A method for predicting composite pallet performance based on machine learning

Abstract

The present invention relates to the technical field of pallet performance prediction, and discloses a composite material pallet performance prediction method based on machine learning, the method comprising: collecting composite material pallet parameter data and performing vectorization processing to obtain vectorized pallet parameters, splicing the vectorized pallet parameters to obtain a fused pallet parameter vector and performing compression processing to obtain a simplified pallet parameter vector; constructing a composite material pallet performance prediction model, the model takes the simplified pallet parameter vector as input and takes the pallet comprehensive performance evaluation result as output, the pallet performance prediction model uses a gradient boosting decision tree based on importance sampling for prediction; and performing performance prediction and evaluation on the pallet according to the constructed composite material pallet performance prediction model to achieve efficient and accurate performance prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of pallet performance prediction, and in particular to a composite material pallet performance prediction method based on machine learning. Background Art

[0002] In recent years, the use of composite materials has expanded to many fields, including aerospace, automobile manufacturing, construction and logistics. Especially in the field of logistics, pallets are indispensable tools in the transportation, storage and handling of goods. Composite pallets are gradually replacing traditional wooden, plastic and metal pallets due to their advantages such as light weight, high strength, corrosion resistance and fatigue resistance. However, the design and manufacturing process of composite pallets is relatively complicated, and their performance depends not only on the composition and structure of the material, but also on the manufacturing process, use conditions and other factors. Therefore, the prediction of the performance of composite pallets, especially the prediction of mechanical properties, has important engineering value. Traditional prediction methods usually rely on theoretical models and experimental verification. Although these methods can guide the design of composite pallets to a certain extent, due to the complexity of the composite materials themselves and the uncertainty in the manufacturing process, theoretical models are often difficult to accurately reflect the actual situation, and the experiment is costly, time-consuming and labor-intensive. Summary of the invention

[0003] In view of this, the present invention proposes a composite material pallet performance prediction method based on machine learning. By analyzing a large amount of experimental data, a mapping relationship between the performance of the composite material pallet and the influencing factors is established, thereby achieving efficient and accurate performance prediction, which is helpful to improve the design level of the pallet, optimize the use of materials, reduce production costs, and shorten the design cycle.

[0004] To achieve the above object, the present invention provides a composite material pallet performance prediction method based on machine learning, comprising the following steps:

[0005] S1: Collecting composite material pallet parameter data and performing vectorization processing to obtain vectorized pallet parameters, wherein the parameter data includes material composition parameters, manufacturing process parameters, geometric size parameters, mechanical parameters and use load parameters;

[0006] S2: concatenate the vectorized pallet parameters to obtain a fused pallet parameter vector and perform compression processing to obtain a simplified pallet parameter vector;

[0007] S3: Construct a composite material pallet performance prediction model, the model takes a simplified pallet parameter vector as input and a pallet comprehensive performance evaluation result as output, and the pallet performance prediction model uses a gradient boosting decision tree based on importance sampling for prediction;

[0008] S4: The performance of the pallet is predicted and evaluated based on the constructed composite pallet performance prediction model.

[0009] As a further improvement method of the present invention:

[0010] Optionally, in step S1, collecting composite material pallet parameter data and performing vectorization processing to obtain vectorized pallet parameters includes:

[0011] The material composition parameters include matrix material type, fiber type, fiber volume fraction and filler type, wherein the matrix material types include epoxy resin, polyester resin, polyethylene, polypropylene and polyurethane, which are vectorized by one-hot coding, the fiber types include glass fiber, carbon fiber and aramid fiber, which are vectorized by one-hot coding, the fiber volume fraction is the proportion of each fiber component in the composite material, and the proportion value is vectorized, and the filler types include glass microspheres, nanoclay, talcum powder and calcium carbonate, which are vectorized by one-hot coding;

[0012] The manufacturing process parameters include the fiber material laying angle, laying sequence, curing temperature and curing time of the composite laminate, wherein the fiber material laying angle converts the laying angle of each layer into a corresponding numerical code to form a coding vector, and the fiber material laying sequence converts the laying material of each layer into a corresponding numerical code to form a coding vector, and the vector dimension is the preset maximum number of laying layers;

