Construction method of high-quality research and development data set

Abstract

The invention discloses a construction method of a high-quality research and development data set, which comprises the following steps: acquiring a physical field type preset by a user, and calling corresponding test data and simulation data in a to-be-processed database according to the physical field type; selecting a proper data assimilation method according to the physical field type, and performing data assimilation on the called test data and simulation data according to the selected data assimilation method to obtain an assimilation data set; calling a corresponding mechanism model in a model database according to the physical field type to obtain a target physical law; converting the assimilation data set into a complete physical field data set by using a physical neural network deep learning algorithm according to a target physical law; according to the invention, unification of test data and simulation data is realized, and support is provided for generative artificial intelligence and data and scene decoupling.

Description

Technical Field

[0001] The present invention belongs to the field of computer application technology, and in particular relates to a method for constructing a high-quality research and development data set. Background Art

[0002] In the modern R&D process, in order to solve the data bottleneck in the application of physical simulation and digital-physical integration technologies in test identification, the existing method is to further explore and release the value of physical data and simulation data in the entire life cycle of the equipment, combined with the characteristics of easy repeatability of simulation data, large test boundary range, and high confidence in the accuracy of simulation data after verification by physical data, and focus on the research of physical simulation and digital-physical integration, so that the accumulation of test data and simulation data is increasing. In order to make use of them, the existing technology usually integrates the data sets separately and then uses them. This leads to the existing technology often relying on a single data source in the R&D process, lacking the comprehensive use of multi-source data, resulting in insufficient quality and integrity of the data set. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a method for constructing a high-quality R&D dataset to solve the above technical problems.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A method for constructing a high-quality R&D dataset, including:

[0006] Obtain the physical field type preset by the user, and call the corresponding test data and simulation data in the database to be processed according to the physical field type;

[0007] Select a suitable data assimilation method according to the physical field type, and perform data assimilation on the called test data and simulation data according to the selected data assimilation method to obtain an assimilated data set;

[0008] According to the physical field type, the corresponding mechanism model is called in the model database to obtain the target physical law;

[0009] According to the target physical laws, the assimilated dataset is transformed into a complete physical field dataset using a physical neural network deep learning algorithm.

[0010] Furthermore, the database to be processed includes test data and simulation data of multiple physical fields involved in the research and development process and simulation process of the target research and development project.

[0011] Furthermore, the called test data and simulation data are consistent in time and space.

[0012] Furthermore, the data assimilation method includes a three-dimensional variational data assimilation method suitable for processing continuous data, and an ensemble Kalman filter assimilation method suitable for processing sequential data.

[0013] Furthermore, if there are multiple suitable data assimilation methods at the same time, the overall relative error of each suitable data assimilation method is calculated, and the data set obtained by the data assimilation method with the smallest overall relative error is used as the assimilated data set.

[0014] Furthermore, the obtained mechanism model includes key characteristic data of the corresponding physical field type.

[0015] Furthermore, a method for constructing a high-quality R&D dataset also includes:

[0016] Step 1, obtaining a large set of historical call data of a physical field data set; the large set of historical call data includes multiple small sets of historical call data, each small set of historical call data includes multiple historical call data sorted according to the call order, and the data call pause time between each two small sets of historical call data is greater than the preset pause time;

[0017] Step 2: extract the key attributes of each historical call data in each historical call data set, and mark the key attributes of the data with serial numbers according to the call sequence to generate the historical call intent in the historical call data;

[0018] Step 3: Obtain the first call request currently faced by the physical field data set, extract the second data key attribute of the call data corresponding to the first call request, and match it with the historical call intention to obtain multiple predicted call intentions;

[0019] Step 4: Based on the predicted call intention, the corresponding data key attributes are obtained to construct index information for all the associated data in the physical field data set; the index information contains a data key attribute tag of a type of data;

[0020] Step 5, obtain the nth call request currently facing the physical field dataset, and determine whether the key attributes of the data carried in the nth call request are consistent with the index information; if they are consistent, go to step 6; if not, go to step 8; the initial value of n is 2;

[0021] Step 6: extract the predicted call intent corresponding to the index information consistent with the key attribute of the data carried by the nth call request as the nth predicted call intent;

[0022] Step 7: Based on the nth predicted call intention, obtain the corresponding data key attributes and update the index information of all related data in the physical field data set for the n-1th time; go to step 5 and perform n=n+1 operation;

[0023] Step 8: Collect multiple call data of this call and re-update the historical call intention.

