Knowledge base-based incomplete knowledge reasoning method and system
Through incomplete knowledge reasoning methods based on the knowledge base, the problem of low knowledge management efficiency in the knowledge base is solved, the effective reuse and automatic update of knowledge is achieved, and the design efficiency and knowledge quality of the knowledge base are improved.
Patent Information
- Application Number
- CN202411093776.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-11-15
AI Technical Summary
The existing technology cannot effectively organize and manage existing empirical knowledge, which affects the design speed and efficiency of the knowledge base.
Using incomplete knowledge inference method based on the knowledge base, the knowledge of device design parameters is obtained, characterized and stored in the database, assigned weighting, calculate the similarity, and set the threshold. If the similarity is high, the knowledge will be multiplexed, otherwise it will be divided into the training set. Perform deep learning inference with the test set and update the parameter library.
It realizes effective management and reuse of knowledge, reduces dependence on professionals, reduces production costs, ensures continuous accumulation and inheritance of knowledge, avoids knowledge loss, and improves design efficiency.
Smart Images

Figure CN118966340A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of knowledge engineering, and in particular to an incomplete knowledge reasoning method and system based on a knowledge base. Background Art
[0002] The concept of knowledge engineering was first proposed by Edward Feigenbaum at the Fifth International Conference on Artificial Intelligence held at Stanford University in 1977. He described knowledge engineering as using artificial intelligence principles and methods to solve practical application problems that are difficult to deal with in the expert field with traditional methods. The knowledge base is one of the core modules of knowledge engineering. It is based on the traditional database concept and created using artificial intelligence technology, covering functions such as knowledge acquisition, storage and management. The knowledge base is essentially a collection of knowledge, containing concrete data information and abstract instance information, where the data information is similar to the tuples that describe the geometric information of products in the database. The knowledge base system uses the ideas of artificial intelligence to study databases, logical reasoning and application technologies to realize the intelligent management of database systems. This allows users to reuse the knowledge in the typical model knowledge base when designing different solutions, and pass on product feature information and expert experience through knowledge acquisition technology, so as to apply this knowledge in new product design.
[0003] Since the concept of knowledge engineering was proposed, scholars at home and abroad have conducted extensive research on its key technologies. Xu Baoxiang et al. pointed out that knowledge representation is one of the key technologies of knowledge engineering, and a suitable knowledge representation method is the premise for building an efficient and reliable knowledge engineering system. Nian Zhigang et al. studied and summarized the commonly used knowledge representation methods, and analyzed their respective characteristics and advantages and disadvantages. Hatzilygeroudis I et al. combined production rules with neural algorithms to reduce conflicts between rules. Tu Chengyi applied case-based reasoning technology to tooling design and established a case library. Qiu Junping et al. investigated the research status of knowledge engineering in my country and pointed out that technical research mainly focuses on knowledge acquisition, knowledge representation, knowledge reasoning, expert systems, knowledge bases and knowledge management, while practical application research mainly focuses on computer science and mechanical engineering. Liu Qi et al. studied the application of knowledge engineering technology in the field of traditional Chinese medicine, analyzed the construction method, knowledge representation and reasoning method of traditional Chinese medicine knowledge base, and believed that word vector knowledge representation method and case-based reasoning method have good application prospects in the field of traditional Chinese medicine. In summary, the research and application of knowledge engineering technology at home and abroad have developed rapidly, and the research on key technologies mainly focuses on knowledge representation, knowledge reasoning and knowledge base construction.
[0004] However, in the process of building a knowledge base, it is not enough to just store knowledge in a computer. The organization, management, quantity and update of knowledge directly affect the speed and efficiency of knowledge-based design. How to effectively organize and manage existing experience knowledge is an urgent problem to be solved. Summary of the invention
[0005] 1. Technical issues to be resolved
[0006] In view of the deficiencies of the prior art, the present invention provides an incomplete knowledge reasoning method and system based on a knowledge base, which solves the technical problem that the existing knowledge engineering technology cannot effectively organize and manage the existing empirical knowledge.
[0007] (II) Technical solution
[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:
[0009] In a first aspect, the present invention provides an incomplete knowledge reasoning method based on a knowledge base, comprising:
[0010] Acquire knowledge of the device’s design parameters;
[0011] Characterize parameter knowledge, store the characterized parameter knowledge in a database, and establish a parameter library;
[0012] Assigning weights to various design parameters of the target device to be designed;
[0013] Calculate the similarity between the weighted design parameters and the existing parameter knowledge in the parameter library;
[0014] Set a similarity threshold. If the similarity is greater than the similarity threshold, use the existing parameter knowledge in the parameter library to reuse effective knowledge. Otherwise, divide the existing knowledge into training set and test set according to the ratio.
[0015] Deep learning is performed based on the divided training set and test set, incomplete knowledge reasoning is performed, and the reasoning results are obtained and output.