[0013] The geometric dimension parameters include the pallet size and the pallet wall thickness, and the values ​​are concatenated and vectorized;

[0014] The mechanical parameters include tensile strength, compressive strength, bending strength and impact strength, wherein tensile strength indicates the maximum tensile stress of the pallet, compressive strength indicates the maximum compressive stress of the pallet, bending strength indicates the ability of the pallet to resist damage under bending load, and impact strength indicates the maximum energy absorption capacity of the pallet under impact load, and the values ​​are spliced ​​and vectorized;

[0015] The load parameters include static load and dynamic load, wherein the static load indicates the maximum weight that the pallet can bear under static conditions, and the dynamic load indicates the maximum weight that the pallet can bear under moving conditions, and the values ​​are concatenated and vectorized.

[0016] Optionally, in step S2, the vectorized pallet parameters are concatenated to obtain a fused pallet parameter vector and compressed to obtain a simplified pallet parameter vector, including:

[0017] S21: Convert the fused pallet parameter vector into a semantic vector through a linear layer , the conversion formula is:

[0018] ;

[0019] in:

[0020] Represents the semantic vector dimension;

[0021] represents the weight matrix of the encoder network;

[0022] represents the bias vector of the encoder network;

[0023] represents the output fusion pallet parameter vector;

[0024] S22: Calculate the attention score of the semantic vector. The calculation formula is:

[0025] ;

[0026] in:

[0027] represents the query vector;

[0028] represents the key vector;

[0029] Represents the semantic vector dimension;

[0030] S23: Normalize the attention scores to obtain normalized attention weights, so that the sum of all attention weights is 1. The normalization calculation formula is:

[0031] ;

[0032] in:

[0033] represents the attention score of the i-th dimension;

[0034] represents the normalized attention weight of the i-th dimension;

[0035] S24: Use the normalized attention weights to perform weighted summation on the semantic value vector to obtain a simplified pallet parameter vector. The calculation formula is:

[0036] ;

[0037] in:

[0038] represents the i-th semantic value vector;

[0039] represents the i-th value weight matrix;

[0040] S25: Decode the simplified pallet parameter vector through the decoder network to reconstruct the fused pallet parameter vector. The decoding and reconstruction calculation formula is:

[0041] ;

[0042] in:

[0043] represents the weight matrix of the decoder network;

[0044] represents the bias vector of the decoder network;

[0045] S26: Construct a target loss function based on the reconstructed fusion pallet parameter vector and the original fusion pallet parameter vector. The target loss function calculation formula is:

[0046] ;

[0047] in:

[0048] Represents the number of pallet parameter vectors;

[0049] represents the parameter vector of the i-th fusion pallet;

[0050] represents the parameter vector of the fusion pallet after the i-th reconstruction;

[0051] S27: Optimize and solve the target loss function to obtain the neural coding model parameters and perform vector compression processing, wherein mixed conjugate gradient is the implementation method of the optimization solution.

[0052] Optionally, in step S27, optimizing and solving the target loss function to obtain neural coding model parameters includes:

[0053] S27.1: Initialize the neural encoding model parameters and calculate the target loss function gradient under the current model parameters. The calculation formula is:

[0054] ;

[0055] in:

[0056] Represents the target loss function gradient of the kth iteration;

[0057] represents the model parameters of the kth iteration;

[0058] Indicates that the parameters The objective loss function for optimizing variables;

[0059] S27.2: Calculate the conjugate gradient direction, and perform weighted processing on the conjugate gradient direction and the gradient direction to obtain the hybrid search direction of the current round. The hybrid search direction calculation formula is:

[0060] ;

[0061] in:

[0062] Represents the dynamic weight parameter calculated in the kth round;

[0063] Represents the gradient direction of the target loss function;

[0064] Represents the conjugate gradient direction obtained by the k-th round of iteration;

[0065] Indicates mixed search direction;

