A multi-parallel computing method for discretized capability space

Abstract

The present invention is applicable to the field of data processing technology and provides a multi-parallel computing method for discretized capability space, including: obtaining a specific capability region space and a specific capability body element set; discretizing the specific capability region space; standardizing capability body data of the specific capability body elements; abstracting GPU computing logic; inputting data and computing logic into the GPU and computing to obtain computing results; compared with existing serial computing methods, the method can effectively improve computing efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a multi-parallel computing method for discretized capability space. Background Art

[0002] Discretized capability space represents the capabilities of specific elements within a specific regional space, including but not limited to image processing, electrical charge, and field data modeling. Depending on the application scenario, this can encompass population density and transportation accessibility in geographic information systems, as well as detection and command communication capabilities in command information systems. By calculating and analyzing the capabilities of specific elements, the strengths and weaknesses of these elements within each regional space can be determined, effectively demonstrating the performance of these specific elements within that regional space.

[0003] Discretization is a key concept in the multi-parallel computation of spatial capabilities based on discretized regional centers. It decomposes a specific region or regional space into subdivided regional spaces of a specific granularity. This concept can be used to decompose a complete, complex, and repetitive computational task into multiple, or even several, simple steps or tasks with low computational complexity. This approach is suitable for splitting complex, repetitive tasks into smaller steps, enabling rapid task resolution.

[0004] Therefore, research on efficient processing methods is crucial in areas such as large-scale data processing, large-scale modeling, implementing heat maps in geographic information systems, representing and modeling multidimensional capabilities, and modeling and computing detection and command communication capabilities in command information systems. Existing computational methods, most of which are based on serial computational logic, require improvement in computational efficiency, thus impacting the overall efficiency of tasks or systems.

[0005] In short, the computational efficiency of existing computing methods in areas such as spatial modeling, heat map calculation, and large-scale data processing is difficult to meet the current efficiency requirements for responding to real-time computing data, thus affecting the subsequent tasks of the overall system. Summary of the Invention

[0006] Since existing serial computing efficiency is difficult to meet the needs of real-time computing in specific scenarios, the purpose of the embodiments of the present invention is to provide a multiple parallel computing method for discretized capability space to meet the requirements of computing efficiency in various system tasks.

[0007] The embodiment of the present invention is implemented as follows: a multi-parallel computing method for discretized capability space, comprising:

[0008] Step 1: Obtain the specific capability region space and specific capability body element set;

[0009] Specific capability area spaces include but are not limited to two-dimensional plane map areas in geographic information systems, three-dimensional spaces in information command systems, large-scale data processing task sets, and field data spaces;

[0010] The specific capability element set includes but is not limited to population distribution capability, traffic information capability, obstacle capability, typhoon wind field capability in geographic information systems, and data abstraction capability in large-scale data processing tasks;

[0011] Step 2: Discretize the specific capability area space, including but not limited to the discretization of two-dimensional plane map areas in geographic information systems, the discretization of three-dimensional space in information command systems, and the discretization of task decomposition in large-scale data processing;

[0012] Its mathematical logic can be expressed as: the specific ability area space S(x i ,y i , z i ), i = (1 ... L); where: (x, y, z) represents the coordinate point of the specific ability area space. When the z coordinate is a constant value, it can be represented as a two-dimensional plane space. In a two-dimensional plane task or a three-dimensional space task, it can refer to the coordinate position. In a large-scale data processing task set, it can refer to its repeated complex logical calculation task.

[0013] Step 3: Data standardization of the capability body of specific capability body elements, including but not limited to numerical standardization of population distribution capability in geographic information systems, numerical standardization of traffic information capability, standardization of obstacle capability data, standardization of typhoon wind field capability, and standardization of standard data abstraction capability body data in large-scale data processing tasks;

[0014] Step 4: Abstract the GPU computing logic. Leveraging the high efficiency of GPU parallel computing, we abstract the general computing logic of the specific capability elements within the specific capability region, breaking down complex, repetitive tasks into single computing tasks. We also process the logical relationship between each specific capability element and the discretized region, implementing the relevant computing logic based on the GPU's parallel computing capabilities.

