Intelligent optimization control system based on machine self-learning

Abstract

The present invention discloses an intelligent optimization control system based on machine self-learning, which relates to the technical field of machine self-learning. It solves the technical problem that the training data is too much and too chaotic, resulting in the machine being unable to fully receive and train in the first time during specific training, resulting in poor overall training effect. Parameter optimization processing is performed on parameter data within a learning data set and several groups of data packets, a large number of repeated parameters are optimized, and the optimized learning data set is transmitted to a learning database for storage. In this way, the training steps are optimized in sequence, and the overall capacity of the optimization steps is fully reduced, which facilitates the corresponding device end to train independently, and avoids the corresponding training machine from reacting slowly due to the chaos of the training data during the training process. By optimizing the corresponding training parameters, the training machine can obtain better training effects.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine self-learning, and in particular to an intelligent optimization control system based on machine self-learning. Background Art

[0002] Machine learning is a science of artificial intelligence. The main research object of this field is artificial intelligence, especially how to improve the performance of specific algorithms through empirical learning. Machine learning is the study of computer algorithms that can automatically improve through experience. Machine learning uses data or past experience to optimize the performance standards of computer programs.

[0003] The invention with publication number CN107272465B discloses a restaurant robot management and control system, including a central console scheduling system and an interactive platform. The central console scheduling system and the interactive platform transmit information to each other via WiFi wireless communication technology. This system can manage the robots through the central console, assign them tasks to complete meal delivery and tableware collection, optimize the restaurant management process, enable the robots to complete tasks intelligently and efficiently, and improve the customer experience while improving the quality of restaurant management, saving a lot of labor costs.

[0004] During the self-learning process, the machine needs to adopt the corresponding intelligent optimization control system to control the machine. In the specific control process, because the training data is too much and too chaotic, the machine cannot fully receive and train it in the first time during the specific training, resulting in poor overall training effect. If the repeated data within a large amount of data is optimized, the machine can receive the corresponding training control data in the first time, and then train the machine, the corresponding training effect can be improved and the optimization control effect can be achieved. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an intelligent optimization control system based on machine self-learning, which is used to solve the technical problem that the training data is too much and too chaotic, resulting in the machine being unable to fully receive and train in the first time during specific training, resulting in poor overall training effect.

[0006] To achieve the above-mentioned object, an embodiment according to a first aspect of the present invention provides an intelligent optimization control system based on machine self-learning, comprising a data input terminal, a self-learning master control terminal, and an energy input terminal;

[0007] The self-learning master control terminal includes a data packet screening unit, a learning data analysis unit, a data management optimization unit, a learning database and an autonomous operation unit;

[0008] The data input terminal is used to input a large amount of learning data sets, and the self-learning main control terminal receives the input learning data sets;

[0009] The data packet screening unit within the self-learning master terminal performs screening processing on different learning data sets, partitions the different learning data sets according to the same initial training parameters, and integrates the learning data sets of different partitions;

[0010] The learning data analysis unit analyzes and processes the partitioned learning data set, wherein the learning data set includes learning data and other data, wherein the learning data includes parameter data and video data, and the learning data analysis unit needs to analyze and process the video data to obtain a corresponding data packet to be trained;

[0011] The data management optimization unit performs parameter optimization processing on the parameter data within the learning data set and the data packet to be trained, optimizes a large number of repeated parameters, and transmits the optimized learning data set to the learning database for storage.

[0012] Preferably, the data packet screening unit performs screening processing on different learning data sets in the following specific manner:

[0013] Obtain the front-end training data of the learning dataset, which is 30% of the front-end data of the learning dataset;

[0014] Several groups of front-end training data are merged, analyzed and compared to obtain the corresponding repeatability parameter CFi, where i represents different front-end training data. The repeatability parameter CFi is compared with the preset parameter Y1. When CFi ≥ Y1, the merged and compared learning data sets are classified into the same group of partitions. Otherwise, the merged partitions are not performed.

[0015] Preferably, the learning data analysis unit analyzes and processes the partitioned learning data set in the following manner:

[0016] Get the turning point of the corresponding moving part from the video data, bind the turning point to the corresponding three-dimensional coordinate system, and mark the corresponding turning point as ZZ k (X k , Y k , Z k ), where k represents different turning points, where k=1, 2, ..., n. When k=1, it means that this turning point is the initial point, and the initial point is the origin of the three-dimensional coordinate system. When k=2, it means that this turning point moves to the second set of points, and so on. When k=n, it means that this turning point moves to the point at n;

[0017] Calculate the distance between multiple turning points k from 1 to n, using Get the distance parameter JL between the two points, merge the initial turning point ZZ1 (X1, Y1, Z1) and the second set of turning points ZZ2 (X2, Y2, Z2) and the distance parameter JL between them to get the first set of data packets. Based on the value of k, and so on, get multiple turning points and the distance parameter JL between them and merge them to get n-1 data packets;

[0018] Delete the video data of this partition, retain n-1 data packets, and transmit the learning data set after deleting the video data to the data management optimization unit.