[0066] The conjugate gradient calculation formula is:

[0067] ;

[0068] in: represents the adjustment factor, and the calculation formula of the adjustment factor is:

[0069] ;

[0070] S27.3: Determine the update step size by line search method, the calculation formula is:

[0071] ;

[0072] in:

[0073] represents the step size parameter variable;

[0074] Represents the update step size calculated in the kth round;

[0075] S27.4: Update the neural encoding model parameters according to the calculated update step size. The parameter update calculation formula is:

[0076]

[0077] ;

[0078] in:

[0079] represents the model parameters obtained in the k+1th iteration;

[0080] S27.5: Iterate steps S27.1 to S27.4 until the maximum number of iterations is reached, and output the final model parameters as the optimal neural coding model parameters.

[0081] Optionally, the dynamic weight parameter includes:

[0082] S51: performing incremental calculation based on the conjugate gradient and gradient of the current round to obtain the conjugate gradient increment and gradient increment, and the calculation formula is:

[0083] ;

[0084] ;

[0085] in:

[0086] and They represent the conjugate gradient increment and gradient increment of the kth round respectively;

[0087] S52: Perform weight calculation based on the calculated conjugate gradient increment and gradient increment to obtain the weight parameter of the current round. The weight calculation formula is:

[0088] ;

[0089] in:

[0090] Represents a small positive number, and its value is , used to prevent the denominator from being 0.

[0091] Optionally, in step S3, a composite material pallet performance prediction model is constructed, wherein the model takes a simplified pallet parameter vector as input and takes a pallet comprehensive performance evaluation result as output, including:

[0092] S31: Randomly initialize the gradient boosting decision tree model and calculate the initial deviation. The initial deviation calculation formula is:

[0093] ;

[0094] in:

[0095] represents the initial residual of the i-th reduced pallet parameter vector sample data;

[0096] Represents the model prediction value With the true value The loss function between

[0097] represents the sample data of the i-th simplified pallet parameter vector;

[0098] Represents the predicted value of the model on the i-th streamlined pallet parameter vector sample data;

[0099] represents the true value of the i-th simplified pallet parameter vector sample data;

[0100] represents the predicted value of the model at the initial moment on the i-th streamlined pallet parameter vector sample data;

[0101] represents the partial derivative operation;

[0102] S32: Use the calculated deviation as the target variable to perform decision tree fitting;

[0103] S33: Perform comprehensive performance evaluation of the pallet based on the calculated final model.

[0104] Optionally, in step S32, using the calculated deviation as the target variable to perform decision tree fitting includes:

[0105] S32.1: Calculate the gradient of all sample data. The calculation formula is:

[0106] ;

[0107] in:

[0108] represents the gradient of the sample data of the i-th reduced pallet parameter vector;

[0109] represents the predicted value of the t-1 round model on the i-th streamlined pallet parameter vector sample data;

[0110] S32.2: Sort the sample data according to the absolute value of the calculated gradient, and select the one with the largest absolute value of gradient. samples as the important sample set A, and randomly sample from the remaining samples samples as non-important sample set B;

[0111] in:

[0112] Represents the proportion of important sample sets, with a value range of (0,1);

[0113] Represents the proportion of non-important sample sets, with a value range of (0,1);

[0114] Represents the total amount of sample data of simplified pallet parameter vector;

[0115] S32.3: The gradient of non-important samples is amplified, and important samples are not processed. The gradient processing calculation formula is:

[0116] ;

[0117] in:

[0118] represents the gradient of the sample data of the i-th streamlined pallet parameter vector after adjustment;

[0119] S32.4: Use the adjusted gradients of all samples as classification targets for the sample set Building a decision tree ;

[0120] in:

[0121] represents the decision tree constructed in round t;

[0122] x represents the reduced pallet parameter vector;

[0123] S32.5: Update the model based on the decision tree constructed in the current round. The calculation formula is:

[0124] ;

[0125] in:

[0126] represents the learning rate;

[0127] Represents the model after the tth iteration;

[0128] Represents the model after the t-1th iteration;

[0129] S32.6: Calculate the deviation value based on the updated model, and iterate steps S32.1 to S32.5 until the maximum number of iterations is reached.

