Design scheme joint decision multi-caudal neural network model and training method

Abstract

The invention relates to the technical field of design scheme intelligent decision, and discloses a design scheme joint decision multi-caudal neural network model and a training method, and the method comprises the steps: generating a target data set based on a historical design case; performing staged joint training on a multi-caudal neural network model through the target data set to obtain a scheme decision model; and according to the scheme decision model, performing joint decision on the plurality of design parameters to obtain a target design scheme. The technical problems of small scale, multiple types and non-uniform distribution of the training samples in the scheme type selection service can be effectively solved, and the efficiency and accuracy of scheme decision making are effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent decision-making of design solutions, and in particular to a multi-tail neural network model and training method for joint decision-making of design solutions. Background Art

[0002] With the rapid development of intelligent technology, the design of railway bridge span schemes has been developing in the direction of integration, digitization, visualization and intelligence. The intelligent design of bridge substructures, especially bridge pile group foundations, is an important part of the intelligent design of bridge span schemes. Rapid, accurate and efficient decision-making on bridge pile group foundation schemes will help to automatically search and optimize bridge span schemes. Pile group foundation is a commonly used foundation form in railway bridge engineering, especially in large-span structures that need to bear large loads. The pile group foundation is a pile group composed of multiple piles, which jointly bear the load of the superstructure and transfer these loads to the deep soil or rock layer. Commonly used pile diameters include 1.0m, 1.25m, 1.5m, 1.8m, 2.0m, 2.2m, 2.5m, 3.0m, etc., and commonly used numbers of pile groups include 8, 9, 10, 11, 12, 15, 16, 18, etc. It can be seen that there are more than 10 types of pile diameters and more than 10 types of pile numbers, and there are many combinations. This leads to a "combination explosion" problem in pile foundation scheme decision-making, resulting in low efficiency and poor accuracy in pile foundation scheme decision-making.

[0003] Therefore, how to effectively improve the efficiency and accuracy of program decision-making is a problem that needs to be solved urgently.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a multi-tail neural network model and training method for joint decision-making of design schemes, aiming to solve the technical problem of how to effectively improve the efficiency and accuracy of scheme decision-making.

[0006] To achieve the above objectives, the present application proposes a design scheme joint decision multi-tail neural network model and training method, wherein the method comprises:

[0007] Generate target datasets based on historical design cases;

[0008] The multi-tail neural network model is jointly trained in stages using the target data set to obtain a solution decision model;

[0009] A joint decision is made on multiple design parameters according to the scheme decision model to obtain a target design scheme.

[0010] In one embodiment, generating a target data set based on historical design cases includes:

[0011] Extracting characteristic parameters of the historical design cases to obtain an initial data set;

[0012] The initial data set is augmented and cleaned according to the associated parameters to obtain a target data set.

[0013] In one embodiment, the multi-tail neural network model is jointly trained in stages using the target data set to obtain a solution decision model, including:

[0014] Dividing the target data set into a training set and a test set according to a preset ratio, wherein the target data set includes a plurality of samples, and the samples include different feature parameters and corresponding actual design solutions;

[0015] The multi-tail neural network model is jointly trained in stages by using the training set to obtain a trained multi-tail neural network model, wherein the multi-tail neural network model includes a data embedding layer, a common feature extraction layer, and a multi-tail decision layer;

[0016] Performing a performance test on the trained multi-tail neural network model using the test set to obtain a loss value of the test set;

[0017] The parameters of the trained multi-tail neural network model are updated according to the loss value of the test set to obtain a solution decision model.

[0018] In one embodiment, the multi-tail neural network model is jointly trained in stages using the training set to obtain a trained multi-tail neural network model, including:

[0019] Inputting the characteristic parameters of each sample in the training set into the data embedding layer for embedding processing to obtain a target embedding vector;

[0020] Inputting the target embedding vector into the common feature extraction layer for feature extraction to obtain a feature vector;

[0021] Inputting the feature vector into the multi-tail decision layer for joint decision making to obtain a predicted design solution;

[0022] Comparing the predicted design solution with the corresponding actual design solution to obtain a prediction error;

[0023] The multi-tail neural network model is back-propagated and trained according to the prediction error, and the model weights are updated to obtain a trained multi-tail neural network model.

[0024] In one embodiment, the step of inputting the feature parameters into the data embedding layer for embedding processing to obtain a target embedding vector includes:

[0025] Determine discrete features and continuous features according to the feature parameters;

[0026] Mapping the discrete features into a continuous vector space to obtain a first embedding matrix;

[0027] The continuous features are automatically discretized, discretely embedded, and aggregated to obtain a second embedding matrix.

[0028] A target embedding vector is generated according to the first embedding matrix and the second embedding matrix.

[0029] In one embodiment, the step of inputting the target embedding vector into the common feature extraction layer for feature extraction to obtain a feature vector includes:

[0030] Get the weight matrix and bias vector of the fully connected neural network;

[0031] Feature extraction is performed on the target embedding vector based on the weight matrix and the bias vector to obtain a feature vector.

[0032] In one embodiment, the step of inputting the feature vector into the multi-tail decision layer for joint decision making to obtain a predicted design solution includes:

[0033] Making a decision on a first design parameter according to the characteristic vector to obtain a first decision vector;

[0034] Concatenate the first decision vector and the feature vector to obtain a concatenated vector;

[0035] Making a decision on a second design parameter according to the splicing vector to obtain a second decision vector;

[0036] Performing decision generation of a third design parameter based on the second decision vector and the common feature vector until generation of decision vectors for each design parameter is completed;

[0037] The decision vector generation of each design parameter is summarized to obtain the predicted design solution.

[0038] In addition, to achieve the above purpose, the present application also proposes a design scheme joint decision multi-tail neural network model and training device, the design scheme joint decision multi-tail neural network model and training device comprising:

[0039] A generation module, used to generate a target data set based on historical design cases;

[0040] A training module, used for performing phased joint training on the multi-tail neural network model through the target data set to obtain a solution decision model;

[0041] The decision module is used to make joint decisions on multiple design parameters according to the scheme decision model to obtain a target design scheme.

[0042] In addition, to achieve the above-mentioned objectives, the present application also proposes a design scheme joint decision-making multi-tail neural network model and training device, the device comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the design scheme joint decision-making multi-tail neural network model and training method as described above.

[0043] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the design scheme joint decision multi-tail neural network model and training method as described above are implemented.

[0044] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the design scheme joint decision multi-tail neural network model and training method as described above.

[0045] The present application provides a multi-tail neural network model for joint decision-making on design schemes and a training method. The present application can improve data balance by first generating a target data set based on historical design cases; the multi-tail neural network model is jointly trained in stages using the target data set to obtain a scheme decision model, which can effectively improve the model accuracy and generalization ability; a plurality of design parameters are jointly decided according to the scheme decision model to obtain a target design scheme, which can effectively improve the efficiency and accuracy of scheme decision-making.

