CN111767338A - Distributed data storage method and system for online ultra-real-time simulation of power system - Google Patents

Abstract

The present invention provides a distributed data storage method and system for power system online ultra-real-time simulation, including: constructing a distributed power system online ultra-real-time simulation system including metadata nodes and data storage nodes; scheduling requests for storage resources according to simulation tasks and the available storage space of each data storage node, the metadata node allocates two independent data storage nodes as the original storage node and the summary storage node; continuously and periodically collects the status data of the power grid operation from the power grid, and sends the status data to The original data storage node and data storage node, each data storage node runs the online ultra-real-time simulation of the power system according to the simulation task and state data, and obtains local ultra-real-time simulation results; summarizes the local simulation results of each data storage node to the summary storage node, and obtains the power grid. full hyperreal-time simulation results. The present invention can provide data storage management for simulation while running power system simulation.

Description

Distributed data storage method and system for online ultra-real-time simulation of power system

technical field

The invention relates to the field of on-line super-real-time simulation of power systems in system simulation, and particularly relates to a method and system for on-line super-real-time simulation of power systems based on distributed data.

Background technique

Online ultra-real-time simulation of power system refers to the continuous periodic (10-second cycle) collection of power grid operating status data from the interconnected large power grid through SCADA, RTU and other online measurement equipment, and transmitted back to the data center through high-speed interconnection network, and then through the power grid. The system simulation software system quickly calculates and analyzes the online measurement data, and then judges the stability and safety of the current operating state of the interconnected large power grid, and accurately predicts the operating state for a short period of time (within a few seconds) in the future. Generally, the regional backbone power grid generates hundreds of megabytes of measurement data in each measurement cycle. The calculation and analysis of the power system simulation software will generate gigabytes of calculation result data. There is significant value in management, so efficient storage management of this data is required. At present, disk arrays are mostly used in the storage of online ultra-real-time simulation big data of power systems, and disk arrays are generally expensive and have limited scalability. Although the existing distributed storage systems on the market are hot, there are A small number of applications, but there is a big bottleneck in storage I/O efficiency, and general-purpose commercial products are generally less cost-effective.

In the face of massive and rapidly growing power system simulation data, the scale and cost of data storage is one of the key issues to be solved. In order to solve reliability problems, traditional distributed storage systems generally improve data storage reliability through data redundancy (such as multiple copies, erasure codes, etc.), which further reduces storage space efficiency and increases storage costs. In the online ultra-real-time simulation environment of the power system, in addition to the higher requirements for the storage I/O performance, the storage space efficiency also needs to be considered. There are thousands of files in the simulation input and output data. The content and value of the data stored in each file are different. For example, some files store relatively static configuration information, and some files store relatively static configuration information. It is periodically generated online measurement data such as voltage, current, power, etc. The data stored in some files needs to be stored for a long time for subsequent analysis and mining, and the data stored in some files can be repeatedly generated through simulation calculations. Therefore, it is generally required that historical data can be stored for different periods of time according to different value levels, such as three months or six months, or even one year. In addition, in order to make the online ultra-real-time simulation of the power system run stably for a long time, in the event of a single server failure, single disk failure or full disk storage space, the system can have fault isolation and self-healing capabilities to ensure that the system can continue to Runs online without being affected.

The existing distributed storage systems on the market are rarely used in the field of power system simulation. On the one hand, it is due to its high cost and high maintenance threshold; on the other hand, its I/O throughput is difficult to meet the requirements of online ultra-real-time simulation. Performance requirements often become the performance bottleneck of simulation computing. This is because, during the calculation process of the power system simulation software system, a large number of intermediate results and final calculation results will be generated, and these data are often output in the same time period, that is, a high I/O load is generated in a short period of time. Some distributed storage systems are generally difficult to deal with such application challenges. In addition, the independent distributed storage cluster occupies a large amount of computer room space, which greatly increases the energy consumption of the computer room. Therefore, it is generally rarely used in actual production.