[0130] In order to solve the above problem, the present invention provides an electronic device, the electronic device comprising:

[0131] A memory storing at least one instruction;

[0132] Communication interface, enabling electronic equipment to communicate; and

[0133] A processor executes instructions stored in the memory to implement the above-mentioned composite material pallet performance prediction method based on machine learning.

[0134] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned composite material pallet performance prediction method based on machine learning.

[0135] Compared with the prior art, the present invention proposes a composite material pallet performance prediction method based on machine learning, which has the following advantages:

[0136] (1) This scheme ensures the comprehensiveness and diversity of data by systematically vectorizing five aspects of composite pallets (material composition, manufacturing process, geometric dimensions, mechanical parameters, and service load), providing a rich information basis for subsequent analysis.

[0137] (2) This scheme introduces a neural encoding model that integrates the attention mechanism during the compression process, which can dynamically focus on the importance of different parameters, thereby improving the compression effect and information retention rate, and can effectively extract key features and reduce redundant information.

[0138] (3) This scheme uses a gradient boosting decision tree based on importance sampling as a performance prediction model, which can effectively handle high-dimensional data and nonlinear relationships and improve the robustness and predictive ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0139] Figure 1 A schematic flow chart of a composite material pallet performance prediction method based on machine learning provided in one embodiment of the present invention;

[0140] Figure 2 A schematic structural diagram of an electronic device for implementing a composite material pallet performance prediction method based on machine learning provided by an embodiment of the present invention.

[0141] In the figure: 1 electronic device, 10 processor, 11 memory, 12 program, 13 communication interface.

[0142] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0143] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0144] The embodiment of the present application provides a composite material pallet performance prediction method based on machine learning. The execution subject of the composite material pallet performance prediction method based on machine learning includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided in the embodiment of the present application. In other words, the composite material pallet performance prediction method based on machine learning can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0145] Embodiment 1:

[0146] S1: Collecting composite material pallet parameter data and performing vectorization processing to obtain vectorized pallet parameters, wherein the parameter data includes material composition parameters, manufacturing process parameters, geometric dimension parameters, mechanical parameters and use load parameters.

[0147] In the step S1, the composite material pallet parameter data is collected and vectorized to obtain vectorized pallet parameters, including:

[0148] The material composition parameters include matrix material type, fiber type, fiber volume fraction and filler type, wherein the matrix material types include epoxy resin, polyester resin, polyethylene, polypropylene and polyurethane, which are vectorized by one-hot coding, the fiber types include glass fiber, carbon fiber and aramid fiber, which are vectorized by one-hot coding, the fiber volume fraction is the proportion of each fiber component in the composite material, and the proportion value is vectorized, and the filler types include glass microspheres, nanoclay, talcum powder and calcium carbonate, which are vectorized by one-hot coding;

[0149] The manufacturing process parameters include the fiber material laying angle, laying sequence, curing temperature and curing time of the composite laminate, wherein the fiber material laying angle converts the laying angle of each layer into a corresponding numerical code to form a coding vector, and the fiber material laying sequence converts the laying material of each layer into a corresponding numerical code to form a coding vector, and the vector dimension is the preset maximum number of laying layers;

[0150] The geometric dimension parameters include the pallet size and the pallet wall thickness, and the values ​​are concatenated and vectorized;

[0151] The mechanical parameters include tensile strength, compressive strength, bending strength and impact strength, wherein tensile strength indicates the maximum tensile stress of the pallet, compressive strength indicates the maximum compressive stress of the pallet, bending strength indicates the ability of the pallet to resist damage under bending load, and impact strength indicates the maximum energy absorption capacity of the pallet under impact load, and the values ​​are spliced ​​and vectorized;

[0152] The load parameters include static load and dynamic load, wherein the static load indicates the maximum weight that the pallet can bear under static conditions, and the dynamic load indicates the maximum weight that the pallet can bear under moving conditions, and the values ​​are concatenated and vectorized.

