Intelligent numerical calculation methods, equipment and systems

Abstract

The present invention discloses an intelligent numerical calculation method and system. The method comprises: symbolizing the calculation data to obtain a symbolized calculation data sequence; using a knowledge engineering state machine to perform knowledge assignment on the symbolized calculation data sequence, and establishing a corresponding relationship between the symbolized calculation data sequence and a solution space region division pattern; based on the corresponding relationship between the symbolized calculation data sequence and the solution space region division pattern, performing a parallel search in a knowledge engineering engine to search for a corresponding solution space region division pattern; and obtaining a numerical calculation result according to the solution space region division pattern obtained by the search. The present invention performs mapping through a symbolized calculation data knowledge engineering state machine to complete the fusion of numerical calculation and symbolic calculation. Through a high-performance parallel search process, the information acquisition of the numerical calculation is completed to obtain the numerical calculation result value. Thus, the efficiency and performance in the calculation tasks with higher complexity are improved.

Description

Technical Field

[0001] The present invention relates to the field of parallel computing technology in the field of information systems, and in particular to an intelligent numerical computing method, device and system. Background Art

[0002] Numerical computing is the method and process of effectively using digital computers to find approximate solutions to mathematical problems, as well as the related theoretical aspects. It primarily studies how to use computers to better solve various mathematical problems, including discretizing continuous systems and solving equations in discrete form, while considering issues such as error, convergence, and stability.

[0003] Numerical computation methods play a vital role in modern scientific research and engineering calculations. Speed ​​and accuracy are two key indicators of a numerical computation method's quality. A good numerical computation method is not only highly accurate but also fast. While this advantage is largely negated by computer power for moderately sized problems, speed remains a crucial factor for large-scale problems, and slow numerical computation methods are often eliminated as impractical. Traditional numerical computation methods commonly used include the finite element method, multigrid method, finite difference method, finite volume method, error estimation methods for approximate solutions, and multiscale computation methods. These numerical computation methods must rely on arithmetic and logical operations, otherwise the computer will be unable to perform the calculations, which directly affects the computer's ability to solve practical problems. Therefore, solving complex problems inevitably faces computing bottlenecks, resulting in low efficiency and performance for complex computational tasks.

[0004] Based on this, a new solution is needed. Summary of the Invention

[0005] The main purpose of the present invention is to provide an intelligent numerical calculation method, device and system to address the computing power bottleneck problem and the shortcomings of low performance and efficiency of traditional numerical calculation methods.

[0006] To achieve the above object, the present invention provides an intelligent numerical calculation method, comprising:

[0007] Symbolize the calculation data to obtain a symbolic calculation data sequence;

[0008] Using a knowledge engineering state machine to assign knowledge to the symbolic calculation data sequence, associating the symbolic calculation data sequence with knowledge of a corresponding sequence in a knowledge base, and establishing a corresponding relationship between the symbolic calculation data sequence and a solution space region division pattern;

[0009] Based on the correspondence between the symbolic calculation data sequence and the solution space region division pattern, a parallel search is performed in the knowledge engineering engine to search for the corresponding solution space region division pattern;

[0010] The numerical calculation results are obtained according to the solution space region division pattern obtained by the search.

[0011] In the intelligent numerical calculation method provided by the present invention, before the steps of assigning knowledge to the symbolic calculation data sequence using a knowledge engineering state machine, associating the symbolic calculation data sequence with knowledge of a corresponding sequence in a knowledge base, and establishing a correspondence between the symbolic calculation data sequence and a solution space region division pattern, the method further includes:

[0012] Based on the decomposition, identification and reconstruction of the numerical calculation system, a calculation process state machine is constructed, wherein the calculation process state machine includes elementary calculation steps and intermediate states;

[0013] Based on the relationship between the computing process state machine and the knowledge attributes, the knowledge engineering state machine is constructed. The knowledge engineering state machine is used to perform knowledge assignment on the symbolic sequence of data to obtain a corresponding solution space region division pattern.

[0014] In the intelligent numerical calculation method provided by the present invention, before the step of performing a parallel search in a knowledge engineering engine based on the correspondence between the symbolic calculation data sequence and the solution space region division pattern and searching for the corresponding solution space region division pattern, the method further includes:

[0015] Determine whether the partitioning granularity of the solution space region partitioning pattern obtained by the current mapping meets the calculation accuracy requirements;

[0016] When the division granularity of the solution space region division pattern obtained by the current mapping does not meet the calculation accuracy requirements, the solution space region obtained by the current mapping continues to be fine-grained to obtain sub-intervals with fine resolution that meet the calculation accuracy requirements and refine and expand the corresponding knowledge graph relationship in the knowledge engineering engine.