SUMMARY OF THE INVENTION

The purpose of the present invention is to solve the problems of I/O performance and high cost in the field of power system simulation with the above-mentioned existing distributed storage technology, and proposes an online ultra-real-time simulation big data based on computing-storage coupling architecture power system A method and system for distributed storage and management.

In view of the deficiencies of the prior art, the present invention proposes a distributed data storage method for on-line ultra-real-time simulation of power systems, including:

Step 1. Build a distributed power system online ultra-real-time simulation system including a metadata node and multiple data storage nodes;

Step 2. According to the storage resource scheduling request of the simulation task and the available storage space of each data storage node, the metadata node allocates two mutually independent data storage nodes as the original storage node and the summary storage node respectively;

Step 3. Continuously and periodically collect the status data of the power grid operation from the power grid through the online measurement equipment, and send the status data to the original data storage node and the data storage node. Each data storage node is based on the simulation task and the status data. Run the online ultra-real-time simulation of the power system to obtain local ultra-real-time simulation results;

Step 4: Aggregate the local simulation results of each data storage node to the aggregated storage node to obtain a complete ultra-real-time simulation result of the power grid.

The distributed data storage method for online ultra-real-time simulation of the power system, wherein the data storage node uses distributed message queues to realize real-time transmission of real-time data, and uses a distributed storage system based on a centralized structure to complete non-real-time data storage. Persistent storage and access of data.

In the distributed data storage method for on-line ultra-real-time simulation of power systems, the data storage nodes use TCP protocol to exchange data.

In the distributed data storage method for online ultra-real-time simulation of power systems, the metadata node includes a metadata management module, which is used for the elimination management of data in the disk and the elimination of data in the memory.

The distributed data storage method for online ultra-real-time simulation of power systems, wherein the metadata management module will update the used space of the corresponding disk in the database after receiving the signal that data persistence is completed, and when the used space is larger than a preset When the value is set, query the database table to find the data with the earliest storage time; send a signal to the data storage node where the earliest data is located; after receiving the signal, the data storage node deletes the relevant data according to the instruction, and sends a successful deletion to the metadata management module instruction.

The present invention also proposes a distributed data storage system for online ultra-real-time simulation of power systems, including:

Module 1. Build a distributed power system online ultra-real-time simulation system including a metadata node and multiple data storage nodes;

Module 2. According to the storage resource scheduling request of the simulation task and the available storage space of each data storage node, the metadata node allocates two mutually independent data storage nodes as the original storage node and the summary storage node respectively;

Module 3. Continuously and periodically collect the status data of the power grid operation from the power grid through the online measurement equipment, and send the status data to the original data storage node and the data storage node, and each data storage node is based on the simulation task and the status data. Run the online ultra-real-time simulation of the power system to obtain local ultra-real-time simulation results;

Module 4: Aggregate local simulation results of each data storage node to the aggregated storage node to obtain a complete ultra-real-time simulation result of the power grid.

The distributed data storage system for the online ultra-real-time simulation of the power system, wherein the data storage node uses distributed message queues to realize real-time transmission of real-time data, and uses a distributed storage system based on a centralized structure to complete non-real-time data. Persistent storage and access of data.

In the distributed data storage system for online super-real-time simulation of the power system, the data storage nodes use the TCP protocol to exchange data.

In the distributed data storage system for the online ultra-real-time simulation of the power system, the metadata node includes a metadata management module, which is used for the elimination management of the data in the disk and the elimination management of the data in the memory.

The distributed data storage system for the online super-real-time simulation of the power system, wherein the metadata management module will update the used space of the corresponding disk in the database after receiving the signal that the data persistence is completed. When the used space is larger than the preset When the value is set, query the database table to find the data with the earliest storage time; send a signal to the data storage node where the earliest data is located; after receiving the signal, the data storage node deletes the relevant data according to the instruction, and sends a successful deletion to the metadata management module instruction.