[0046] In summary, this application conducts staged joint training of the multi-tail neural network model through a target data set generated based on historical design cases, and makes joint decisions through the pile foundation scheme decision model obtained through training to obtain a pile foundation design scheme, thereby overcoming the technical defects of low efficiency and poor accuracy of design scheme decision-making, and can effectively improve the efficiency and accuracy of scheme decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0048] In order to more clearly illustrate the embodiments of the present application 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, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0049] Figure 1 A flowchart diagram of a joint decision-making multi-tail neural network model and training method embodiment 1 of the design scheme of this application;

[0050] Figure 2 A schematic diagram of a pile group foundation for a joint decision-making multi-tail neural network model and a training method for the first embodiment of the present application;

[0051] Figure 3 A schematic diagram of data distribution of main beam types for designing a joint decision-making multi-tail neural network model and training method for the first embodiment of the present application;

[0052] Figure 4 A distribution diagram of pile foundation types for the joint decision-making multi-tail neural network model and training method for the first embodiment of the present application;

[0053] Figure 5 A network architecture diagram of a pile foundation solution decision model for a multi-tail neural network model and a training method for a joint decision-making design solution of the first embodiment of the present application;

[0054] Figure 6 A flowchart diagram of the joint decision-making multi-tail neural network model and training method embodiment 2 of the design scheme of this application;

[0055] Figure 7 A schematic diagram of the AutoDis principle of the joint decision-making multi-tail neural network model and training method designed for the second embodiment of the present application;

[0056] Figure 8 A schematic diagram of a loss value curve of a joint decision-making multi-tail neural network model and a training method for the second embodiment of the present application;

[0057] Fig. 9 A schematic diagram of a confusion matrix of a joint decision-making multi-tail neural network model and a training method for the second embodiment of the present application;

[0058] Fig.10 A schematic diagram of the curves of accuracy and recall rate versus training batch deformation of the joint decision-making multi-tail neural network model and training method designed in the second embodiment of the present application;

[0059] Fig.11 A schematic diagram of the module structure of a joint decision-making multi-tail neural network model and a training device designed for an embodiment of the present application;

[0060] Fig.12 A schematic diagram of the device structure of the hardware operating environment involved in the design of a joint decision-making multi-tail neural network model and training method for an embodiment of the present application.

[0061] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0062] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0063] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0064] The main solution of the embodiment of the present application is: generate a target data set based on historical design cases; conduct phased joint training of the multi-tail neural network model through the target data set to obtain a scheme decision model; make joint decisions on multiple design parameters according to the scheme decision model to obtain a target design scheme.

[0065] The pile group foundation is composed of multiple piles, which jointly bear the load of the superstructure and transfer these loads to the deep soil or rock layer. Commonly used pile diameters are 1.0m, 1.25m, 1.5m, 1.8m, 2.0m, 2.2m, 2.5m, 3.0m, etc., and the commonly used number of piles in the pile group foundation is 8, 9, 10, 11, 12, 15, 16, 18, etc. It can be seen that there are more than 10 types of pile diameters and more than 10 types of pile numbers. There are many combinations, and the problem of "combination explosion" occurs when making decisions on pile foundation schemes, resulting in low efficiency and poor accuracy in pile foundation scheme decisions. Therefore, how to effectively improve the efficiency and accuracy of scheme decisions is a problem that needs to be solved urgently.

[0066] This application conducts staged joint training of the multi-tail neural network model based on a target data set generated by historical design cases, and makes joint decisions through the pile foundation scheme decision model obtained through training to obtain a pile foundation design scheme, thereby overcoming the technical defects of low efficiency and poor accuracy in design scheme decision-making, and can effectively improve the efficiency and accuracy of scheme decision-making.

[0067] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a design scheme joint decision multi-tail neural network model and a training device, etc. The following takes the design scheme joint decision multi-tail neural network model and the training device as an example to illustrate this embodiment and the following embodiments.

[0068] Based on this, the embodiment of the present application provides a design scheme joint decision multi-tail neural network model and training method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the joint decision-making multi-tail neural network model and training method designed for this application.

[0069] In this embodiment, the design scheme joint decision multi-tail neural network model and training method include steps S10 to S30:

[0070] Step S10, generating a target data set based on historical design cases.

[0071] It should be noted that in this embodiment, a variety of decision-making schemes can be designed, such as pile foundation scheme design, bridge pier scheme design, etc. This embodiment does not specifically limit this. This embodiment takes pile foundation scheme design as an example for explanation. Pile foundation scheme design is pile foundation intelligent selection. Pile foundation intelligent selection is only a preferred application example of the present invention. Other more fields can be applied. By reducing the number of labels of combined classification problems, the sample size can be reduced.

[0072] It is understandable that for pile foundation scheme design, historical design cases are historical pile foundation design cases, which refer to design cases of existing railway bridge pile group foundations. Pile group foundations are composed of multiple piles, which jointly bear the load of the superstructure and transfer these loads to the deep soil or rock layer. The commonly used railway pile diameters are 1.0m, 1.25m, 1.5m, 1.8m, 2.0m, 2.2m, 2.5m, 3.0m, etc. The commonly used number of piles in pile group foundations are 8, 9, 10, 11, 12, 15, 16, 18, etc. The commonly used pile group layout methods include determinant, plum blossom and custom. The pile group foundation is composed of pile diameters, layout types and number of piles to form different types of pile foundation configuration types, such as Figure 2 As shown, Figure 2 Schematic diagram of pile group foundation.

[0073] It can be understood that the target data set is composed of multiple samples, and the combination of the characteristic parameters of each bridge type and the corresponding actual number of piles and actual pile diameter is a sample.

[0074] It is worth noting that, in this embodiment, the design of the pile foundation is mainly aimed at the combination of the diameter of the foundation pile and the number of foundation piles. The diameter of the foundation pile and the number of foundation piles are mainly affected by multiple factors such as design load factors, geological conditions, economic factors, construction technology factors, environmental impact factors, durability and maintenance factors, and regional factors. Design load factors refer to the weight of the bridge, expected traffic load, dynamic load, wind load, temperature change and other design loads that will affect the design of the pile foundation. The greater the load, the greater the number of piles and pile diameters required. Geological conditions refer to the type of soil, layer thickness, bearing capacity, friction angle, cohesion, compressibility, groundwater level and other geological conditions that will affect the design of the pile foundation. For example, in soft soil areas, more piles may be needed to disperse the load, while in hard rock layers, fewer piles may be needed. Economic factors refer to the designer's need to find the most cost-effective balance between the number and diameter of piles under the premise of meeting safety and functional requirements. Construction technology factors refer to the available construction technology and equipment that will also affect the design of the pile foundation. For example, if the construction equipment limits the diameter of the pile foundation, then the number of piles may need to be increased during design to meet the bearing requirements. Environmental impact factors refer to the impact of construction activities on the environment, which also needs to be considered during design. For example, if there is a sensitive ecological environment near the construction area, special design may be required to reduce the impact. Durability and maintenance factors refer to the durability and long-term maintenance costs of the pile foundation, which also need to be considered during design. In some cases, using larger diameter piles may reduce long-term maintenance costs. Regional factors refer to different regions have different bridge design specifications and standards. These specifications will put forward specific requirements for the design of pile foundations, including the number and diameter of piles.

