Data mining system, method and program product

Abstract

The invention discloses a data mining system and method and a program product, and belongs to the technical field of artificial intelligence. In the system, a task distributor is configured to distribute data mining tasks to N intelligent data mining models, and a decision device is configured to judge data results mined by the intelligent data mining models, send true data results to a data fusion device for data fusion and discard false data results; a global parameter calculator calculates global parameters according to data results, data mining strategies and rewards reported by the N intelligent data mining models; and the nth intelligent data mining model updates own parameters according to the global parameters, performs data mining from the data source according to the distributed task, generates a data result for the distributed task, and reports the data result, a data mining strategy and an award to the central processing unit. The method does not need to use an experience pool to store historical samples, saves the storage space, improves the data sampling efficiency, and improves the training speed.

Description

Technical Field

[0001] The present invention relates to a data mining system, method and program product, belonging to the technical field of artificial intelligence. Background Art

[0002] The Chinese patent application with the publication number CN109597839A discloses a data mining method based on avionics combat situation. The method includes the following three aspects: establishing a multi-platform avionics system resource cloud platform; generating fire control strike training data based on STK; establishing a fire control strike trajectory planning algorithm based on deep reinforcement learning. Specifically as follows: 1.1) Establishing a multi-platform avionics system resource cloud platform. Each aircraft platform is equipped with different system functions and resources. The avionics systems of each aircraft form a "cyber" avionics system through a data link. The avionics system of each aircraft can be divided into multiple integrated application function areas. The avionics systems of each platform can dynamically combine different application function areas on different platforms according to different combat mission requirements, and realize system information integration to form a task-oriented "logical avionics system". The multi-platform avionics system resource cloud platform provides platform support for the generation and application of avionics data. 1.2) Generating fire control strike training data based on STK. Using the STK tool to model a complex battlefield environment to realize the scenario setting of combat missions in a collaborative combat scenario; and by the STK tool, the fire control program is called in real time to record the attitude of the aircraft when launching a missile each time and whether the target is hit in real time, accumulating training data to prepare for the data mining part of the data mining system based on avionics combat situation. 1.3) Establishing a fire control strike trajectory planning algorithm based on deep reinforcement learning. On the basis of obtaining battlefield situation data, training a deep reinforcement learning neural network to extract data features for route planning; at the same time, on the basis of the flight simulation data generated by the STK tool for the strike effect, establishing a route planning evaluation model based on deep reinforcement learning; on the trained route planning evaluation model, reading the current battlefield situation data, and using the deep reinforcement learning neural network to realize route planning and output the flight state parameters of the aircraft.

[0003] However, this patent requires a replay memory unit and needs a large storage space. Summary of the Invention

[0004] To overcome the disadvantages existing in the prior art, the object of the present invention is to provide a data mining system, method and computer program product, which do not need to use an experience pool to store historical samples, save storage space, improve the sampling efficiency of data, and thus improve the training speed.

[0005] To achieve the above-mentioned invention object, the present invention provides a data mining system, which includes a central processing unit and N intelligent data mining models. Among them, the central processing unit includes a task allocator, a judge, a data fuser, and a global parameter calculator. Among them, the task allocator is configured to allocate data mining tasks to N intelligent data mining models. The judge is configured to judge the data results mined by the intelligent data mining models, transmit the true data results to the data fuser for data fusion, and discard the false data results; the data fuser is configured to perform data fusion on the true data results mined by N intelligent data mining models; the global parameter calculator calculates global parameters according to the data results reported by N intelligent data mining models, data mining strategies, and rewards; the nth intelligent data mining model updates its own parameters according to the global parameters, and performs data mining from the data source according to the allocated tasks, generates data results for the allocated tasks, and reports the data results, data mining strategies, and rewards to the central processing unit, where n = 1, 2,..., N.