As can be seen from the above scheme, the advantages of the present invention are:

First, it can make full use of the computing and storage capabilities of the server, provide storage management of massive data for the simulation while running the power system simulation calculation, save the computer room space, and reduce energy consumption. Second, the memory cache and the principle of locality can be used to greatly reduce the storage I/O load during the running of the emulator, shorten the storage I/O path, and improve the running efficiency of the emulator. Third, it is possible to customize the storage reliability strategy of file granularity according to the value level of the data, utilize the storage space reasonably and effectively, comprehensively consider the value level of the data, the number of copies, the generation time and other indicators, and dynamically calculate the data elimination priority to ensure the storage system. The data can be automatically eliminated according to reasonable rules to provide usable storage capacity and ensure the continuous online operation of power system simulation.

Description of drawings

Figure 1 is a software architecture diagram of a storage system;

Fig. 2 is the composition diagram of power system simulation data;

Fig. 3 is the logical diagram of distributed storage system deployment;

Fig. 4 is the working flow chart of data storage operation;

Fig. 5 is the original data storage work flow chart;

Fig. 6 is the summary data storage work flow chart;

Fig. 7 is the data storage state change diagram on the summary data storage node;

FIG. 8 is a diagram of data storage state changes on a non-aggregated data storage node.

Detailed ways

In order to realize the above-mentioned technical effect, the present invention includes the following three key points:

Key point 1, power system simulation computing and storage service symbiosis and resource coordinated scheduling technology. Technical effect: Simultaneous calculation and data storage management can be performed on the same high-performance server cluster, and each server runs both the simulation calculation program and the storage service program. This can bring two advantages: in terms of economy, it can make full use of the computing and storage capabilities of the server, reduce the scale of server clusters, and reduce the space occupation and energy consumption of the computer room; in terms of performance, it can improve the simulation data storage I// The locality of O shortens the data movement path and improves the I/O performance. At the same time, through the coordinated scheduling of computing and storage resources, the processes responsible for the storage I/O input/output function in the simulation program are distributed to different computing nodes. Run, distribute the I/O load of each server, avoid I/O hotspots on a few servers, and affect the simulated I/O performance;

Key point 2, memory-based storage I/O acceleration technology; technical effect: During the periodic operation of the power system simulation program, it is necessary to periodically read the original input data, configuration information, exchange intermediate result data, and output result data. I/O-intensive operations such as disk I/O occupy a large proportion of the running time of the simulation program, which has a non-negligible impact on the performance of the simulation program. In order to solve this problem, this scheme uses the memory file system as input and output. The data transfer cache layer temporarily stores the original input data and configuration information in the memory file system for the simulation program to call. The intermediate result data and calculation result data generated during the running of the simulation program are also written to the memory file system. In , the storage service asynchronously writes the data that needs to be persisted to the disk of the local or other servers, thereby avoiding I/O, which is called the bottleneck of the simulation program computing process. In order to ensure the running safety and performance of the emulator, the upper limit of memory usage is set for the memory file system. When the delay of data writing to the disk is large and the memory file system is about to reach the upper limit of memory usage), the data that can be discarded will be calculated through a certain strategy, thereby freeing up memory space and ensuring the continuity of the power system online ultra-real-time simulation program. Stable operation;

Key point 3, the fine-grained hybrid reliability strategy and data elimination dynamic priority mechanism based on files; technical effect: different reliability guarantee strategies can be designed according to the value level of the power system simulation data itself, and these different reliability Sexual assurance policies coexist in the same data storage management system and can be set and adjusted as needed. For example, for the simulation original input data, its value level is high, in order to ensure its reliability, it can be stored in three copies. For the simulation calculation result data, since it can be calculated according to the corresponding original input data, therefore, The reliability requirements for storage are low, and a single copy can be used for storage. For a specific power system simulation platform, since the storage space capacity of the server cluster is limited, the data that can be stored is also limited. In order to always ensure that the newer simulation data can be stored in a timely and reliable manner, it is necessary to Eliminate old data stored in the system. However, due to the different value levels, usage requirements, and usage methods of different data, the corresponding elimination priorities are also different. For example, the value level of the simulation input data is higher, and the old data should be eliminated as much as possible to release more available storage space. Retain such data, data generated within 1 day should be saved as much as possible due to the high probability of being accessed, and so on. To this end, the dynamic priority of data elimination is calculated by comprehensively considering the indicators of the three dimensions of data value level, number of copies, and generation time, and the data is eliminated according to the priority.

