An intelligent optimization method for heterogeneous data-driven models
By combining greedy strategies and genetic algorithms in heterogeneous data-driven models, data deduplication, missing and abnormalities are handled, and the application difficulties of heterogeneous data models in practical applications are solved, and the accuracy of data processing is achieved. Security ensures global optimal solution selection of data and prevents data theft.
Patent Information
- Application Number
- CN202411295359.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-18
- Publication Date
- 2024-12-27
- Estimated Expiration
- 2044-09-18
AI Technical Summary
The existing heterogeneous data-driven model has difficulties in application in actual social production, especially in terms of data processing and optimization.
Using intelligent optimization methods in the process of data collection, preprocessing, transformation, integration and calling, including the combination of greedy strategies and genetic algorithms, through data deduplication, missing and exception processing, the mutation algorithm is used to avoid local optimal solution convergence, and in Signal cut-off protection is realized in the data security module.
It improves the data processing accuracy and security of heterogeneous data models, ensures global optimal solution selection and prevents data theft, and enhances the flexibility and security of the system.
Smart Images
Figure CN118820923B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to an intelligent optimization method for a heterogeneous data driven model, and in particular to an intelligent optimization method for a heterogeneous data driven model applied in the field of digital data processing. Background Art
[0002] The artificial intelligence big model refers to a machine learning model with super-large-scale parameters and super-powerful computing resources, which can process massive data and complete various complex tasks, such as natural language processing and image recognition. Heterogeneous data refers to data of different types, different versions or data with different structures. This data can be reflected at multiple levels, including the heterogeneity of computer architecture, the heterogeneity of operating system, the heterogeneity of data format, the heterogeneity of data storage location and the heterogeneity of logical model of data storage. It is the intelligent model.
[0003] The invention patent 202010957696.2 specification discloses a method for modeling a physical model based on heterogeneous data source connections. The invention includes accessing heterogeneous data sources, establishing a unified physical model object, rule binding and algorithm analysis. Heterogeneous data sources include relational databases, structured files, structured distributed file directories and streaming data. Access a relational database through the JDBC driver, upload or access structured files / directories, configure a message server that sends structured streaming data, establish a physical model based on business objects, describe model content, functions and external information, define access data source rules, bind data formats and physical model attributes, and write SQL scripts to achieve the ability to process multiple heterogeneous data sources. The present invention solves the problems of data heterogeneity and standardization, ensures system correctness and flexibility, and realizes data sharing.
[0004] The above invention only systematically discloses heterogeneous data and standardization issues, but does not disclose in detail the heterogeneous data optimization method, which makes it still difficult to apply the heterogeneous data driven model to actual social production. Summary of the invention
[0005] In view of the above-mentioned prior art, the technical problem to be solved by the present invention is that the existing heterogeneous data-driven model still has certain difficulties in applying it to actual social production.
[0006] To solve the above problems, the present invention provides an intelligent optimization method for a heterogeneous data-driven model, which mainly includes the following steps:
[0007] S1. Data collection: staff enters the collected data through the data entry module;
[0008] S2. Data preprocessing: the data entered by the data entry module is uploaded to the preprocessing module and preprocessed in the preprocessing module, wherein the data preprocessing includes removing duplicates, missing data and anomalies;
[0009] S3, data conversion, after the data is pre-processed in the pre-processing module, it is uploaded to the data processing terminal, which performs directional conversion of the data, converts the data into the format and structure supported by the data storage module, and converts the data from the relational format to the star model or snowflake model according to actual use needs, and the converted data is loaded into the data storage module for storage;
[0010] S4, data integration, classifying the transformed data in the data storage module and forming multiple data sets to facilitate subsequent search and call work;
[0011] S5. Data call. When calling driving data, heterogeneous data models need to select the best data from the data storage module. The selection methods of the best data mainly include greedy strategy and genetic algorithm.
[0012] In the process of selecting driving data, there are multiple selection criteria, and they are sorted according to the weight ratio of the multiple selection criteria. The selection criteria with higher weight ratios are executed more advanced. The greedy strategy is to select the data that is the optimal solution in all the above multiple selection criteria, and to select the optimal driving data by selecting multiple local optimal solutions.