[0075] In a feasible implementation, step 10 may include: extracting characteristic parameters of the historical design cases to obtain an initial data set; and performing data augmentation and data cleaning on the initial data set according to associated parameters to obtain a target data set.

[0076] It should be noted that for the pile foundation scheme design, in order to make full use of the design cases of the existing railway bridge pile group foundation and realize the new design auxiliary decision, it is necessary to effectively characterize the design cases. On the one hand, it can provide a basis for the characterization and quantification of bridge cases, and on the other hand, the effective characterization of design cases can provide data support for the retrieval of similar cases and the reasoning decision of neural network models. In this embodiment, nine characteristic attributes are selected to characterize the cases. These attributes can be divided into numerical type and enumeration type. Among them, the numerical type includes the small mileage side span s1, the large mileage side span s2, the pier height s3, and the average ground slope s4; the enumeration type includes the seismic intensity s5, the small mileage side beam type s6, the large mileage side beam type s7, the pier type s8, and the geological label s9. The pile group scheme is characterized by two parts, namely the pile diameter target1 and the number of piles target2. It can be understood that, in this embodiment, data is extracted from existing cases, and the collected data set contains 1014 records, that is, 1014 samples, each sample record describes the relevant information of a pile group foundation, that is, characteristic parameters and sample labels, wherein the sample label is the type of pile foundation (including quantity and diameter) corresponding to the characteristic parameter, as shown in Table 1. Table 1 is a sample example data table, which includes multiple sample cases, each case consisting of characteristic parameters and sample labels, and the characteristic parameters include small mileage side beam type, large mileage side beam type, small mileage side span, large mileage side span, pier model, pier height, etc. The sample labels include pile foundation type, including side beam type for small mileage, such as simply supported beam, continuous beam, etc.; side beam type for large mileage, such as simply supported beam; side span for small mileage, such as 80m, 32m; side span for large mileage, such as 32m, 24m, etc.; pier type, such as hollow pier and solid pier; pier height, such as 28m, 21m, etc.; seismic fortification level, such as C; ground slope, such as 0%; geological label, such as A; pile foundation type, such as quantity 11, diameter 1.50m, quantity 12, diameter 1.00m, etc.

[0077] Table 1

[0078]

[0079] In the specific implementation, the initial data set collected from the existing cases contains 1014 samples, among which the continuous beam type bridges are the largest, with a total of 455 bridges, while the simply supported beam type bridges have 328 bridges, the continuous beam arch type bridges have 69 bridges, the T-shaped rigid frame type bridges have 64 bridges, the continuous rigid frame type bridges have 46 bridges, the low tower cable-stayed bridge type bridges have 35 bridges, and the simply supported tied arch type bridges have 17 bridges. Figure 3 As shown, Figure 3The data distribution diagram of the main beam type is shown in Figure 2. The data distribution is the same as the current design situation. The main types of bridges currently using pile group foundations are continuous beams and simply supported beams, and there are fewer bridges with other special structures. There are 37 different types of pile foundations in the initial data set, but their distribution in the data set is not uniform. For example, the number of bridges with pile foundations of 8-1.0m is the largest, with 1014 bridges, while the number of other types of pile foundations is relatively small, such as 129 bridges with pile foundations of 10-1.0m, 114 bridges with pile foundations of 10-1.25m, etc. Figure 4 As shown, Figure 4 This is the distribution map of pile foundation types.

[0080] It is understandable that from the distribution of samples in the initial data set, most of the bridges in the initial data set use pile foundation types of 8-1.0m (i.e., 8 pile foundations with a diameter of 1.0m), and the number of other pile foundation types is relatively small. It can be seen that the data distribution in the initial data set is unbalanced, which will lead to the following problems in the subsequent model training of the model: 1. Some pile foundation types (such as 8-1.0m) in the data set are far more than other types, which may cause the model to tend to identify the most common pile foundation type during training. For a few pile foundation types in the data set, the model may not be able to effectively learn their features, thereby affecting the generalization ability of the model; 2. Due to the large number of samples of certain pile foundation types in the data set, the model may be biased towards recognizing the most common pile foundation type during training. 1. Over-focusing on these types of features and ignoring other types of features may result in the model being unable to accurately identify those types of pile foundations with a small number of samples in practical applications; 2. Due to the large number of samples of certain types of pile foundations, the model may overfit on these types, resulting in poor performance on the test set; 3. Due to the uneven distribution of data, it is difficult to determine which features contribute most to the identification of different types of pile foundations, which may lead to difficulties in the feature selection process, because some features may not have enough samples on a few types of pile foundations to verify their effectiveness; 4. The distribution of pile foundation types in the dataset may not be sufficient to cover all possible practical application scenarios, that is, new pile foundation types encountered in actual engineering may not be accurately identified by the model. Therefore, in order to solve the problem of unbalanced data, the initial dataset needs to be augmented and cleaned according to the pile top area and pile group stiffness to obtain the target dataset for training.

[0081] It is worth noting that for pile foundation design, the associated parameters refer to the parameters associated with the diameter of the foundation pile and the number of foundation piles. The pile top area refers to the total area occupied by all pile tops on the cap in the pile foundation design. It is usually used to calculate the size of the pile foundation cap and evaluate the bearing capacity of the cap. In the case of a multi-pile cap (i.e., a pile group cap), the sum of the pile top areas affects the size and shape of the cap, as well as the distribution of the load by the cap. The role of the cap is to evenly distribute the load of the superstructure to each pile. The stiffness of the pile group refers to the overall stiffness exhibited when multiple piles work together in the pile foundation design. When a pile group is subjected to a load, the interaction between each pile will affect the response of the entire pile group system, including settlement and bearing capacity. The change in stiffness caused by the pile group effect is due to the interaction between piles, between piles and soil, and between soil and soil. These interactions may make the stiffness of the pile group different from that of a single pile. Usually, the stiffness of the pile group is smaller than that of a single pile because of the stress dispersion between piles and the coordinated deformation of the soil.

[0082] In the specific implementation, the specific implementation steps of data augmentation and data cleaning of the initial data set according to the pile top area and pile group stiffness are as follows: for the five special beam types with a small number, namely continuous beam arch, T-shaped rigid frame, continuous rigid frame, low-tower cable-stayed bridge, and simply supported tie arch, design cases are added through redesign by design experts; other pile foundation design schemes are added based on the principle that the pile top area and overall stiffness of the pile group foundation do not change by more than 10%; the data is cleaned, and only one of the relatively similar design cases is retained, so as to obtain the final target data set.

[0083] Step S20, performing joint training of the multi-tail neural network model in stages using the target data set to obtain a solution decision model.