In order to make the above-mentioned features and effects of the present invention more clearly and comprehensible, embodiments are given below, and detailed descriptions are given below in conjunction with the accompanying drawings.

The online ultra-real-time power system simulation data distributed storage management system is mainly divided into three parts: data storage layer, communication service layer and interface layer, as shown in Figure 1.

(1) Data storage layer

Since the data to be stored is divided into two types, real-time data and non-real-time data, it will correspond to two different data storage services. For real-time messages, a message queue will be used to realize real-time data transmission, and the message queue can be a distributed message queue based on the memory of multiple storage nodes; for non-real-time messages, a distributed storage system based on a centralized structure will be used to complete the Persistent storage and access of data.

(2) Communication service layer

In order to ensure the reliability of the data, the TCP protocol is used for data exchange.

(3) Interface layer

In order to facilitate the human-computer interaction terminal to read data, a series of data access interfaces will be provided.

A power system simulation job will include multiple computing tasks, denoted as N computing tasks, and the calculation results will form N simple summaries, N detailed results, N final summaries, and N difference files. In addition, N tasks share the same input data, as shown in Figure 2.

The deployment logic of the distributed storage system is shown in Figure 3. The system includes two parts: metadata nodes and storage nodes. The storage of metadata is based on relational databases, such as MySQL. In order to ensure the reliability of metadata storage management, metadata storage can be Using the HA architecture, the storage node also acts as a computing node running the simulation program, that is, the storage management system and the simulation computing system run on the same physical computing node. The metadata node records metadata information such as the storage node where the data is located, the disk (or storage path) of the storage node, and is also responsible for scheduling storage requests, that is, the metadata node decides which storage node to store the data on. .

The data storage operation workflow of the distributed storage system is mainly divided into the following steps:

1) Request storage resources

The simulation calculation scheduling node sends a storage resource scheduling request to the metadata node, and requests the metadata node to allocate storage space for the data of this simulation calculation. The metadata node manages storage resources in units of disks, and selects the disk with the most free space according to the current status of the available disks in the storage cluster and the available storage space. When the metadata node allocates storage resources, it needs to allocate two independent disks, one for storing summary data packets and the other for storing original data packets. After the selection is completed, the metadata management module sends the <summary data disk ID, task ID> to the summary data storage node, and simultaneously sends the <original data disk ID, task ID> to the original data storage node. After the relevant nodes receive the notification , confirm your role, and confirm the final persistent path of the data according to the disk ID, and then enter the data aggregation stage.

2) Packet summary

Since a copy of the original data is cached in the in-memory file system of each compute node, no aggregation is required. After receiving the notification <summary data disk ID, task ID>, the original data storage node sends a confirmation message to the metadata management module of the primary data node, and the metadata management module sets the read status of the task data in the data table to ramfs available state, after which the metadata management module synchronously updates the database tables from the metadata node. The raw data storage workflow is shown in Figure 5.

The summary process of the summary data packet is as follows, as shown in Figure 6:

(1) After the summary storage node receives the summary command issued by the metadata management module, it sends a collection data packet instruction to the storage node where the IO process is located;

(2) The summary node waits to recycle the data, and the storage node where the IO process is located sends the data to the summary node;

(3) The collection of the summary node is completed, and a summary completion signal is sent to the metadata management module on the main metadata node;

(4) The metadata management module updates the data to the available state of ramfs, and synchronizes the data to the slave metadata node.

3) Data persistence.

After the summary data storage node and the original data storage node are persisted to the specified disk directory, the storage node updates the available status of the data package (the data is accessible in the disk) to the primary metadata node, and at the same time updates the used space of the corresponding disk.

In order to automatically maintain the availability of storage space, the system provides a data elimination function, which involves two aspects: elimination management of data in disk and elimination management of data in memory. In order to realize the automatic and reasonable elimination of the aggregated data, the storage state of the aggregated data is managed, as shown in Figure 7. Figure 8 shows the status changes of related aggregated data on the non-aggregated data storage node (mainly related to the data elimination management of the memory space).

