Nuclear power simulation model development data management method and system
By dividing nuclear power simulation model data into subdomains and managing them according to access frequency and compression ratio, the problems of I/O load imbalance and low cache hit rate during nuclear power simulation model data storage are solved, achieving efficient data management.
Patent Information
- Application Number
- CN202511492943.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-20
- Publication Date
- 2025-11-18
- Estimated Expiration
- 2045-10-20
AI Technical Summary
Managing nuclear power simulation model data in a uniform format leads to I/O load imbalance during storage and reduced cache hit rate during data retrieval.
The simulation mesh is divided into different subdomains, and I/O bandwidth is allocated according to the number of cells in each subdomain. The data preloading method is determined by calculating the effective access frequency, hot spot centers, and predicted access probability. Data management is carried out in combination with dynamic compression ratio, and data storage and retrieval are optimized by using multi-layer storage.
It enables dynamic adjustment and load balancing of I/O load, improves the accuracy and efficiency of data management, and increases cache hit rate.
Smart Images

Figure CN120973478A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, in particular to a nuclear power simulation model development data management method and system. BACKGROUND
[0002] Nuclear power simulation model data management refers to a series of operations such as effective organization, storage, access, update and maintenance of various data related to simulation models in the process of nuclear power plant simulation, so as to ensure the accuracy, integrity and availability of data, thereby supporting the development, operation, maintenance and optimization of simulation models. Generally, a simulation calculation needs to iterate hundreds of thousands of time steps on a grid of hundreds of millions of levels, resulting in super large data. When managing and storing such super large data, all original data sets of subdomains are usually written into one or a few super large files, and are compressed and stored in a unified fixed format for long-term preservation.
[0003] However, there are spatial and temporal differences between the local data of nuclear power simulation model data. Specifically, the cell size, physical field gradient and access frequency of different subdomains often cause spatial differences between local data, and the data change amplitude of each subdomain between adjacent time steps often causes temporal differences between local data. However, the nuclear power simulation model data management directly manages the data of different subdomains according to the same format, which may cause I / O load imbalance during storage, and also reduce cache hit rate during retrieval. SUMMARY
[0004] The present application provides a nuclear power simulation model development data management method and system to solve the problem that the data of nuclear power simulation model is managed according to a unified format, resulting in I / O load imbalance during data storage and reducing cache hit rate during data retrieval. The technical solution adopted is as follows: In a first aspect, an embodiment of the present application provides a nuclear power simulation model development data management method, which includes the following steps: Extracting the development data of the nuclear power simulation model, dividing the simulation grid into different subdomains, and allocating the I / O bandwidth of the subdomain according to the number of cells of the subdomain; According to the number of times the subdomain is accessed in all time steps within a preset time window before the current time step and the current time step, determining the effective access frequency of the subdomain in the current time step, combining the number of times all time steps in the time window are accessed to determine the weighted effective access frequency of the time window, and calculating the hot spot center according to the effective access frequency, determining the predicted access probability of future time steps according to the distance between the geometric center coordinates of the subdomain and the hot spot center, and the interval between the current time step and the future time step, and determining the data of the time step preloaded to the first storage layer according to the predicted access probability; Based on the differences between data in adjacent time steps and the volume differences between different subdomains, the dynamic compression ratio of the subdomain in the current time step is determined. Based on the dynamic compression ratio of the subdomain in the time step, the compression method of the data in the subdomain in adjacent time steps is determined, and the compression result is written to the second storage layer. Combining the effective access frequency of the subdomain in the current time step, a portion of the data in the second storage layer is selected for compression and written to the third storage layer, thereby realizing the management of nuclear power simulation model development data.
[0005] Furthermore, the formula for calculating the effective access frequency of the subdomain at the current time step is: in, Indicates the first Each subdomain at the current time step Valid access frequency; Indicates the length of the preset time window; Indicates the current time step; Indicates the first The subdomain in the first The number of times a step is accessed; Represents the natural constant; This indicates the preset attenuation coefficient.
[0006] Furthermore, the specific method for obtaining the weighted effective access frequency of the time window is as follows: The effective access frequency of a time step is used as the weight of the number of times a subdomain is accessed in a time step. The number of times all time steps are accessed within a preset time window is weighted and summed to obtain the weighted effective access frequency of the time window.