[0084] It should be noted that the multi-tailed neural network model is a deep learning model specially designed to handle combined classification problems. Its unique structure can simultaneously learn and optimize multiple classification tasks and share knowledge and features between these tasks.

[0085] It can be understood that for pile foundation scheme design, the scheme decision model is the pile foundation scheme decision model, which is the model ultimately used to predict and evaluate the effectiveness of the pile foundation design scheme. The pile foundation scheme decision model can automatically extract key features from the data through the training of the multi-tail neural network model, and learn the relationship between different feature parameters and the pile diameter and the number of piles.

[0086] It is worth mentioning that phased joint training refers to dividing the entire training process into multiple stages. For pile foundation scheme design, the entire training process is divided into two stages to conduct targeted learning and optimization of the pile diameter and the number of piles, and finally obtain the scheme decision model.

[0087] Step S30, making joint decisions on multiple design parameters according to the solution decision model to obtain a target design solution.

[0088] It should be noted that according to the scheme decision model, the first design parameter, the second design parameter and the Nth design parameter are jointly decided in turn, that is, all design parameters are decided in turn. For the pile foundation scheme design, there are 2 design parameters in total, namely the first design parameter and the second design parameter. The first design parameter is the number of piles, and the second design parameter is the pile diameter. The target design scheme is the pile foundation design scheme. The pile foundation scheme decision model splits the combined decision problem of pile diameter and pile number into two problems: pile diameter decision and pile number decision. By classifying the pile diameter and pile number respectively, the number of decision categories of a single model is reduced. At the same time, in order to ensure that there is still some connection between the two decisions, the decision models of the two problems share some weights. At the same time, the decision result of the pile diameter will be used as the input of the pile number decision model, which directly affects the decision of the pile number, thereby obtaining the pile foundation design scheme.

[0089] It can be understood that the pile foundation scheme decision model is divided into three parts in total, namely the data embedding layer, the common feature extraction layer, and the multi-tail decision layer, and each layer is composed of an MLP fully connected neural network.

[0090] like Figure 5 As shown, Figure 5The network architecture diagram of the pile foundation scheme decision model. The pile foundation scheme decision model includes a data embedding layer, a common feature extraction layer (HIDDEN LAYERS), a multi-tail decision layer (Design-Making LAYERS), and an output layer (OUTPUT LAYER-A, OUTPUT LAYER-B). In the data embedding layer, the input data (INPUT DATA) of the model is divided into continuous data (cont data) and discrete data (dis data). The continuous data (cont data) is automatically discretized (Automatic Dicretization), discrete embedding (Meta Embedding) and aggregation (Aggregation) are performed through the AutoDis module to obtain an embedding matrix. Numerical data (numeric data) and enumerated data (enumerated data) are generated according to the discrete data (dis data). Embedding (Embedding) is performed on the numeric data (numeric data) and enumerated data (enumerated data) to obtain an embedding matrix. Then, an embedding vector is generated according to the two embedding matrices, and the embedding vector is input into the common feature extraction layer (HIDDEN LAYERS). Feature extraction is performed through the fully connected neural network in the common feature extraction layer (HIDDEN LAYERS) to obtain a feature vector. The feature vector is input into the multi-tail decision layer (Design-Making LAYERS) to make a decision on the number of foundation piles. The output layer (OUTPUT LAYER-A) outputs the pile quantity decision vector, which is then concatenated with the feature vector to make a pile diameter decision. The pile diameter decision vector is then output through the output layer (OUTPUTLAYER-B).

[0091] In the specific implementation, the pile foundation scheme decision model is combined with the automatic hole layout algorithm and scheme evaluation algorithm of the railway bridge to obtain the intelligent decision algorithm service for the bridge span scheme. By deploying microservices and accessing the professional beam design system, the overall bridge design scheme can be quickly generated and multiple schemes can be evaluated and compared.

[0092] The present embodiment provides a multi-tail neural network model and training method for joint decision-making of design schemes. The present embodiment can improve data balance by first generating a target data set based on historical design cases; the multi-tail neural network model is jointly trained in stages using the target data set to obtain a scheme decision model, which can effectively improve the model accuracy and generalization ability; a plurality of design parameters are jointly decided according to the scheme decision model to obtain a target design scheme, which can effectively improve the efficiency and accuracy of scheme decision-making.

[0093] In summary, this embodiment conducts stage-by-stage joint training of the multi-tail neural network model through the target data set generated based on historical design cases, and makes joint decisions through the pile foundation scheme decision model obtained through training to obtain the pile foundation design scheme, thereby overcoming the technical defects of low efficiency and poor accuracy of design scheme decision-making, and can effectively improve the efficiency and accuracy of scheme decision-making.

[0094] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can refer to the above introduction, and will not be repeated later. Figure 6 , the step S20 further includes steps S201-S204:

[0095] Step S201, dividing the target data set into a training set and a test set according to a preset ratio, wherein the target data set includes a plurality of samples, and the samples include different feature parameters and corresponding actual design solutions.

[0096] It should be noted that the preset ratio may be 8:2 or other ratios, and this embodiment does not impose any specific limitation on this. This embodiment is described by taking 8:2 as an example.

[0097] In the specific implementation, 1800 samples were selected from the data set as the training set and 229 samples were selected as the test set. The Adam optimizer was used with a learning rate of 0.001 to train the multi-tailed neural network model.

[0098] It can be understood that for pile foundation design, the target data set includes multiple samples, each sample is composed of characteristic parameters of different bridge types and corresponding actual pile foundation types, and the actual pile foundation type refers to a combination of the actual number of piles and the actual pile diameter.

[0099] Step S202, performing joint training on the multi-tail neural network model in stages through the training set to obtain a trained multi-tail neural network model, wherein the multi-tail neural network model includes a data embedding layer, a common feature extraction layer and a multi-tail decision layer.

[0100] It should be noted that the multi-tail neural network model is divided into three parts, namely the data embedding layer, the common feature extraction layer, and the multi-tail decision layer. Each layer is composed of an MLP fully connected neural network. The data embedding layer is used to convert the input feature parameters into a numerical vector that the model can understand, the common feature extraction layer is used to extract the key features of the input data, and the multi-tail decision layer makes decisions based on the extracted features and outputs the pile foundation design plan.

[0101] In a feasible implementation, step 202 may include: inputting the feature parameters of each sample in the training set into the data embedding layer for embedding processing to obtain a target embedding vector; inputting the target embedding vector into the common feature extraction layer for feature extraction to obtain a feature vector; inputting the feature vector into the multi-tail decision layer for joint decision-making to obtain a predicted design scheme; comparing the predicted design scheme with the corresponding actual design scheme to obtain a prediction error; performing back-propagation training on the multi-tail neural network model according to the prediction error and updating the model weights to obtain a trained multi-tail neural network model.

[0102] It should be noted that the common feature extraction layer includes one common fully connected neural network, the multi-tail decision layer includes multiple independent fully connected neural networks connected to the common fully connected neural network, and the N-1th decision vectors of the multiple independent fully connected neural networks are concatenated with the common feature vector to generate the Nth decision vector.