[0015] Step 5: Input the data and computing logic into the GPU for calculation. The standardized data prepared in the above steps and the abstract computing logic implemented are loaded into the GPU through memory cache loading and other methods, and the computing logic is activated to obtain the calculation results.

[0016] As a further solution of the present invention, the step 2 of discretizing the specific capability region space to obtain a discretized region center coordinate set or a discretized task sequence number comprises the following steps:

[0017] Step 2.1: Theoretically analyze and determine the number N of operations required to disassemble the specific region space or repeat the operations for discretization (N represents the discretization accuracy of the specific capability region space);

[0018] Step 2.2: Split the specific area space or the repeatedly executed operation into N single-execution computing tasks;

[0019] Step 2.3, splitting the specific two-dimensional area plane into n=N*N subdivided two-dimensional area planes;

[0020] Step 2.4, splitting the specific three-dimensional region space into n=N*N*N subdivided three-dimensional region spaces;

[0021] In step 2.5, the specific large-scale data processing task can be split into n=N separate tasks.

[0022] As a further solution of the present invention, the mathematical logic for obtaining the discretized region center coordinate set or the discretized task sequence number from the discretized specific capability region space is as follows:

[0023] According to the starting point (x0, y0, z0) and the end point (x e ,y e , z e ) and the precision N that needs to be discretized do the ability space discretization;

[0024] When y and z are constant values, they can refer to specific large-scale data processing tasks or repetitive complex computing tasks;

[0025] When z is a constant value, it can refer to a two-dimensional plane or a two-dimensional related computing task;

[0026] The discretization step size is determined by subtracting the starting point from the ending point and dividing it by the precision N.

[0027]

[0028]

[0029]

[0030] According to its step size, discretize the regional space,

[0031] S={(x i ,y i , z i )},

[0032]

[0033]

[0034]

[0035] A set S of capability space points is obtained, the number of elements of which is n=N*N*N.

[0036] As a further solution of the present invention, step 3 of normalizing the capability data of the capability body includes the following steps:

[0037] Step 3.1: standardize the coordinate data of the capability body;

[0038] Step 3.2: Standardize the data of the ability body's influence distance and radiation range;

[0039] Step 3.3: perform data standardization on the capability values ​​of the capability body;

[0040] Step 3.4: Perform data standardization on the abstract method and abstract parameters of the capability modeling of the capability body.

[0041] As a further solution of the present invention, the implementation of the calculation logic in step 4 includes the following steps:

[0042] 4.1, General calculation logic of the central coordinate point after the abstract capability body and regional space discretization;

[0043] 4.2, realize the logical relationship between the capability body and the central coordinate point after discretization of the regional space;

[0044] 4.3, mapping the relationship between the GPU parallel task representation itself and the discretized spatial region center coordinate set or the discretized task sequence number;

[0045] 4.4. After completing the above steps, the standardized data processing capability body data and its necessary parameters are input into the GPU parallel computing task. Based on the GPU parallel task's own identification, the center coordinates of the discretized spatial region or the discretized task number referred to by the parallel computing task are inferred. The corresponding calculation logic is executed individually, and the result value of its specific element capability body affecting the spatial region or the task is output.

[0046] As a further solution of the present invention, the mathematical logic of the calculation logic is as follows:

[0047] By the ability set O m Make a Cartesian set with the capability space point set S to get the process quantity

[0048]

[0049] The number of its elements is M*n;

[0050] Based on the concurrent thread capability of GPU, M*n computing tasks are started to calculate each element in T (O i , S i ) and its corresponding capability relationship f (λ) , function method f (λ) It can refer to the computational logic after abstraction is achieved;

[0051] Get the calculation result of T *

[0052] By T * Mapping relationship with capability space S

[0053]

[0054] Generate a set S of all capability values ​​contained in each capability space point * , whose size is n based on the concurrent thread capability of the GPU. Its n computing tasks, within each computing unit, traverse the capabilities of all capability bodies acting on that capability space point, classify them according to the capability λ, and sum up all capability values:

[0055]

[0056] Get the final result R (S，O) .