[0016] Preferably, the inference result is also used to update the parameter library.
[0017] Preferably, the parameter knowledge is characterized, and the characterized parameter knowledge is stored in a database to establish a parameter library, including:
[0018] Store in a MySQL database in a deterministic storage format and create a parameter library;
[0019] The qualitative storage format includes ID, name and formula.
[0020] Preferably, the step of assigning weights to various design parameters of the target device to be designed includes:
[0021] Obtain experts' scoring data on the importance of various design parameters of the target device. The scoring data constitutes a data set, including:
[0022] Obtain scores from n experts engaged in phased array antenna design on the importance of each design parameter, and the scoring data constitute a data set A = {a1, a2, a3, ..., a n}, sort the data in the set in descending order, starting with the subscript 0, and obtain a new data set B = {b0, b1, b2, ..., b n-1};
[0023] Calculate the weight of the combination of scoring data, including:
[0024] Use the number of combinations to calculate the weight of the scoring data and get the weighted vector θ f+1 ,have
[0025]
[0026] In the formula, is the number of combinations of removing k data from n-1 data, and
[0027] According to the combination weight of the scoring data and the data set, the absolute weight of the design parameter is obtained, and the absolute weight is normalized to obtain the relative weight of the parameter as the weight value, including:
[0028] The data set A is weighted by the combination number weight to obtain the absolute weight of a single design parameter:
[0029]
[0030] Similarly, the absolute weights of multiple design parameters can be calculated, and the relative weights of the design parameters can be normalized and calculated as weights:
[0031]
[0032] Among them, w j That is the weight of the design parameter j.
[0033] Preferably, the similarity calculation between the weighted design parameters and the existing parameter knowledge in the parameter library includes:
[0034] The design parameters are taken as the parent sequence X0, as the data sequence that can reflect the design characteristics, and the existing parameter sequence is taken as the subsequence X p , as a data sequence composed of factors affecting the system behavior, the antenna parameter variables are then preprocessed:
[0035]
[0036] Calculate the correlation coefficient between each parameter in the subsequence and the parent sequence using the following formula:
[0037] X0=(x0(1),x0(2),...,x0(j)) T
[0038] X1=(x1(1),x1(2),...,x1(j)) T
[0039] …
[0040] X q =(x q (1),x q (2),...x q (j)) T
[0041] Then the correlation coefficient is |X0-X p |=|x0(k)-x p (k)
[0042] The minimum difference between the two poles is a=min p min k |x0(k)-x p (k)|, the maximum difference between the two poles is b=max p max k |x0(k)-x p (k)|, then the method of calculating the correlation degree is:
[0043]
[0044] Among them, ρ is the resolution coefficient;
[0045] w represents the weight of the design parameter, and the sum of the weights of each design parameter is equal to 1, and the value of each attribute is between (0, 1);
[0046] w={w1,w2,…w j ,…w m} j=1,2,……m
[0047] Then the correlation coefficient is |X0-X p |=w*|x0(k)-x p (k)|;
[0048] The minimum difference between the two poles is a=min p min k w*|x0(k)-xp (k)|;
[0049] The maximum difference between the two poles is b=max p max k w*|x0(k)-x p (k)|;
[0050] The similarity between the weighted design parameters and the existing parameters in the parameter library is calculated by the formula:
[0051]
[0052] Preferably, the deep learning is performed according to the divided training set and test set, and incomplete knowledge reasoning is performed to obtain the reasoning result, including:
[0053] Build a multi-layer perceptron model, train the multi-layer perceptron model through the divided training set and test set, and obtain the inference model;
[0054] The target device to be designed is inferred and predicted through the inference model to obtain the inference result.
[0055] In a second aspect, the present invention provides an incomplete knowledge reasoning system based on a knowledge base, comprising:
[0056] A knowledge acquisition module is used to acquire the design parameter knowledge of the device;
[0057] The parameter library establishment module is used to characterize parameter knowledge, store the characterized parameter knowledge in the database, and establish a parameter library;
[0058] An assignment module, used for assigning weights to various design parameters of the target device to be designed;
[0059] A similarity calculation module is used to calculate the similarity between the weighted design parameters and the existing parameter knowledge in the parameter library;
[0060] The judgment module is used to set a similarity threshold. If the similarity is greater than the similarity threshold, the existing parameter knowledge in the parameter library is used for effective knowledge reuse; otherwise, the existing knowledge is divided into a training set and a test set according to the ratio;
[0061] The reasoning module is used to perform deep learning based on the divided training set and test set, execute incomplete knowledge reasoning, obtain reasoning results and output them.
[0062] Preferably, the inference result is also used to update the parameter library.
[0063] In a third aspect, the present invention provides a computer-readable storage medium storing a computer program for incomplete knowledge reasoning based on a knowledge base, wherein the computer program enables a computer to execute the incomplete knowledge reasoning method based on a knowledge base as described above.