[0013] The basis of genetic algorithm is selection algorithm. Under a selection criterion, the optional data are numbered. The formula is:
[0014] P n (i)=F(n) / ∑F(x);
[0015] P n (i) represents the probability of data group numbered n being selected under the i-th selection criterion, F(n) represents the selection weight function value of data group numbered n under the i-th selection criterion, and ∑F(x) represents the sum of the selection weight function values of all optional data under the i-th selection criterion. This formula embodies the natural law of "survival of the fittest", that is, individuals with higher fitness are more likely to be selected and continue to participate in the subsequent selection process. After all selection criteria are executed, the P formed by the corresponding data group n (x) to obtain the final selection weight, and the data with the largest selection weight is the optimal solution data, and the selection of the optimal driving data is achieved in a global way.
[0016] However, in the above selection algorithm, there is still a problem that the data converges to the local optimal solution too early, so a mutation algorithm needs to be added to the above selection algorithm:
[0017] P ogn (i) represents the initial probability of the data set numbered n being selected under the i-th selection criterion in the genetic algorithm, I mr Indicates P ogn (i) The mutation rate is a number between 0 and 1, which is used to control the probability and amplitude of mutation. max and S min Then they are P ogn (i) The maximum and minimum value ranges are used to constrain the mutated P ogn (i) value, I random The random value representing the mutation operation is a random number between -1 and 1. The randomness of the mutation operation obtained is as follows:
[0018] P newn =P ogn (i) + I mr *(S max -S min )*I random .
[0019] In the above-mentioned intelligent optimization method of heterogeneous data-driven model, when data needs to be called to provide driving assistance for the heterogeneous data model, data will be selected in the data storage module. When there are fewer selection criteria, a greedy strategy is selected to use multiple local optimal solutions to quickly select the overall optimal solution. When there are more selection criteria, a genetic algorithm is selected to determine the final global optimal solution by calculating the selection weight. At the same time, on the basis of the genetic algorithm, a mutation algorithm is added to avoid the premature convergence of the selection weight during the calculation process due to a single selection criterion, thereby increasing the accuracy of the final selection weight.
[0020] As a further improvement of the present application, data deduplication refers to the process of eliminating duplicate records in a data set. The main reason for the duplication is errors in data collection, storage or processing. Its existence will waste storage space and cause distortion of data analysis results. In the data deduplication work, it is necessary to set a deduplication threshold according to the actual situation of the data. Once the same parts of multiple groups of data reach or exceed the set deduplication threshold, the data containing the most same parts will be retained and the rest of the data will be deleted.
[0021] As a further improvement of the present application, data missing refers to the failure to record or missing values of certain fields in a data set. Missing data will cause incomplete or biased data analysis results. A missing threshold is set for missing data. When the missing portion of the missing data exceeds the missing threshold, it is determined that the data is severely missing and the missing data is directly deleted. When the missing portion of the missing data is less than the missing threshold, it is determined that the data is repairable and the mean, median or mode or other statistics of the data are filled. When the repaired missing data and the same part of the existing data exceed the deduplication threshold, the repaired missing data is deleted first to increase the accuracy of the data.
[0022] As a further improvement of the present application, data anomaly refers to the range of values of certain values in the data set that significantly deviates from normal values or has obvious differences with other data. It is mainly caused by data collection errors, measurement errors or special events. In data anomaly judgment, a normal threshold is set according to the distribution of the data, and values exceeding the normal threshold are treated as abnormal data for processing. For abnormal data, when the data input volume is sufficient, it is directly deleted. When the data input volume is small, data smoothing techniques, such as moving average, exponential smoothing, etc., can be used to correct the abnormal values to make them closer to normal values and de-abnormalize the data.