[0006] To achieve the above-mentioned invention object, the present invention also provides a data mining method, which includes the following steps: Step 1: The task allocator of the central processing unit allocates data mining tasks to N intelligent data mining models; Step 2: The nth intelligent data mining model performs data mining from the data source according to the allocated tasks, generates data results for the allocated tasks, and reports the data results, data mining strategies, and rewards to the central processing unit, where n = 1, 2,..., N Step 3: The judge of the central processing unit judges the data results mined by the intelligent data mining models, transmits the true data results to the data fuser for data fusion, and discards the false data results; Step 4: The data fuser performs data fusion on the true data results mined by N intelligent data mining models; Step 5: The global parameter calculator calculates global parameters according to the data results reported by N intelligent data mining models, data mining strategies, and rewards; Step 6: The nth intelligent data mining model updates its own parameters according to the global parameters.

[0007] To achieve the above-mentioned invention object, the present invention also provides a computer program product, which includes computer program code. The computer program code can be called by a processor to execute the above-mentioned method.

[0008] Compared with the prior art, the data mining system, method, and computer program product provided by the present invention have the following beneficial effects: Historical samples that do not need to be stored in the experience pool are used, saving storage space and improving the sampling efficiency of data, thereby enhancing the training speed. At the same time, samples are collected from multiple different training environments, and the distribution of the samples is more uniform, which is more conducive to the training of the neural network. Brief Description of the Drawings

[0009] Figure 1 is a block diagram of the data mining system provided by the first embodiment of the present invention. Detailed Description of the Embodiment

[0010] In order to make the technical problems, technical solutions, and beneficial effects to be solved by 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 used to limit the present invention.

[0011] First Embodiment

[0012] Figure 1 is a block diagram of the data mining system provided by the first embodiment of the present invention. As Figure 1 shown, the data mining system provided by the first embodiment of the present invention includes a central processor and N intelligent data mining models. Among them, the central processor includes a task allocator, a decision maker, a data integrator, and a global parameter calculator. Among them, the task allocator is configured to allocate data mining tasks to N intelligent data mining models. The decision maker is configured to make a decision on the data results mined by the intelligent data mining models, transmit the true data results to the data integrator for data integration, and discard the false data results; the data integrator is configured to perform data integration on the true data results mined by N intelligent data mining models; the global parameter calculator calculates global parameters according to the data results reported by N intelligent data mining models, data mining strategies, and rewards; the nth intelligent data mining model updates its own parameters according to the global parameters, and performs data mining from the data source according to the allocated tasks, generates data results for the allocated tasks, and reports the data results, data mining strategies, and rewards to the central processor, where n = 1, 2,..., N.

[0013] In the first embodiment, the nth intelligent data mining model includes a data mining strategy network and an action value function , is the parameter to be optimized of the data mining strategy network of the nth intelligent data mining model, and ρ n is the parameter of the value function. The training of the nth intelligent data mining model includes the following process: The nth intelligent data mining model performs a series of data mining actions according to the data source , and obtains a series of data results , obtain a series of rewards , and , and are reported to the global parameter calculator of the central processing unit. t is the number of times, and T is a positive integer greater than or equal to 2; System data mining is performed according to the data source, including data collection, web crawling, and application programming interfaces, etc.

[0014] The global parameter calculator calculates the global parameter w according to Formula 1 and broadcasts the global parameter w to N intelligent data mining models: Formula 1: , In the formula, is the discount factor; The nth intelligent data mining model updates the parameters of the data mining policy network through Formula 2: Formula 2: , In the formula, is the current parameter of the data mining policy network of the nth intelligent data mining model; is the changed parameter of the data mining policy network of the nth intelligent data mining model; is the gradient function of , is the learning coefficient.

[0015] In the first embodiment, the nth intelligent data mining model updates the parameters of the action value function through Formula 3 as follows of: Formula 3: .

[0016] Optionally, the nth intelligent data mining model includes a data mining policy network and an action value function , is the parameter to be optimized of the data mining policy network of the nth intelligent data mining model, and ρ n is the parameter of the value function. The training of the nth intelligent data mining model includes the following process: The nth intelligent data mining model performs a series of data mining actions according to the data source , and obtains a series of data results , obtain a series of rewards , and , and are reported to the global parameter calculator of the central processing unit; The global parameter calculator calculates the global parameter w according to Formula 1 and broadcasts the global parameter w to N intelligent data mining models; The nth intelligent data mining model updates the parameters of the data mining policy network through Formula 4: Formula 4: , , wherein, is the learning coefficient, , is and the Euclidean distance of, is the distance threshold; The nth intelligent data mining model updates the parameters of the action value function by Formula 3.