[0064] In a fourth aspect, the present invention provides an electronic device, comprising:
[0065] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include methods for executing the incomplete knowledge reasoning method based on the knowledge base as described above.
[0066] (III) Beneficial effects
[0067] The present invention provides an incomplete knowledge reasoning method and system based on a knowledge base. Compared with the prior art, it has the following beneficial effects:
[0068] The knowledge engineering technology used in the embodiment of the present invention effectively promotes the management and reuse of various device design knowledge. Through the design method based on knowledge engineering, all the detailed knowledge and expert experience in the design process can be stored. This not only reduces the dependence on professionals and reduces production costs, but also ensures the continuous accumulation and inheritance of design knowledge and avoids knowledge loss caused by personnel turnover. BRIEF DESCRIPTION OF THE DRAWINGS
[0069] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0070] Figure 1 It is a block diagram of an incomplete knowledge reasoning method based on a knowledge base;
[0071] Figure 2 It is a specific flow chart of an incomplete knowledge reasoning method based on a knowledge base;
[0072] Figure 3 The parameter library diagram of the constructed phased array antenna;
[0073] Figure 4 This is the structure diagram of the constructed instance library data table;
[0074] Figure 5a-5c It is the model file diagram of antenna array 4*4, 8*8, 12*16;
[0075] Figure 6 This is the training flowchart for the inference model;
[0076] Figure 7 It is a graph of changes in training loss on the training set;
[0077] Figure 8 It is a graph of training loss changes on the test set;
[0078] Fig. 9 It is a graph of the accuracy change on the training set;
[0079] Fig.10 It is a graph of the accuracy change on the test set. DETAILED DESCRIPTION
[0080] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0081] The embodiments of the present application provide an incomplete knowledge reasoning method and system based on a knowledge base to solve the technical problem that the existing knowledge engineering technology cannot effectively organize and manage the existing empirical knowledge, and realize that the adopted knowledge engineering technology effectively promotes the management and reuse of various device design knowledge.
[0082] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:
[0083] At present, the research on knowledge engineering technology at home and abroad mainly focuses on knowledge representation, knowledge reasoning and knowledge base construction. However, in the process of knowledge base construction, it is not enough to just store knowledge inside the computer. The organization, management, quantity and update methods of knowledge directly affect the speed and efficiency of knowledge-based design. How to effectively organize and manage existing empirical knowledge, and quickly and intelligently update knowledge when knowledge is insufficient, is of great significance to the theoretical development and practical application of knowledge engineering. Therefore, in terms of engineering applications, for phased array antenna design knowledge, research on knowledge classification, knowledge acquisition, knowledge representation, creation and automatic update of knowledge base,
[0084] In order to solve the above problems, the embodiment of the present invention proposes an effective knowledge classification method and knowledge management means. At the same time, in order to solve the problem of automatic knowledge update when knowledge experience is insufficient, the present invention combines knowledge update with deep learning and proposes an incomplete knowledge reasoning method based on a knowledge base. The method aims to effectively reuse existing knowledge, automatically update knowledge, improve the efficiency and reusability of mature data, and thus improve the efficiency of antenna design.
[0085] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0086] The embodiment of the present invention provides an incomplete knowledge reasoning method based on a knowledge base, such as Figure 1 As shown, including:
[0087] S1. Obtaining knowledge of device design parameters
[0088] S2, characterize the parameter knowledge, store the characterized parameter knowledge in a database, and establish a parameter library;
[0089] S3, assigning weights to various design parameters of the target device to be designed;
[0090] S4, calculating the similarity between the weighted design parameters and the existing parameter knowledge in the parameter library;
[0091] S5. Set a similarity threshold. If the similarity is greater than the similarity threshold, use the existing parameter knowledge in the parameter library to reuse effective knowledge; otherwise, divide the existing knowledge into a training set and a test set according to the ratio;
[0092] S6. Perform deep learning based on the divided training set and test set, execute incomplete knowledge reasoning, obtain the reasoning results and output them.
[0093] The knowledge engineering technology used in the embodiment of the present invention effectively promotes the management and reuse of various device design knowledge. Through the design method based on knowledge engineering, all the detailed knowledge and expert experience in the design process can be stored. This not only reduces the dependence on professionals and reduces production costs, but also ensures the continuous accumulation and inheritance of design knowledge and avoids knowledge loss caused by personnel turnover.
[0094] Combine the following Figure 2 A detailed description of each step:
[0095] In one embodiment, S1, obtaining the design parameter knowledge of the device. The specific implementation process is as follows:
[0096] In the embodiment of the present invention, in addition to obtaining the design parameter knowledge of the device, the instance knowledge of the device is also obtained, and the design parameter knowledge of the device and the instance knowledge of the device are classified to obtain parameter class knowledge and instance class knowledge. The specific implementation process is as follows:
[0097] It should be noted that in the embodiment of the present invention, the device takes a phased array antenna as an example.