[0024] The beneficial effects of the present invention are:

[0025] The present invention proposes a method for constructing a high-quality R&D data set, which combines the experimental data generated in the R&D process, the test data generated in the test identification process, the operation data during the service of the equipment, etc. with mechanism models such as mathematical physics, and converts them into complete physical field data through a physical neural network deep learning algorithm, thereby realizing the unification of experimental data and simulation data, and providing support for generative artificial intelligence and the decoupling of data and scenarios.

[0026] Other advantages, objectives and features of the present invention will be described in the following description and will be apparent to those skilled in the art to some extent, or they may be taught from the practice of the present invention. The purposes and other advantages of the present invention may be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0027] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0029] Figure 1 A method flow chart of a method for constructing a high-quality research and development data set in an embodiment of the present invention;

[0030] Figure 2 Another method flow chart of a method for constructing a high-quality research and development dataset in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0032] like Figure 1 As shown, the present invention proposes a method for constructing a high-quality research and development data set, comprising:

[0033] S101, obtaining a physical field type preset by a user, and calling corresponding test data and simulation data in a database to be processed according to the physical field type;

[0034] S102, selecting a suitable data assimilation method according to the physical field type, and performing data assimilation on the called test data and simulation data according to the selected data assimilation method to obtain an assimilated data set;

[0035] S103, calling a corresponding mechanism model in a model database according to the physical field type to obtain a target physical law;

[0036] S104. According to the target physical law, the assimilated data set is converted into a complete physical field data set using a physical neural network deep learning algorithm;

[0037] The working principle of the above technical solution is as follows: the present invention proposes a method for constructing a high-quality R&D data set, targeting the data bottlenecks in the application of physical simulation, digital-physical combination and other technologies in test identification, further exploring and releasing the value of physical data and simulation data in the entire life cycle of the equipment, combining the characteristics of easy repeatability of simulation data, large test boundary range, and high confidence in the accuracy of simulation data after verification by physical data, to carry out research on deep learning inversion algorithms, break through the assimilation technology of simulation data and test data, form a hierarchical and scenario-based high-quality R&D data set, establish a standard, annotation and identification system, serve the training needs of deep learning algorithms driven by mechanism and data fusion, and provide data support for the subsequent formation of multi-task thread test identification capabilities and the creation of an identification environment combining physical objects with simulations; specifically, first obtain the physical field type pre-set by the user, which refers to the distribution of the value of a physical quantity in time and space. For example, an electric field can be regarded as an extension of an electric field radiating from a charge in space, and each point in space can have a value of a physical quantity related to the position. ; and according to the type of physical field, the corresponding test data and simulation data are called in the database to be processed, and the database to be processed includes several test data and simulation data in the process of research and development and simulation; then, a suitable data assimilation method is selected according to the type of physical field, and the called test data and simulation data are assimilated according to the selected data assimilation method to obtain an assimilated data set. When assimilating data, selecting an assimilation method with high adaptability is beneficial to reducing system errors and improving data assimilation accuracy. Since the specific data assimilation process belongs to the usual means of personnel in this field, it will not be repeated here. It is worth noting that when assimilating, the test data and simulation data are assimilated in turn, not the sum of the test data and simulation data; at the same time, according to the type of physical field, the corresponding mechanism model is called in the model database to obtain the target physical law. The source of the mechanism model in the model database includes but is not limited to the generation from relevant technical data in existing scientific literature; finally, according to the target physical law, a physical neural network deep learning algorithm (i.e., PINN algorithm - Physics Informed Neural Network) converts the assimilated data set into a complete physical field data set; it is worth noting that when training PINN, the design of the loss function is preferably associated with the selected target physical law to ensure that it contains the assimilated data error term and the physical information error term, so that the final prediction result can provide data support for the identification environment combining the real object and the simulation;

[0038] The beneficial effects of the above technical solution are: through the above technical solution, the test data generated during the research and development process, the test data generated during the test identification process, the operation data during the service of the equipment, etc., are combined with mathematical physics and other mechanism models, and converted into complete physical field data through the physical neural network deep learning algorithm, so as to achieve the unification of test data and simulation data, and provide support for generative artificial intelligence and the decoupling of data and scenarios.