[0103] It can be understood that for pile foundation design, the predicted design scheme is the predicted pile foundation type, and the actual design scheme is the actual pile foundation type. The pile foundation scheme recommendation problem can be abstracted as the problem of classifying input data composed of characteristic parameters such as main beam type, main span length, span layout, pier height, pier location, terrain, etc. The classification label is the actual pile foundation type. The common feature extraction layer is used to extract useful information from the input data, or it can be a deformation or remapping of the input features. The common feature extraction layer includes multiple fully connected neural networks. The fully connected layer can be used as a data feature extraction layer. By embedding the attributes of the previous layer into a vector and remapping it to a new vector space, the final decision on the pile foundation scheme is achieved.

[0104] It is understandable that since there are many influencing variables and complex influencing factors in the pile foundation scheme, the input parameters include both enumerated variables and numerical variables, and the dimensions of the numerical variables are also different. The role of the data embedding layer is to embed the input information, that is, to uniformly map the feature attributes of different types, dimensions, and numerical ranges into a 128-dimensional vector space for subsequent feature learning and feature extraction processes. The multi-tail decision layer is used to convert the features extracted by the previous layer into outputs that are useful for specific tasks. That is, after receiving the feature information given by the common feature extraction layer, the multi-tail decision layer will first make a decision on the number of piles. The decision result will be spliced ​​with the feature information to make a decision on the pile diameter, and finally the decision result of the entire pile foundation scheme will be obtained.

[0105] It is worth noting that when making decisions, the decision on the pile diameter will be affected by the decision on the number of piles. However, during the training process, the decision result on the number of piles is unknown and needs to be replaced by the decision result on the number of piles generated by the model itself. This difference will cause errors to accumulate during the reasoning process and affect the model performance. If the correct decision result on the number of piles is used for training during the training process, the data received by the pile diameter decision will be different from the data received during the actual reasoning process, affecting the robustness and accuracy of the model. In order to solve this problem, a probability value ε(i) that decays with the training process can be used to choose whether to use the real decision result on the number of piles for pile diameter training. In the early stage of training, there is a greater probability of using the completely accurate decision result on the number of piles for training. As the training progresses, this probability will be gradually reduced and the data obtained by self-reasoning will be used for training. A linear decay function is selected to control the probability value ε(i), that is, ε(i) = max(ε,k-ci), where ε is the probability of using the minimum true decision solution, and k and c control the decay rate.

[0106] It is worth noting that the predicted pile foundation type is compared with the corresponding actual pile foundation type, that is, the predicted number of piles is compared with the actual number of piles, and the predicted pile diameter is compared with the actual pile diameter. The cross entropy loss function is used to measure the difference between the predicted result and the actual result. The quantity difference and diameter difference are back-propagated and the model weight is updated respectively. The cross entropy loss function is as follows:

[0107] L = -∑[y×log(f(x))]

[0108] Among them, L is the loss value, y is the true result, and f(x) is the predicted result.

[0109] In the specific implementation, when training the model, the number of piles is first predicted, the error is back-propagated, and the model weight is updated. Then the predicted result of the number of piles is concatenated with the feature vector and used as input to predict the pile diameter, and the error is back-propagated again to update the model weight, thus completing one training.

[0110] In a feasible implementation manner, the feature parameters are input into the data embedding layer for embedding processing to obtain a target embedding vector, including: determining discrete features and continuous features according to the feature parameters; mapping the discrete features to a continuous vector space to obtain a first embedding matrix; automatically discretizing, discretely embedding, and aggregating the continuous features to obtain a second embedding matrix; and generating a target embedding vector according to the first embedding matrix and the second embedding matrix.

[0111] It should be noted that the characteristic parameters of the input model are divided into discrete data and continuous data, namely, discrete features and continuous features. Discrete features refer to features with a finite number of possible values ​​in the feature space, such as the five node attributes of seismic intensity s5, small mileage side beam type s6, large mileage side beam type s7, pier type s8, and geological label s9; while continuous features refer to features whose characteristic values ​​change continuously within the real number range, such as the four node attributes of small mileage span s1, large mileage span s2, pier height s3, and average ground slope s4. In order to uniformly process these two types of features, corresponding embedding processing is required.

[0112] It is understandable that for discrete features, since they are usually represented as text or classification labels, they cannot be directly input into the neural network for processing. Therefore, these discrete features need to be mapped into a continuous vector space and converted into continuous features that can be processed by the neural network.

[0113] In the specific implementation, the input values ​​of the five node attributes in the discrete features, such as seismic intensity s5, small mileage side beam type s6, large mileage side beam type s7, pier type s8, and geological label s9, are all enumerated integers. Therefore, each variable is mapped to a 128-dimensional continuous vector through the Embedding module, which includes: randomly initializing a matrix E, E∈R V×D , where R is a real number set, V is the number of enumeration categories, and D is the embedding dimension, which is 128. Each row of the matrix represents the embedding vector of a category in the enumeration category. When an index is input, the Embedding module extracts the corresponding row from the matrix E to obtain a high-dimensional vector of variable mapping. The five discrete feature attributes all have their own embedding matrix E. i , and finally get a 5×128-dimensional embedding matrix, where, during the training process, this embedding matrix E i It will be updated through back propagation to optimize the performance of the model. The final embedding matrix of the discrete features is as follows:

[0114]

[0115] Among them, A is the final embedding matrix of discrete features, and E is the embedding matrix corresponding to the five discrete feature attributes.

[0116] It is understandable that for continuous features, such as short mileage span s1, long mileage span s2, pier height s3, average ground slope s4, etc., since they are themselves non-discrete continuous data, the AutoDis module is used to perform end-to-end automatic encoding processing on continuous values ​​and encode continuous data into high-dimensional vectors. The AutoDis module can automatically divide continuous features, realize unique coding representation for continuous features, map high-dimensional sparse attribute features into low-dimensional dense vectors, and provide data enhancement effects for continuous data in the feature learning process of downstream tasks. The AutoDis module is a deep learning framework used to process Embedding learning of continuous numerical features. It improves the expressiveness and generalization capabilities of the model by introducing three core components: Meta-Embedding, Auto-Discretization, and Aggregation Function.

[0117] like Figure 7 As shown, Figure 7 This is the schematic diagram of the AutoDis principle. The continuous variable x with a total length of Total Length = 128 i Perform automatic discretization to obtain multiple discrete features Multiple discrete features After Meta-Embedding, aggregation is performed through the aggregation function to obtain the final discrete result of the continuous variable.

[0118] It is worth mentioning that automatic discretization is to automatically discretize a continuous numerical feature into 64 buckets through a two-layer neural network, that is, to obtain a matrix of size (1,64). The data features are then normalized and the mapping results are normalized using SoftMax to obtain the final discrete result of the continuous variable. This discrete result is similar to the onehot encoding of the discrete variable, which is used to select the embedding vector from the embedding matrix.