[0017] In addition, to achieve the above-mentioned object, the present invention further provides an intelligent numerical calculation device, comprising:

[0018] A data processing module is used to symbolize the calculation data to obtain a symbolic calculation data sequence;

[0019] a mapping module for assigning knowledge to the symbolic calculation data sequence using a knowledge engineering state machine, associating the symbolic calculation data sequence with the knowledge of the corresponding sequence in the knowledge base, and establishing a corresponding relationship between the symbolic calculation data sequence and the solution space region division pattern;

[0020] A search module is used to perform a parallel search in the knowledge engineering engine based on the correspondence between the symbolic calculation data sequence and the solution space region division pattern, and search for the corresponding solution space region division pattern;

[0021] The calculation module is used to obtain numerical calculation results based on the solution space area division pattern obtained by the search.

[0022] The intelligent numerical computing device provided by the present invention further includes a knowledge engineering state machine construction module and a computing process state machine generation module:

[0023] A computational process state machine generation module, configured to construct a computational process state machine based on the decomposition, identification, and reconstruction of a numerical computation system, wherein the computational process state machine includes elementary computational steps and intermediate states;

[0024] The knowledge engineering state machine construction module is used to construct the knowledge engineering state machine based on the relationship between the calculation process state machine and the knowledge attributes. The knowledge engineering state machine is used to assign knowledge to the symbolic sequence of data to obtain the corresponding solution space area division pattern.

[0025] The intelligent numerical computing device provided by the present invention also includes a knowledge expansion module, which is used to determine whether the division granularity of the solution space area division pattern obtained by the current mapping meets the calculation accuracy requirements. When the division granularity of the solution space area division pattern obtained by the current mapping does not meet the calculation accuracy requirements, the solution space area obtained by the current mapping continues to be fine-grained to obtain sub-intervals with fine resolution that meet the calculation accuracy requirements and refine and expand the corresponding knowledge graph relationship in the knowledge engineering engine.

[0026] The present invention also provides an intelligent numerical calculation device, comprising a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the intelligent numerical calculation method described above are implemented.

[0027] The intelligent numerical computing device and method provided by the present invention have the following beneficial effects: Based on the natural knowledge-based numerical property that "natural number" data has "knowledge-based symbol" representation values, the present invention completes the fusion of data and symbolic computing in numerical computing. Based on the support of a complex pattern recognition algorithm system that integrates numerical computing and intelligent computing, the computing workflow is decomposed, identified, and reconstructed through a knowledge engineering engine, and a knowledge engineering state machine is constructed with data as the center. The symbolic computing data is mapped to the knowledge engineering state machine to complete the fusion of numerical computing and symbolic computing. Through a high-performance parallel search process, the information acquisition of the numerical computing is completed, and the numerical computing result value is obtained. As a result, the efficiency and performance of computing tasks with higher complexity are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only embodiments of the present invention. Those skilled in the art can also derive other drawings based on the provided drawings without inventive work.

[0029] Figure 1 FIG2 is a flow chart of an intelligent numerical calculation method provided by an embodiment of the present invention;

[0030] Figure 2 FIG2 is a diagram showing the composition of an intelligent numerical calculation device provided by an embodiment of the present invention;

[0031] Figure 3 FIG. 1 is a logical architecture diagram of an intelligent numerical computing system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0032] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate exemplary embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the present invention.

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0034] Figure 1 FIG. 1 is a flow chart of an intelligent numerical calculation method provided by an embodiment of the present invention. Figure 1 As shown, the intelligent numerical calculation method includes the following steps:

[0035] Step S1: symbolize the calculation data to obtain a symbolized calculation data sequence;

[0036] Specifically, in one embodiment of the present invention, the data of the computing task is first serialized by time to obtain a time series , the time series is symbolized by wavelet space method, and the symbolized sequence is obtained . The so-called symbolization is to transform the real number sequence into a symbol sequence. The research object of the knowledge engineering state machine is a finite symbol set. The time series obtained by sampling cannot be processed directly. The time series needs to be symbolized first, and then analyzed using the knowledge engineering state machine. Symbolization requires the ability to remove irrelevant redundant information while retaining the information of the key features of the original sequence. The wavelet transform can adopt related wavelet transform or wavelet fast transform algorithm, such as Mallat algorithm. Therefore, the wavelet space method for data symbolization processing includes the following steps:

[0037] 1) For the preprocessed sequence Calculate the scale space of wavelet decomposition and determine the scale space for decomposing the signal , let the minimum decomposition scale , the maximum decomposition scale The calculation formula is as follows:

[0038]

[0039] Where N is the time series to be analyzed length.