[0019] Preferably, the data management optimization unit optimizes a large number of repeated parameters in the following specific manner:

[0020] Merge and analyze the parameter data and the data packets to be trained, check for duplicate data, extract different duplicate data, and transfer them to the learning database for storage;

[0021] After storage, the duplicate data are marked with a to-be-learned marker, and the to-be-learned marker is in the form of Dt, where t = 1, 2, ..., m, and D1 and D2 represent different duplicate data.

[0022] The tags to be learned and the duplicate data are bundled together, and a data table to be extracted is generated (the corresponding duplicate data in the data table to be extracted all correspond to the corresponding tags to be learned), and the data table to be extracted is transferred to the learning database for storage, and redundant duplicate data is deleted.

[0023] Preferably, the autonomous operation unit receives the training data that has been replaced with the to-be-learned marker, extracts duplicate data using the to-be-learned marker, completes the training data, and then performs training processing on its own using the training data.

[0024] Preferably, the energy input end is used to transmit energy to the autonomous operation unit.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: a large number of learning data sets are input in advance, the data packet screening unit then screens different learning data sets, partitions different learning data sets according to the same initial training parameters, and integrates the learning data sets of different partitions, and then analyzes and processes the learning data sets that have completed the partitioning. The learning data set includes learning data and other data, and the learning data analysis unit needs to analyze and process the video data to obtain the corresponding data packet to be trained and processed, and then optimizes the parameter data inside the learning data set and the data packet to be trained and processed, optimizes a large number of repeated parameters, and transmits the optimized learning data set to the learning database for storage. In this way, the training steps are optimized in sequence, and the overall capacity of the optimization steps is fully reduced, which is convenient for the corresponding device end to train by itself, avoiding the corresponding training machine from responding slowly due to the clutter of the training data during the training process. By optimizing the corresponding training parameters, the training machine can obtain better training effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a schematic diagram of the principle framework of the present invention;

[0027] Figure 2 Schematic diagram of the video data analysis and processing flow of the present invention. DETAILED DESCRIPTION

[0028] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0029] Example 1

[0030] See also Figure 1 , the present application provides an intelligent optimization control system based on machine self-learning, including a data input terminal, a self-learning master control terminal and an energy input terminal;

[0031] The data input terminal and the energy input terminal are both electrically connected to the input terminal of the self-learning main control terminal;

[0032] The self-learning master control terminal includes a data packet screening unit, a learning data analysis unit, a data management optimization unit, a learning database and an autonomous operation unit;

[0033] The output end of the data packet screening unit is electrically connected to the input end of the learning data analysis unit, the output end of the learning data analysis unit is electrically connected to the input end of the data management optimization unit, and the data management optimization unit is bidirectionally connected to the learning database and the autonomous operation unit respectively;

[0034] The data input terminal is used to input a large amount of learning data sets, and the self-learning main control terminal receives the input learning data sets;

[0035] The data packet screening unit within the self-learning master terminal performs screening processing on different learning data sets, partitions the different learning data sets according to the same initial training parameters, and integrates the learning data sets of different partitions. The specific method of the screening processing is as follows:

[0036] Obtain the front-end training data of the learning dataset, which is 30% of the front-end data of the learning dataset;

[0037] Several groups of front-end training data are merged, analyzed and compared to obtain the corresponding repeatability parameter CFi, where i represents different front-end training data. The repeatability parameter CFi is compared with the preset parameter Y1. When CFi ≥ Y1, the merged and compared learning data sets are classified into the same group of partitions (indicating that the overlap parameter is very high, where Y1 is the preset parameter factor, and the specific value of Y1 is determined by the operator based on experience). Otherwise, the partitions are not merged.