[0119] Discrete embedding (Meta-Embedding) is to balance the model capacity and the number of model parameters. For each feature, a set of shared Meta-Embeddings is defined. Meta-Embedding is a 128x 64 vector space. Each Meta-Embedding is regarded as a subspace of the latent space. These subspaces can improve the model's ability to express each continuous numerical feature. During the training process, Meta-Embedding also belongs to a variable vector space. During the calculation process, the input continuous data is dimensioned and upsampled, and the resulting size is (128,64). By performing dot product multiplication with Meta-Embedding, the calculated result size is (128,64). This matrix of size (128,64) is the final embedding matrix of the continuous variable.

[0120] Aggregation means that after the input continuous numerical features are calculated by the previous two modules, the probability distribution of this continuous numerical feature is obtained in 64 discrete embeddings (Meta-Embedding). The result after discrete embedding (Meta-Embedding) processing and the corresponding probability calculated by softmax are weighted summed to obtain the final discrete vector of size 1x 128.

[0121] In the specific implementation, four kinds of continuous data, including short mileage span s1, long mileage span s2, pier height s3, and average ground slope s4, are automatically discretized and mapped into a 64-dimensional vector, which are aggregated with the 64x 128 vector space discrete embedding Meta-Embedding to form a 128-dimensional vector. In the subsequent training process, the vectors in Meta-Embedding and the weights of the fully connected neural network are trained through the back-propagation algorithm. Finally, a continuous data can be discretized into a 1x 128 vector through AutoDis, thereby giving each continuous value an independent embedded vector representation, so that the network can effectively extract the feature information of the continuous value.

[0122] In a feasible implementation, the target embedding vector is input into the common feature extraction layer for feature extraction to obtain a common feature vector, including: obtaining a weight matrix and a bias vector of a fully connected neural network; and performing feature extraction on the target embedding vector based on the weight matrix and the bias vector to obtain a common feature vector.

[0123] It should be noted that the common feature extraction layer includes one MLP fully connected neural network, namely, a common fully connected neural network, and the multi-tail decision layer includes multiple MLP fully connected neural networks, namely, multiple independent fully connected neural networks connected to the common fully connected neural network, and the multiple independent fully connected neural networks are multiple MLP fully connected neural networks with different numbers of nodes, namely, the MLP fully connected neural network can be used as a data feature extraction layer, and the final decision on the pile foundation scheme is achieved by embedding the attributes of the previous layer into a vector and remapping it to a new vector space.

[0124] It can be understood that in this embodiment, the common feature extraction layer may include a 6-layer MLP fully connected neural network with 1024, 512, 256, 256, 128, and 128 nodes respectively. After the feature parameters are embedded in the data embedding layer, the embedding vector A∈R of the data is obtained. 9×128 After flattening the embedding vector, we get the final input A of the common feature extraction layer flatten ∈R 9×1152 , and then sent to the six-layer MLP fully connected neural network containing 1024, 512, 256, 256, 128, and 128 nodes respectively for feature extraction. After calculation at each layer, the final feature extraction output A is obtained. 6 ∈R 1×128 , the feature extraction of MLP fully connected neural network is as follows:

[0125] Z l =W l ·A l-1 +b l

[0126] A l =σ(Z l )

[0127]

[0128] in, is the weight matrix of layer l, is the bias vector of layer l, s l and l-1 are the number of nodes in this layer and the number of nodes in the previous layer, respectively. is the activation output of the previous layer, i.e., the l-1 layer, and σ is the activation function.

[0129] It is worth noting that the weights and parameters of the common feature extraction layer are globally shared parameters, that is, the decisions on the diameter and number of piles will adopt the features extracted by the common feature extraction layer, reflecting the common characteristics of the diameter and number of piles in the business.

[0130] In a feasible implementation manner, the feature vector is input into the multi-tail decision layer for joint decision-making to obtain a predicted design scheme, including: making a decision on a first design parameter according to the feature vector to obtain a first decision vector; splicing the first decision vector with the feature vector to obtain a spliced ​​vector; making a decision on a second design parameter according to the spliced ​​vector to obtain a second decision vector; making a decision on a third design parameter based on the second decision vector and the common feature vector, until the decision vector generation of each design parameter is completed; and summarizing the decision vector generation of each design parameter to obtain a predicted design scheme.

[0131] It should be noted that the first design parameter, the second design parameter and the Nth design parameter are decided in turn according to the characteristic vector to obtain the decision vectors of all design parameters. For the pile foundation scheme design, the design parameters include 2 in total, namely the first design parameter and the second design parameter. The first design parameter is the number of piles, and the first decision vector is the number of piles decision vector. The second design parameter is the pile diameter, and the second decision vector is the pile diameter decision vector. That is, the number of piles is decided according to the characteristic vector to obtain the number of piles decision vector, the number of piles decision vector is spliced ​​with the characteristic vector, and the pile diameter is decided according to the spliced ​​vector to obtain the pile diameter decision vector. The predicted pile foundation type is generated according to the pile number decision vector and the pile diameter decision vector. The characteristic vector A∈R 1×128 After being input into the multi-tail decision layer, the feature vector is calculated using a four-layer MLP fully connected neural network with 128, 64, 32, and 9 nodes respectively, to obtain the final pile quantity decision vector O pileNurmber ∈R 1×9 .

[0132] It can be understood that the final decision vector of the number of piles is concatenated with the features extracted from the feature layer to obtain the input of the pile diameter decision module, that is, the concatenated vector A dia ∈R 1×137 .

[0133] It is worth noting that according to the splicing vector A dia ∈R 1×137 The concatenated vector is calculated using a four-layer MLP fully connected neural network with 128, 64, 32, and 8 nodes respectively, to obtain the final pile diameter decision vector O pileDia ∈R 1×8 .

[0134] Step S203, performing a performance test on the trained multi-tail neural network model using the test set to obtain a loss value of the test set.

[0135] It should be noted that by performing performance testing on the trained multi-tail neural network model through the test set and obtaining the loss value of the test set, the change of the loss value of the test set in different training times can be determined, and then whether the trained multi-tail neural network model is overfitting.

[0136] It is understandable that in the process of model training, in addition to paying attention to the performance of the model, it is also necessary to pay attention to the generalization ability of the model. Generalization ability refers to the model's ability to adapt to new data, that is, the model can maintain a high accuracy rate on unseen data. In order to improve the generalization ability of the model, some common techniques can be used according to the performance of the model on the test set, such as regularization, data enhancement and early stopping strategy, etc., to prevent the model from overfitting.

[0137] Step S204, updating the parameters of the trained multi-tail neural network model according to the loss value of the test set to obtain a solution decision model.

[0138] It should be noted that during the entire model training process, when the number of iterations was 500, the loss value of the model prediction result and the true label was calculated, and then the loss value curve was drawn. The gradient of the loss value curve in the training set and the prediction set decreased very quickly, and the gradient descent curves of the training set and the test set were close, and the convergence was very good, as shown in Figure 2. Figure 8 As shown, Figure 8 Schematic diagram of the curve of loss function value changing with training batch, including (a) the tail of pile foundation type, that is, the curve of loss function value of pile foundation type changing with training batch, and (b) the tail of pile number, that is, the curve of loss function value of pile number changing with training batch.