[0153] S2: concatenate the vectorized pallet parameters to obtain a fused pallet parameter vector and perform compression processing to obtain a simplified pallet parameter vector.

[0154] In the step S2, the vectorized pallet parameters are concatenated to obtain a fused pallet parameter vector and compressed to obtain a simplified pallet parameter vector, including:

[0155] S21: Convert the fused pallet parameter vector into a semantic vector through a linear layer , the conversion formula is:

[0156] ;

[0157] in:

[0158] Represents the semantic vector dimension;

[0159] represents the weight matrix of the encoder network;

[0160] represents the bias vector of the encoder network;

[0161] represents the output fusion pallet parameter vector;

[0162] S22: Calculate the attention score of the semantic vector. The calculation formula is:

[0163] ;

[0164] in:

[0165] represents the query vector;

[0166] represents the key vector;

[0167] Represents the semantic vector dimension;

[0168] S23: Normalize the attention scores to obtain normalized attention weights, so that the sum of all attention weights is 1. The normalization calculation formula is:

[0169] ;

[0170] in:

[0171] represents the attention score of the i-th dimension;

[0172] represents the normalized attention weight of the i-th dimension;

[0173] S24: Use the normalized attention weights to perform weighted summation on the semantic value vector to obtain a simplified pallet parameter vector. The calculation formula is:

[0174] ;

[0175] in:

[0176] represents the i-th semantic value vector;

[0177] represents the i-th value weight matrix;

[0178] S25: Decode the simplified pallet parameter vector through the decoder network to reconstruct the fused pallet parameter vector. The decoding and reconstruction calculation formula is:

[0179] ;

[0180] in:

[0181] represents the weight matrix of the decoder network;

[0182] represents the bias vector of the decoder network;

[0183] S26: Construct a target loss function based on the reconstructed fusion pallet parameter vector and the original fusion pallet parameter vector. The target loss function calculation formula is:

[0184] ;

[0185] in:

[0186] Represents the number of pallet parameter vectors;

[0187] represents the parameter vector of the i-th fusion pallet;

[0188] represents the parameter vector of the fusion pallet after the i-th reconstruction;

[0189] S27: Optimize and solve the target loss function to obtain the neural coding model parameters and perform vector compression processing, wherein mixed conjugate gradient is the implementation method of the optimization solution.

[0190] In the step S27, the objective loss function is optimized and solved to obtain the neural coding model parameters, including:

[0191] S27.1: Initialize the neural encoding model parameters and calculate the target loss function gradient under the current model parameters. The calculation formula is:

[0192] ;

[0193] in:

[0194] Represents the target loss function gradient of the kth iteration;

[0195] represents the model parameters of the kth iteration;

[0196] Indicates that the parameters The objective loss function for optimizing variables;

[0197] S27.2: Calculate the conjugate gradient direction, and perform weighted processing on the conjugate gradient direction and the gradient direction to obtain the hybrid search direction of the current round. The hybrid search direction calculation formula is:

[0198] ;

[0199] in:

[0200] Represents the dynamic weight parameter calculated in the kth round;

[0201] Represents the gradient direction of the target loss function;

[0202] Represents the conjugate gradient direction obtained by the k-th round of iteration;

[0203] Indicates mixed search direction;

[0204] The conjugate gradient calculation formula is:

[0205] ;

[0206] in: represents the adjustment factor, and the calculation formula of the adjustment factor is:

[0207] ;

[0208] S27.3: Determine the update step size by line search method, the calculation formula is:

[0209] ;

[0210] in:

[0211] represents the step size parameter variable;

[0212] Represents the update step size calculated in the kth round;

[0213] S27.4: Update the neural encoding model parameters according to the calculated update step size. The parameter update calculation formula is:

[0214] ;

[0215] in:

[0216] represents the model parameters obtained in the k+1th iteration;

[0217] S27.5: Iterate steps S27.1 to S27.4 until the maximum number of iterations is reached, and output the final model parameters as the optimal neural coding model parameters.

