Key-value memory network-based active recommendation method for design knowledge of complex mechatronic systems

Abstract

The application discloses a kind of based on key value memory network's complex electromechanical system design knowledge active recommendation method.First, the scene feature semantic information extraction of software platform log file in the design process of complex electromechanical system is carried out, and scene ontology is established based on scene feature semantic information, and system scene knowledge base is formed by scene ontology and original knowledge base;Then, the scene maximum frequent sequence is used to describe the scene sequence feature similarity between designers;Again, the knowledge item interaction sequence of all designers is learned, and the knowledge item sequence preference vector corresponding to all designers is obtained;Further, input to key value memory network, and the initial design knowledge active recommendation sequence is obtained;Finally, after knowledge item selection, the final design knowledge active recommendation sequence of each designer is obtained.The application obtains design knowledge active recommendation sequence more in line with the needs of designers, and then improves the efficiency of complex electromechanical system design.

Description

A proactive knowledge recommendation method for complex electromechanical system design based on key-value memory networks Technical Field

[0001] This invention belongs to the field of digital design technology and is a method for proactively recommending design knowledge for complex electromechanical systems. Specifically, it involves a method for proactively recommending design knowledge for complex electromechanical systems based on key-value memory networks. Background Technology

[0002] In the design of complex electromechanical systems, designers have access to a wealth of knowledge and experience, as well as advanced software in various disciplines to assist in the design process. However, overall efficiency remains low. The main reasons for this are the numerous software tools available for electromechanical system design and the fragmented organizational structure of design knowledge, leading to platform heterogeneity and knowledge disorientation. This, in turn, creates difficulties in collaboration and knowledge services for complex electromechanical system design. The diverse sources and varying management methods of electromechanical system design knowledge, coupled with its dispersed storage across different information platforms, result in multiple knowledge silos, making it difficult to express them using a unified semantic framework.

[0003] Complex electromechanical system design teams typically utilize various specialized tools, such as 3D design, component design, simulation calculations, stress calculations, and digital simulations. While each discipline's design software and tools are highly targeted, they suffer from numerous hidden costs and risks, including inconsistent formats, information discrepancies, poor data correlation, time-consuming iterative communication, and difficulties in tracing changes. When electromechanical system designers need to coordinate their work, the lack of proactive recommendations based on professional experience and knowledge leads to relatively low collaboration efficiency, casting doubt on the reliability and feasibility of proposed design solutions. Furthermore, the lack of full utilization of the electromechanical system design scenario and personnel preferences makes parallel design difficult, forcing designers to increase the number of design iterations and severely impacting the efficiency of complex electromechanical system design. Summary of the Invention

[0004] To address the problems and needs existing in the background technology, this invention provides a proactive knowledge recommendation method for complex electromechanical system design based on key-value memory networks. It analyzes the characteristics of software platforms in the field of complex electromechanical system design and generates a standardized scene ontology representation. By learning the knowledge sequence preference representation of designers as a basic profile of electromechanical system designers, it integrates key-value memory networks to learn the knowledge feature preference representation of designers, achieving a deep characterization of designers' preferences. Finally, it utilizes the most frequent sequence of scenes to characterize the similarity of designers and proactively recommends complex electromechanical system design knowledge items, thereby reducing the difficulty and complexity for designers to acquire relevant knowledge and improving the work efficiency of complex electromechanical system design.

[0005] The technical solution adopted in this invention is:

[0006] I. An Active Knowledge Recommendation Method for Complex Electromechanical System Design Based on Key-Value Memory Networks

[0007] Step 1: Extract scene feature semantic information from the software platform log files during the design process of complex electromechanical systems to obtain scene feature semantic information of complex electromechanical systems. Based on the scene feature semantic information, establish a design scene ontology of complex electromechanical systems. The system scene knowledge base is composed of the design scene ontology of complex electromechanical systems and the original knowledge base of complex electromechanical systems.

[0008] Step 2: Based on the software platform log files, use the most frequent sequence of scenarios to characterize the similarity of scenario sequence features among designers of complex electromechanical systems;

[0009] Step 3: Perform preference learning on the knowledge item interaction sequences of all complex electromechanical system designers to obtain the knowledge item sequence preference vectors corresponding to all designers;

[0010] Step 4: Input the knowledge item sequence preference vectors of all designers and the system scenario knowledge base into the key-value memory network, and output the initial design knowledge active recommendation sequence corresponding to all designers;

[0011] Step 5: Based on the similarity of scene sequence features among designers of complex electromechanical systems, select knowledge items for the initial design knowledge proactive recommendation sequence of all designers to obtain the final design knowledge proactive recommendation sequence for each designer.