[0039] In one embodiment, the database to be processed includes test data and simulation data of multiple physical fields involved in the research and development process and simulation process of the target research and development project;

[0040] The working principle and beneficial effects of the above technical solution are as follows: In real life, within a research and development organization, multiple research and development projects are usually developed at the same time, but sometimes there are similar or identical research and development processes in some research and development processes of multiple research and development projects. At this time, if a separate database is designed for each research and development project, this part of the test data and simulation data will be lost when constructing a data set for another related research and development process. For the data assimilation process in the data set construction, adding multi-source test data and simulation data can make up for the shortcomings of different test data and simulation data. Therefore, a unified database to be processed is designed for all target research and development projects to store the test data and simulation data of multiple physical fields involved in the research and development process and the simulation process. When calling, the relevant data is called for subsequent data set construction according to the label associated with the physical field type, and the consistency of parameters such as the timestamp and spatial position of the data, which is beneficial to improve the reliability of the subsequent constructed data set. It is worth noting that when storing data, it is preferred to perform quality control on the stored data, such as eliminating outliers and inaccurate data.

[0041] In one embodiment, the called test data and simulation data are consistent in time and space;

[0042] The working principle and beneficial effects of the above technical solution are as follows: when calling data, the called data needs to be preprocessed. The preprocessing is to ensure the consistency of the test data and simulation data in time and space, for example, to ensure the consistency of the timestamp, spatial position and other key parameters of the acquired data, which is beneficial to the subsequent effective assimilation of the test data and simulation data and improve the reliability of data assimilation.

[0043] In one embodiment, the data assimilation method includes a three-dimensional variational data assimilation method suitable for processing continuous data, and an ensemble Kalman filter assimilation method suitable for processing sequential data;

[0044] The working principle and beneficial effects of the above technical solution are as follows: In the prior art, according to the association mechanism between the data assimilation method and the data model, the data assimilation algorithm can be divided into two categories: sequential data assimilation algorithm and continuous data assimilation algorithm, or divided into two categories: linear system assimilation algorithm and nonlinear system assimilation algorithm. In the present application, in order to improve the overall effect of data assimilation and provide a basis for subsequent data set construction, the present application preferably uses a data assimilation algorithm corresponding to the physical field data type to perform data assimilation on the test data and simulation data during data assimilation. It is worth noting that the data assimilation method suitable for processing continuous data also includes a four-dimensional variational data assimilation method, and the data assimilation method suitable for processing sequential data also includes a particle filter algorithm, but since the assimilation methods that are currently widely used and have better effects are the three-dimensional variational data assimilation method and the ensemble Kalman filter assimilation method, the present application prefers to use these two assimilation methods to perform data assimilation operations on the test data and simulation data.

[0045] In one embodiment, if there are multiple suitable data assimilation methods at the same time, the overall relative error of each suitable data assimilation method is calculated, and the data set obtained by the data assimilation method with the smallest overall relative error is used as the assimilated data set;

[0046] The working principle and beneficial effects of the above technical solution are as follows: in actual operation, if there are too many data assimilation methods pre-deployed in the data processing system corresponding to the present method, and the original data to be assimilated happens to be suitable for multiple data assimilation methods, in order to select the optimal assimilated data set so as to improve the reliability of subsequent high-quality research and development data sets, the present application proposes a method of using overall relative error comparison to calculate the relative errors of the assimilated data obtained by all data assimilation methods, and finally to determine the overall error of each data assimilation method by counting all relative errors. It is worth noting that when performing error comparison, similar to the data assimilation operation, the error comparison is performed on the two types of data in turn, rather than on the sum of the two types of data.