[0040] 2) Use Mallat algorithm to Scale-based Perform discrete wavelet transform to obtain the scale space The specific calculation method for multiple groups of wavelet coefficients on is as follows:

[0041] like is the discrete sampling data of the signal, if , the multi-scale Mallat algorithm calculates the wavelet coefficients through discrete convolution and second-order downsampling process, the formula is as follows:

[0042]

[0043] Where N is the number of discrete sampled data; h, g are the filter impulse responses, i.e., the filter coefficients of each column decomposed; for exist The approximation at the resolution is called the low-frequency coefficient of the signal, which is the approximate coefficient at scale j and j+1 and needs to be further decomposed; for exist The discrete details at the resolution are called the high-frequency coefficients of the signal. This part, as the detail value coefficients at scales j and j+1, maintains the information we need in the sequence.

[0044] For example, starting from layer j=1, the wavelet coefficient scale sequence obtained after J-layer decomposition is .

[0045] 3) In scale space In the following example, assuming that N wavelet transforms are performed, the following N groups of wavelet coefficient scale sequences will be obtained:

[0046]

[0047]

[0048]

[0049] ,

[0050] The scale sequence obtained by combination is: .

[0051] 4) Perform static division on the scale sequence to obtain the symbol sequence.

[0052] It requires first determining the threshold, and the threshold remains unchanged during the processing. Therefore, the selection of the threshold is the most critical. The value that can fully reflect the trend of the real number sequence should be selected. The degree of detail of the division can also be selected by changing the number of thresholds. The division is as follows:

[0053]

[0054] Where: s i Indicates the symbol, x t Indicates the amplitude of the time series at time t, x1~x N represents the set threshold, and N represents the size of the symbol set. The threshold can be selected using the equal interval method or the equal probability method.

[0055] Step S2: Use the knowledge engineering state machine to assign knowledge to the symbolic calculation data sequence, associate the symbolic calculation data sequence with the knowledge of the corresponding sequence in the knowledge base, and establish a corresponding relationship between the symbolic calculation data sequence and the solution space area division pattern.

[0056] Specifically, in one embodiment of the present invention, through symbolization, based on the natural knowledge-based numerical property of "natural number" data with "knowledge-based symbol" representation value, the fusion of data and symbolic calculation in numerical calculation is completed; then, knowledge attributes are assigned to the symbol sequence, and a knowledge engineering state machine for calculation is constructed with data as the center; the knowledge engineering state machine of the symbolic calculation process is then used to promote the underlying mapping of data and knowledge fusion. Therefore, before step S2, it also includes:

[0057] Step S11: Based on the decomposition, identification and reconstruction of the numerical calculation system, a calculation process state machine is constructed, wherein the calculation process state machine includes elementary calculation steps and intermediate states;

[0058] Specifically, in one embodiment of the present invention, based on the requirements of a specific numerical computation task, the system selectively determines the corresponding algorithm, and the computational steps and processes are subsequently determined, thereby defining the computational engineering flow. Therefore, by hierarchically decomposing, identifying, and reconstructing the numerical computation workflow of an existing complex numerical computation system using algorithms (including but not limited to the Euler method, Runge-Kutta method, Lagrange interpolation method, and Adams method), a computational process state machine consisting of elementary computational steps and intermediate states can be obtained. By deconstructing the numerical computation system, the complex computational process is converted into elementary computational processes such as area approximation and polynomial approximation, resulting in multiple solution space regions, facilitating the subsequent construction of the knowledge engineering state machine.

[0059] Step S12: Based on the relationship between the computational process state machine and the knowledge attributes, the knowledge engineering state machine is constructed. The knowledge engineering state machine is used to perform knowledge assignment on the symbolic sequence of data to obtain a corresponding solution space region division pattern.