[0218] The dynamic weight parameters include:

[0219] S51: performing incremental calculation based on the conjugate gradient and gradient of the current round to obtain the conjugate gradient increment and gradient increment, and the calculation formula is:

[0220] ;

[0221] ;

[0222] in:

[0223] and They represent the conjugate gradient increment and gradient increment of the kth round respectively;

[0224] S52: Perform weight calculation based on the calculated conjugate gradient increment and gradient increment to obtain the weight parameter of the current round. The weight calculation formula is:

[0225] ;

[0226] in:

[0227] Represents a small positive number, and its value is , used to prevent the denominator from being 0.

[0228] S3: Construct a composite material pallet performance prediction model, which takes a simplified pallet parameter vector as input and a pallet comprehensive performance evaluation result as output. The pallet performance prediction model uses a gradient boosting decision tree based on importance sampling for prediction.

[0229] In the step S3, a composite material pallet performance prediction model is constructed. The model takes a simplified pallet parameter vector as input and takes a pallet comprehensive performance evaluation result as output, including:

[0230] S31: Randomly initialize the gradient boosting decision tree model and calculate the initial deviation. The initial deviation calculation formula is:

[0231] ;

[0232] in:

[0233] represents the initial residual of the i-th reduced pallet parameter vector sample data;

[0234] Represents the model prediction value With the true value The loss function between

[0235] represents the sample data of the i-th simplified pallet parameter vector;

[0236] Represents the predicted value of the model on the i-th streamlined pallet parameter vector sample data;

[0237] represents the true value of the i-th simplified pallet parameter vector sample data;

[0238] represents the predicted value of the model at the initial moment on the i-th streamlined pallet parameter vector sample data;

[0239] represents the partial derivative operation;

[0240] S32: Use the calculated deviation as the target variable to perform decision tree fitting. The specific process includes:

[0241] S32.1: Calculate the gradient of all sample data. The calculation formula is:

[0242] ;

[0243] in:

[0244] represents the gradient of the sample data of the i-th reduced pallet parameter vector;

[0245] represents the predicted value of the t-1 round model on the i-th streamlined pallet parameter vector sample data;

[0246] S32.2: Sort the sample data according to the absolute value of the calculated gradient, and select the one with the largest absolute value of gradient. samples as the important sample set A, and randomly sample from the remaining samples samples as the non-important sample set B;

[0247] in:

[0248] Represents the proportion of important sample sets, with a value range of (0,1);

[0249] Represents the proportion of non-important sample sets, with a value range of (0,1);

[0250] Represents the total amount of sample data of simplified pallet parameter vector;

[0251] S32.3: The gradient of non-important samples is amplified, and important samples are not processed. The gradient processing calculation formula is:

[0252] ;

[0253] in:

[0254] represents the gradient of the sample data of the i-th streamlined pallet parameter vector after adjustment;

[0255] S32.4: Use the adjusted gradients of all samples as classification targets for the sample set Building a decision tree ;

[0256] in:

[0257] represents the decision tree constructed in round t;

[0258] x represents the reduced pallet parameter vector;

[0259] S32.5: Update the model based on the decision tree constructed in the current round. The calculation formula is:

[0260] ;

[0261] in:

[0262] represents the learning rate;

[0263] Represents the model after the tth iteration;

[0264] Represents the model after the t-1th iteration;

[0265] S32.6: Calculate the deviation value according to the updated model, and iterate steps S32.1 to S32.5 until the maximum number of iterations is reached;

[0266] S33: Perform comprehensive performance evaluation of the pallet based on the calculated final model.

[0267] S4: Predict and evaluate the performance of the pallet based on the constructed composite pallet performance prediction model

[0268] Embodiment 2:

[0269] like Figure 2 , is a schematic diagram of the structure of an electronic device for implementing a composite material pallet performance prediction method based on machine learning provided by an embodiment of the present invention.

[0270] The electronic device 1 may include a processor 10 , a memory 11 , a communication interface 13 and a bus, and may also include a computer program stored in the memory 11 and executable on the processor 10 , such as a program 12 .