[0007] Furthermore, the specific method for calculating hotspot centers based on effective access frequency includes: Calculate the geometric center coordinates of the subdomains, use the effective access frequency of the time step as the weight of the geometric center coordinates of the subdomains, and sum the weighted geometric center coordinates of all subdomains to obtain the hot spot center.
[0008] Furthermore, the formula for calculating the predicted access probability at the future time step is: in, Indicates the first Each subdomain at the current time step The next future Predicted access probability at each time step; Indicates the current time step The next future One time step; Indicates the first Each subdomain at the current time step a Euclidean distance between the geometric center coordinate of the time step and the hotspot center; represents a normalization function.
[0009] Further, the method of determining the data to be preloaded to the first storage layer according to the predicted access probability comprises: a third quartile of the predicted access probabilities of all future time steps after the current time step is recorded as a second division threshold, and the data of the future time steps after the current time step with the predicted access probability greater than the second division threshold is preloaded to the first storage layer.
[0010] Further, the specific method of obtaining the dynamic compression ratio of the subdomain at the current time step comprises: a difference square sum of the corresponding field variables of the corresponding cells of the same subdomain at the current time step and the previous adjacent time step of the current time step is recorded as a first square sum of the corresponding cells of the same subdomain at the current time step, the volume of the cell is taken as a weight of the first square sum of the cell, the first square sums of all cells of the same subdomain at the current time step are weighted and summed, and a normalized value of the result of the weighted sum is recorded as a relative adjacent time step difference of the same subdomain at the current time step. a median of the relative adjacent time step differences of all subdomains at the current time step is recorded as a sensitivity parameter of the current time step, and a sum of the relative adjacent time step difference of the subdomain at the current time step and the sensitivity parameter of the current time step is recorded as a relative relative adjacent time step difference of the subdomain at the current time step. a maximum value of the relative relative adjacent time step difference of the subdomain at the current time step and the third division threshold is recorded as the dynamic compression ratio of the subdomain at the current time step.
[0011] Further, the specific method of determining the compression mode of the data of the subdomain at the adjacent time step according to the dynamic compression ratio of the subdomain at the time step comprises: a difference of the corresponding field variables of the same subdomain at the current time step and the previous adjacent time step of the current time step is recorded as a difference data block of the same subdomain at the current time step. when the dynamic compression ratio of the subdomain at the current time step is greater than or equal to the fourth division threshold, the difference data block is losslessly compressed; when the dynamic compression ratio of the subdomain at the current time step is greater than or equal to the third division threshold and less than the fourth division threshold, the dynamic compression ratio of the subdomain at the current time step is taken as a value of the compression ratio, the difference data block is lossily quantized, and the lossily quantized difference data block is losslessly compressed.
[0012] Further, the specific method of selecting part of the data of the second storage layer for compression comprises: The first sixth quantile of the effective access frequency of all sub-domains at the current time step is recorded as a fifth division threshold, and the data of the sub-domains with the effective access frequency less than the fifth division threshold at the corresponding time step and all time steps before the corresponding time step is compressed.
[0013] In a second aspect, the embodiments of the present application also provide a nuclear power simulation model development data management system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of the method according to any one of the preceding aspects when executing the computer program.
[0014] The present application has the following beneficial effects: The present application first allocates I / O bandwidth to the sub-domains according to the number of units of the sub-domains, realizes dynamic adjustment of I / O load, eliminates file lock competition and realizes load balancing, and due to the time locality of the access heat of the nuclear power simulation model, the read-write performance closer to the current step can represent the access tendency of the next step, the weighted effective access frequency of the time window is determined by weighting the access frequency according to the interval between different time steps in the time window, and the region that has cooled down is avoided from being misjudged as a hot spot, then, in order to avoid blind prefetching and waste of cache, the prediction access probability of the future time step is determined according to the distance between the geometric center coordinate of the sub-domain and the hot spot center and the interval between the current time step and the future time step, so that the prediction access probability meets the condition that the units closer to the current hot spot center in space have a higher probability of being accessed in the next step, and the units closer to the current time in time have a higher probability of being accessed, and the data of the time step preloaded to the first storage layer is determined according to the prediction access probability; further, the compression ratio of each time step is adaptively determined according to the difference between the development data of adjacent time steps, that is, the dynamic compression ratio of the sub-domain at the current time step is determined, and compression is performed according to the adaptively determined compression ratio, and the compression result is written into the second storage layer; finally, when the data of the sub-domain is not frequently accessed for a long time, the necessity of real-time calculation and high-frequency analysis of the data of the time step corresponding to the sub-domain is relatively low, in order to speed up the read-write of other sub-domain data with relatively high necessity, and at the same time, in order to save storage space, the historical differential data of these sub-domains is more strongly compressed, and the compression result is stored in the third storage layer, realizing the management of the nuclear power simulation model development data, solving the problem that the data of the nuclear power simulation model is managed according to a unified format, resulting in I / O load imbalance during data storage and low cache hit rate during data retrieval, and improving the management accuracy and management efficiency of the nuclear power simulation model data. BRIEF DESCRIPTION OF DRAWINGS
[0015] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description only only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor.