[0047] In one embodiment, the acquired mechanism model includes key feature data of the corresponding physical field type; the key feature data is related to the physical field type, such as relevant conservation laws, set boundary conditions and other feature data, which are used to perform feature matching on the mechanism model to be acquired to obtain an accurate mechanism model;

[0048] Furthermore, the mechanism models corresponding to multiple physical fields are obtained, such as thermodynamics-mechanics, water flow-mechanics, and other mechanism models involving multiple physical fields, and the assimilated data sets of multiple corresponding physical fields are obtained. Then, according to the target physical laws of multi-physical fields, the physical neural network deep learning algorithm is used to convert multiple assimilated data sets into complete physical field data sets, that is, the target high-quality R&D data sets, to provide support for generative artificial intelligence and decoupling of data and scenarios.

[0049] like Figure 2 As shown, in one embodiment, a method for constructing a high-quality research and development dataset further includes:

[0050] Step 1, obtaining a large set of historical call data of a physical field data set; the large set of historical call data includes multiple small sets of historical call data, each small set of historical call data includes multiple historical call data sorted according to the call order, and the data call pause time between each two small sets of historical call data is greater than the preset pause time;

[0051] Step 2: extract the key attributes of each historical call data in each historical call data set, and mark the key attributes of the data with serial numbers according to the call sequence to generate the historical call intent in the historical call data;

[0052] Step 3: Obtain the first call request currently faced by the physical field data set, extract the second data key attribute of the call data corresponding to the first call request, and match it with the historical call intention to obtain multiple predicted call intentions;

[0053] Step 4: Based on the predicted call intention, the corresponding data key attributes are obtained to construct index information for all the associated data in the physical field data set; the index information contains a data key attribute tag of a type of data;

[0054] Step 5, obtain the nth call request currently facing the physical field dataset, and determine whether the key attributes of the data carried in the nth call request are consistent with the index information; if they are consistent, go to step 6; if not, go to step 8; the initial value of n is 2;

[0055] Step 6: extract the predicted call intent corresponding to the index information consistent with the key attribute of the data carried by the nth call request as the nth predicted call intent;

[0056] Step 7: Based on the nth predicted call intention, obtain the corresponding data key attributes and update the index information of all related data in the physical field data set for the n-1th time; go to step 5 and perform n=n+1 operation;

[0057] Step 8: Collect multiple call data of this call and re-update the historical call intention;

[0058] The working principle and beneficial effects of the above technical solution are as follows: Under normal circumstances, the call for data in a data set is carried out by determining its storage address and then calling the data according to the storage address, or by determining the key features of a certain type of data and matching and calling the data according to the key features, etc., but these methods are usually used for small data sets. For large data sets, the commonly used means is to speed up data retrieval by establishing indexes, thereby improving data calling efficiency.

[0059] In the actual R&D process, after a R&D dataset is established, it is often not only used in one test scenario, or even in one R&D institution. In collaborative R&D, the existing dataset sharing method is often to share directly through data transmission. However, in the initial stage of dataset construction, the dataset index is often established on the database or data storage system. Therefore, when data is transmitted, the transmitted dataset and index file may not be applicable to the database system of the other party. At this time, the database system in the R&D system needs to be optimized to adapt to the shared data files, which greatly reduces the applicable scenarios of the dataset.

[0060] In order to solve the above technical problems, the present application proposes a method for constructing a data set module, which forms a constructed data set into an independent module with a calling interface and a universal interface. When data transmission is shared, the data set module is shared. The data set module is connected to the database system of each R&D system according to its own universal interface. There is no need to optimize the calling algorithm of the database system in the R&D system to adapt to the index calling method of the data set, thereby improving the universality of the data set.