[0060] Specifically, in one embodiment of the present invention, knowledge engineering, the discipline of knowledge information processing involving the representation, expression, sharing, retrieval, and application of knowledge, primarily includes the construction and reasoning of ontology libraries and knowledge graphs. A computational workflow state machine is a state machine composed of elementary computational steps and intermediate states, while a knowledge engineering state machine is a state machine that assigns knowledge to symbolic sequences of data. This assignment process is the mapping of symbols to knowledge. Using the resulting sequence of data symbolization, the corresponding sequence of knowledge in the knowledge base is associated and mapped, thereby establishing a one-to-one correspondence between the symbolic sequence of data and the partitioning pattern of the solution space.

[0061] Step S3: Based on the correspondence between the symbolic calculation data sequence and the solution space region division pattern, a parallel search is performed in the knowledge engineering engine to search for the corresponding solution space region division pattern;

[0062] Specifically, in one embodiment of the present invention, the knowledge represented by the knowledge engineering engine is mapped with the sequence knowledge converted from data symbolization, thereby establishing a correspondence between the data sequence and the solution space region division pattern; thus, the solution process is transformed into a search process for the solution space region division pattern corresponding to the solution space.

[0063] Specifically, in one embodiment of the present invention, the parallel search algorithm can adopt the "roll call" method, and the specific implementation process is as follows: each knowledge storage unit has its own information comparator, which can only store one piece of knowledge, and obtains relevant information through the bus (search system), and also transmits the knowledge through this bus (search / broadcast system). A certain number of such storage units constitute a knowledge storage area. When the system needs to search for knowledge in this knowledge storage area, the corresponding information will be broadcast through the bus. At this time, each knowledge storage unit compares the transmitted information together. If it is related to the knowledge of its own unit, the knowledge will be uploaded to the bus. In this way, the system can quickly find the target knowledge.

[0064] Specifically, in one embodiment of the present invention, the parallel computing models adopted include a shared memory model, a message passing model, a data flow model, and a data parallel model, wherein:

[0065] Shared memory model: multiple processors share the same memory space and communicate and synchronize by reading and writing shared memory;

[0066] Message passing model: Multiple processors communicate by sending and receiving messages. Each processor has its own local memory and uses message passing for synchronization and data transfer.

[0067] Dataflow model: Programs are centered around data flow, and parallel computations are described by operations on data flow. Computational nodes in the dataflow model can execute in parallel and are automatically scheduled based on input and output availability. This model is suitable for data-intensive computing and stream processing.

[0068] Data parallel model: Divide data into multiple parts, and different processors perform parallel calculations on different data parts; the data parallel model is suitable for parallel image processing, matrix operations and other applications.

[0069] Step S4: Obtain numerical calculation results based on the solution space region division pattern obtained by the search.

[0070] Specifically, in one embodiment of the present invention, the corresponding solution space partitioning pattern is searched based on the corresponding relationship, i.e., the corresponding knowledge is searched for, i.e., the required calculation result. For example, the 9x9 multiplication table is the knowledge, and calculating the multiplication result of each number is to search the multiplication table according to the position of the two multipliers, and the corresponding value obtained is the required multiplication result.

[0071] Furthermore, in one embodiment of the present invention, based on the computational accuracy requirements, if the granularity of the solution space region partition obtained by the current mapping is insufficient, further fine-grained partitioning is performed into sub-intervals with finer resolutions, and the corresponding knowledge graph relationships are further refined and expanded, and computational processing is performed on the further refined and expanded knowledge levels. Therefore, before step S3, the following is also included:

[0072] Step S21: determine whether the partition granularity of the solution space region partition mode obtained by the current mapping meets the calculation accuracy requirement;

[0073] Step S22: When the division granularity of the solution space area division pattern obtained by the current mapping does not meet the calculation accuracy requirements, continue to fine-grainedly divide the solution space area obtained by the current mapping to obtain sub-intervals with fine resolution that meet the calculation accuracy requirements and refine and expand the corresponding knowledge graph relationship in the knowledge engineering engine.

[0074] In this embodiment, by expanding the "content" and "extension" data of knowledge, knowledge attribute big data is created based on the characteristics of knowledge, and through multiple iterative data-knowledge-data-knowledge cycle processes, new knowledge reconstruction of deep and fine-grained knowledge attributes is carried out.