[0012] Step 3 specifically involves:

[0013] The Item2vec algorithm is used to learn the knowledge item interaction sequences of all designers of complex electromechanical systems. The knowledge item embedding representation vectors corresponding to all designers are generated and then input into the corresponding bidirectional long short-term memory networks to obtain the knowledge item sequence preference vectors corresponding to all designers.

[0014] In step 4, the value matrix and key matrix of the key-value memory network are specifically as follows:

[0015] The value memory matrix V consists of the knowledge item sequence preference vectors of all designers. u Let K be the value matrix. Each attribute in the system scene knowledge base is used as a key vector, and the knowledge feature instances corresponding to each attribute in the system scene knowledge base are used as value vectors. The shared key memory matrix K is composed of all the key vectors and the corresponding value vectors, and is denoted as the key matrix.

[0016] Step 5 specifically involves:

[0017] Based on the similarity of scene sequence features among designers of complex electromechanical systems, determine the nearest neighbor set of each designer. Sort all knowledge items of the initial design knowledge proactive recommendation sequence in the nearest neighbor set of each designer in descending order of quantity, and select the top K knowledge items to form the final design knowledge proactive recommendation sequence of each designer.

[0018] II. A computer device

[0019] The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method.

[0020] III. A computer-readable storage medium

[0021] It stores a computer program that, when executed by a processor, implements the steps of the method.

[0022] The beneficial effects of this invention are:

[0023] 1. In the design of complex electromechanical systems, different system platforms have different knowledge organization frameworks, structures, and design feature types, leading to heterogeneity and diversity of scene feature types. This invention captures the semantic information of the features of complex electromechanical system design scenarios, generating a standardized scene ontology representation, thus weakening the information exchange barriers between various system platforms in interactive design. Considering the complexity of complex electromechanical system design, this invention integrates the multi-platform characteristics of complex electromechanical system design scenarios into the proactive knowledge recommendation process, further improving and refining the performance of proactive knowledge recommendation methods.

[0024] 2. Traditional basic profiles of designers are weak in capturing fine-grained priorities of knowledge item feature dimensions. This invention stores attribute and feature information from the knowledge base of complex electromechanical system design in a key-value memory network. The resulting knowledge feature preferences of designers greatly improve the interpretability and performance of the proactive recommendation system, making the proactively recommended knowledge items more in line with the needs of designers, thereby improving the efficiency of complex electromechanical system design. Attached Figure Description

[0025] Figure 1 is a general flowchart of the method of the present invention.

[0026] Figure 2 is a representation model of the ontology of a complex electromechanical system design scenario in an embodiment of the present invention.

[0027] Figure 3 shows a basic portrait of a designer of a complex electromechanical system in an embodiment of the present invention.

[0028] Figure 4 is a diagram illustrating the evolution of knowledge feature preferences of designers of complex electromechanical systems in an embodiment of the present invention. Detailed Implementation

[0029] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] As shown in Figure 1, this embodiment discloses a method for proactively recommending design knowledge for complex electromechanical systems based on key-value memory networks. The specific implementation steps are as follows:

[0031] Step 1: Extract scene feature semantic information from the software platform log files involved in the design process of complex electromechanical systems to obtain scene feature semantic information of complex electromechanical systems. Based on the scene feature semantic information, establish a complex electromechanical system design scene ontology, which contains instances with assigned values. The system scene knowledge base consists of the complex electromechanical system design scene ontology and the original knowledge base of the complex electromechanical system. The original knowledge base of the complex electromechanical system is specifically a database that stores knowledge in the form of entity-relation-entity triples. At the same time, the complex electromechanical system design scene ontology is used as the instance in the original knowledge base that is closest to the ontology concept, and data type attributes and object attributes are added, as shown in Figure 2.

[0032] In practice, the knowledge organization framework, structure and design feature types of different software platforms are not the same in the design process of complex electromechanical systems, which leads to the heterogeneity and diversity of scene feature types. Therefore, it is necessary to build a corresponding feature ontology for each software platform.

[0033] By utilizing the data interfaces provided by various software platforms for complex electromechanical system design, scene features and related information are extracted, including the name, type, and annotations of the scene features. Then, the dependencies between scene features and information such as sketch names, geometric information, and boundary representations are extracted. Each software platform feature instance corresponds to a local data source for the complex electromechanical system design and is associated with the complex electromechanical system design scene ontology.