[0016] Figure 1 A flowchart of a nuclear power simulation model development data management method provided by an embodiment of the present application; Figure 2 An effective access frequency acquisition flowchart provided by an embodiment of the present application. DETAILED DESCRIPTION
[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.
[0018] Please refer to Figure 1 which shows a flowchart of a nuclear power simulation model development data management method provided by an embodiment of the present application. The method comprises the following steps: Step S001, extracting development data of a nuclear power simulation model, dividing a simulation grid into different subdomains, and distributing I / O bandwidth of the subdomains according to the number of units of the subdomains.
[0019] Extracting development data of a nuclear power simulation model. If the development data of the nuclear power simulation model is read and written according to a centralized single large file, all processes will compete for the same POSIX file lock, resulting in queuing and waiting. In order to shorten I / O delay and eliminate competition for file locks, a parallel way is adopted to read and write the development data of each subdomain. Specifically, the simulation grid is divided into different subdomains according to spatial connectivity, and the development data of different subdomains is written into different files.
[0020] Among them, I / O load imbalance refers to the uneven distribution of data input and output operations among storage devices or hardware components in a computer system, resulting in some devices becoming bottlenecks due to excessive workload, while other devices are not fully utilized.
[0021] In the finite element solver, the number of field variables and the storage bytes of each element are basically fixed, so the number of elements and the final I / O byte number are in linear relationship, the number of elements can be used as weight to allocate I / O task, and the number of elements is an existing integer field in the grid file, without sampling. In order to ensure that the reading and writing between the subdomains are conflict-free, the simulation grid is divided according to the spatial connectivity of the grid, the grid is regarded as an undirected graph in the pre-processing stage, and the graph partitioning algorithm is used to obtain the element-level partition mapping table, and then the elements are sent to each subdomain according to the partition mapping table in the solver.
[0022] Specifically, preferably, as an embodiment of the present application, the element file is exported in the pre-processing tool, the adjacent surface of each element is parsed using PyMesh, the undirected graph of the element and the adjacent element is established, the partition is performed according to the spatial connectivity using ParMETIS, all subdomains are obtained, a separate binary file is created for each subdomain, and a mapping file corresponding to the element and the subdomain is generated. In the embodiment, ANSYS Meshing is selected as the pre-processing tool; the number of subdomains is set to 128, the number of subdomains is set according to the number of elements in the file, so that there are elements in each subdomain; 128 MPI processes are started in the solver, each process reads and writes the file according to the mapping file for different subdomains; the adjacent surface of each element is parsed using PyMesh, and the partition is performed according to the spatial connectivity using ParMETIS, which are both known technologies and will not be described in detail.
[0023] The I / O bandwidth is allocated according to the number of elements of each subdomain, and the I / O bandwidth of the subdomain is proportional to the number of elements of the subdomain in the cumulative sum of the number of elements of all subdomains. Therefore, the more the number of elements of the subdomain, the greater the I / O bandwidth.
[0024] After each time step or working condition is completed, the actual time consumption of each subdomain is extracted, if the actual time consumption of the subdomain is greater than the second quartile of the time consumption average, the subdomain is re-partitioned and the mapping file is rewritten before the next iteration using ParMETIS, to continuously maintain the I / O load balance.
[0025] It can be understood that the re-partitioning of the subdomain and the rewriting of the mapping file can realize dynamic adjustment of the I / O load, eliminate file lock competition and realize load balancing.
[0026] Thus, the development data of the nuclear power simulation model, the subdomains and the I / O bandwidth of the subdomains are obtained.