[0098] Some design knowledge of phased array antennas is classified. Parameter knowledge includes radiation phase, radiation amplitude, half-power beam width, polarization mode, cross-polarization discrimination, directivity coefficient, antenna gain, main lobe and side lobe width, beam angle and other parameters. Instance knowledge mainly includes models of known parts, design technical parameters of phased array antennas, components, general assembly files, etc.
[0099] In one embodiment, S2, parameter knowledge is characterized, and the characterized parameter knowledge is stored in a database to establish a parameter library. The specific implementation process is as follows:
[0100] The characterization management of different types of knowledge is divided into the characterization management of parameter knowledge and instance knowledge. For parameter knowledge, that is, the design parameters of each component, such as the size, power, gain, weight, material and other knowledge of the array antenna, these are deterministic knowledge. This type of knowledge is stored in the MySQL database in the storage format of ID, name, formula, etc. to establish a parameter library. For instance knowledge, that is, the existing phased array antenna instance model, instance knowledge can be divided into whole machine instance and component instance. Instance knowledge is stored in the MySQL database in the form of instance number ID, instance document, component number ID, component document, quantity and description to establish an instance library.
[0101] In one embodiment, S3, weighting is performed on each design parameter of the target device to be designed. The specific implementation process is as follows:
[0102] S301, obtaining expert scoring data on the importance of each design parameter of the target device, the scoring data constitutes a data set. Including:
[0103] Invite n experts engaged in phased array antenna design to rate the importance of the above design parameters on a 10-point scale (of course, other rating systems can also be used in the specific implementation process). The rating data constitute a data set A = {a1, a2, a3, ..., a n}, sort the data in the set in descending order, starting with the subscript 0, and obtain a new data set B = {b0, b1, b3, ..., b n-1}.
[0104] S302, calculating the weight of the combination number of scoring data, including:
[0105] Use the number of combinations to calculate the weight of the scoring data, that is, the number of combinations determines the data b i The weight of the weight vector θ f+1 ,have
[0106]
[0107] In the formula, is the number of combinations of removing k data from n-1 data, and
[0108] S303, according to the combination weight of the scoring data and the data set, the absolute weight of the design parameter is obtained, and the absolute weight is normalized to obtain the relative weight of the parameter as the weight value. Including:
[0109] The data set A is weighted using the combination number weight calculated in S302 to obtain the absolute weight of the parameter:
[0110]
[0111] Similarly, the absolute weights of multiple design parameters can be calculated, and the relative weights of the design parameters can be normalized and calculated as weights:
[0112]
[0113] Among them, w j That is the weight of the design parameter j.
[0114] In one embodiment, S4, similarity calculation is performed between the weighted design parameters and the existing parameter knowledge in the parameter library. The specific implementation process is as follows:
[0115] S401, taking the design parameters as the parent sequence X0, as the data sequence that can reflect the design features, and taking the existing parameter sequence as the subsequence X p , as a data sequence composed of factors affecting the system behavior, the antenna parameter variables are then preprocessed:
[0116]
[0117] S402: Calculate the correlation coefficient between each parameter in the subsequence and the parent sequence, using the following formula:
[0118] X0=(x0(1),x0(2),...,x0(j)) T
[0119] X1=(x1(1),x1(2),...,x1(j)) T
[0120] …
[0121] X q =(x q (1),x q (2),...x q (j)) T
[0122] Then the correlation coefficient is |X0-X p |=|x0(k)-x p (k)|
[0123] S403, record the minimum difference between the two poles as a=min p min k |x0(k)-x p (k)|, the maximum difference between the two poles is b=max p max k |x0(k)-x p (k)|, then the method of calculating the correlation degree is:
[0124]
[0125] Here, ρ is the resolution coefficient, which is usually taken as 0.5.
[0126] S404, w is used to represent the weight of the design parameter, and each design parameter satisfies that the sum of the weights is equal to 1, and the value of each attribute is between (0, 1).
[0127] w={w1,w2,…w j ,…w m} j=1,2,……m
[0128] Then the correlation coefficient is |X0-X p |=w*|x0(k)-x p (k)|;
[0129] The minimum difference between the two poles is a=min p min k w*|x0(k)-x p (k)|;
[0130] The maximum difference between the two poles is b=max p max k w*|x0(k)-x p (k)|.