[0139] In the specific implementation, according to the loss value curve, after about 100 iterations of training, the loss value of the test set tends to change smoothly, but the loss function of the training set continues to decrease, indicating that the model has entered an overfitting state. Therefore, the model parameters during 100 iterations of training are selected to update the current multi-tail neural network model and obtain the solution decision model.

[0140] In the specific implementation, after the model training is completed, the confusion matrix of the model is finally obtained, which shows the classification performance of the model on the test set. The rows represent the true labels, the columns represent the predicted labels, and each grid shows the number of samples corresponding to the true labels and predicted labels. The numbers on the diagonal represent the number of correctly classified samples, and the numbers on the non-diagonal represent the number of incorrectly classified samples. The accuracy of the final model is about 90%, as shown in Fig. 9 As shown, Fig. 9 A diagram of the confusion matrix.

[0141] It is worth noting that for pile foundation design, through the analysis of the cases where the prediction failed, it was found that the decision of the pile foundation scheme is essentially controlled by the pile top area. In the cases where the prediction failed, the pile top area predicted by the model was equal to the pile top area of ​​the actual case, but the pile foundation scheme selected was different. The scheme predicted by the model is still feasible. If the prediction results with a difference of less than 5% between the pile top area predicted by the model and the pile top area of ​​the actual case are listed as correct results, the accuracy of the model prediction can reach 95.6%.

[0142] like Fig.10 As shown, Fig.10 Schematic diagram of the curves of accuracy and recall rate changing with the deformation of training batch, including (a) the curve of accuracy index changing with the deformation of training batch and (b) the curve of recall rate index changing with the deformation of training batch.

[0143] In this embodiment, the multi-tail neural network model is jointly trained and performance tested in stages through the training set and the test set, and then the model parameters are updated according to the loss value of the test set to obtain a scheme decision model, which can effectively improve the prediction accuracy and generalization ability of the pile foundation scheme decision model.

[0144] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the joint decision-making multi-tail neural network model and training method of the design scheme of the present application. More forms of simple transformations based on this technical concept are all within the scope of protection of the present application.

[0145] This application also provides a design solution for joint decision-making multi-tail neural network model and training device, please refer to Fig.11 The design scheme combines the decision-making multi-tail neural network model and the training device to include:

[0146] The generating module 10 is used to generate a target data set based on historical design cases.

[0147] The training module 20 is used to perform phased joint training on the multi-tail neural network model using the target data set to obtain a solution decision model.

[0148] The decision module 30 is used to make a joint decision on multiple design parameters according to the scheme decision model to obtain a target design scheme.

[0149] The present embodiment provides a multi-tail neural network model for joint decision-making on design schemes and a training device. The present embodiment can improve data balance by first generating a target data set based on historical design cases; the multi-tail neural network model is jointly trained in stages using the target data set to obtain a scheme decision model, which can effectively improve the model accuracy and generalization ability; a plurality of design parameters are jointly decided according to the scheme decision model to obtain a target design scheme, which can effectively improve the efficiency and accuracy of scheme decision-making.

[0150] In summary, this embodiment conducts stage-by-stage joint training of the multi-tail neural network model through the target data set generated based on historical design cases, and makes joint decisions through the pile foundation scheme decision model obtained through training to obtain the pile foundation design scheme, thereby overcoming the technical defects of low efficiency and poor accuracy of design scheme decision-making, and can effectively improve the efficiency and accuracy of scheme decision-making.

[0151] Optionally, the generating module 10 is further used to extract characteristic parameters of the historical design cases to obtain an initial data set; and perform data augmentation and data cleaning on the initial data set according to associated parameters to obtain a target data set.

[0152] Optionally, the training module 20 is also used to divide the target data set into a training set and a test set according to a preset ratio, wherein the target data set includes multiple samples, and the samples include different feature parameters and corresponding actual design schemes; the multi-tail neural network model is jointly trained in stages through the training set to obtain a trained multi-tail neural network model, wherein the multi-tail neural network model includes a data embedding layer, a common feature extraction layer and a multi-tail decision layer; the performance of the trained multi-tail neural network model is tested through the test set to obtain a loss value of the test set; the parameters of the trained multi-tail neural network model are updated according to the loss value of the test set to obtain a scheme decision model.

[0153] Optionally, the training module 20 is also used to input the feature parameters of each sample in the training set into the data embedding layer for embedding processing to obtain a target embedding vector; input the target embedding vector into the common feature extraction layer for feature extraction to obtain a feature vector; input the feature vector into the multi-tail decision layer for joint decision-making to obtain a predicted design scheme; compare the predicted design scheme with the corresponding actual design scheme to obtain a prediction error; perform back-propagation training on the multi-tail neural network model according to the prediction error and update the model weights to obtain a trained multi-tail neural network model.

[0154] Optionally, the training module 20 is also used to determine discrete features and continuous features based on the feature parameters; map the discrete features to a continuous vector space to obtain a first embedding matrix; automatically discretize, discretely embed and aggregate the continuous features to obtain a second embedding matrix; generate a target embedding vector based on the first embedding matrix and the second embedding matrix.

[0155] Optionally, the training module 20 is also used to obtain a weight matrix and a bias vector of a fully connected neural network; and perform feature extraction on the target embedding vector based on the weight matrix and the bias vector to obtain a feature vector.

[0156] Optionally, the training module 20 is also used to make a decision on a first design parameter based on the feature vector to obtain a first decision vector; concatenate the first decision vector with the feature vector to obtain a concatenated vector; make a decision on a second design parameter based on the concatenated vector to obtain a second decision vector; make a decision on a third design parameter based on the second decision vector and the common feature vector until the decision vectors for each design parameter are generated; and summarize the decision vectors for each design parameter to obtain a predicted design solution.

[0157] The design scheme joint decision-making multi-tail neural network model and training device provided in the present application adopt the design scheme joint decision-making multi-tail neural network model and training method in the above-mentioned embodiment, which can solve the technical problem of how to effectively improve the efficiency and accuracy of scheme decision-making. Compared with the prior art, the beneficial effects of the design scheme joint decision-making multi-tail neural network model and training device provided in the present application are the same as the beneficial effects of the design scheme joint decision-making multi-tail neural network model and training method provided in the above-mentioned embodiment, and other technical features in the design scheme joint decision-making multi-tail neural network model and training device are the same as the features disclosed in the above-mentioned embodiment method, which will not be repeated here.

[0158] The present application provides a design scheme joint decision-making multi-tail neural network model and training equipment, and the design scheme joint decision-making multi-tail neural network model and training equipment include: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the design scheme joint decision-making multi-tail neural network model and training method in the above-mentioned embodiment one.