[0027] Step S002, determining the effective access frequency of the sub-domain at the current time step according to the number of times that the sub-domain is accessed at all time steps within a preset time window before the current time step and at the current time step, determining the weighted effective access frequency of the time window in combination with the number of times that all time steps within the time window are accessed, and calculating the hotspot center according to the effective access frequency, determining the predicted access probability of the future time step according to the distance between the geometric center coordinate of the sub-domain and the hotspot center and the interval between the current time step and the future time step, and determining the data of the time step preloaded to the first storage layer according to the predicted access probability.
[0028] Each development data has strong spatiotemporal continuity, and when the solver calculates a current cell, the values of all adjacent cells of the current cell must be accessed. If a cell is accessed, its adjacent cells will also be accessed at the same time step or the next step. Therefore, the current access hotspot region may still be a hotspot at the next time step.
[0029] The time advancement of the nuclear power simulation model usually adopts small step display or semi-implicit recommendation. The solutions of the same physical field are almost continuous within adjacent time steps, so the solver needs to repeatedly read the field variables of the same batch of cells and their adjacent cells, which leads to a much higher access frequency of some sub-domains than that of other sub-domains. If the traditional on-demand reading is adopted, since the adjacent development data may be written into different sub-domain files, each reading will cause high latency due to cross-node data pulling. Therefore, the access of each sub-domain and the spatiotemporal characteristics of the sub-domain are needed to be statistically calculated to calculate the access probability of the next step. When the probability exceeds a threshold, the data of the sub-domain is pulled into the cache in advance, and the cache capacity is managed by LRU.
[0030] Since the access hotness of the nuclear power simulation model has obvious temporal locality, the read-write performance closer to the current step can better represent the access tendency of the next step. Therefore, the access frequency needs to be weighted according to the interval between different time steps within a time window to avoid misjudging a cooled region as a hotspot. Specifically, the application adopts a weight that decays over time, so that the contribution weight of the historical access times quickly decays, thereby retaining the contribution weight of the recent access times, so that the effective access frequency calculation is more in line with the actual situation.
[0031] The effective access frequency of the sub-domain at the current time step is determined according to the number of times that the sub-domain is accessed at all time steps within a preset time window before the current time step and at the current time step.
[0032] wherein, represents the number of times that the i-th sub-domain is accessed at the current time step t and at all time steps within a preset time window before the current time step t. Valid access frequency; This indicates the length of the preset time window. In this embodiment, the length of the time window is set to 100. Indicates the current time step; Indicates the first The subdomain in the first The number of times a step is accessed; Represents the natural constant; This represents the preset attenuation coefficient, which is used to control the rate at which historical access counts are forgotten. The value of the attenuation coefficient should be greater than or equal to 0.05 and less than or equal to 0.1. In this embodiment, the attenuation coefficient is set to 0.05.
[0033] Understandably, the current time step When the time window length is greater than the current time step, the effective access frequency is calculated. The larger the decay coefficient, the faster the historical access count is forgotten, and the more the calculation of the effective access frequency depends on recent data.
[0034] The flowchart for obtaining the effective access frequency is as follows: Figure 2 As shown.
[0035] During the solver's operation, a counter is assigned to each subdomain. Each time a subdomain is read or written, the counter count is incremented and a timestamp is recorded. When the counter count reaches the length of the time window, the effective access frequency of the time step is used as the weight of the number of times the subdomain is accessed in the time step. The number of times all time steps are accessed within the preset time window is weighted and summed to obtain the weighted effective access frequency of the time window.
[0036] The geometric center coordinates of a subdomain are calculated based on its vertex coordinates. The effective access frequency of each time step is used as the weight of the geometric center coordinates of the subdomain. The geometric center coordinates of all subdomains are then summed using weighted averages to obtain the hotspot centers. The third quartile of all time steps within a preset time window is designated as the first partitioning threshold. Subdomains whose cumulative access frequency across all time steps within the preset time window exceeds the first partitioning threshold are designated as hotspot subdomains.
[0037] The vertex coordinates of the subdomain can be directly extracted from the He Dian simulation model. The acquisition of the vertex coordinates of the subdomain is a well-known technique and will not be elaborated further.
[0038] In the nuclear power simulation model, physical quantities exhibit the characteristics of continuous diffusion in space and time, that is, the closer the unit is to the current hotspot center in space, the higher the probability of being accessed in the next step, and the closer the unit is to the current time in time, the higher the probability of being accessed. Therefore, in order to avoid blind prefetching and waste of cache, according to the distance between the geometric center coordinates of the sub-domain and the hotspot center, and the interval between the current time step and the future time step, the predicted access probability of the future time step is determined.