[0061] In order to make the data set module meet the calling requirements of various database systems, this solution predicts the future data calling requirements by analyzing the calling patterns of historical data, and builds dynamic indexes accordingly to speed up data calling and enhance the adaptability of indexes to different R&D systems. The specific solution is as follows;

[0062] First, obtain a large set of historical call data of the physical field data set, which is all the call data requirements in the historical call data; the large set of historical call data includes multiple small sets of historical call data, each small set of historical call data includes multiple historical call data sorted according to the call order, and the data call pause time between each two small sets of historical call data is greater than the preset pause time. For example, starting from time t, the physical field data set is called 5 times, and the called data are ABCDE respectively. After E is called, there is no call request within the preset time. Starting from the kth time after the preset time, the physical field data set is called again, and the called data is F; then at this time, the set composed of ABCDE is a small set of historical call data, and the call order is sorted as A, B, C, D, E, and the preset pause time is the aforementioned preset time, which is user-defined and preferably determined according to the R&D time of each stage in the R&D process; it is worth noting that ABCDEF do not refer to one data, but to all data called at a certain call;

[0063] Then, the key data attributes of each historical call data in each historical call data set are extracted. The key data attributes are preferably key features in time and space. Then, the key data attributes are serially marked according to the calling sequence. The calling sequence here is the calling sequence of ABCDE described above, for example, the serial number of A is marked as 1, the serial number of B is marked as 2, and so on; finally, according to the serial number marks in the historical call data set and the corresponding key data attributes, the historical call intention of the historical call data set is determined. The historical call intention can be expressed as the calling sequence of each key data attribute in a historical call data set;

[0064] According to the above method, after all the historical call data sets in the historical call data set generate corresponding historical call intentions, the preliminary data sorting work of the data set module intended to be constructed by this application is preliminarily completed. In specific use, for example, when the physical field data set obtains the first call request issued by the external database system, it first responds to the first call request to satisfy the external database system's call to the data in the physical data set. The calling method can be to call by scanning the entire data set, and then obtain the data key attribute of the call data corresponding to the first call request. In order to distinguish it from the aforementioned technical features, it is referred to as the second data key attribute here and hereinafter. According to the second data key attribute, it is matched with the previously sorted historical call intentions. The matching method is to match the second data key attribute with the data key attribute marked as 1 in each historical call intention to determine whether they are the same. If they are the same, it is determined that the match is successful, and the successfully matched historical call intention is used as the predicted call intention. The number of predicted call intentions can be multiple or zero. If it is zero, it is determined that the match is unsuccessful, and step 8 is executed to generate a small set of historical call data according to the data called this time, and then update the historical call intention according to the newly added small set of historical call data;

[0065] After obtaining the predicted call intention, obtain all the related data in the physical field data set for the corresponding data key attributes to construct index information based on these related data; the number of index information corresponding to the number of predicted call intentions is the number of corresponding data key attributes; it is worth noting that the corresponding data key attribute refers to the data key attribute with the serial number marked as 2 in the predicted call intention, and the value of the serial number mark is determined by adding one to the number of updates of the predicted call intention; at the same time, because the physical field data often called during research and development work usually have temporal and spatial consistency, the index information is usually a data key attribute tag containing only one type of data, and the tag is used for the database system to quickly match the index information according to the tag when calling the data next time, thereby improving the index matching efficiency and thus accelerating the data positioning efficiency;

[0066] Since the database system of the R&D system has a different number of calls each time it uses the physical field data set, in order to complete the description of this solution in a shorter expression, this solution introduces the concept of an integer n and constructs a loop for the subsequent calling process so that those skilled in the art can better understand it. For example, after the first call, the nth call request currently facing the physical field data set is obtained. The initial value of n here is 2, but it will increase with the loop, and it is determined whether the key attributes of the data carried by the nth call request are consistent with the index information. When judging, the key attributes of the data carried by the nth call request are compared with the multiple index information obtained above. If any index information is consistent with it, it is determined to be one. If they are consistent, otherwise it is determined that the key attributes of the data carried by the nth call request are inconsistent with the index information. When they are inconsistent, it means that there is no same historical call intention in the current physical field data set module, then a small set of historical call data is generated according to the data of this call, and then the historical call intention is updated according to the newly added small set of historical call data; if they are consistent, the corresponding data is extracted according to the corresponding index information to respond to the nth call request issued by the database system, and the predicted call intention corresponding to the index information consistent with the key attributes of the data carried by the nth call request is extracted as the nth predicted call intention, and other inconsistent predicted call intentions are eliminated. It is worth noting that when they are consistent, the number of nth predicted call intentions is at least 1;