[0038] The learning data analysis unit analyzes and processes the partitioned learning data set, wherein the learning data set includes learning data and other data, wherein the learning data includes parameter data and video data, and the learning data analysis unit needs to analyze and process the video data to obtain a corresponding data packet to be trained. The specific method of performing the analysis and processing is as follows:

[0039] Get the turning point of the corresponding moving part from the video data, bind the turning point to the corresponding three-dimensional coordinate system, and mark the corresponding turning point as ZZ k (X k , Y k , Z k ), where k represents different turning points, where k=1, 2, ..., n. When k=1, it represents that the turning point is the initial point, and the initial point is the origin of the three-dimensional coordinate system. When k=2, it represents that the turning point moves to the second set of points, and so on. When k=n, it represents that the turning point moves to the point at n (such as Figure 2 shown);

[0040] Calculate the distance between multiple turning points k from 1 to n, using Get the distance parameter JL between the two points, merge the initial turning point ZZ1 (X1, Y1, Z1) and the second set of turning points ZZ2 (X2, Y2, Z2) and the distance parameter JL between them to get the first set of data packets. Based on the value of k, and so on, get multiple turning points and the distance parameter JL between them and merge them to get n-1 data packets;

[0041] Delete the video data of this partition, retain n-1 data packets, and transmit the learning data set after deleting the video data to the data management optimization unit.

[0042] The data management and optimization unit performs parameter optimization processing on the parameter data within the learning data set and the data packet to be trained, optimizes a large number of repeated parameters, and transmits the optimized learning data set to the learning database for storage. The specific method of optimization is:

[0043] Merge and analyze the parameter data and the data packets to be trained, check for duplicate data, extract different duplicate data, and transfer them to the learning database for storage;

[0044] After storage, the duplicate data are marked with a to-be-learned marker, and the to-be-learned marker is in the form of Dt, where t = 1, 2, ..., m, and D1 and D2 represent different duplicate data.

[0045] Bundle the tags to be learned and the duplicate data, and generate a table of data to be extracted (the duplicate data in the table of data to be extracted all correspond to the tags to be learned), and transfer the table of data to be extracted to the learning database for storage, and delete the redundant duplicate data;

[0046] The energy input terminal is used to transmit energy to the autonomous operation unit;

[0047] The autonomous operation unit receives the training data that has been replaced with the to-be-learned marker, extracts duplicate data using the to-be-learned marker, completes the training data, and then performs training processing on its own using the training data;

[0048] In this way, the training steps are optimized sequentially, and the overall capacity of the optimization steps is fully reduced, which makes it easier for the corresponding device to train on its own. This avoids the corresponding training machine from responding slowly due to the clutter of training data during the training process. By optimizing the corresponding training parameters, the training machine can achieve better training results.

[0049] Example 2

[0050] In the specific implementation process, this embodiment is different from the first embodiment in that:

[0051] The data packet screening unit within the self-learning master terminal performs screening processing on different learning data sets, partitions the different learning data sets according to the same initial training parameters, and integrates the learning data sets of different partitions. The specific method of the screening processing is as follows:

[0052] Obtain the front-end training data of the learning dataset, which is about 40% of the front-end data of the learning dataset;

[0053] Merge and analyze several groups of front-end training data to obtain the corresponding repeatability parameter CFi, where i represents different front-end training data. Compare the repeatability parameter CFi with the preset parameter Y1. When CFi ≥ Y1, the merged and compared learning data sets are classified into the same group of partitions (indicating that the overlap parameter is very high, where Y1 is a preset parameter factor, and the specific value of Y1 is determined by the operator based on experience). Otherwise, the partitions are not merged;

[0054] experiment

[0055] The operators spread the specific contents of Example 1 and Example 2 in specific experiments to experience them, and obtain the experience data after the corresponding personnel use them, wherein the experience data includes experience evaluation parameters. The specific experience evaluation parameters are shown in the following table:

[0056]

[0057] It can be seen from the data in the table that the experience data of Example 1 is better than the experience data of Example 2, so the operator can select a suitable embodiment according to actual application requirements.

[0058] Specifically, the present invention solves the technical problem that the training data is too much and too messy, resulting in the machine being unable to fully receive and train in the first time during specific training, resulting in poor overall training effect. Parameter optimization processing is performed on the parameter data within the learning data set and the data packet to be trained, a large number of repeated parameters are optimized, and the optimized learning data set is transferred to the learning database for storage. In this way, the training steps are optimized in sequence, and the overall capacity of the optimization steps is fully reduced, which is convenient for the corresponding device end to train by itself, avoiding the corresponding training machine from responding slowly due to the disorder of the training data during the training process. By optimizing the corresponding training parameters, the training machine can achieve better training results.

