0, ΔN is a positive value, which can be expressed as an absolute value, and the following expression holds true for the increase in judgment time due to the shortened delay value :

$<span\; class="notranslate">\; |</span>\frac{\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}}{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; \&Delta;</span>}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}$

情况就会相反，因此设置太小的延迟判断时间值也不可取。Because the time of the entry and exit judgment process is generally at the level of milliseconds, there is ΔT _ω <<T ₀ , so reducing T ₀ to a certain extent will improve the cascade calculation efficiency, but if it is set too small, it will cause

The situation will be the opposite, so it is not advisable to set too small a delay judgment time value.

In a single transient stability calculation process, in order to judge whether the transient stability calculation process is over, it is recommended that the empirical value of the detection cycle be set to 1 second to prevent the impact on the entire calculation time caused by too long or too short.

(d) Accumulate the detection time for each judgment and compare it with the set maximum transient stability calculation time-consuming threshold. If it exceeds the preset threshold, the transient stability calculation process is forcibly terminated, and the The forced operation process is recorded in the fault calculation result. The purpose of setting the maximum time-consuming threshold for transient stability calculation is to prevent the rigidity of the transient stability calculation program. There may be problems in solving the matrix, data integration, matching, and statistics, which will lead to a deadlock in the transient stability calculation, making the cascade calculation impossible. Ongoing. Since the dead threshold of transient stability calculation is set, the overall failure probability of cascade calculation is technically effectively solved, and the reliability of the entire calculation is improved.

(e) After the transient stability calculation is completed, the results need to be extracted for stability judgment analysis, and the stability judgment conclusion includes the judgment of stability and instability. According to the type of power grid instability, the instability properties include angle-of-attack instability, dynamic instability, voltage instability, and frequency instability. The completion of the stable result judgment indicates that a fault calculation is completed, and the relevant conclusions are placed in the result list, and this fault calculation is exited.

(f) Extract the data item of the fault data queue, if there is still fault data, enter the calculation process of the next fault item, and return to step (a). If all the faults in the current calculation job content have been calculated, summarize all the results in this job to form a statistical summary, and then exit the job calculation.

Parallel batch computing mode with specified number of cores. This method supports multi-core computer transient stability concurrent calculation. Before realizing parallel calculation, the number of CPU cores used for transient stability calculation must be specified within the range of the maximum number of cores allowed by the computer. Computation is only related to the number of cores of the computer, and has nothing to do with the type of computer and the number of CPUs. It supports a single personal desktop, notebook and server. You must specify more than one core number to theoretically start batch parallel computing. Its sub-process steps are as follows:

1) Judging that multi-core exists and the setting value is greater than 1. If it exists and the set value is greater than 1, jump directly to step 2); otherwise, enter the basic cascade serial batch calculation process.

2) Establish a data queue equal to the specified number of cores. In the parallel batch calculation process, data import and export, transient stability calculation, fault set decomposition, transient stability result analysis, and transient stability calculation data preparation are sequenced in order of time consumption. In order to avoid waiting for transient stability calculation and improve the overall calculation efficiency, the transient stability calculation must be started as soon as a computer core is vacant. Therefore, all parallel data queues are always filled before starting the transient stability calculation. Required calculation parameter data. Because the fault data preparation time is shorter than the transient stability calculation, there will be no data calculation core vacancy, and the data is not yet fully prepared. In normal operation, it always waits for the completion of transient stability calculation after the parallel data queue is ready.

3) Judging whether there is an idle queue in the computing data queue. If there is an idle queue, immediately take a group of single fault data from the fault data queue decomposed in step (4) and fill it into the idle data queue, and keep it in the queue until it is specified to be applied by the transient stability calculation program. If there is no idle data queue, it enters the task waiting, and waits for any one that has started the transient stability calculation process to end, and then releases the data queue.

4) Judging whether all the queues have been filled with data, if they have been filled, then go to step 5); otherwise, go to step 9).