[0023] As another improvement of the present application, it includes an authentication module plugged into a heterogeneous data model terminal, the authentication module includes a working body, the working part of the authentication module is concentrated on the working body, a communication slot is drilled on the upper end of the working body, a mounting platform is drilled at the opening of the communication slot, a signal relay device is fixedly connected to the bottom plate of the communication slot, a sealing cover matching itself is fixedly connected to the mounting platform, a data security module is fixedly connected to one end of the sealing cover close to the signal relay device, the data security module includes an elastic capsule, a through hole is drilled on the elastic capsule, the position of which matches the signal relay device, the elastic capsule is a hollow structure, the elastic capsule is filled with an air-hardening gel material, the air-hardening gel material is doped with a signal shielding material, a through hole is drilled on the inner wall of the through hole, a hot-melt adhesive block matching itself is fixedly connected in the through hole, an electric heating network is buried in the outer wall of the elastic capsule, and the elastic The capsule is made of high thermal conductivity material. When the heterogeneous data model terminal and the authentication module are working normally, the heterogeneous data model terminal can realize signal connection with the data storage module through the signal relay device. The signal of the data storage module is transmitted to the heterogeneous data model terminal through the signal relay device, and provides driving data for the heterogeneous data model. When the heterogeneous data model terminal is in an abnormal working state and tries to obtain a large amount of data in a short time, the electric heating network will start to heat the air-hardening gel material and the hot-melt adhesive block until the hot-melt adhesive block melts, the air-hardening gel material flows out, and covers the signal relay device. The signal shielding material doped in the air-hardening gel material will affect the operation of the signal relay device, thereby cutting off the signal connection between the data storage module and the heterogeneous data model terminal, preventing the data in the data storage module from being further stolen.
[0024] As another improved supplement of the present application, a plurality of thermally conductive fibers are filled in the aerohardening gel material, and the outer walls of the plurality of thermally conductive fibers are coated with a homopolar magnetic coating. The heat generated by the electric heating network will produce a temperature difference in the aerohardening gel material, and the presence of the thermally conductive fibers will increase the overall temperature transfer effect of the aerohardening gel material, while the homopolar magnetic coating will make the plurality of thermally conductive fibers more evenly distributed in the aerohardening gel material.
[0025] As another improvement supplement to the present application, the working status of the working subject and the electric heating network is strongly correlated, that is, the working subject can work normally only when the electric heating network can work normally, thereby increasing the safety of use of heterogeneous data model terminals and authentication modules.
[0026] In summary, in this application, after the data is collected and entered through the data entry module, it is uploaded to the data processing terminal after deduplication, missing and exception processing in the preprocessing module for data conversion, and the data is converted into the format and structure supported by the data storage module. According to actual usage needs, the data is converted from a relational format to a star model or a snowflake model, etc., and classified in the data storage module to form multiple data sets to facilitate subsequent search and call work.
[0027] When data needs to be called to provide driving assistance for heterogeneous data models, data will be selected in the data storage module. When there are fewer selection criteria, a greedy strategy is selected to use multiple local optimal solutions to quickly select the overall optimal solution. When there are more selection criteria, a genetic algorithm is selected to determine the final global optimal solution by calculating the selection weight. At the same time, a mutation algorithm is added on the basis of the genetic algorithm to avoid premature convergence of the selection weight during the calculation process due to a single selection criterion, thereby increasing the accuracy of the final selection weight.
[0028] At the same time, the specific structure of the authentication module is disclosed, which enables the signal connection between the data storage module and the heterogeneous data model terminal to be cut off in time when the heterogeneous data model terminal is in an abnormal working state and attempts to obtain a large amount of data in a short period of time, thereby preventing the data in the data storage module from being further stolen, and effectively protecting the security of the data in the data storage module. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 This is a schematic diagram of the structure of an intelligent optimization system for a heterogeneous data-driven model according to the first embodiment of the present application;
[0030] Figure 2 This is a main flow chart of the intelligent optimization method of a heterogeneous data-driven model according to the first embodiment of the present application;
[0031] Figure 3 This is a schematic diagram of a data preprocessing method according to a first embodiment of the present application;
[0032] Figure 4 A schematic diagram of driving data selection according to a first embodiment of the present application;
[0033] Figure 5 A pedigree diagram for selecting driving data for the first embodiment of the present application;
[0034] Figure 6 This is a schematic diagram of the structure of a heterogeneous data model terminal and an authentication module according to the second embodiment of the present application;
[0035] Figure 7 This is a schematic diagram of the structure of the authentication module of the second implementation mode of this application;
[0036] Figure 8 An exploded view of the main structure of the authentication module of the second implementation mode of the present application;
[0037] Fig. 9 This is a schematic side cross-sectional view of the authentication module of the second embodiment of the present application;
[0038] Fig.10 This is a schematic diagram of the structure of a data security module according to the second embodiment of the present application.