[0159] Reference below Fig.12, which shows a schematic diagram of the structure of the multi-tail neural network model and training device for joint decision-making of the design scheme of the embodiment of the present application. The multi-tail neural network model and training device for joint decision-making of the design scheme in the embodiment of the present application may include but are not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions: tablet computers), PMPs (Portable Media Players: portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.12 The design scheme of joint decision-making multi-tail neural network model and training equipment shown is only an example and should not bring any limitation to the function and scope of use of the embodiments of the present application.

[0160] like Fig.12 As shown, the design scheme joint decision multi-tail neural network model and training device may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM: Read Only Memory) 1002 or the program loaded from the storage device 1003 to the random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the design scheme joint decision multi-tail neural network model and training device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the design solution joint decision multi-tail neural network model and the training device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a design solution joint decision multi-tail neural network model and a training device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.

[0161] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0162] The design scheme joint decision-making multi-tail neural network model and training device provided in the present application adopt the design scheme joint decision-making multi-tail neural network model and training method in the above-mentioned embodiment, which can solve the technical problem of how to effectively improve the efficiency and accuracy of scheme decision-making. Compared with the prior art, the beneficial effects of the design scheme joint decision-making multi-tail neural network model and training device provided in the present application are the same as the beneficial effects of the design scheme joint decision-making multi-tail neural network model and training method provided in the above-mentioned embodiment, and other technical features in the design scheme joint decision-making multi-tail neural network model and training device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0163] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0164] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0165] The present application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the design scheme joint decision-making multi-tail neural network model and training method in the above-mentioned embodiments.

[0166] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0167] The above-mentioned computer-readable storage medium can be included in the design scheme joint decision-making multi-tail neural network model and training device; or it can exist independently without being assembled into the design scheme joint decision-making multi-tail neural network model and training device.

[0168] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the design scheme joint decision multi-tail neural network model and training device, the design scheme joint decision multi-tail neural network model and training device: generate a target data set based on historical design cases; perform phased joint training on the multi-tail neural network model through the target data set to obtain a scheme decision model; perform joint decision on multiple design parameters according to the scheme decision model to obtain a target design scheme.

[0169] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, C#, python, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0170] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0171] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0172] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned design scheme joint decision-making multi-tail neural network model and training method, and can solve the technical problem of how to effectively improve the efficiency and accuracy of scheme decision-making. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the design scheme joint decision-making multi-tail neural network model and training method provided in the above-mentioned embodiment, and will not be repeated here.

[0173] The present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the above-mentioned design scheme joint decision multi-tail neural network model and training method.

[0174] The computer program product provided by this application can solve the technical problem of how to effectively improve the efficiency and accuracy of solution decision-making. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the design solution joint decision multi-tail neural network model and training method provided in the above embodiment, which will not be repeated here.

[0175] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A design scheme joint decision-making multi-tail neural network model and training method, characterized in that: The method comprises: Generate target datasets based on historical design cases; The multi-tail neural network model is jointly trained in stages using the target data set to obtain a solution decision model; A joint decision is made on multiple design parameters according to the scheme decision model to obtain a target design scheme.

2. The method according to claim 1, characterized in that The generating of the target data set based on the historical design cases includes: Extracting characteristic parameters of the historical design cases to obtain an initial data set; The initial data set is augmented and cleaned according to the associated parameters to obtain a target data set.

3. The method according to claim 1, characterized in that The multi-tail neural network model is jointly trained in stages by the target data set to obtain a solution decision model, including: Dividing the target data set into a training set and a test set according to a preset ratio, wherein the target data set includes a plurality of samples, and the samples include different feature parameters and corresponding actual design solutions; The multi-tail neural network model is jointly trained in stages by using the training set to obtain a trained multi-tail neural network model, wherein the multi-tail neural network model includes a data embedding layer, a common feature extraction layer, and a multi-tail decision layer; Performing a performance test on the trained multi-tail neural network model using the test set to obtain a loss value of the test set; The parameters of the trained multi-tail neural network model are updated according to the loss value of the test set to obtain a solution decision model.

4. The method according to claim 3, characterized in that The multi-tail neural network model is jointly trained in stages by the training set to obtain a trained multi-tail neural network model, including: Inputting the characteristic parameters of each sample in the training set into the data embedding layer for embedding processing to obtain a target embedding vector; Inputting the target embedding vector into the common feature extraction layer for feature extraction to obtain a common feature vector; Inputting the common feature vector into the multi-tail decision layer for joint decision making to obtain a predicted design solution; Comparing the predicted design solution with the corresponding actual design solution to obtain a prediction error; The multi-tail neural network model is back-propagated and trained according to the prediction error, and the model weights are updated to obtain a trained multi-tail neural network model.

5. The method according to claim 4, characterized in that The step of inputting the feature parameters into the data embedding layer for embedding processing to obtain a target embedding vector includes: Determine discrete features and continuous features according to the feature parameters; Mapping the discrete features into a continuous vector space to obtain a first embedding matrix; Automatically discretizing, discretely embedding, and aggregating the continuous features to obtain a second embedding matrix; A target embedding vector is generated according to the first embedding matrix and the second embedding matrix.

6. The method according to claim 4, characterized in that The step of inputting the target embedding vector into the common feature extraction layer for feature extraction to obtain a feature vector comprises: Get the weight matrix and bias vector of the fully connected neural network; Feature extraction is performed on the target embedding vector based on the weight matrix and the bias vector to obtain a feature vector.

7. The method according to claim 4, characterized in that The step of inputting the common feature vector into the multi-tail decision layer for joint decision making to obtain a prediction design solution includes: Making a decision on a first design parameter according to the characteristic vector to obtain a first decision vector; Concatenating the first decision vector with the common feature vector to obtain a concatenated vector; Making a decision on a second design parameter according to the splicing vector to obtain a second decision vector; Performing decision generation of a third design parameter based on the second decision vector and the common feature vector until generation of decision vectors for each design parameter is completed; The decision vector generation of each design parameter is summarized to obtain the predicted design solution.

8. A design scheme joint decision multi-tail neural network model and training device, characterized in that: The design scheme combines the decision-making multi-tail neural network model and training device, including: A generation module, used to generate a target data set based on historical design cases; A training module, used for performing phased joint training on the multi-tail neural network model through the target data set to obtain a solution decision model; The decision module is used to make joint decisions on multiple design parameters according to the scheme decision model to obtain a target design scheme.

9. A design scheme joint decision multi-tail neural network model and training device, characterized in that: The design scheme joint decision-making multi-tail neural network model and training device include: a memory, a processor, and a design scheme joint decision-making multi-tail neural network model and training program stored in the memory and executable on the processor. The design scheme joint decision-making multi-tail neural network model and training program are configured to implement the design scheme joint decision-making multi-tail neural network model and training method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that: The storage medium stores a design scheme joint decision-making multi-tail neural network model and a training program. When the design scheme joint decision-making multi-tail neural network model and the training program are executed by the processor, the design scheme joint decision-making multi-tail neural network model and the training method as described in any one of claims 1 to 7 are implemented.

Images