5) Judging whether all the designated calculation cores are busy, if there are vacant calculation cores, immediately take a set of parameters from the data queue, start the transient stability calculation process on the idle cores, and perform the transient stability calculation of the fault data taken . As long as the transient stability calculation is started, a transient stability calculation monitoring thread will be established to detect the end state of the calculation in real time through polling with a certain detection time.

6) In a detection cycle, when the transient stability calculation is detected, the result analysis process is immediately started, and the power grid accident calculation results are classified and analyzed, including stability qualitative judgment, unstable classification judgment, and minimum voltage point statistics Analysis, statistical analysis of the largest change in power angle, etc. The result analysis is completed, indicating the completion of a fault calculation, exit this fault calculation, and jump to 8). If no end of transient stability calculation is detected within a detection period, the time-consuming time of transient stability calculation will be recorded.

7) Judging whether the time-consuming time of all parallel transient stability calculations exceeds the specified dead limit threshold, if it exceeds the dead limit threshold, the transient stability calculation process exceeding the threshold is forcibly terminated, and jumps to 8). If the limit threshold is not exceeded, enter the next detection cycle and jump to 6). When each transient stability calculation is started, a counter T _i is opened, and i represents the program number of the started transient stability calculation. There is a T _i =K×ΔT for each calculation process, each calculation process has its own K value, and K starts to count only after it is started, when K×ΔT is greater than T _setting , T _setting is the dead threshold value, indicating When the transient stability calculation times out, the calculation termination process will be initiated. This method has requirements for setting the dead threshold. It does not rule out that the transient stability calculation is carried out normally but takes too long, so there may be a possibility of terminating the normal calculation. Therefore, when setting the time threshold, a temporary calculation of the research grid can State stability calculation trial calculation, calculate the calculation time, and then add a certain margin as the dead limit threshold. In general, the time margin can be 5 to 10 minutes, and the deadlock limit threshold is generally less than 30 minutes. The dead limit threshold cannot be set too small. If it is set too small, a large number of transient stability calculations will fail, which is not recommended by this method. The main purpose of this method is to quickly calculate a large number of grid faults in batches.

8) Either the failure calculation ends normally, or the overtime limit is forced to end. As long as the calculation is completed, the occupied calculation core and data parameter queue will be released immediately.

9) Continuing from 4), in the process of filling the parallel data queue, if the current fault job still has unfilled fault data, it will be filled into the vacant data queue. If there is no fault data to be calculated, it will enter to judge whether all transient stability calculations are over. If it is over, this operation will end, otherwise wait for all calculation cores to be vacant.

Automatic load balancing calculation mode. The automatic load balancing mode decides whether to start the transient stability calculation process according to the dynamic load of the computer. During the transient stability calculation process, other time-consuming business startups may be involved. If the business priority is higher than the calculation mode, if there is no load A balanced dynamic mechanism will lead to an equal distribution of resources, which will inevitably affect the operator's feeling of use and work progress. The automatic load balancing mode is very similar to the parallel batch computing mode with a specified number of cores, and the execution process and steps are basically carried out according to the steps of the parallel batch computing mode with a specified number of cores. The main differences are:

I. In the automatic load balancing mode, it is not judged in advance whether the specified number of cores must be greater than 1. As long as there is a specified number of cores, even if 1 core is specified, and the minimum calculation requirement is set, a data queue will be established. That is to say, this mode does not perform step 1) judgment.

II. In step 5), dynamic load judgment conditions are added to start the transient stability calculation process according to the load judgment results, which makes more scientific use of computing system resources than simply considering the idleness of computing cores. The principle of dynamic load judgment is as follows:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; cpu</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; idle</span>}}}{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; period</span>}}}\\ {\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; meme</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; use</span>}}}{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; total</span>}}}\\ {\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; cpu</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; setting</span>}}\\ {\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; meme</span>}}<span\; class="notranslate">\; \&le;</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; setting</span>}}\end{array}\right.$