[0075] This invention, based on the inherent knowledge-based numerical property of "natural number" data, which possesses "knowledge-based symbols" as representational values, achieves the fusion of data and symbolic computation in numerical computation. Supported by a complex pattern recognition algorithm system that integrates numerical computation and intelligent computing, the knowledge engineering engine decomposes, identifies, and reconstructs computational workflows, constructing a data-centric knowledge engineering state machine. The knowledge engineering state machine then maps symbolic computational data to achieve the fusion of numerical and symbolic computation. Through a high-performance parallel search process, numerical computation information is acquired to obtain the numerical computation result. This improves efficiency and performance in computational tasks with high complexity.

[0076] The present invention also provides an intelligent numerical calculation device, the composition of which is shown in the figure Figure 2 Shown, including:

[0077] The data processing module includes two functional submodules: a symbolization module and a computational process state machine generation module. The symbolization module is used to symbolize computational data to generate a symbolic computational data sequence. The computational process state machine generation module is used to construct a computational process state machine based on the decomposition, identification, and reconstruction of the numerical computation system. The computational process state machine includes elementary computational steps and intermediate states.

[0078] The knowledge processing module is used for knowledge editing, knowledge engineering state machine generation and other processing, including a knowledge engineering state machine construction module and a knowledge expansion module. The knowledge engineering state machine construction module is used to construct the knowledge engineering state machine based on the relationship between the computational process state machine and the knowledge attributes. The knowledge engineering state machine is used to perform knowledge assignment on the symbolic sequence of data to obtain the corresponding solution space region division pattern. The knowledge expansion module is used to determine whether the division granularity of the solution space region division pattern obtained by the current mapping meets the computational accuracy requirements. When the division granularity of the solution space region division pattern obtained by the current mapping does not meet the computational accuracy requirements, the solution space region obtained by the current mapping is continued to be fine-grained to obtain fine-resolution sub-intervals that meet the computational accuracy requirements and refine and expand the corresponding knowledge graph relationships in the knowledge engineering engine.

[0079] a mapping module for assigning knowledge to the symbolic calculation data sequence using a knowledge engineering state machine, associating the symbolic calculation data sequence with the knowledge of the corresponding sequence in the knowledge base, and establishing a corresponding relationship between the symbolic calculation data sequence and the solution space region division pattern;

[0080] The parallel computing node module is used for parallel computing and processing, and mainly includes a multi-computer parallel search module and a multi-processor parallel computing module.

[0081] A multi-computer parallel search module is used to perform a parallel search in the knowledge engineering engine based on the correspondence between the symbolic calculation data sequence and the solution space region division pattern, and search for the corresponding solution space region division pattern;

[0082] A multi-processor parallel computing module is used to obtain numerical calculation results based on the solution space region partitioning pattern obtained by the search;

[0083] The multi-computer parallel search module utilizes multiple von Neumann machines to form a multi-machine system, supporting a node-level parallel computing model. However, the computational process fundamentally changes the control flow-driven approach of the von Neumann machine. Instead, it employs a data flow-driven approach. As long as the data is ready, relevant instructions can be executed in parallel, resulting in a truly non-Neumann computing process.

[0084] This module has four computing models, including shared memory model, message passing model, data flow model and data parallel model.

[0085] Shared memory model: Multiple processors share the same memory space and communicate and synchronize by reading and writing shared memory.

[0086] Message passing model: Multiple processors communicate by sending and receiving messages. Each processor has its own local memory and uses message passing for synchronization and data transfer.

[0087] Dataflow model: Programs are centered around data flow, and parallel computations are described through operations on that data flow. Computational nodes in the dataflow model can execute in parallel and are automatically scheduled based on input and output availability. This model is suitable for data-intensive computing and stream processing.

[0088] Data parallel model: Splits data into multiple parts, and different processors perform parallel computations on different parts of the data. The data parallel model is suitable for applications such as parallel image processing and matrix operations.

[0089] The multi-computer parallel module adopts a computer cluster mode, and each computer is installed with a parallel network operating system. In the form of a network operating system, each high-performance computing node is distributed and aggregated for parallel tasks, thereby realizing cluster-level network parallel processing.

[0090] The computers in the multi-computer parallel module's computer cluster are linked via a high-speed network, forming a highly reliable, high-bandwidth communications network. This high-speed network primarily utilizes a deterministic computing power network. Based on a deterministic network and centered around computing power, it integrates computing power and networking, achieving unified scheduling and collaboration between computing and networking. This provides a new type of infrastructure for low-cost, high-performance, and highly secure integrated computing and network services. This includes a computing network collaboration gateway that processes service identifiers rather than traditional IP / port numbers. By sensing service identifiers, various computing network collaboration policies (such as access control, load balancing, and traffic engineering) can be implemented with high precision. Furthermore, when applications are scheduled across regions by the computing network collaboration scheduling engine, the service identifier remains unchanged, fundamental to achieving the "location transparency" and "migration transparency" goals of the distributed operating system. High-speed routing and switching equipment connects all computing nodes, enabling high-speed communication between high-performance computing nodes.