[0017] The data mining system provided by the first embodiment further includes an input unit through which a user inputs instructions, data, etc.

[0018] The data mining system provided by the first embodiment further includes an output unit. The output unit includes a display, etc., and the display is used to display data policies, data results, data resources, etc.

[0019] In the first embodiment of the present invention, data mining is performed through an intelligent data mining model, which can learn autonomously and has the ability to adapt to new data resources and new data mining tasks.

[0020] The first embodiment of the present invention does not need to use an experience pool to store historical samples, saving storage space and improving the sampling efficiency of data, thereby enhancing the training speed. At the same time, samples are collected from multiple different training environments, and the distribution of samples is more uniform, which is more conducive to the training of neural networks.

[0021] Second Embodiment Only the content different from the first embodiment is described in the second embodiment of the present invention, and the same content will not be described repeatedly.

[0022] The second embodiment of the present invention provides a data mining method, which includes the following steps: Step 1: The task allocator of the central processing unit allocates data mining tasks to N intelligent data mining models; Step 2: The nth intelligent data mining model performs data mining from the data source according to the allocated task, generates a data result for the allocated task, and reports the data result, data mining strategy, and reward to the central processing unit, where n = 1, 2,..., N Step 3: The decider of the central processing unit decides on the data results mined by the intelligent data mining model, transmits the true data results to the data fusion unit for data fusion, and discards the false data results; Step 4: The data fusion unit performs data fusion on the true data results mined by N intelligent data mining models; Step 5: The global parameter calculator calculates global parameters based on the data results reported by N intelligent data mining models, data mining strategies, and rewards; Step 6: The nth intelligent data mining model updates its own parameters according to the global parameters.

[0023] In the second embodiment of the present invention, data mining is performed through an intelligent data mining model, which can learn autonomously and has the ability to adapt to new data resources and new data mining tasks.

[0024] The second embodiment of the present invention does not need to use an experience pool to store historical samples, saving storage space and improving the sampling efficiency of data, thereby enhancing the training speed. At the same time, samples are collected using multiple different training environments, and the distribution of samples is more uniform, which is more conducive to the training of neural networks.

[0025] Third Embodiment

[0026] To achieve the above-mentioned invention objective, the present invention also provides a computer program product, which includes computer program code that can be called by a processor to execute the method described in the second embodiment.

[0027] The beneficial effects of the third embodiment of the present invention are the same as those of the second embodiment and will not be repeated here.

[0028] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means two or more, unless otherwise specifically defined. "Several" means one or more, unless otherwise specifically defined.

[0029] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only to illustrate the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A data mining system, characterized in that: It includes a central processing unit and N intelligent data mining models, wherein the central processing unit includes a task distributor, a judger, a data fusion unit and a global parameter calculator, wherein the task distributor is configured to distribute data mining tasks to the N intelligent data mining models, the judger is configured to judge the data results mined by the intelligent data mining models, transmit the true data results to the data fusion unit for data fusion, and discard the false data results; the data fusion unit is configured to perform data fusion on the true data results mined by the N intelligent data mining models; the global parameter calculator calculates global parameters according to the data results, action value functions and rewards reported by the N intelligent data mining models; the nth intelligent data mining model updates its own parameters according to the global parameters, and performs data mining from the data source according to the assigned tasks, generates data results for the assigned tasks, and reports the data results, action value functions and rewards to the central processing unit, n=1,2,…,N, N is a positive integer greater than or equal to 2.

2. The data mining system according to claim 1, characterized in that: The nth intelligent data mining model includes a data mining strategy network and the action-value function , is the parameter to be optimized of the data mining strategy network of the nth intelligent data mining model, ρ n is the parameter of the value function. The training of the nth intelligent data mining model includes the following process: The nth intelligent data mining model performs a series of data mining actions based on the data source , get the series data results , get a series of rewards , and , , The global parameter calculator reported to the CPU, t is the number of times, T is a positive integer greater than or equal to 2; The global parameter calculator calculates the global parameter w according to the following formula and broadcasts the global parameter w to N intelligent data mining models: ， In the formula, is the discount factor; The nth intelligent data mining model updates the parameters of the data mining strategy network through the following formula: ， In the formula, is the current parameter of the data mining strategy network of the nth intelligent data mining model; is the changed parameter of the data mining strategy network of the nth intelligent data mining model; Yes The gradient function of is the learning coefficient.