The automatic retirement process for data stored on disk is as follows:

Details such as the number of disks, disk space, and used disk capacity of each node are recorded in the database. The cleanup of disk space will be initiated by the metadata management module. The conditions for triggering the metadata management module to initiate data cleaning include the following strategies:

(1) Calculate the space utilization of each disk, when the average value of the disk utilization of all nodes exceeds the configured threshold;

(2) Take all disks as a whole and calculate the overall disk utilization, when the utilization exceeds the configured threshold;

After receiving the signal that data persistence is complete, the metadata management module will update the used space of the corresponding disk in the database. If it is found that the trigger condition is met, the following steps will be performed in sequence:

(1) Query the database table to find out the global oldest N (configurable) pieces of data.

(2) The metadata management module sends a signal to the node where the oldest data is located. This signal is <task data ID list to delete>.

(3) After the relevant node receives the signal, it deletes the relevant data according to the instruction, and sends a successful deletion instruction to the metadata management module.

(4) The metadata management module updates the database table: set the data to a clean state, and update the related disk utilization.

(5) Synchronize data to the slave metadata node.

The automatic elimination process of data cached in memory is as follows:

Taking into account the normal operation of the system, the proportion of the physical space occupied by the mounting space of the memory file system will be configured according to the configuration options in the configuration file. At the same time, there will be a memory file system cleaning module to clean up related data, and the cleaning cycle is 8s (optional).

The main functions of the memory file system cleaning module are as follows:

(1) Record the status of all data in the memory file system. Summary data has three states: summary data receiving, summary data persisting, and summary data disk available; raw data has three states: original data receiving, in RAMFS, original data persistence, and original data disk available. (A separate file is maintained for each piece of data on the storage node, and the state of the corresponding data is saved in the file. When modifying the file, that is, modifying/deleting the data, you need to lock the file first. , which guarantees that the modification operation is atomic.)

(2) This module will detect the memory file system space to ensure that the usage rate (used space/total space) of the memory file system does not exceed 75% (this item can be configured).

(3) After reaching the threshold, the oldest aggregated data (the aggregated data must be older than the output data that has not been aggregated), the oldest original data and the oldest output data will be deleted directly.

The following are system embodiments corresponding to the foregoing method embodiments, and this implementation manner may be implemented in cooperation with the foregoing implementation manners. The related technical details mentioned in the foregoing embodiment are still valid in this embodiment, and are not repeated here in order to reduce repetition. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied to the above-mentioned embodiments.

The present invention also proposes a distributed data storage system for online ultra-real-time simulation of power systems, including:

Module 1. Build a distributed power system online ultra-real-time simulation system including a metadata node and multiple data storage nodes;

Module 2. According to the storage resource scheduling request of the simulation task and the available storage space of each data storage node, the metadata node allocates two mutually independent data storage nodes as the original storage node and the summary storage node respectively;

Module 3. Continuously and periodically collect the status data of the power grid operation from the power grid through the online measurement equipment, and send the status data to the original data storage node and the data storage node, and each data storage node is based on the simulation task and the status data. Run the online ultra-real-time simulation of the power system to obtain local ultra-real-time simulation results;

Module 4: Aggregate local simulation results of each data storage node to the aggregated storage node to obtain a complete ultra-real-time simulation result of the power grid.

The distributed data storage system for the online ultra-real-time simulation of the power system, wherein the data storage node uses distributed message queues to realize real-time transmission of real-time data, and uses a distributed storage system based on a centralized structure to complete non-real-time data. Persistent storage and access of data.

In the distributed data storage system for online super-real-time simulation of the power system, the data storage nodes use the TCP protocol to exchange data.

In the distributed data storage system for on-line ultra-real-time simulation of the power system, the metadata node includes a metadata management module, which is used for the elimination management of data in the disk and the elimination of data in the memory.