[0059] Some of the data in the above formula are calculated by removing the dimensions and taking their numerical values. The formula is a formula that is closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0060] The working principle of the present invention is as follows: a large number of learning data sets are input in advance, and the data packet screening unit then screens different learning data sets, partitions different learning data sets according to the same initial training parameters, and integrates the learning data sets of different partitions, and then analyzes and processes the learning data sets that have completed the partitioning. The learning data set includes learning data and other data, and the learning data analysis unit needs to analyze and process the video data to obtain the corresponding data packet to be trained and processed, and then optimizes the parameter data inside the learning data set and the data packet to be trained and processed, optimizes a large number of repeated parameters, and transmits the optimized learning data set to the learning database for storage. In this way, the training steps are optimized in sequence, and the overall capacity of the optimization steps is fully reduced, which is convenient for the corresponding device end to train by itself, and avoids the corresponding training machine from responding slowly due to the clutter of the training data during the training process. By optimizing the corresponding training parameters, the training machine can obtain better training effects.

[0061] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. Intelligent optimization control system based on machine self-learning, characterized by: It includes a data input terminal, a self-learning main control terminal and an energy input terminal; the energy input terminal is used to transmit energy to the autonomous operation unit; The self-learning master control terminal includes a data packet screening unit, a learning data analysis unit, a data management optimization unit, a learning database and an autonomous operation unit; The data input terminal is used to input a large amount of learning data sets, and the self-learning main control terminal receives the input learning data sets; The data packet screening unit within the self-learning master terminal performs screening processing on different learning data sets, partitions the different learning data sets according to the same initial training parameters, and integrates the learning data sets of different partitions; The learning data analysis unit analyzes and processes the partitioned learning data set, wherein the learning data set includes learning data, and the learning data includes parameter data and video data. The learning data analysis unit needs to analyze and process the video data to obtain a corresponding data packet to be trained; The specific method of the learning data analysis unit for analyzing and processing the partitioned learning data set is as follows: Get the turning point of the corresponding moving part from the video data, bind the turning point to the corresponding three-dimensional coordinate system, and mark the corresponding turning point as ZZ k (X k , Y k , Z k ), where k represents different turning points, where k=1, 2, ..., n. When k=1, it means that this turning point is the initial point, and the initial point is the origin of the three-dimensional coordinate system. When k=2, it means that this turning point moves to the second set of points, and so on. When k=n, it means that this turning point moves to the point at n; Calculate the distance between multiple turning points k from 1 to n, using Get the distance parameter JL between the two points, merge the initial turning point ZZ1 (X1, Y1, Z1) and the second set of turning points ZZ2 (X2, Y2, Z2) and the distance parameter JL between them to get the first set of data packets. Based on the value of k, and so on, get multiple turning points and the distance parameter JL between them and merge them to get n-1 data packets; Delete the video data of this partition, retain n-1 data packets, and transfer the learning data set after deleting the video data to the data management optimization unit; The data management optimization unit performs parameter optimization processing on the parameter data within the learning data set and the data packet to be trained, optimizes a large number of repeated parameters, and transmits the optimized learning data set to the learning database for storage.

2. The intelligent optimization control system based on machine self-learning according to claim 1 is characterized in that: The specific method of the data packet screening unit for screening different learning data sets is as follows: Obtain the front-end training data of the learning dataset, which is 30% of the front-end data of the learning dataset; Several groups of front-end training data are merged, analyzed and compared to obtain the corresponding repeatability parameter CFi, where i represents different front-end training data. The repeatability parameter CFi is compared with the preset parameter Y1. When CFi ≥ Y1, the merged and compared learning data sets are classified into the same group of partitions. Otherwise, the merged partitions are not performed.

3. The intelligent optimization control system based on machine self-learning according to claim 2 is characterized in that: The specific method in which the data management optimization unit optimizes a large number of repeated parameters is as follows: Merge and analyze the parameter data and the data packets to be trained, check for duplicate data, extract different duplicate data, and transfer them to the learning database for storage; After storage, the duplicate data are marked with a to-be-learned marker, and the to-be-learned marker is in the form of Dt, where t = 1, 2, ..., m, and D1 and D2 represent different duplicate data. The tags to be learned and the duplicate data are bundled together to generate a table of data to be extracted. The table of data to be extracted is transferred to the learning database for storage, and redundant duplicate data is deleted.

4. The intelligent optimization control system based on machine self-learning according to claim 3 is characterized in that: The autonomous operation unit receives the training data that has been replaced with the to-be-learned marker, extracts duplicate data using the to-be-learned marker, completes the training data, and then performs training processing on its own using the training data.

Images