[0057] Compared with the prior art, the beneficial effect of the present invention is that the method of the present invention can effectively improve the calculation efficiency compared with the existing serial calculation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A flowchart of a multi-parallel computing method for discretized capability space provided by an embodiment of the present invention;

[0059] Figure 2 A flowchart of a method for rapid deduction and modeling of a football match provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0061] The specific implementation of the present invention is described in detail below with reference to specific embodiments.

[0062] like Figure 1 and Figure 2 FIG. 1 is an embodiment of the present invention, a multi-parallel computing method for discretized capability space, comprising:

[0063] Step 1: Obtain the specific capability region space and specific capability body element set;

[0064] Specific capability area spaces include but are not limited to two-dimensional plane map areas in geographic information systems, three-dimensional spaces in information command systems, large-scale data processing task sets, and field data spaces;

[0065] The specific capability element set includes but is not limited to population distribution capability, traffic information capability, obstacle capability, typhoon wind field capability in geographic information systems, and data abstraction capability in large-scale data processing tasks;

[0066] Step 2: Discretize the specific capability area space, including but not limited to the discretization of two-dimensional plane map areas in geographic information systems, the discretization of three-dimensional space in information command systems, and the discretization of task decomposition in large-scale data processing;

[0067] Its mathematical logic can be expressed as: the specific ability area space S(x i ,y i , z i ), i = (1 ... N); where (x, y, z) represents the coordinate points of the capability space. When the z coordinate is constant at 0, it can be represented as a two-dimensional plane space. In two-dimensional plane tasks or three-dimensional space tasks, it can refer to the coordinate position. In a large-scale data processing task set, it can refer to its repeated and complex logical calculation tasks.

[0068] Step 3: Data standardization of the capability body of specific capability body elements, including but not limited to numerical standardization of population distribution capability in geographic information systems, numerical standardization of traffic information capability, standardization of obstacle capability data, standardization of typhoon wind field capability, and standardization of standard data abstraction capability body data in large-scale data processing tasks;

[0069] Step 4: Abstract the GPU computing logic. Leveraging the high efficiency of GPU parallel computing, we abstract the general computing logic of the specific capability elements within the specific capability region, breaking down complex, repetitive tasks into single computing tasks. We also process the logical relationship between each specific capability element and the discretized region, implementing the relevant computing logic based on the GPU's parallel computing capabilities.

[0070] Step 5: Input the data and computing logic into the GPU for calculation. The standardized data prepared in the above steps and the abstract computing logic implemented are loaded into the GPU through memory cache loading and other methods, and the computing logic is activated to obtain the calculation results.

[0071] The step 2 is to discretize the specific capability region space to obtain a discretized region center coordinate set or a discretized task sequence number, and the operation steps include:

[0072] Step 2.1: Theoretically analyze and determine the number N of operations required to disassemble a specific region of space or to perform repeated operations (N represents the discretization accuracy of the capability space);

[0073] Step 2.2: Split the specific area space or the repeatedly executed operation into N single-execution computing tasks;

[0074] Step 2.3, splitting the specific two-dimensional area plane into n=N*N subdivided two-dimensional area planes;

[0075] Step 2.4, splitting the specific three-dimensional region space into n=N*N*N subdivided three-dimensional region spaces;

[0076] Step 2.5: Split the specific large-scale data processing task into n = N individual tasks;

[0077] The mathematical logic for obtaining the discretized region center coordinate set or the discretized task sequence number for the discretized specific capability region space is as follows:

[0078] According to the starting point (x0, y0, z0) and the end point (x e ,y e , z e ) and the precision N that needs to be discretized do the ability space discretization;

[0079] When y and z are constant values, they can refer to specific large-scale data processing tasks or repetitive complex computing tasks;

[0080] When z is a constant value, it can refer to a two-dimensional plane or a two-dimensional related computing task;

[0081] The discretization step size is determined by subtracting the starting point from the ending point and dividing it by the precision N.