In the above formula, R _cpu represents the computing core occupancy rate, and R _mem represents the memory occupancy rate, which are two key indicators for automatic load balancing monitoring, because in the process of transient stability calculation, the calculation speed and efficiency are mainly affected by the computing core Main frequency, memory usage. T _period is an extremely short monitoring period during automatic load calculation, and T _idle is the idle time of the calculation core during the load monitoring period. Their ratio is the current idle rate of the calculation core. After normalization, it is subtracted by 1 to obtain timely Calculate the core load rate of . M _total is the total amount of memory during automatic load calculation, M _use is the total amount of memory usage during automatic load calculation, and their ratio is the memory usage. R _c-setting is the upper limit of the calculation core occupancy, and R _m-setting is the upper limit of the memory occupancy. If the calculation core occupancy and memory occupancy are less than or equal to the upper limit, and there is an idle computing core, a single temporary Calculation of state stability. Generally, the value of R _c-setting is 1/N (N is the total number of cores of the computer), and R _m-setting is generally 0.8. The above fixed values are related to the architecture and type of the computer CPU, and the fixed values can be properly adjusted according to the specific models.

(6) Judging whether all the fault data in the calculation operation have been calculated. If the calculation is completed, the single fault calculation and analysis results are summarized to form a comprehensive summary report of the transient stability calculation of the entire fault set, which is saved and displayed in the report. If the calculation is not completed, continue to step (5).

(7) Judging whether all transient stability calculation operations are completed, if not, enter the next operation calculation and return to step (3), otherwise, end all operation calculations.

2. According to the asynchronous parallel calculation method of single-machine multi-core speed-up power grid transient stability calculation efficiency described in requirement 1 of the power letter, it is characterized in that it realizes three calculation modes of batch calculation of multiple groups of operations and multiple faults, that is, basic cascade series Row batch computing mode, parallel batch computing mode with specified number of cores, batch computing mode with dynamic automatic load balancing. The basic cascade serial batch calculation method realizes the serial serialized batch calculation of transient stability faults, which is the basic calculation form; the parallel batch calculation mode with a specified number of cores realizes the asynchronous calculation of multiple transient stability calculations under multiple computing cores Start-up and parallel computing can satisfy unattended fully automated high-efficiency computing analysis; the batch computing mode of dynamic automatic load balancing is an automatic working mode that considers coordination with other non-computing tasks and satisfies computing comfort and work selectivity. Here In this way, the automatic conversion of computing sleep, cascading batch computing and parallel batch computing can be realized. In the actual calculation process of faulty operations, the three calculation modes can be freely switched without interrupting the entire calculation process of all operations through manual intervention.

The present invention has been developed and verified in a new generation power grid simulation analysis platform based on the PSD-BPA core, and the formed independent components have been integrated into the power grid simulation analysis platform, and can be used in the research on the safety and stability adaptability of a certain power grid UHV, the security of a certain interconnected large power grid It has been widely used in in-depth research, research on wind power access system and operation control of a wind power plant, and calculation and analysis of power grid security and stability control strategies. In order to test its function and speed-up efficiency, a working computer is randomly selected for transient stability calculation test. The process is as follows:

1. Select a working computer at random.

Basic(EM)32-Bit，处理器为双计算核2Quad CPU Q9400，主频率为2.66GHz，安装内存为2.00GB，硬盘450G。The test selects Dell OPTIPLEX 960 desktop, and the operating system uses 32 bits

Basic (EM) 32-Bit, the processor is a dual computing core 2Quad CPU Q9400, the main frequency is 2.66GHz, the installed memory is 2.00GB, and the hard disk is 450G.

2. Determine the test case.

Selecting the model data of a regional power grid, there are a total of 2332 bus nodes, 1731 line branches and 1738 transformer branches. Prepare tidal current data files, stable data files, and 4 fault set files including 10 faults, 50 faults, 100 faults and 300 faults respectively.