[0067] Based on the obtained nth predicted call intention, the corresponding data key attributes are obtained for all related data in the physical field data set, so as to update the index information according to these related data, and the index information that has not been updated is deleted, thereby realizing dynamic adjustment of the index information; it is worth noting that the corresponding data key attribute refers to the data key attribute marked with serial number 3 in the predicted call intention. From the aforementioned "the value of the serial number mark is determined by adding one to the number of updates of the predicted call intention", it can be seen that the number of updates of the predicted call intention in the nth predicted call intention is n+1, but the number of updates of the index information is n-1; finally, after updating the index information, the aforementioned steps of obtaining the call request are repeated, and a loop of n=n+1 is executed until the loop is jumped out; there are three ways to jump out of the loop, two of which are that the second data key attribute of the data called by the first call request described above does not match the historical call intention and the data key attribute carried by the nth call request is inconsistent with the index information, and the last one is that no call request is received after the preset time, and it is determined that this call is ended, all index information is deleted, and the call of this physical field data set is completed.

[0068] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A method for constructing a high-quality R&D dataset, characterized in that: include: Obtain the physical field type preset by the user, and call the corresponding test data and simulation data in the database to be processed according to the physical field type; Select a suitable data assimilation method according to the physical field type, and perform data assimilation on the called test data and simulation data according to the selected data assimilation method to obtain an assimilated data set; According to the physical field type, the corresponding mechanism model is called in the model database to obtain the target physical law; According to the target physical laws, the assimilated dataset is transformed into a complete physical field dataset using a physical neural network deep learning algorithm.

2. A method for constructing a high-quality R&D dataset according to claim 1, characterized in that: The database to be processed includes test data and simulation data of multiple physical fields involved in the research and development process and simulation process of the target research and development project.

3. The method for constructing a high-quality R&D dataset according to claim 1, characterized in that: The called test data and simulation data are consistent in time and space.

4. The method for constructing a high-quality R&D dataset according to claim 1, characterized in that: Data assimilation methods include three-dimensional variational data assimilation methods suitable for processing continuous data, and ensemble Kalman filtering assimilation methods suitable for processing sequential data.

5. The method for constructing a high-quality R&D dataset according to claim 1, characterized in that: If there are multiple suitable data assimilation methods at the same time, the overall relative error of each suitable data assimilation method is calculated, and the data set obtained by the data assimilation method with the smallest overall relative error is used as the assimilated data set.

6. The method for constructing a high-quality R&D dataset according to claim 1, characterized in that: The obtained mechanism model includes key characteristic data of the corresponding physical field type.

7. The method for constructing a high-quality R&D dataset according to claim 1, characterized in that: Also includes: Step 1, obtaining a large set of historical call data of a physical field data set; the large set of historical call data includes multiple small sets of historical call data, each small set of historical call data includes multiple historical call data sorted according to the call order, and the data call pause time between each two small sets of historical call data is greater than the preset pause time; Step 2: extract the key attributes of each historical call data in each historical call data set, and mark the key attributes of the data with serial numbers according to the call sequence to generate the historical call intent in the historical call data; Step 3: Obtain the first call request currently faced by the physical field data set, extract the second data key attribute of the call data corresponding to the first call request, and match it with the historical call intention to obtain multiple predicted call intentions; Step 4: Based on the predicted call intention, the corresponding data key attributes are obtained to construct index information for all the associated data in the physical field data set; The index information contains a data key attribute tag of a type of data; Step 5: Obtain the nth call request currently facing the physical field dataset, and determine whether the key attributes of the data carried in the nth call request are consistent with the index information; If they are consistent, go to step 6; if they are inconsistent, go to step 8; the initial value of n is 2; Step 6: extract the predicted call intent corresponding to the index information consistent with the key attribute of the data carried by the nth call request as the nth predicted call intent; Step 7: Based on the nth predicted call intention, obtain the corresponding data key attributes and update the index information of all related data in the physical field data set for the n-1th time; go to step 5 and perform n=n+1 operation; Step 8: Collect multiple call data of this call and re-update the historical call intention.

Images