[0039] Description of the numbers in the figure:
[0040] 1 data storage module, 2 data processing terminal, 3 preprocessing module, 4 data entry module, 5 heterogeneous data model terminal, 6 authentication module, 601 working body, 602 mounting platform, 603 signal relay device, 604 sealing cover, 7 data security module, 701 elastic bag, 702 through hole, 703 hot melt adhesive block, 704 electric heating network, 705 thermal conductive fiber. DETAILED DESCRIPTION
[0041] Two implementation modes of the present application are described in detail below with reference to the accompanying drawings.
[0042] The first implementation method:
[0043] Figure 1 The intelligent optimization system of the heterogeneous data-driven model shown includes a data storage module 1, the data storage module 1 is signal-connected to a data processing terminal 2, the data processing terminal 2 is signal-connected to a plurality of pre-processing modules 3, the plurality of pre-processing modules 3 are signal-connected to a data entry module 4, the data storage module 1 is also signal-connected to a plurality of heterogeneous data model terminals 5, and a plurality of authentication modules are connected between the data storage module 1 and the plurality of heterogeneous data model terminals 5;
[0044] See also Figure 2-Figure 5 , an intelligent optimization method for heterogeneous data-driven models, mainly including the following steps:
[0045] S1, data collection, the staff enters the collected data through the data entry module 4;
[0046] S2, data preprocessing, the data entered by the data entry module 4 is uploaded to the preprocessing module 3, and preprocessed in the preprocessing module 3, wherein the data preprocessing includes removing duplicates, missing and abnormalities;
[0047] Data deduplication refers to the process of eliminating duplicate records in a data set. The main reason for the occurrence of duplicate records is errors in data collection, storage or processing. The existence of duplicate records will waste storage space and cause distortion of data analysis results. In the data deduplication work, it is necessary to set a deduplication threshold according to the actual situation of the data. Once the same part of multiple sets of data reaches or exceeds the set deduplication threshold, the data containing the most same parts is retained, and the rest of the data is deleted;
[0048] Missing data refers to the failure to record or missing values of certain fields in a data set. Missing data can lead to incomplete or biased data analysis results. For missing data, a missing threshold is set. When the missing portion of the missing data exceeds the missing threshold, the data is considered severely missing and deleted directly. When the missing portion of the missing data is less than the missing threshold, the data is considered repairable and the mean, median or mode or other statistics of the data are filled. When the same portion of the repaired missing data and the existing data exceeds the deduplication threshold, the repaired missing data is deleted first to increase data accuracy.
[0049] Data anomaly refers to the range of values of certain values in the data set that significantly deviates from normal values or has obvious differences with other data. It is mainly caused by data collection errors, measurement errors or special events. In data anomaly determination, a normal threshold is set according to the distribution of the data, and values exceeding the normal threshold are treated as abnormal data for processing. For abnormal data, when the data input volume is sufficient, it is directly deleted. When the data input volume is small, data smoothing techniques such as moving average and exponential smoothing can be used to correct the abnormal values to make them closer to normal values and de-abnormalize the data.
[0050] In particular, in this application, the setting and use of the deduplication threshold, the missing threshold and the normal threshold all require those skilled in the art to make reasonable designs according to the real-time data acquisition accuracy and application requirements. This is a well-known technology of those skilled in the art, so it is not disclosed in detail in this application. Those skilled in the art can make reasonable designs according to the existing technology;
[0051] S3, data conversion. After the data is pre-processed in the pre-processing module 3, it is uploaded to the data processing terminal 2. The data processing terminal 2 performs directional conversion of the data and converts the data into the format and structure supported by the data storage module 1. According to actual use needs, the data is converted from a relational format to a star model or a snowflake model, etc. During the conversion process, it is necessary to formulate unified data mapping and conversion rules to ensure that the data is consistent during the conversion process between different systems. In addition, it is also necessary to ensure the integrity and accuracy of the data to avoid data loss or damage. The converted data is loaded into the data storage module 1 for storage;
[0052] S4, data integration, classifying the transformed data in the data storage module 1 and forming multiple data sets to facilitate subsequent search and call work;
[0053] S5, data call, when calling the driving data, the heterogeneous data model needs to select the optimal data from the data storage module 1, and the selection methods of the optimal data mainly include two categories: greedy strategy and genetic algorithm;
[0054] In the process of selecting driving data, there are multiple selection criteria, and they are sorted according to the weight ratio of the multiple selection criteria. The selection criteria with higher weight ratios are executed more advanced. The greedy strategy is to select the data that is the optimal solution in all the above multiple selection criteria, and to select the optimal driving data by selecting multiple local optimal solutions.