3. The data mining system according to claim 2, characterized in that: The nth intelligent data mining model updates the action value function by the following formula Parameters: 。 4. The data mining system according to claim 1, characterized in that: The nth intelligent data mining model includes a data mining strategy network and the action-value function , is the parameter of the data mining strategy network of the nth intelligent data mining model, ρ n is the parameter of the value function. The training of the nth intelligent data mining model includes the following process: The nth intelligent data mining model performs serial data mining based on the data source , get the series data results , get a series of rewards , and , and Global parameter calculator reporting to the CPU; The global parameter calculator calculates the global parameter w according to the following formula and broadcasts the global parameter w to N intelligent data mining models: , where is the discount factor; The nth intelligent data mining model updates the parameters of the data mining strategy network through the following formula: ， ， In the formula, is the learning coefficient, , yes and The second norm distance, is the distance threshold, .

5. The data mining system according to claim 4, characterized in that: The nth intelligent data mining model updates the action value function by the following formula Parameters: 。 6. A data mining method, characterized in that: The steps include: Step 1: Assign data mining tasks to N intelligent data mining models through the task allocator of the central processing unit; Step 2: The nth intelligent data mining model mines data from the data source according to the assigned task, generates data results for the assigned task, and reports the data results, data mining strategy and reward to the central processor, n=1,2,…,N; Step 3: The decision device of the central processing unit determines the data results mined by the intelligent data mining model, transmits the true data results to the data fusion device for data fusion, and discards the false data results; Step 4: Use the data fusion device to fuse the real data results mined by N intelligent data mining models; Step 5: Calculate global parameters through a global parameter calculator based on the data results, data mining strategies and rewards reported by N intelligent data mining models; Step 6: The nth intelligent data mining model updates its own parameters according to the global parameters.

7. The data mining method according to claim 6, characterized in that: The training of the nth intelligent data mining model includes the following process: S2-01: The nth intelligent data mining model performs a series of data mining actions based on the data source , get the series data results , get a series of rewards , and , and The global parameter calculator reported to the central processor, the nth intelligent data mining model includes a data mining strategy network and the action-value function , is the parameter to be optimized of the data mining strategy network of the nth intelligent data mining model, ρ n is the parameter of the value function, t is the number of times, and T is a positive integer greater than or equal to 2; S2-02: The global parameter calculator calculates the global parameter w according to the following formula and broadcasts the global parameter w to N intelligent data mining models: S2-03: The nth intelligent data mining model updates the parameters of the data mining strategy network through the following formula: ， In the formula, is the current parameter of the data mining strategy network of the nth intelligent data mining model; is the changed parameter of the data mining strategy network of the nth intelligent data mining model; Yes The gradient function of is the learning coefficient; S2-04: The nth intelligent data mining model updates the action value function by the following formula Parameters: 。 8. The data mining method according to claim 6, characterized in that: The training of the nth intelligent data mining model includes the following process: S3-01: The nth intelligent data mining model performs a series of data mining actions based on the data source , get the series data results , get a series of rewards , and , and The global parameter calculator reported to the central processor, the nth intelligent data mining model includes a data mining strategy network and the action-value function , is the parameter of the data mining strategy network of the nth intelligent data mining model, ρ n is the parameter of the value function, S3-02: The global parameter calculator calculates the global parameter w according to the following formula and broadcasts the global parameter w to N intelligent data mining models: , where is the discount factor; is the discount factor; S3-03: The nth intelligent data mining model updates the parameters of the data mining strategy network through the following formula: ， ， In the formula, is the learning coefficient, , yes and The second norm distance, is the distance threshold; S3-04: The nth intelligent data mining model updates the action value function by the following formula Parameters: 。 9. A computer program product, characterized in that The method comprises a computer program code, which can be called by a processor to execute the method according to any one of claims 6 to 8.

Images