3. Conduct the test and record the test results.

The test is carried out in two periods, and the same test method is used for each period. That is to say, it is carried out according to single-task serial batch processing, dual-core parallel batch processing, and batch processing with load balancing, and records the start and stop time of job calculation. When carrying out the load balancing test, operate on the computer according to the normal working method, such as writing reports, browsing materials, sending and receiving emails, etc. The test results are shown in the table below:

Table 1 Test time 1 Test results

Table 2 Test time 2 Test results

4. Analyze the test results.

Through the data analysis of test results, the method of the present invention can meet the batch calculation requirements of massive power grid faults, and can achieve obvious speed-up effects when implementing power grid transient stability calculations. The test process and data show that:

Unattended transient stability parallel batch calculation can be realized;

To further improve the calculation speed of transient stability, parallel batch calculation can be realized by increasing the number of computer cores under the support of the computing hardware architecture.

It can realize the synchronization of transient stability calculation and result statistical analysis;

Parallel batch transient stability calculation can greatly improve the fault calculation speed, and the speedup ratio of parallel batch calculation is greater than or equal to 3.1 times than serial batch calculation;

Parallel batch processing has good speed-up and stability, and the time-consuming of the same fault set is close to that of two repeated calculation example tests;

Batch calculation with load balancing, its speed is limited by the working conditions of the computer, when the calculation load is heavy, it will affect the overall calculation efficiency, but through the test case analysis, even under heavy load conditions, computer resources can be fully utilized , improve the calculation speed of transient stability, and can adapt to batch automatic calculation of transient stability calculation under various working conditions.

Claims (9)

1. based on unit multinuclear pattern speed-raising electrical network transient stability asynchronous parallel computing method, it is characterized in that described method comprises the steps:

(1) determines the ground state method of operation that grid simulation calculates, set up the electrical network transient stability model parameter that ground state trend level adapts, and needs research electrical network is carried out the ground state mode prepare; Set up different classes of failure collection according to the difference of analysis purpose, and adjust electrical network transient stability model;

(2) determine ground state trend, fault collection and transient stability model corresponding relation, make up transient stability computational tasks task groups, with the TU task unit form, form the job patch task groups of whole evaluating objects, be job task batch queue data, the job batch formation is carried out by the serial sequence pattern, and it is right promptly to form transient stability compute associations data successively by the priority level of calculating;

(3) integrality of judgment data and validity are opened the transient stability operation and are calculated;

(4) be single fault data formation with current operation fault collection data decomposition;

(5) implementing selected fault collection transient stability according to the computation schema of appointment calculates in batches;

(6) transient stability that forms whole fault collection calculates the comprehensive summary form;

(7) finishing whole operations after all the transient stability computational tasks is finished calculates.

2. computing method as claimed in claim 1 is characterized in that, described integrality judgement comprises the existence of ground state data existence, fault collection data, the existence of transient Model data;

Described validity is judged rationality and the validity judgement that comprises single failure in the failure collection.

3. computing method as claimed in claim 1 is characterized in that, described computation schema is divided into:

Basic cascade serial is account form in batches;

Band is specified the parallel batch computation schema of check figure;

Dynamic auto load balance computation schema.

4. computing method as claimed in claim 3 is characterized in that, the step of described basic cascade serial batch account form is as follows:

(a) from the fault data formation that described step (4) decomposition obtains, get one group of single failure data, form the computational data group with mode data, the pairing of transient stability model data;

(b) start transient stability with (a) computational data group and calculate nuclear;

(c) transient stability jumps to (e) after calculating and finishing;

(d) each judgement detect cycle is added up, and compares, surpass predefined threshold values, just stop transient stability computation process by force with the maximum transient stability Calculation threshold values of setting consuming time, and by force operation process recording in the calculation of fault result;

(e) the extraction result carries out the judgement of stability analysis, differentiates to be stabilization result, and promptly primary fault calculates and finishes, and related conclusions is placed in the results list, withdraws from this calculation of fault;

(f) extract fault data queuing data item, after the fault in the current computational tasks content has been calculated and finished, gather all result of calculations in this subjob, form statistical abstract, and withdraw from this subjob and calculate.