[0039] The calculation formula of the predicted access probability of the future time step is: wherein, represents the predicted access probability of the i-th sub-domain in the future j-th time step after the current time step t; represents the future j-th time step after the current time step t, wherein, represents the Euclidean distance between the geometric center coordinates of the i-th sub-domain at the current time step t and the hotspot center; represents a normalization function, which is used to normalize the value in the parentheses. The third quartile of the predicted access probability of all future time steps after the current time step is recorded as the second division threshold, and the data of the future time step after the current time step with a predicted access probability greater than the second division threshold is preloaded to the first storage layer, and the cache is managed using the LRU strategy.
[0040] wherein, the carrier of the first storage layer is an NVMe SSD or NVRAM local to the server node, which is used to store the data of the hotspot sub-domain with high probability of access in the next step; the cache data is managed using the LRU strategy, which is a well-known technology and will not be described again. In addition, in addition to the first storage layer, there are also a second storage layer and a third storage layer; the carrier of the second storage layer is a parallel system file or a distributed SSD pool, and the second storage layer is used to store the compressed differential data of the sub-domain and the complete data of the nearest time step; the carrier of the third storage layer is object storage or tape library, which is used to store the compressed package after difference merging. Among them, the parallel system file is Lustre, GPFS, BeeGFS, and the object storage is S3, Ceph.
[0041] At this point, the data of the preloaded time step is determined.
[0042]
[0043] In step S003, the dynamic compression ratio of the sub-domain at the current time step is determined according to the difference between the data of the adjacent time steps of the sub-domain and the volume difference of different sub-domains, the compression mode of the data of the adjacent time steps of the sub-domain is determined according to the dynamic compression ratio of the sub-domain at the time step, and the compression result is written into the second storage layer. In combination with the value of the effective access frequency of the sub-domain at the current time step, part of the data of the second storage layer is selected for compression and written into the third storage layer, so as to realize the management of the development data of the nuclear power simulation model.
[0044] A simulation calculation needs to iterate hundreds of thousands to millions of time steps on a grid of hundreds of millions, generating massive data, and most of the field variables in the region change very little in the adjacent time steps. If direct lossless saving is performed, a large amount of storage space will be consumed, and I / O delay will also be caused. However, if all data is uniformly compressed in a format, the accuracy of the hot area cannot be guaranteed. Therefore, the compression ratio of each time step needs to be adaptively determined according to the difference between the development data of the adjacent time steps, and compression needs to be performed according to the adaptively determined compression ratio.
[0045] The dynamic compression ratio of the sub-domain at the current time step is determined according to the difference between the data of the adjacent time steps of the sub-domain and the volume difference of different sub-domains.
[0046] The sum of squares of the corresponding field variables of the corresponding cells of the same sub-domain at the current time step and the previous adjacent time step of the current time step is denoted as the first sum of squares of the corresponding cells of the same sub-domain at the current time step. The volume of the cell is used as the weight of the first sum of squares of the cell, and the first sum of squares of all cells of the same sub-domain at the current time step is weighted and summed. The normalized value of the result of the weighted sum is denoted as the adjacent time step difference of the same sub-domain at the current time step.
[0047] In this embodiment, the normalized value is calculated by using a sigmoid function. The sigmoid function is a known technology and will not be described here. As other embodiments, the implementer can use other methods of the prior art, such as a tanh function. The field variable of the time step is the development data of the corresponding cell of the time step. The volume of the cell is obtained by using a known technology, which will not be described here.
[0048] The median of the adjacent time step differences of all sub-domains at the current time step is denoted as the sensitivity parameter of the current time step. The sum of the adjacent time step difference of the sub-domain at the current time step and the sensitivity parameter of the current time step is denoted as the relative adjacent time step difference of the sub-domain at the current time step. The maximum value of the relative adjacent time step difference of the sub-domain at the current time step and the third preset division threshold value is denoted as the dynamic compression ratio of the sub-domain at the current time step.
[0049] The sensitivity parameter of the current time step can make the dynamic compression ratio of nearly half of the subdomains approach 0.5, so as to ensure that these subdomains are neither excessively compressed nor excessively reserved.
[0050] The third division threshold is a preset parameter, and the value of the third division threshold should be greater than or equal to 0.2 and less than or equal to 0.5. In this embodiment, the value of the third division threshold is 0.2.