[0082]

[0083]

[0084]

[0085] According to its step size, discretize the regional space,

[0086] S={(x i ,y i , z i )},

[0087]

[0088]

[0089]

[0090] Get the set S of capability space points, whose number of elements is n = N*N*N;

[0091] Step 3 is to standardize the capability data of the capability body, and the steps include:

[0092] Step 3.1: standardize the coordinate data of the capability body;

[0093] Step 3.2: Standardize the data of the ability body's influence distance and radiation range;

[0094] Step 3.3: perform data standardization on the capability values ​​of the capability body;

[0095] Step 3.4: Standardize the data of abstract methods and abstract parameters for capability modeling of capability bodies;

[0096] The implementation of the calculation logic in step 4 includes the following steps:

[0097] 4.1, General calculation logic of the central coordinate point after the abstract capability body and regional space discretization;

[0098] 4.2, realize the logical relationship between the capability body and the central coordinate point after discretization of the regional space;

[0099] 4.3, mapping the relationship between the GPU parallel task representation itself and the discretized spatial region center coordinate set or the discretized task sequence number;

[0100] 4.4. After completing the above steps, the standardized data processing capability body data and its necessary parameters are input into the GPU parallel computing task. Based on the GPU parallel task's own identification, the center coordinates of the discretized spatial region or the discretized task number referred to by the parallel computing task are deduced. The corresponding calculation logic is executed individually, and the result value of the specific element capability body affecting the spatial region or task is output;

[0101] The mathematical logic of the calculation logic is as follows:

[0102] By the ability set O m Make a Cartesian set with the capability space point set S to get the process quantity

[0103]

[0104] The number of its elements is M*n;

[0105] Based on the concurrent thread capability of GPU, M*n computing tasks are started to calculate each element in T (O i , S i ) and its corresponding capability relationship f (λ) , function method f (λ) It can refer to the computational logic after abstraction is achieved;

[0106] Get the calculation result of T *

[0107] By T * Mapping relationship with capability space S

[0108]

[0109] Generate a set S of all capability values ​​contained in each capability space point * , whose size is n based on the concurrent thread capability of the GPU. Its n computing tasks, within each computing unit, traverse the capabilities of all capability bodies acting on that capability space point, classify them according to the capability λ, and sum up all capability values:

[0110]

[0111] Get the final result R (S，O) ;

[0112] A method for rapid deduction and modeling of a football match based on the present invention includes:

[0113] Step 1: Get the size of the football field and the information of the players on both sides;

[0114] In this step, the size of the football field: describes the activity area space of the task, usually represented by a two-dimensional coordinate system;

[0115] Player information: including parameters such as player running speed, offensive ability, defensive ability, organizational ability, shooting accuracy, and player position on the field.

[0116] Using public information, FIFA stipulates that the standard size of a football field is 105m*68m. Based on public player information and querying previous match information, player-related information such as running speed, offensive ability, defensive ability, playmaking ability, and shot accuracy can be obtained. Player position information can also be obtained based on real-time match data.

[0117] Step 2: Discretize the football field;

[0118] In this step, the accuracy of the discretized football field can be determined according to the required level of precision recommended by the review, and the discretization can be performed in units of 1m, 0.5m, or 0.1m.

[0119] Select 1m as the unit of discretization area space. According to the size of a standard football field, the modeled football field is a two-dimensional array from (0,0) to (105,68). The coordinate point represented by each array is the center point of the discretization area space.

[0120] Taking Python code as an example, we can use the Numpy module to create an ndarray with a shape of (105*68, 2) and fill it with the discretized data as the modeling data after the discretization of the football field.

[0121] Step 3: Standardized data processing of player information and ability information

[0122] In this step, data pre-processing is performed on the player's running speed, offensive ability, defensive ability, organizational ability, and shooting accuracy, including filling in missing values ​​and removing abnormal data.

[0123] The program represents a player's running speed, offensive ability, defensive ability, organizational ability, and shooting accuracy as numerical values. Missing values ​​are filled with 0, and abnormal values ​​are normalized to be within a reasonable range. As a player's ability changes over time, an attenuation coefficient is added, with time as a parameter.