5. computing method as claimed in claim 3 is characterized in that, described parallel batch computation schema step is as follows:

1) multinuclear existence and setting value jump to 2 greater than 1);

2) set up the data queue that equates with it according to the check figure of appointment;

When 3) having idle queues in the computational data formation, from the fault data formation that described step (4) decomposition obtains, get one group of single failure data and be filled in the idle data formation, specify to be retained in the formation before the transient stability computation program application always;

4) data have been filled up in all formations, change the 5th over to) step;

5) calculating of appointment nuclear has vacant calculating nuclear, gets one group of parameter from data queue, starts transient stability computation process on the nuclear of free time, and the transient stability that carries out the fault data of getting calculates;

6) in detect cycle, calculate end when detecting transient stability, just start resolving as a result immediately, to implement power grid accident result of calculation and carry out classification analysis, interpretation of result is finished, promptly finishing of a calculation of fault withdrawed from this calculation of fault, jumps to 8);

7) all parallel transient stabilities calculate the ossified qualification threshold values that duration consuming time surpasses appointment, stop surpassing the transient stability computation process of threshold values by force, jump to 8);

8) finish to calculate the calculating nuclear of release busy immediately and data parameters formation;

9) fill in the parallel data formation process,, just fill it in the vacant data queue when the prior fault operation also has unfilled fault data;

6. computing method as claimed in claim 3 is characterized in that, described ALB calculation procedure is:

I. set up the data queue that equates with it according to the check figure of appointment;

When II. having idle queues in the computational data formation, from the fault data formation that described step (4) decomposition obtains, get one group of single failure data and be filled in the idle data formation, specify to be retained in the formation before the transient stability computation program application always;

III. data have been filled up in all formations, change the V step over to;

IV. the calculating of appointment nuclear has vacant calculating nuclear, gets one group of parameter from data queue, starts transient stability computation process according to the load judgment result on the nuclear of free time, and the transient stability that carries out the fault data of getting calculates, and the dynamic load judgment formula is as follows:

$\left\{\begin{array}{c}{R}_{\mathrm{cpu}}=1-\frac{T_{\mathrm{idle}}}{T_{\mathrm{period}}}\\ R_{\mathrm{mem}}=1-\frac{M_{\mathrm{use}}}{M_{\mathrm{total}}}\\ R_{\mathrm{cpu}}\&le;R_{c-\mathrm{setting}}\\ R_{\mathrm{mem}}\&le;R_{m-\mathrm{setting}}\end{array}\right.$

7. computing method as claimed in claim 1 is characterized in that, the substep of described step (2) comprising:

1. select to calculate convergent ground state mode data;

2. select to be used for the failure collection data that transient stability calculates;

3. select to be matched with the transient stability data of ground state trend mode;

4. will be 1.-data message that 3. obtains is integrated into an independently computational tasks T _i

5. repeat 1.-4. process, set up independent computational tasks more than two, P sorts to task priority by weight factor, forms to calculate priority sequence, according to calculating priority level, arranges the sequencing of calculation task; For complete operation, with weight factor P zero setting, when new operation begins to carry out, will be again by weight factor P to task ranking, because of as calculated operation weight factor P is 0, be placed to the descending end;

Set up deque's administrative power repeated factor, the weight factor that one of them queue management is provided with, the variation of the weight factor when another is used for Management Calculation.

8. computing method as claimed in claim 7 is characterized in that, described step 5. in according to dynamic change priority weight system, realize the serializing cascade dynamic execution of computing operation in batches, it carries out weight by the [p of deque ₁, p ₂...., p _N] ^TColumn vector control, static reaction is manual sets control, in dynamic response execution and the computation process to the not dynamic execution sequence change of computational tasks, the performance coeffcient p that complete transient stability is calculated _iZero setting to unenforced operation, when determining operation next time, realizes that by the column vector descending bubbling of computational tasks is selected.

9. computing method as claimed in claim 4 is characterized in that, the transient stability that provides in the described step (c) calculates and finishes to judge that the time set of single detect cycle is 1 second.

Images