The distributed data storage system for the online super-real-time simulation of the power system, wherein the metadata management module will update the used space of the corresponding disk in the database after receiving the signal that the data persistence is completed. When the used space is larger than the preset When the value is set, query the database table to find the data with the earliest storage time; send a signal to the data storage node where the earliest data is located; after receiving the signal, the data storage node deletes the relevant data according to the instruction, and sends a successful deletion to the metadata management module instruction.

Claims (10)

1. a distributed data storage method of power system online ultra-real-time simulation, is characterized in that, comprises: Step 1. Build a distributed power system online ultra-real-time simulation system including a metadata node and multiple data storage nodes; Step 2. According to the storage resource scheduling request of the simulation task and the available storage space of each data storage node, the metadata node allocates two mutually independent data storage nodes as the original storage node and the summary storage node respectively; Step 3. Continuously and periodically collect the status data of the power grid operation from the power grid through the online measurement equipment, and send the status data to the original data storage node and the data storage node. Each data storage node is based on the simulation task and the status data. Run the online ultra-real-time simulation of the power system to obtain local ultra-real-time simulation results; Step 4: Aggregate the local simulation results of each data storage node to the aggregated storage node to obtain a complete ultra-real-time simulation result of the power grid. 2. The distributed data storage method of power system online ultra-real-time simulation as claimed in claim 1, is characterized in that, this data storage node adopts distributed message queue to realize real-time transmission to real-time data, adopts centralized structure-based The distributed storage system completes the persistent storage and access functions for non-real-time data. 3 . The distributed data storage method for on-line ultra-real-time simulation of a power system according to claim 1 , wherein the data storage nodes adopt TCP protocol to exchange data. 4 . 4. The distributed data storage method of power system online ultra-real-time simulation as claimed in claim 1, it is characterized in that, this metadata node comprises metadata management module, is used for the elimination management of data in disk and the elimination of data in memory manage. 5. The distributed data storage method of power system online ultra-real-time simulation as claimed in claim 4, it is characterized in that, this metadata management module can update the use of corresponding disk in database after receiving the signal that data persistence is completed space, when the used space is greater than the preset value, query the database table to find the data with the earliest storage time; send a signal to the data storage node where the earliest data is located; after the data storage node receives the signal, delete the relevant data according to the instructions, And send a deletion success instruction to the metadata management module. 6. A distributed data storage system of power system online ultra-real-time simulation, is characterized in that, comprises: Module 1. Build a distributed power system online ultra-real-time simulation system including a metadata node and multiple data storage nodes; Module 2. According to the storage resource scheduling request of the simulation task and the available storage space of each data storage node, the metadata node allocates two mutually independent data storage nodes as the original storage node and the summary storage node respectively; Module 3. Continuously and periodically collect the status data of the power grid operation from the power grid through the online measurement equipment, and send the status data to the original data storage node and the data storage node, and each data storage node is based on the simulation task and the status data. Run the online ultra-real-time simulation of the power system to obtain local ultra-real-time simulation results; Module 4: Aggregate local simulation results of each data storage node to the aggregated storage node to obtain a complete ultra-real-time simulation result of the power grid. 7. The distributed data storage system of power system online ultra-real-time simulation as claimed in claim 6, is characterized in that, this data storage node adopts distributed message queue to realize real-time transmission of real-time data, adopts centralized structure-based The distributed storage system completes the persistent storage and access functions for non-real-time data. 8 . The distributed data storage system for on-line ultra-real-time simulation of power systems according to claim 6 , wherein the data storage nodes use TCP protocol to exchange data. 9 . 9. The distributed data storage system of power system on-line ultra-real-time simulation as claimed in claim 6, it is characterized in that, this metadata node comprises metadata management module, is used for the elimination management of data in disk and the elimination of data in memory manage. 10. The distributed data storage system for online ultra-real-time simulation of power systems as claimed in claim 9, wherein the metadata management module updates the usage of the corresponding disks in the database after receiving the signal that data persistence is completed. space, when the used space is greater than the preset value, query the database table to find the data with the earliest storage time; send a signal to the data storage node where the earliest data is located; after the data storage node receives the signal, delete the relevant data according to the instructions, And send a deletion success instruction to the metadata management module.

Images