[0051] The greater the difference between the subdomain in the current time step and the adjacent time step, the greater the change of the physical field generated in the current time step and the previous adjacent time step of the current time step, the more important the data in the current time step, the more necessary it is to retain high-precision data in the compression process, and the greater the compression ratio should be selected for compression. At this time, the greater the dynamic compression ratio of the subdomain in the current time step.
[0052] According to the dynamic compression ratio of the subdomain in the time step, the compression mode of the data of the subdomain in the adjacent time step is determined, and the compression result is written into the second storage layer.
[0053] The difference between the corresponding field variables of the same subdomain in the current time step and the previous adjacent time step of the current time step is recorded as the difference data block of the same subdomain in the current time step. When the dynamic compression ratio of the subdomain in the current time step is greater than or equal to the fourth division threshold, the difference data block is losslessly compressed by using the LZ4-HC level 9 algorithm; when the dynamic compression ratio of the subdomain in the current time step is greater than or equal to the third division threshold and less than the fourth division threshold, the dynamic compression ratio of the subdomain in the current time step is taken as the value of the compression ratio, and the difference data block is lossy quantized and losslessly compressed by using the LZ4 level 6 algorithm. The compression result of the difference data block is written into the second storage layer.
[0054] The fourth division threshold is a preset parameter, and the value of the fourth division threshold should be greater than 0.2 and less than or equal to 1. In this embodiment, the value of the fourth division threshold is 0.9. The lossless compression by using the LZ4-HC level 9 algorithm and the LZ4 level 6 algorithm are both known technologies, and will not be described herein.
[0055] When the data of the subdomain is not frequently accessed for a long time, it indicates that the necessity of real-time calculation and high-frequency analysis on the data of the time step corresponding to the subdomain is low. In order to speed up the reading and writing of other subdomain data with strong necessity, and at the same time, in order to save storage space, the historical difference data of these subdomains can be more strongly compressed, and the compression result is stored in the third storage layer.
[0056] The first sixth quantile of the effective access frequency of all sub-domains at the current time step is recorded as a fifth division threshold, the differential data blocks of the sub-domains with the effective access frequency less than the fifth division threshold at the current time step at all time steps before the current time step are selected, the selected differential data blocks are arranged in chronological order, complete sub-domain data is obtained, the complete sub-domain data is compressed by using a compression method based on wavelet change, and the compressed data is written into the third storage layer, and the storage space of the data in the second storage layer is released.
[0057] Thus, the management of the nuclear power simulation model development data is realized.
[0058] Based on the same inventive concept as the above method, the embodiments of the present application also provide a nuclear power simulation model development data management system, which comprises a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of any one of the methods in the above nuclear power simulation model development data management method when executing the computer program.
[0059] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. within the principles of the present application shall be included in the protection scope of the present application.
Claims
1. A data management method for developing nuclear power simulation models, characterized in that, The method includes the following steps: Extract the development data of the nuclear power simulation model, divide the simulation mesh into different subdomains, and allocate the I / O bandwidth of the subdomains according to the number of cells in each subdomain. Based on the number of times the subdomain is accessed in the current time step and all time steps within the preset time window before the current time step, the effective access frequency of the subdomain in the current time step is determined. Combining the number of times the subdomain is accessed in all time steps within the time window, the weighted effective access frequency of the time window is determined. Based on the effective access frequency, the hot spot center is calculated. Based on the distance between the geometric center coordinates of the subdomain and the hot spot center, and the interval between the current time step and future time steps, the predicted access probability of future time steps is determined. Based on the predicted access probability, the data of the time steps to be preloaded to the first storage layer is determined. Based on the differences between data in adjacent time steps and the volume differences between different subdomains, the dynamic compression ratio of the subdomain in the current time step is determined. Based on the dynamic compression ratio of the subdomain in the time step, the compression method of the data in the subdomain in adjacent time steps is determined, and the compression result is written to the second storage layer. Combining the effective access frequency of the subdomain in the current time step, a portion of the data in the second storage layer is selected for compression and written to the third storage layer, thereby realizing the management of nuclear power simulation model development data.
2. The nuclear power simulation model development data management method according to claim 1, characterized in that, The formula for calculating the effective access frequency of the subdomain at the current time step is: in, Indicates the first Each subdomain at the current time step Valid access frequency; Indicates the length of the preset time window; Indicates the current time step; Indicates the first The subdomain in the first The number of times a step is accessed; Represents the natural constant; This indicates the preset attenuation coefficient.