[0091] The multi-processor parallel computing module is a modification of the traditional von Neumann machine, such as using multiple processing components to form pipeline processing, relying on time overlap to improve processing efficiency; or forming an array machine structure to form a single instruction stream and multiple data streams to improve processing speed.

[0092] The present invention also discloses an intelligent numerical calculation system, the logical structure of which is as follows: Figure 3 As shown. Figure 3As shown in the figure, the overall architecture of the system consists of five layers: the application layer, the expert system layer (AIGC-ES), the parallel operating system layer, the high-speed network layer, and the node layer. The application layer handles the input and output of various computing tasks in various fields, interacting with users. The expert system layer (AIGC-ES) is responsible for learning and unifying symbolic representations of computational data in various fields, as well as decomposing, identifying, and reconstructing computational workflows. Based on artificial intelligence technologies such as machine learning, computer vision, natural language processing, optimization algorithms, and large-scale pre-trained models, the system automatically and intelligently generates big data on relevant knowledge attributes through learning and pattern recognition from existing data with appropriate generalization capabilities. The parallel operating system layer manages parallel computing resources, schedules and distributes computing tasks, and aggregates and reports results. The high-speed network layer flexibly provides high-speed network bandwidth and resources on demand to meet real-time computing requirements. The node layer implements high-performance numerical computation, including finite element, streaming, multi-scale, and symbolic computation.

[0093] Specifically, the node layer can realize high-performance node numerical calculation functions, including: finite element calculation: providing finite element calculation functions, including integral equation construction, regional unit decomposition, unit basis function determination, unit analysis, overall synthesis, boundary condition processing, finite element equation solving and other functions; finite difference calculation: providing forward difference, backward difference, central difference and other calculation functions; data streaming calculation: data flow calculation driver, providing data streaming calculation function, supporting big data streaming calculation; multiple grid division: providing grid division function according to different calculation accuracy and convergence speed; multi-scale calculation: providing different scale operators and basis function construction required for multi-resolution analysis and other functions; communication function: local communication function, connecting multiple local processors; at the same time, providing high-speed network interface to support cluster communication of multiple computing nodes.

[0094] Furthermore, the node computing flow is as follows: The entire computing process consists of multiple global phases separated by a series of synchronization barriers; within each global phase, each processor runs its local program asynchronously; the last instruction in each local program is a synchronization barrier instruction; each processor can read and write shared memory asynchronously, but two processors are not allowed to access the same storage unit in the same global phase; the classification of instructions within the program: global synchronization, global read / write, local operation.

[0095] Specifically, the intelligent expert system (AIGC-ES) can achieve the following functions:

[0096] (1) AIGC Learning Engine: Based on artificial intelligence technologies such as machine learning, computer vision, natural language processing, optimization algorithms, and large pre-trained models, it automatically and intelligently generates knowledge attribute big data with appropriate generalization capabilities through learning and pattern recognition of existing data, supporting the reconstruction of new knowledge;

[0097] (2) Knowledge engineering engine, which performs computational workflow decomposition, identification, and reconstruction based on the knowledge engineering of Reflective General Artificial Intelligence (R-AGI), and computational knowledge engineering construction;

[0098] (3) Knowledge graph ontology library, based on the AIGC engine, provides knowledge graphs for problem solving;

[0099] (4) Expert System (ES), knowledge reasoning module: knowledge-based causal reasoning module; knowledge expert system: AGI knowledge modeling, storage, and query;

[0100] Specifically, a parallel operating system can achieve the following functions:

[0101] (1) Parallel computing resource management: register, activate, and deregister the high-performance computing node resources under its jurisdiction;

[0102] (2) Parallel algorithm library, 1) General algorithm library; provides general algorithms required for calculation; 2) Special algorithm library; provides special parallel algorithms; partition algorithm, divide and conquer algorithm, balanced tree algorithm, doubling algorithm, pipeline algorithm parallel scheduling algorithm library;

[0103] (3) Parallel search engine: multiple search algorithms: support high-performance indexing and parallel retrieval, multi-granularity indexing and retrieval, general retrieval, classification retrieval, quantum search algorithm, etc.;

[0104] Search core components: abstract and componentize basic core capabilities for easy expansion, such as index calculation and retrieval core;

[0105] Search Basic Services: A layered retrieval service based on core search components, including offline indexing, online retrieval, and retrieval access. Supports distributed indexing and retrieval environments, including memory and disk indexing.