[0034] Step 2: In different complex electromechanical system design scenarios, the behaviors of designers exhibit a sequential pattern. Based on the software platform log files, the similarity of scenario sequence features among complex electromechanical system designers is characterized using the maximum frequent sequence of scenarios. Specifically: based on the software platform log files, the access frequency of complex electromechanical system designers for each scenario is statistically analyzed. Then, scenarios with low access frequency are filtered according to a frequency threshold to form a header item set. Frequent sequences are constructed based on the header item set, and the similarity of the maximum frequent sequence is calculated based on the longest common subsequence.

[0035] Step 3: As shown in Figure 3, the Item2vec algorithm is used to learn the knowledge item interaction sequences of all complex electromechanical system designers. This generates knowledge item embedding vectors for all designers, which are then input into the corresponding bidirectional long short-term memory (BiLSTM) networks to obtain knowledge item sequence preference vectors for all designers. In other words, the implicit knowledge item sequence preferences of complex electromechanical system designers are represented from the interaction records. The Item2vec algorithm embeds knowledge items according to their order in the interaction logs of the complex electromechanical system designers, allowing the resulting knowledge item embedding vectors to be clustered based on similarity in a low-dimensional space, with knowledge items sharing a common context having closer embedding distances. The forward and backward context of the sequence data is represented using a Bidirectional Long-Short Term Memory (BiLSTM) network to capture and characterize the temporal dependencies of the designers' interaction sequence data, abstractly representing the knowledge sequence preference profile of the complex electromechanical system designers. At each time step of the training model, the transmission results of each input parameter are used as the output of the hidden layer.

[0036] Step 4: Input the knowledge item sequence preference vectors of all designers and the system scenario knowledge base into the key-value memory network, and output the initial design knowledge active recommendation sequence corresponding to all designers;

[0037] Key-value memory networks are used to model the knowledge feature preferences of designers. By storing design attribute information of complex electromechanical systems in key vectors and the designers' knowledge feature preferences in value vectors, the evolution of preferences at the feature level of designers in complex electromechanical systems is modeled, resulting in a designer knowledge feature preference vector. Key-value memory networks are better able to capture and model data features by splitting memory slots into value vectors and key vectors, and then associating key vectors with value vectors in memory. In step 4, the value matrix and key matrix of the key-value memory network are specifically as follows:

[0038] The value memory matrix V consists of the knowledge item sequence preference vectors of all designers. u Let k be the value matrix, and let k be the key vector of each attribute in the system scenario knowledge base. A The knowledge feature instances corresponding to each attribute in the system scenario knowledge base are used as value vectors. Key-value memory networks use a set of vectors This is used to model the preferences of designers of complex electromechanical systems. A shared key memory matrix K, denoted as the key matrix, is formed by all key vectors and their corresponding value vectors.

[0039] At time t, the knowledge item sequence preference vector from the BiLSTM model of the complex electromechanical system designer is used... This means that, as a query on the key-value pair memory network, it is used to address and access the memory of the key vectors in the key matrix K, and to obtain the weights of different attributes assigned by the designers.

[0040] Designer's knowledge item sequence preference vector It may be incompatible with the key matrix K (e.g., dimensionality, etc.), so a nonlinear transformation is achieved through a multilayer perceptron, i.e.

[0041]

[0042] Use the transformed vector The read operation can be represented by the following formula:

[0043]

[0044] in, Represents a preference vector for a given sequence of knowledge items. In the case of a complex electromechanical system designer u, the knowledge feature preference vector at time t is represented by R(), which indicates the read operation of the key-value memory network.

[0045] Emphasizing knowledge sequence preference, while This emphasizes knowledge feature preferences, and these two parts complement each other to better represent the profile of complex electromechanical system designers. Once the key-value pair memory network receives a new interaction record between the complex electromechanical system designer and knowledge items, it embeds a reference vector for writing, and then updates the associated complex electromechanical system designer knowledge feature preference value vector according to the following formula:

[0046]

[0047] Among them, e i W() represents the embedded representation of the new interactive knowledge item, and W() represents the write operation of the key-value memory network. This represents the updated knowledge feature preference value vector of the complex electromechanical system designer u. This represents the modeling of the knowledge feature preferences of the designer u of the complex electromechanical system by the key-value pair memory network before the update. The evolution of the designer's knowledge feature preferences can be maintained through read and write operations, as shown in Figure 4.

[0048] Then, the knowledge item sequence preference vector of complex electromechanical system designers is generated. Knowledge Feature Preference To perform the splicing, the formula is:

[0049]

[0050] in, This represents the preference of designers of complex electromechanical systems at time t.