[0124] Taking Python code as an example, we can use the Numpy module to create an ndarray whose dtype is a custom type, including player coordinate information, running speed, offensive ability, defensive ability, organizational ability, and shooting accuracy. This ndarray contains information about all 22 players on the field for both teams.

[0125] Step 4: Abstract the task calculation logic

[0126] In this step, we first need to map the relationship between the coordinate points of the court and the GPU computing task numbers;

[0127] Secondly, the player's offensive ability, defensive ability, organizational ability, and running speed are calculated as the game progresses.

[0128] Finally, based on the influence of various player abilities, the function calculation logic is abstracted. According to the player's position coordinates and the coordinates of the discretized court ability space, the distance between the player and the discretized court space coordinates is calculated. Based on the distance, the player's influence on the area is calculated.

[0129] Discretize the football field according to step 2, obtain the football field coordinate information and the total number of coordinate positions, start a corresponding number of parallel computing tasks in parallel based on the GPU parallel computing logic, and map the football field coordinate information according to the sequence number of the parallel computing tasks;

[0130] The input to the calculation logic is the player information of both teams, including the player coordinates, running speed, offensive ability, defensive ability, playmaking ability, shooting accuracy, and ability attenuation coefficient after the standardized data processing in step 3, as well as the game time, as the calculation logic input parameters. Based on the distance between each player's ability and the current area space and the attenuation coefficient, the impact of the player's ability value on the current area space as the game time changes is calculated, and the results are calculated. The calculation logic and modeling code are analyzed and implemented. The current implementation method is preprocessing.

[0131] The example of CUDA C code required for GPU computing can deduce the coordinate point of the discretized court space represented by the current task based on the thread x,y sequence number of the computing task and the x,y requirements of the thread block;

[0132] Implement the calculation code for various player abilities based on CUDA C code, using the inferred court space coordinates and input player information as parameters to improve the code calculation logic;

[0133] Step 5: Enter GPU parallel computing tasks

[0134] The current step is a specific execution step. Based on the standardized data prepared in steps 3 and 4, as well as the abstracted calculation logic code, it is input into the GPU to calculate the relevant data in parallel.

[0135] Taking the CUDA C code as an example, according to the kernel function code that can be run by the GPU after step 4, it is loaded into the GPU, the ndarray data generated by the player information parameters prepared in step 3 is loaded into the GPU cache, and the GPU kernel function code is executed;

[0136] Step 6: Data modeling and calculation results

[0137] According to step five, the court situation data of the current game process is obtained to characterize the strengths and weaknesses of both teams, so as to provide a review and deduction of the impact of individual player abilities on the overall game process, and to assist the coach in analysis and judgment.

[0138] Summarize:

[0139] In this embodiment, according to Amdahl's law, the method of the present invention can effectively improve the computing efficiency by about 7000 times compared with the existing serial computing method.

[0140] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-parallel computing method for discretized capability space, characterized in that: include: Step 1: Obtain the specific capability region space and specific capability body element set; Specific capability area spaces include but are not limited to two-dimensional plane map areas in geographic information systems, three-dimensional spaces in information command systems, large-scale data processing task sets, and field data spaces; The specific capability element set includes but is not limited to population distribution capability, traffic information capability, obstacle capability, typhoon wind field capability in geographic information systems, and data abstraction capability in large-scale data processing tasks; Step 2: Discretize the specific capability area space, including but not limited to the discretization of two-dimensional plane map areas in geographic information systems, the discretization of three-dimensional space in information command systems, and the discretization of task decomposition in large-scale data processing; Its mathematical logic can be expressed as: specific ability area space Where (x, y, z) represents the coordinate points of a specific capability area. When the z coordinate is a constant value, it can be represented as a two-dimensional plane space. In two-dimensional plane tasks or three-dimensional space tasks, it can refer to the coordinate position. In a large-scale data processing task set, it can refer to its repeated complex logical calculation tasks. Step 3: Data standardization of the capability body of specific capability body elements, including but not limited to numerical standardization of population distribution capability in geographic information systems, numerical standardization of traffic information capability, standardization of obstacle capability data, standardization of typhoon wind field capability, and standardization of standard data abstraction capability body data in large-scale data processing tasks; Step 4: Abstract the GPU computing logic. Leveraging the high efficiency of GPU parallel computing, we abstract the general computing logic of the specific capability elements within the specific capability region, breaking down complex, repetitive tasks into single computing tasks. We also process the logical relationship between each specific capability element and the discretized region, implementing the relevant computing logic based on the GPU's parallel computing capabilities. Step 5: Input the data and computing logic into the GPU for calculation. The standardized data prepared in the above steps and the abstract computing logic implemented are loaded into the GPU through memory cache loading, and the computing logic is activated to obtain the calculation results.