[0106] Search object storage system: including offline index generation, online index rolling update, search intervention, AB testing and other functions;

[0107] Effect extension component: Search results vary with business scenarios. We decouple scoring and sorting, provide built-in basic relevance sorting functions, and also allow for custom sorting.

[0108] Query analysis: In addition to basic word segmentation, it also provides synonyms, error correction, timeliness calculation, intent recognition, component analysis, non-biology, and new word discovery. (Interface with the AIGC engine);

[0109] API component: provides multiple access methods including SDK protocol access and HTTP RESTful interface.

[0110] (4) Parallel scheduling engine: Based on the busy and idle status of computing resources, computing tasks are scheduled and distributed on demand, and the structure is summarized and reported.

[0111] Specifically, high-speed networks can achieve the following functions:

[0112] Collaborative Scheduling: Responsible for implementing control decisions for computing power and networks, including blueprint perception, computing-network collaboration, hierarchical cross-domain, and interface adaptation.

[0113] Closed-loop monitoring: Responsible for collecting and responding to computing power and network status, including real-time collection, event triggering, scheduling feedback, log tracking, and other key capabilities;

[0114] Resource Management: Responsible for the full lifecycle management of computing power and network resources, including resource registration, unified management, certificate permissions, computing network topology, and other key capabilities;

[0115] Identity governance: Responsible for the full life cycle management of service identities, including identity registration, identity allocation, identity resolution, identity strategy and other key capabilities.

[0116] An embodiment of the present invention further provides an intelligent numerical calculation device, which may include:

[0117] memory for storing computer programs;

[0118] The processor, when used to execute the computer program stored in the above-mentioned memory, can implement the following steps:

[0119] The calculation data is symbolized to obtain a symbolic calculation data sequence; the symbolic calculation data sequence is assigned knowledge using a knowledge engineering state machine, the symbolic calculation data sequence is associated with the knowledge of the corresponding sequence in the knowledge base, and a corresponding relationship between the symbolic calculation data sequence and the solution space area division pattern is established; based on the corresponding relationship between the symbolic calculation data sequence and the solution space area division pattern, a parallel search is performed in the knowledge engineering engine to search for the corresponding solution space area division pattern; and a numerical calculation result is obtained according to the solution space area division pattern obtained by the search.

[0120] In the description provided herein, numerous specific details are described. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

[0121] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Accordingly, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of the invention.

[0122] Furthermore, those skilled in the art will appreciate that although some embodiments herein include certain features included in other embodiments but not other features, combinations of features from different embodiments are intended to be within the scope of the present invention and to form different embodiments. For example, in the claims below, any of the claimed embodiments may be used in any combination.

[0123] It should be noted that the above embodiments illustrate rather than limit the invention, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between brackets should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The present invention may be implemented by means of hardware comprising several different elements and by means of appropriately programmed computers. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third etc. does not indicate any order. These words may be interpreted as names.

Claims

1. An intelligent numerical calculation method, characterized in that: The following steps are involved: Symbolize the calculated data to obtain a symbolized calculated data sequence, including serializing the calculated data according to time to obtain time series data, and then symbolizing the time series data by using a wavelet space method, and symbolically converting the time series data into the symbolized calculated data sequence; Based on the decomposition, identification and reconstruction of the numerical calculation system, a calculation process state machine is constructed, wherein the calculation process state machine includes elementary calculation steps and intermediate states; Based on the relationship between the computing process state machine and the knowledge attributes, a knowledge engineering state machine is constructed, wherein the knowledge engineering state machine is used to perform knowledge assignment on the symbolic sequence of data to obtain a corresponding solution space region division mode; Using a knowledge engineering state machine to assign knowledge to the symbolic computing data sequence, associating the symbolic computing data sequence with the knowledge of the corresponding sequence in the knowledge base, and establishing a corresponding relationship between the symbolic computing data sequence and the solution space region division mode; Based on the correspondence between the symbolic calculation data sequence and the solution space region division pattern, a parallel search is performed in the knowledge engineering engine to search for the corresponding solution space region division pattern; According to the solution space area division mode obtained by searching, the numerical calculation results are obtained; Among them, in the step of performing a parallel search in the knowledge engineering engine based on the correspondence between the symbolic calculation data sequence and the solution space region division pattern, searching for the corresponding solution space region division pattern, a multi-machine system composed of multiple von Neumann machines implements and supports a node-level parallel computing model; the adopted parallel computing models include a shared memory model, a message passing model, a data flow model and a data parallel model; the node computing process is composed of multiple global phases separated by multiple synchronization barriers, and two processors are not allowed to access the same storage unit in the same global phase, and each processor can asynchronously read and write the shared storage unit; in each global phase, each processor asynchronously runs a local program, and the last instruction in each local program is a synchronization barrier instruction, and the instructions in the program include global synchronization, global read / write, and local operations.