[0131] The similarity between the weighted design parameters and the existing parameters in the parameter library is calculated by the formula:
[0132]
[0133] In one embodiment, S5, a similarity threshold is set. If the similarity is greater than the similarity threshold, the existing parameter knowledge in the parameter library is used to reuse effective knowledge; otherwise, the existing knowledge is divided into a training set and a test set according to a ratio. The specific implementation process is as follows:
[0134] Set the similarity threshold and select similar knowledge. Set the similarity threshold to 0.9 and select knowledge with a similarity higher than 0.9 for reuse. If the similarity is lower than the similarity threshold, it is considered that the existing parameter knowledge cannot be reused and incomplete rule reasoning is required. Divide the existing knowledge into training set and test set in a ratio of 7:3, and perform data preprocessing to facilitate subsequent incomplete knowledge reasoning based on deep learning.
[0135] In one embodiment, S6, deep learning is performed based on the divided training set and test set, incomplete knowledge reasoning is performed, and reasoning results are obtained and output. The specific implementation process is as follows:
[0136] In the embodiment of the present invention, a multi-layer perceptron (MLP) model is used to construct an inference model, and a multi-layer perceptron (MLP) model is used to perform deep learning on a divided training set and a test set to obtain an inference model. The training process of the model is prior art and will not be described in detail here.
[0137] The multi-layer perceptron (MLP) model includes an input layer, a first hidden layer, a second hidden layer, and an output layer.
[0138] In the embodiment of the present invention, the input layer of the multilayer perceptron model has 5 features, and the first hidden layer has 64 related neurons. In theory, the calculation formula of the linear layer in the first hidden layer is:
[0139] Z1=X*W1+b1
[0140] Where: X is the input data matrix (batch size × number of features); W1 is the weight matrix (number of features × number of hidden layer neurons); b1 is the bias vector (number of hidden layer neurons).
[0141] Select a suitable activation function for the first hidden layer. After the first hidden layer, apply the ReLU activation function to activate the neural network. Here, the ReLU activation function is selected. The formula of this function is:
[0142] ReLU(z)=max(0,z)
[0143] The role of this activation function is to introduce nonlinearity, so that the model can learn more complex features.
[0144] Define a fully connected layer from the first hidden layer to the second hidden layer. Both the first hidden layer and the second hidden layer have 64 neurons. The calculation formula of the linear layer is:
[0145] Z2=a1W2+b2
[0146] Where a1 is the activation output of the first hidden layer. W2 is the weight matrix (number of hidden layer neurons × number of hidden layer neurons). b2 is the bias vector (number of hidden layer neurons).
[0147] The ReLU activation function is applied after the second hidden layer.
[0148] Define a fully connected layer from the second hidden layer to the output layer. The second hidden layer has 64 neurons and the output layer has 1 neuron. The calculation formula for the output layer is:
[0149] y=a2W3+b3
[0150] Perform a complete forward propagation. The forward propagation function applies the linear layer and the ReLU activation function in sequence, uses the trained inference model to perform gain inference prediction on the required designed antenna, and finally outputs the result as the inference result.
[0151] During the specific implementation process, it also includes: reusing the inferred knowledge and storing it in the original knowledge base for updating.
[0152] The effectiveness of the embodiment of the present invention is verified by simulation experiments below:
[0153] 1. Classify antenna design knowledge into parameter knowledge and instance knowledge:
[0154] Here, only the antenna array which is one of the antenna components is taken as an example for description.
[0155] Antenna array fixed numerical data is classified into parameter knowledge, including common antenna array distribution forms, and the main parameters are the number of antenna array element rows and columns, the spacing between antenna array element rows and columns, aperture length, aperture width, antenna operating frequency, and main parameter parameters (here taking gain as an example). It is classified and managed by field name, field, data type, unit, etc. The specific details are shown in Table 1.
[0156] Table 1 Antenna array parameter knowledge classification management table
[0157]
[0158] The existing component models of the antenna array are divided into instance-based knowledge. In system applications, when the existing instances meet the design requirements, they can be directly selected by users. They mainly include the design technical documents, components, and general assembly documents of phased array antennas. They are classified and managed by instance type, instance model, model file, etc. The specific details are shown in Table 2.
[0159] Table 2 Antenna array instance class knowledge classification management table
[0160]
[0161] 2. Establish phased array antenna parameter library and instance library:
[0162] As shown in Table 1, the parameter knowledge is characterized and the parameter table is designed based on the number of element rows and columns of the antenna array, the element row spacing and column spacing of the antenna array, the aperture length, aperture width, antenna operating frequency and main parameter parameters (gain is taken as an example here), and the phased array antenna parameter library is established using the MySQL database as the carrier. Some data in the parameter table are shown in Table 3, and the parameter library is shown in Table 4. Figure 3 shown.
[0163] Table 3 Phased array antenna design parameters
[0164]
[0165] Treat the instance as an object, and the characteristic parameters as its attributes. The instance information is stored in the instance table, and the attribute values are mapped to field attributes. Since the instance is stored in the instance library of the phased array antenna, and the components are stored in the component library, the parent-child relationship cannot be established. Therefore, the relationship between the instance and the component is established through the child ID field. At the same time, each instance item will correspond to its parameter table, and a relationship is established with the parameter table through the instance ID field, thereby realizing the mapping relationship between the instance library and the parameter library. This method can facilitate the storage and query of instances. The structure of the phased array antenna instance information data table is shown in Figures 5a to 5c shown.