[0055] However, as the number of selection criteria increases, the final result of the data selection method of the greedy strategy is likely to deviate from the global optimal solution, and is not suitable for driving heterogeneous data models. Therefore, when the number of selection criteria exceeds the excess threshold preset by the staff according to the actual situation, the driving data selection method is transformed from the greedy strategy to the genetic algorithm, where the basis of the genetic algorithm is the selection algorithm. Under one selection criterion, the optional data are numbered, and the formula is:
[0056] P n (i)=F(n) / ∑F(x);
[0057] P n (i) represents the probability of data group numbered n being selected under the i-th selection criterion, F(n) represents the selection weight function value of data group numbered n under the i-th selection criterion, and ∑F(x) represents the sum of the selection weight function values of all optional data under the i-th selection criterion. This formula embodies the natural law of "survival of the fittest", that is, individuals with higher fitness are more likely to be selected and continue to participate in the subsequent selection process. After all selection criteria are executed, the P formed by the corresponding data group n (x) to obtain the final selection weight, and the data with the largest selection weight is the optimal solution data, and the selection of the optimal driving data is achieved in a global way.
[0058] However, in the above selection algorithm, there is still a problem that the data converges to the local optimal solution too early, so a mutation algorithm needs to be added to the above selection algorithm:
[0059] P ogn (i) represents the initial probability of the data set numbered n being selected under the i-th selection criterion in the genetic algorithm, I mr Indicates P ogn(i) The mutation rate is a number between 0 and 1, which is used to control the probability and amplitude of mutation. max and S min Then they are P ogn (i) The maximum and minimum value ranges are used to constrain the mutated P ogn (i) value, I random The random value representing the mutation operation is a random number between -1 and 1. The randomness of the mutation operation obtained is as follows:
[0060] P newn =P ogn (i) + I mr *(S max -S min )*I random ;
[0061] Among them, I mr The setting determines the probability and magnitude of the mutation operation. The higher the mutation rate, the higher the P ogn (i) The greater the probability of mutation, the greater the diversity of data. However, this may cause the algorithm to be too dispersed in the search space and difficult to converge to the optimal solution. On the contrary, a mutation rate that is too low may cause the algorithm to fall into a local optimal solution too early and lack sufficient exploration ability. Therefore, it is necessary for technicians in this field to select the most appropriate I for the current heterogeneous data model in advance during use. mr , to increase P newn Efficient.
[0062] In particular, when the mutation algorithm is used in the present application, it is necessary to retain the initial probability data calculation data. When the initial selection weights of a data group differ too much, the mutated data is judged to be invalid data, and the initial selection weights are still used for this group of data. When the initial selection weights differ slightly, the selection weight with a larger value is selected as the final selection weight, and the Gates mutation is judged to be effective. The specific difference in data requires technical personnel in this field to make reasonable designs based on the type and usage accuracy of the current heterogeneous data model.
[0063] In the present application, after the data is collected and entered through the data entry module 4, it is uploaded to the data processing terminal 2 after deduplication, missing and exception processing in the preprocessing module 3 for data conversion, and the data is converted into the format and structure supported by the data storage module 1. According to actual usage needs, the data is converted from a relational format to a star model or a snowflake model, etc., and classified in the data storage module 1 to form multiple data sets, which is convenient for subsequent search and call work.
[0064] When it is necessary to call data to provide driving assistance for heterogeneous data models, data will be selected in the data storage module 1. When there are fewer selection criteria, a greedy strategy is selected to use multiple local optimal solutions to quickly select the overall optimal solution. When there are more selection criteria, a genetic algorithm is selected to determine the final global optimal solution by calculating the selection weight. At the same time, based on the genetic algorithm, a mutation algorithm is added to avoid the premature convergence of the selection weight during the calculation process due to a single selection criterion, thereby increasing the accuracy of the final selection weight.