3. The nuclear power simulation model development data management method according to claim 1, characterized in that, The specific method for obtaining the weighted effective access frequency of the time window is as follows: The effective access frequency of a time step is used as the weight of the number of times a subdomain is accessed in a time step. The number of times all time steps are accessed within a preset time window is weighted and summed to obtain the weighted effective access frequency of the time window.
4. The nuclear power simulation model development data management method according to claim 1, characterized in that, The specific method for calculating hotspot centers based on effective access frequency is as follows: Calculate the geometric center coordinates of the subdomains, use the effective access frequency of the time step as the weight of the geometric center coordinates of the subdomains, and sum the weighted geometric center coordinates of all subdomains to obtain the hot spot center.
5. The nuclear power simulation model development data management method according to claim 1, characterized in that, The formula for calculating the predicted access probability at the future time step is: in, Indicates the first Each subdomain at the current time step The next future Predicted access probability at each time step; Indicates the current time step The next future One time step; Indicates the first Each subdomain at the current time step The Euclidean distance between the geometric center coordinates and the hotspot center at time; This represents the normalization function.
6. The nuclear power simulation model development data management method according to claim 1, characterized in that, The specific method for determining the data to be preloaded to the first storage layer at the time step based on the predicted access probability includes: The third quartile of the predicted access probability of all future time steps after the current time step is denoted as the second partitioning threshold. Data of future time steps after the current time step whose predicted access probability is greater than the second partitioning threshold are preloaded into the first storage layer.
7. The nuclear power simulation model development data management method according to claim 1, characterized in that, The specific method for obtaining the dynamic compression ratio of the subdomain at the current time step is as follows: The sum of squares of the differences between the corresponding field variables of the corresponding units in the same subdomain at the current time step and the previous adjacent time step is denoted as the first sum of squares of the corresponding units in the same subdomain at the current time step; the volume of the unit is used as the weight of the first sum of squares of the unit, and the first sum of squares of all units in the same subdomain at the current time step is weighted and summed, and the normalized value of the weighted sum is denoted as the difference between adjacent time steps of the same subdomain at the current time step; The median of the differences between all subdomains at the current time step and their adjacent time steps is denoted as the sensitivity parameter at the current time step; the sum of the differences between the subdomains at the current time step and their adjacent time step differences and the sensitivity parameter at the current time step is denoted as the relative differences between the subdomains at the current time step and their adjacent time steps. The maximum value between the difference between the subdomain and its relative adjacent time steps at the current time step and the preset third partitioning threshold is denoted as the dynamic compression ratio of the subdomain at the current time step.
8. The nuclear power simulation model development data management method according to claim 1, characterized in that, The method for determining the compression method of data in adjacent time steps based on the dynamic compression ratio of the subdomain at each time step includes the following specific methods: The difference between the corresponding field variables of the same subdomain at the current time step and the previous adjacent time step is denoted as the difference data block of the same subdomain at the current time step. When the dynamic compression ratio of the subdomain at the current time step is greater than or equal to the fourth partitioning threshold, lossless compression is performed on the differential data block. When the dynamic compression ratio of the subdomain at the current time step is greater than or equal to the third partition threshold and less than the fourth partition threshold, the dynamic compression ratio of the subdomain at the current time step is used as the value of the compression ratio. Lossy quantization is performed on the differential data block, and lossless compression is performed on the lossily quantized differential data block.
9. The nuclear power simulation model development data management method according to claim 1, characterized in that, The specific method for selecting and compressing a portion of the data in the second storage layer is as follows: The first quartile of the effective access frequency of all subdomains at the current time step is recorded as the fifth partitioning threshold. Data of subdomains whose effective access frequency is less than the fifth partitioning threshold at the corresponding time step and all previous time steps are compressed.
10. A nuclear power plant simulation model development data management system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-9.
Citation Information
Patent Citations
Data storage method and device, computer equipment and storage medium thereof
CN117130549A
Data compression and deduplication aware hierarchy in storage system
CN117677941A
Data full-flash storage optimization method and system based on cloud computing
CN120179176A
Map tile data efficient access method and system based on hybrid storage and intelligent layering
CN120315636A
RAID card static cache management method and device based on data popularity
CN120335719A
Cited By
Machine data management method and device and storage medium
CN122019986A