[0166] The antenna array example library contains common antenna array distributions, including antenna arrays with different numbers of rows and columns. Some of its example data are shown in Table 4, and its corresponding model pictures are shown in Figure 4 .
[0167] Table 4 Phased array antenna array example library
[0168]
[0169] 3. Assign weights to the target antenna parameters to be designed:
[0170] First, the C-OWA operator is used to calculate the weights of the characteristic parameters. Eight experts engaged in phased array antenna design are invited to rate the importance of the three characteristic parameters of antenna array length, array width, and maximum gain on a 10-point scale. The scoring data results are shown in Table 5.
[0171] Table 5 Antenna expert scoring table
[0172]
[0173] First, arrange the expert decision-making score data from large to small, and calculate the weighted vector θ f+1 =(0.0078, 0.0547, 0.1641, 0.2734, 0.2734, 0.1641, 0.0547, 0.0078). Then the absolute weight of the parameter is calculated w′=(4.8359, 4.7734, 8.4375). Finally, the relative weight of the calculated parameter is obtained by normalization processing as w=(0.2680, 0.2645, 0.4675). The antenna characteristic parameter weights are shown in Table 6.
[0174] Table 6 Antenna characteristic parameter weight table
[0175]
[0176] 4. Calculate the similarity between the weighted design parameters and the existing parameter knowledge in the parameter library:
[0177] For example, the design parameters required by the user are 24 rows, 9 row spacing, 18 columns, 9.2 column spacing, and 15.96 operating frequency. After searching the parameter library, it is found that there is parameter knowledge with the same number of rows and columns but different row and column spacing and operating frequency. Since there is no parameter knowledge in the parameter library that is exactly the same as the design requirements, the required design antenna gain is assumed to be 52.041, as shown in Table 7. The existing parameter knowledge queried in the parameter library is organized as shown in Table 8. Therefore, we first calculate the similarity between the design parameters and the existing parameter knowledge in the library.
[0178] Table 7 Required design antenna parameters
[0179]
[0180] The similarity between each instance and the design parameters is calculated by combining the weighted grey correlation degree. The results are shown in Table 9.
[0181] Table 9 Antenna example similarity table
[0182]
[0183] 5. Set a similarity threshold. If the similarity is greater than the similarity threshold, use the existing parameter knowledge in the parameter library for effective knowledge reuse; otherwise, divide the existing knowledge into training set and test set according to the proportion:
[0184] The similarity threshold is set to 0.9, and similar knowledge with a similarity higher than 0.9 is selected for knowledge reuse. As can be seen from Table 9, the antenna parameter knowledge stored in the parameter library has a similarity with the target design parameters lower than the similarity threshold of 0.9. It is considered that the existing parameter knowledge does not meet the design requirements, and knowledge reuse cannot be performed, and incomplete knowledge reasoning is required.
[0185] The existing parameter knowledge is divided into training set and test set in a ratio of 7:3. The number of rows, row spacing, number of columns, column spacing, and operating frequency are selected as learning features. Gain is selected as a different label, as shown in Table 10.
[0186] Table 10 Dataset division
[0187]
[0188] 6. Perform deep learning based on the divided training set and test set, perform incomplete knowledge reasoning, and obtain the reasoning results:
[0189] First, based on the divided data set, this simulation experiment uses StandardScaler to standardize the data and converts the data into PyTorch tensors in preparation for model training.
[0190] Then, a multi-layer perceptron (MLP) model is built, which contains three fully connected layers: the input layer is used for 5 feature inputs; the hidden layer 1 contains 64 neurons and the ReLU activation function, which extracts the initial high-dimensional representation of the input features through linear transformation (weight matrix multiplied by input) and nonlinear activation (ReLU); the hidden layer 2 contains 64 neurons and the ReLU activation function, which further processes the high-dimensional representation of the output of the first layer to capture higher-level features and patterns. Finally, the output layer outputs 1 prediction unit. Through this architecture, the model can effectively learn the complex relationship between the input features and the target values, thereby achieving better performance in regression tasks.
[0191] In the training process of this simulation experiment, the Adam optimizer is used, the learning rate is 0.001, the mean square error (MSE) loss function is used, the number of training rounds is set to 500, and the batch size used for each training is 32. Each round of training takes out the training data in batches for training, including forward propagation calculation output, calculation of loss, and back propagation calculation of gradients and updating model parameters. The loss and accuracy of the training and test sets are printed every 50 rounds, and the loss and accuracy are recorded using TensorBoard.