[0051] For knowledge items related to the design of complex electromechanical systems, the knowledge items of complex electromechanical system designers obtained through the Item2vec algorithm are embedded into the representation vector q. i With entity embedding vector e in the system scenario knowledge base i The formula is expressed by concatenation:

[0052]

[0053] in, This represents the embedded representation vector of knowledge items of designers of complex electromechanical systems after splicing.

[0054] To generate an active recommendation sequence of design knowledge for complex electromechanical systems, the ranking score is calculated according to the following formula:

[0055]

[0056] Where ARP(u, i, t) is the sorting function implemented as the inner product, and MLP(·) is a multilayer perceptron composed of hidden layers, using tanh as the activation function, and through nonlinear transformation... and Mapped to the same space.

[0057] Step 5: Based on the similarity of scene sequence features among designers of complex electromechanical systems, select knowledge items for the initial design knowledge proactive recommendation sequence of all designers to obtain the final design knowledge proactive recommendation sequence for each designer.

[0058] Step 5 specifically involves:

[0059] Based on the similarity of scene sequence features among designers of complex electromechanical systems, determine the nearest neighbor set of each designer. Sort all knowledge items of the initial design knowledge proactive recommendation sequence in the nearest neighbor set of each designer in descending order of quantity, and select the top K knowledge items to form the final design knowledge proactive recommendation sequence of each designer.

[0060] This embodiment discloses a proactive knowledge recommendation method for complex electromechanical system design based on key-value memory networks. By perceiving the large amount of design attribute and model data generated during the interaction process of complex electromechanical system design, it achieves ontology representation of the complex electromechanical system design scenario. Simultaneously, it collects a large amount of log records from designers during the design process and extracts interaction data of existing knowledge items in the system, thereby comprehensively understanding the knowledge sequence preferences and knowledge feature preferences of complex electromechanical system designers. Under different training set ratios, this invention significantly improves the performance of proactive knowledge recommendation for complex electromechanical system design. The key lies in the integration of complex electromechanical system design scenario perception information and attribute and feature information from the design knowledge base during the proactive recommendation process, improving the accuracy of learning designer preferences and thus enhancing the quality of proactive knowledge recommendation for complex electromechanical systems, thereby improving the design efficiency of complex electromechanical systems.

[0061] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for proactively recommending design knowledge for complex electromechanical systems based on key-value memory networks, characterized in that, Includes the following steps: Step 1: Extract scene feature semantic information from the software platform log files during the design process of complex electromechanical systems to obtain scene feature semantic information of complex electromechanical systems. Based on the scene feature semantic information, establish a complex electromechanical system design scene ontology. The system scene knowledge base is composed of the complex electromechanical system design scene ontology and the original knowledge base of the complex electromechanical system. Step 2: Based on the software platform log files, use the maximum frequent sequence of scenes to characterize the scene sequence feature similarity among complex electromechanical system designers. step Step 3: Perform preference learning on the knowledge item interaction sequences of all complex electromechanical system designers to obtain the knowledge item sequence preference vectors corresponding to all designers; Step 3 specifically involves: using the Item2vec algorithm to learn the knowledge item interaction sequences of all complex electromechanical system designers, generating knowledge item embedding representation vectors corresponding to all designers, and then inputting them into the corresponding bidirectional long short-term memory networks to obtain the knowledge item sequence preference vectors corresponding to all designers; Step 4: Input the knowledge item sequence preference vectors of all designers and the system scenario knowledge base into the key-value memory network, and output the initial design knowledge active recommendation sequence corresponding to all designers; In step 4, the value matrix and key matrix of the key-value memory network are specifically: the value memory matrix is ​​composed of the knowledge item sequence preference vectors of all designers. Let be the value matrix. Each attribute in the system scenario knowledge base is used as a key vector, and the knowledge feature instances corresponding to each attribute in the system scenario knowledge base are used as value vectors. The shared key memory matrix is ​​composed of all the key vectors and their corresponding value vectors. Let be the key matrix; Step 5: Based on the similarity of scene sequence features among designers of complex electromechanical systems, select knowledge items for the initial design knowledge proactive recommendation sequence of all designers to obtain the final design knowledge proactive recommendation sequence of each designer.

2. The method for proactively recommending design knowledge for complex electromechanical systems based on key-value memory networks according to claim 1, characterized in that, Step 5 specifically involves: determining the nearest neighbor set for each designer based on the similarity of scene sequence features among designers of complex electromechanical systems; sorting all knowledge items in the initial design knowledge proactive recommendation sequence in the nearest neighbor set of each designer in descending order of quantity; and selecting the top... Each knowledge item constitutes a proactive recommendation sequence of design knowledge for each designer.

3. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 2.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 2.

Images