2. The intelligent numerical calculation method according to claim 1, characterized in that: Before the step of performing parallel search in the knowledge engineering engine based on the correspondence between the symbolic calculation data sequence and the solution space region division pattern to search for the corresponding solution space region division pattern, the method further includes: Determine whether the partition granularity of the solution space region partition mode obtained by the current mapping meets the calculation accuracy requirements; When the division granularity of the solution space region division pattern obtained by the current mapping does not meet the calculation accuracy requirements, the solution space region obtained by the current mapping continues to be fine-grained to obtain fine-resolution sub-intervals that meet the calculation accuracy requirements and refine and expand the corresponding knowledge graph relationship in the knowledge engineering engine.

3. An intelligent numerical computing device, characterized in that: include: The data processing module is used to symbolize the calculation data to obtain a symbolized calculation data sequence, specifically including serializing the calculation data according to time to obtain time series data, and then symbolizing the time series data by wavelet space method, and symbolizing and converting the time series data into the symbolized calculation data sequence; A computing process state machine generation module, used to construct a computing process state machine based on the decomposition, identification and reconstruction of the numerical computing system, wherein the computing process state machine includes elementary computing steps and intermediate states; A knowledge engineering state machine construction module, used to construct a knowledge engineering state machine based on the relationship between the computing process state machine and the knowledge attributes, wherein the knowledge engineering state machine is used to perform knowledge assignment on the symbolic sequence of data to obtain a corresponding solution space region division mode; A mapping module, used to use a knowledge engineering state machine to assign knowledge to the symbolic computing data sequence, associate the symbolic computing data sequence with the knowledge of the corresponding sequence in the knowledge base, and establish a corresponding relationship between the symbolic computing data sequence and the solution space region division mode; A search module, for performing a parallel search in the knowledge engineering engine based on the correspondence between the symbolic calculation data sequence and the solution space region division pattern, to search for the corresponding solution space region division pattern; A calculation module, used for obtaining numerical calculation results according to the solution space region division mode obtained by searching; Among them, the search module is implemented by a multi-machine system composed of multiple von Neumann machines and supports node-level parallel computing models; the adopted parallel computing models include shared memory model, message passing model, data flow model and data parallel model; the node computing process is composed of multiple global phases separated by multiple synchronization barriers, and two processors are not allowed to access the same storage unit in the same global phase, and each processor can read and write the shared storage unit asynchronously; in each global phase, each processor runs a local program asynchronously, and the last instruction in each local program is a synchronization barrier instruction, and the instructions in the program include global synchronization, global read / write, and local operations.

4. The intelligent numerical calculation device according to claim 3, characterized in that: It also includes a knowledge expansion module, which is used to determine whether the division granularity of the solution space area division pattern obtained by the current mapping meets the calculation accuracy requirements. When the division granularity of the solution space area division pattern obtained by the current mapping does not meet the calculation accuracy requirements, the solution space area obtained by the current mapping continues to be fine-grained to obtain sub-intervals with fine resolution that meet the calculation accuracy requirements and refine and expand the corresponding knowledge graph relationship in the knowledge engineering engine.

5. An intelligent numerical calculation system, characterized in that: The method is used to implement the steps of the intelligent numerical calculation method as claimed in any one of claims 1 to 2, wherein the system architecture of the intelligent numerical calculation system consists of five layers: application layer, expert system layer, parallel operating system layer, high-speed network layer, and node layer.

6. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the intelligent numerical calculation method according to any one of claims 1 to 2 are implemented.

Images