[0192] The training process is as follows Figure 6 The final training results are shown in Figure 7 to Figure 10The final results show that the training loss approaches 0 after 100 rounds, and the accuracy on the training set and test set can reach above 0.9 after 100 rounds, and can reach up to 0.99.
[0193] The trained inference model is used to perform gain inference prediction on the required designed antenna, and the results are shown in Table 11.
[0194] Table 11 Design antenna parameters after inference
[0195]
[0196] 7. Use the inference results and update the parameter library using the inference results:
[0197] The inference results are delivered to designers for use, and are input into the knowledge base for knowledge updating.
[0198] The embodiment of the present invention further provides an incomplete knowledge reasoning system based on a knowledge base, comprising:
[0199] A knowledge acquisition module is used to acquire the design parameter knowledge of the device;
[0200] The parameter library establishment module is used to characterize parameter knowledge, store the characterized parameter knowledge in the database, and establish a parameter library;
[0201] An assignment module, used for assigning weights to various design parameters of the target device to be designed;
[0202] A similarity calculation module is used to calculate the similarity between the weighted design parameters and the existing parameter knowledge in the parameter library;
[0203] The judgment module is used to set a similarity threshold. If the similarity is greater than the similarity threshold, the existing parameter knowledge in the parameter library is used for effective knowledge reuse; otherwise, the existing knowledge is divided into a training set and a test set according to the ratio;
[0204] The reasoning module is used to perform deep learning based on the divided training set and test set, execute incomplete knowledge reasoning, obtain reasoning results and output them.
[0205] It can be understood that the incomplete knowledge reasoning system based on knowledge base provided in the embodiment of the present invention corresponds to the above-mentioned incomplete knowledge reasoning method based on knowledge base. The explanations, examples, beneficial effects and other parts of the relevant contents can refer to the corresponding contents in the incomplete knowledge reasoning method based on knowledge base, and will not be repeated here.
[0206] An embodiment of the present invention further provides a computer-readable storage medium storing a computer program for incomplete knowledge reasoning based on a knowledge base, wherein the computer program enables a computer to execute the incomplete knowledge reasoning method based on a knowledge base as described above.
[0207] An embodiment of the present invention also provides an electronic device, comprising: one or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the incomplete knowledge reasoning method based on the knowledge base as described above.
[0208] In summary, compared with the prior art, the present invention has the following beneficial effects:
[0209] 1. The knowledge engineering technology used in the embodiment of the present invention effectively promotes the classification, management and reuse of various machine design knowledge. In the design process of devices (such as phased array antennas), there is a lot of detailed knowledge involved that requires designers to have professional backgrounds. Through the design method based on knowledge engineering, all the detailed knowledge and expert experience in the design process can be stored. This not only reduces dependence on professionals and reduces production costs, but also ensures the continuous accumulation and inheritance of design knowledge, avoiding knowledge loss due to personnel turnover.
[0210] 2. The incomplete knowledge reasoning method based on the knowledge base proposed in the embodiment of the present invention can also realize the automatic updating of knowledge. Combined with deep learning technology, it can not only automatically learn the characteristics of existing antenna design knowledge, but also perform predictive reasoning when the knowledge is insufficient to meet the design requirements, thereby realizing knowledge updating. This method applies rich knowledge resources as much as possible, and maximizes the reasoning and updating of knowledge while realizing knowledge reuse. This not only maximizes the reuse of knowledge, but also improves the knowledge quality and data scale of the phased array antenna knowledge base, which is of great significance for improving design efficiency and product development capabilities.
[0211] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0212] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method of incomplete knowledge reasoning based on knowledge base, It is characterized in that include: Acquire knowledge of the device’s design parameters; Characterize parameter knowledge, store the characterized parameter knowledge in a database, and establish a parameter library; Assigning weights to various design parameters of the target device to be designed; Calculate the similarity between the weighted design parameters and the existing parameter knowledge in the parameter library; Set a similarity threshold. If the similarity is greater than the similarity threshold, use the existing parameter knowledge in the parameter library to reuse effective knowledge. Otherwise, divide the existing knowledge into training set and test set according to the ratio. Deep learning is performed based on the divided training set and test set, incomplete knowledge reasoning is performed, and the reasoning results are obtained and output.
2. The incomplete knowledge reasoning method based on a knowledge base as claimed in claim 1, It is characterized in that The inference results are also used to update the parameter library.
3. The incomplete knowledge reasoning method based on a knowledge base as claimed in claim 1, It is characterized in that Characterize parameter knowledge, store the characterized parameter knowledge in a database, and establish a parameter library, including: Store in a MySQL database in a deterministic storage format and create a parameter library; The qualitative storage format includes ID, name and formula.