[0065] The second implementation method:
[0066] Figure 6-Figure 10 The heterogeneous data model terminal 5 and the authentication module 6 are shown, the authentication module 6 is plugged into the heterogeneous data model terminal 5, the authentication module 6 includes a working body 601, the working part of the authentication module 6 is concentrated on the working body 601, a communication slot is drilled on the upper end of the working body 601, a mounting platform 602 is drilled at the opening of the communication slot, a signal relay device 603 is fixedly connected to the bottom plate of the communication slot, and a sealing cover 604 matching itself is fixedly connected to the mounting platform 602, and the sealing cover 604 is close to the signal relay device 603. One end of the relay device 603 is fixedly connected to the data security module 7, and the data security module 7 includes an elastic capsule 701, and a through hole 702 is drilled on the elastic capsule 701, and the position matches the signal relay device 603. The elastic capsule 701 is a hollow structure, and the elastic capsule 701 is filled with an air-hardening gel material, and the air-hardening gel material is doped with a signal shielding material. A through hole is drilled on the inner wall of the through hole 702, and a hot melt adhesive block 703 matching itself is fixedly connected in the through hole. The elastic capsule 70 1 is embedded with an electric heating network 704, and the elastic capsule 701 is made of a high thermal conductivity material. When the heterogeneous data model terminal 5 and the authentication module 6 are working normally, the heterogeneous data model terminal 5 can realize the signal connection with the data storage module 1 through the signal relay device 603, and the signal of the data storage module 1 is transmitted to the heterogeneous data model terminal 5 through the signal relay device 603, and provides driving data for the heterogeneous data model. When the heterogeneous data model terminal 5 is in an abnormal working state and tries to obtain a large amount of data in a short time, the electric heating network 704 will start to heat the air-hardening gel material and the hot-melt adhesive block 703 until the hot-melt adhesive block 703 melts, the air-hardening gel material flows out, and covers the signal relay device 603, and the signal shielding material doped in the air-hardening gel material will affect the operation of the signal relay device 603, thereby cutting off the signal connection between the data storage module 1 and the heterogeneous data model terminal 5, and preventing the data in the data storage module 1 from being further stolen.
[0067] The aerohardening gel material is filled with multiple thermal conductive fibers 705, and the outer walls of the multiple thermal conductive fibers 705 are coated with homopolar magnetic coatings. The heat generated by the electric heating network 704 will produce a temperature difference in the aerohardening gel material, and the presence of the thermal conductive fibers 705 will increase the overall temperature transfer effect of the aerohardening gel material, and the homopolar magnetic coating will make the multiple thermal conductive fibers 705 more evenly distributed in the aerohardening gel material. The working states of the working body 601 and the electric heating network 704 are strongly correlated, that is, the working body 601 can only work normally when the electric heating network 704 can work normally, thereby increasing the safety of the use of the heterogeneous data model terminal 5 and the authentication module 6.
[0068] Similar to the first embodiment, in this embodiment, the specific structure of the authentication module 6 is disclosed, so that when the heterogeneous data model terminal 5 is in an abnormal working state and attempts to obtain a large amount of data in a short period of time, the signal connection between the data storage module 1 and the heterogeneous data model terminal 5 is cut off in time, preventing the data in the data storage module 1 from being further stolen, thereby effectively protecting the security of the data in the data storage module 1.
[0069] In view of actual needs, the protection scope of the above-mentioned implementation mode adopted in this application is not limited thereto. Various changes made within the knowledge scope of technical personnel in this field without departing from the concept of this application still fall within the protection scope of the present invention.
Claims
1. An intelligent optimization method for a heterogeneous data-driven model, characterized in that: The main steps include: S1. Data collection: the staff enters the collected data through the data entry module (4); S2, data preprocessing, the data entered by the data entry module (4) is uploaded to the preprocessing module (3), and preprocessed in the preprocessing module (3), wherein the data preprocessing includes data deduplication, data missing and data anomaly removal; S3, data conversion, after the data is pre-processed in the pre-processing module (3), it is uploaded to the data processing terminal (2), and the data processing terminal (2) performs directional conversion on the data, and the converted data is loaded into the data storage module (1) for storage; S4, data integration, classifying the transformed data in the data storage module (1) and forming multiple data sets to facilitate subsequent search and call work; S5. Data call. When calling driving data, the heterogeneous data model needs to select the best data from the data storage module (1). The selection methods of the best data include greedy strategy and genetic algorithm. The invention comprises an authentication module (6) plugged into a heterogeneous data model terminal (5), the authentication module (6) comprising a working body (601), the working part of the authentication module (6) being concentrated on the working body (601), the upper end of the working body (601) being provided with a communication slot, the opening of the communication slot being provided with a mounting platform (602), the bottom plate of the communication slot being fixedly connected with a signal relay device (603), the mounting platform (602) being fixedly connected with a sealing cover (604) matching the sealing cover, the sealing cover (604) being fixedly connected with a data security module (7) at one end close to the signal relay device (603), the data security module (7) comprising an elastic capsule (701), the elastic capsule (701) being provided with a through hole having a position matching the signal relay device (603) (702), the elastic capsule (701) is a hollow structure, the elastic capsule (701) is filled with an air-hardening gel material, the air-hardening gel material is doped with a signal shielding material, a through hole is drilled on the inner wall of the through hole (702), a hot melt adhesive block (703) matching the elastic capsule is fixedly connected in the through hole, an electric heating network (704) is buried in the outer wall of the elastic capsule (701), the elastic capsule (701) is made of a high thermal conductivity material, when the heterogeneous data model terminal (5) is in an abnormal working state, when trying to obtain a large amount of data in a short time, the electric heating network (704) will start, heat the air-hardening gel material and the hot melt adhesive block (703), until the hot melt adhesive block (703) melts, the air-hardening gel material flows out, and covers the signal relay device (603).