2. A multi-parallel computing method for discretized capability space according to claim 1, characterized in that: The step 2 is to discretize the specific capability region space to obtain a discretized region center coordinate set or a discretized task sequence number, and the operation steps include: Step 2.1: Theoretically analyze and determine the number N of operations required to disassemble the specific region space or repeat the operations for discretization (N represents the discretization accuracy of the specific capability region space); Step 2.2: Split the specific area space or the repeatedly executed operation into N single-execution computing tasks; Step 2.3, splitting the specific two-dimensional area plane into n=N*N subdivided two-dimensional area planes; Step 2.4, split the specific three-dimensional region space into n=N*N*N subdivided three-dimensional region spaces; In step 2.5, a specific large-scale data processing task can be split into n=N separate tasks.

3. The multi-parallel computing method for discretized capability space according to claim 2, characterized in that: The mathematical logic for obtaining the discretized region center coordinate set or the discretized task sequence number for the discretized specific capability region space is as follows: Based on the starting point of the capability space With end point And the precision N that needs to be discretized is used for discretization of the capability space; When y and z are constant values, they can refer to specific large-scale data processing tasks or repetitive complex computing tasks; When z is a constant value, it can refer to a two-dimensional plane or a two-dimensional related computing task; The discretization step size is determined by subtracting the starting point from the ending point and dividing it by the precision N. ， ， ， According to its step size, discretize the regional space, ， ， ， ， The set S of capability space points is obtained, and the number of its elements is n=N*N*N.

4. The multi-parallel computing method for discretized capability space according to claim 1, characterized in that: Step 3 is to standardize the capability data of the capability body, and the steps include: Step 3.1: standardize the coordinate data of the capability body; Step 3.2: Standardize the data of the ability body's influence distance and radiation range; Step 3.3: perform data standardization on the capability values ​​of the capability body; Step 3.4: Perform data standardization on the abstract method and abstract parameters of capability modeling of capability bodies.

5. The multi-parallel computing method for discretized capability space according to claim 1, characterized in that: The implementation of the calculation logic in step 4 includes the following steps: Step 4.1: General calculation logic of the center coordinate point after the abstract capability body and regional space are discretized; Step 4.2: Realize the logical relationship between the capability body and the central coordinate point after discretization of the regional space; Step 4.3, mapping the relationship between the GPU parallel task representation itself and the discretized spatial region center coordinate set or the discretized task sequence number; Step 4.4: After the above steps are completed, the standardized data processing capability body data and its necessary parameters are input into the GPU parallel computing task. Based on the GPU parallel task's own identifier, the center coordinates of the discretized spatial region or the discretized task number referred to by the parallel computing task are inferred. The corresponding calculation logic is executed individually, and the result value of the specific element capability body's influence on the regional space or the task is output.

6. The multi-parallel computing method for discretized capability space according to claim 5, characterized in that: The mathematical logic of the calculation logic is as follows: Ability set Make a Cartesian set with the capability space point set S to get the process quantity ， The number of its elements is M*n; Based on the concurrent thread capability of the GPU, M*n computing tasks are started to calculate each element in T The corresponding ability relationship , function method It can refer to the computational logic after abstraction is achieved; Get the calculation result of T , pass Mapping relationship with capability space S ， Generate a set of capability values ​​of all capability bodies contained in each capability space point , whose size is n Based on the concurrent thread capability of GPU, its n computing tasks, in each computing unit, traverse the capabilities of all capability bodies acting on the capability space point, according to the capability body capability Classify and sum all ability values: ， Get the final result .

Images