4. The incomplete knowledge reasoning method based on a knowledge base as claimed in claim 1, It is characterized in that The step of assigning weights to various design parameters of the target device to be designed includes: Obtain experts' scoring data on the importance of various design parameters of the target device. The scoring data constitutes a data set, including: Obtain scores from n experts engaged in phased array antenna design on the importance of each design parameter, and the scoring data constitutes a data set A = {a 1 , a 2 , a 3 , …, a n }, sort the data in the set in descending order, starting with the subscript 0, and get a new data set B = {b 0 , b 1 , b 2 , …, b n-1 }; Calculate the weight of the combination of scoring data, including: Use the number of combinations to calculate the weight of the scoring data and get the weighted vector θ f+1 ,have In the formula, is the number of combinations of removing k data from n-1 data, and According to the combination weight of the scoring data and the data set, the absolute weight of the design parameter is obtained, and the absolute weight is normalized to obtain the relative weight of the parameter as the weight value, including: The data set A is weighted by the combination number weight to obtain the absolute weight of a single design parameter: Similarly, the absolute weights of multiple design parameters can be calculated, and the relative weights of the design parameters can be normalized and calculated as weights: Among them, w j That is the weight of the design parameter j.
5. The incomplete knowledge reasoning method based on a knowledge base as claimed in any one of claims 1 to 4, It is characterized in that The similarity calculation between the weighted design parameters and the existing parameter knowledge in the parameter library includes: The design parameters are taken as the parent sequence X 0 , as a data sequence that can reflect the design characteristics, the existing parameter sequence is used as a subsequence X p , as a data sequence composed of factors affecting the system behavior, the antenna parameter variables are then preprocessed: Calculate the correlation coefficient between each parameter in the subsequence and the parent sequence using the following formula: X 0 =(x 0 (1),x 0 (2),...,x 0 (j)) T X 1 =(x 1 (1),x 1 (2),...,x 1 (j)) T …… X q =(x q (1),x q (2),...x q (j)) T Then the correlation coefficient is |X 0 -X p |=|x 0 (k)-x p (k) The minimum difference between the two poles is a=min p min k |x 0 (k)-x p (k)|, the maximum difference between the two poles is b=max p max k |x 0 (k)-x p (k)|, then the method of calculating the correlation degree is: Among them, ρ is the resolution coefficient; w represents the weight of the design parameter, and the sum of the weights of each design parameter is equal to 1, and the value of each attribute is between (0, 1); w={w 1 ,w 2 ,…w j ,…w m }j=1,2,……m Then the correlation coefficient is |X 0 -X p |=w*|x 0 (k)-x p (k)|; The minimum difference between the two poles is a=min p min k w*|x 0 (k)-x p (k)|; The maximum difference between the two poles is b=max p max k w*|x 0 (k)-x p (k)|; The similarity between the weighted design parameters and the existing parameters in the parameter library is calculated by the formula:
6. The incomplete knowledge reasoning method based on a knowledge base as claimed in any one of claims 1 to 4, It is characterized in that The deep learning is performed based on the divided training set and test set, and incomplete knowledge reasoning is performed to obtain reasoning results, including: Build a multi-layer perceptron model, train the multi-layer perceptron model through the divided training set and test set, and obtain the inference model; The target device to be designed is inferred and predicted through the inference model to obtain the inference result.
7. An incomplete knowledge reasoning system based on knowledge base, It is characterized in that include: A knowledge acquisition module is used to acquire the design parameter knowledge of the device; The parameter library establishment module is used to characterize parameter knowledge, store the characterized parameter knowledge in the database, and establish a parameter library; An assignment module, used for assigning weights to various design parameters of the target device to be designed; A similarity calculation module is used to calculate the similarity between the weighted design parameters and the existing parameter knowledge in the parameter library; The judgment module is used to set a similarity threshold. If the similarity is greater than the similarity threshold, the existing parameter knowledge in the parameter library is used for effective knowledge reuse; otherwise, the existing knowledge is divided into a training set and a test set according to the ratio; The reasoning module is used to perform deep learning based on the divided training set and test set, execute incomplete knowledge reasoning, obtain reasoning results and output them.
8. The incomplete knowledge reasoning system based on the knowledge base as claimed in claim 7, It is characterized in that The inference results are also used to update the parameter library.
9. A computer-readable storage medium, It is characterized in that It stores a computer program for incomplete knowledge reasoning based on a knowledge base, wherein the computer program enables a computer to execute the incomplete knowledge reasoning method based on a knowledge base as claimed in any one of claims 1 to 6.
10. An electronic device, It is characterized in that include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the incomplete knowledge reasoning method based on a knowledge base as described in any one of claims 1 to 6.
Citation Information
Patent Citations
Intelligent design system and method for gun barrel
CN105138802A
Method for transforming detection data of skin parts into knowledge
CN107704927A
Aircraft structural part design method based on machine learning
CN114564793A
Method for automatically complementing and updating knowledge graph
CN115033706A
Process database construction method based on additive manufacturing and medium
CN115237878A