2. The intelligent optimization method of a heterogeneous data-driven model according to claim 1, characterized in that: The basis of the genetic algorithm is the selection algorithm. Under a selection criterion, the optional data are numbered. The formula is: P n (i)=F(n) / ∑F(x); P n (i) represents the probability of data group numbered n being selected under the i-th selection criterion, F(n) represents the selection weight function value of data group numbered n under the i-th selection criterion, and ∑F(x) represents the sum of the selection weight function values of all optional data under the i-th selection criterion; Based on the above selection algorithm, a mutation algorithm needs to be added: P ogn (i) represents the initial probability of the data set numbered n being selected under the i-th selection criterion in the genetic algorithm, I mr Indicates P ogn (i) The mutation rate is a number between 0 and 1, which is used to control the probability and amplitude of mutation. max and S min Then they are P ogn (i) The maximum and minimum value ranges are used to constrain the mutated P ogn (i) value, I random The random value representing the mutation operation is a random number between -1 and 1. The randomness of the mutation operation obtained is as follows: P newn =P ogn (i)+I mr *(S max -S min )*I random 。 3. The intelligent optimization method of a heterogeneous data-driven model according to claim 1, characterized in that: Data deduplication refers to the process of eliminating duplicate records in a data set. The main reason for the duplication is errors in data collection, storage or processing. Its existence will waste storage space and cause distortion of data analysis results. In the data deduplication work, it is necessary to set a deduplication threshold according to the actual situation of the data. Once the same parts of multiple groups of data reach or exceed the set deduplication threshold, the data containing the most same parts will be retained and the rest of the data will be deleted.
4. The intelligent optimization method of a heterogeneous data driven model according to claim 1, characterized in that: The data missing refers to the failure to record or missing values of certain fields in the data set. Missing data will cause incomplete or biased data analysis results. A missing threshold is set for missing data. When the missing part of the missing data exceeds the missing threshold, it is determined that the data is severely missing and the missing data is directly deleted. When the missing part of the missing data is less than the missing threshold, it is determined that the data is repairable and the mean, median or mode or other statistics of the data are filled. When the same part of the repaired missing data and the existing data exceeds the deduplication threshold, the repaired missing data is deleted first.
5. The intelligent optimization method of a heterogeneous data driven model according to claim 1, characterized in that: The data anomaly refers to the range of values of certain values in the data set that significantly deviates from the normal value or has obvious differences with other data. It is mainly caused by data collection errors, measurement errors or special events. In data anomaly judgment, a normal threshold is set according to the distribution of the data, and values exceeding the normal threshold are treated as abnormal data for processing. For abnormal data, when the data input volume is sufficient, it is directly deleted. When the data input volume is small, the abnormal value is corrected through data smoothing technology: moving average and exponential smoothing.
6. The intelligent optimization method of a heterogeneous data driven model according to claim 1, characterized in that: The air-hardening gel material is filled with a plurality of heat-conducting fibers (705), and the outer walls of the plurality of heat-conducting fibers (705) are coated with a magnetic coating of the same polarity.
7. The intelligent optimization method of a heterogeneous data driven model according to claim 1, characterized in that: The working states of the working body (601) and the electric heating network (704) are strongly correlated, that is, the working body (601) works normally only when the electric heating network (704) works normally.