[0271] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 may not only be used to store application software and various types of data installed in the electronic device 1, such as the code of the program 12, etc., but also be used to temporarily store data that has been output or is to be output.

[0272] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 10 is the control core (ControlUnit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as a program 12 for realizing performance prediction of composite material pallets based on machine learning), and calls data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0273] The communication interface 13 may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices, and to achieve connection and communication between internal components of the electronic device.

[0274] The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0275] Figure 2 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 2 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0276] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0277] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0278] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0279] The program 12 stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0280] Collecting composite material pallet parameter data and performing vector processing to obtain vectorized pallet parameters;

[0281] The vectorized pallet parameters are concatenated to obtain a fused pallet parameter vector and compressed to obtain a simplified pallet parameter vector;

[0282] A composite material pallet performance prediction model is constructed, wherein the model takes a simplified pallet parameter vector as input and a pallet comprehensive performance evaluation result as output, and the pallet performance prediction model uses a gradient boosting decision tree based on importance sampling for prediction;

[0283] The performance of the pallet is predicted and evaluated based on the constructed composite pallet performance prediction model.

[0284] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figure 1 to Figure 2 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0285] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method 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, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0286] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0287] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A composite material pallet performance prediction method based on machine learning, characterized in that: The method comprises: S1: Collecting composite material pallet parameter data and performing vectorization processing to obtain vectorized pallet parameters, wherein the parameter data includes material composition parameters, manufacturing process parameters, geometric size parameters, mechanical parameters and use load parameters; S2: concatenate the vectorized pallet parameters to obtain a fused pallet parameter vector and perform compression processing to obtain a simplified pallet parameter vector; S3: Construct a composite material pallet performance prediction model, the model takes a simplified pallet parameter vector as input and a pallet comprehensive performance evaluation result as output, wherein a gradient boosting decision tree based on importance sampling is an implementation method of the pallet performance prediction model; S4: Predict and evaluate the performance of the pallet based on the constructed composite pallet performance prediction model; In the step S1, the composite material pallet parameter data is collected and vectorized to obtain vectorized pallet parameters, including: The material composition parameters include matrix material type, fiber type, fiber volume fraction and filler type, wherein the matrix material types include epoxy resin, polyester resin, polyethylene, polypropylene and polyurethane, which are vectorized by one-hot coding, the fiber types include glass fiber, carbon fiber and aramid fiber, which are vectorized by one-hot coding, the fiber volume fraction is the proportion of each fiber component in the composite material, and the proportion value is vectorized, and the filler types include glass microspheres, nanoclay, talcum powder and calcium carbonate, which are vectorized by one-hot coding; The manufacturing process parameters include the fiber material laying angle, laying sequence, curing temperature and curing time of the composite laminate, wherein the fiber material laying angle converts the laying angle of each layer into a corresponding numerical code to form a coding vector, and the fiber material laying sequence converts the laying material of each layer into a corresponding numerical code to form a coding vector, and the vector dimension is the preset maximum number of laying layers; The geometric dimension parameters include the pallet size and the pallet wall thickness, and the values ​​are concatenated and vectorized; The mechanical parameters include tensile strength, compressive strength, bending strength and impact strength, wherein tensile strength indicates the maximum tensile stress of the pallet, compressive strength indicates the maximum compressive stress of the pallet, bending strength indicates the ability of the pallet to resist damage under bending load, and impact strength indicates the maximum energy absorption capacity of the pallet under impact load, and the values ​​are spliced ​​and vectorized; The load parameters include static load and dynamic load, wherein the static load indicates the maximum weight that the pallet can bear under static conditions, and the dynamic load indicates the maximum weight that the pallet can bear under moving conditions, and the values ​​are concatenated and vectorized; In the step S2, the vectorized pallet parameters are concatenated to obtain a fused pallet parameter vector and compressed to obtain a simplified pallet parameter vector, including: S21: Convert the fused pallet parameter vector into a semantic vector through a linear layer , the conversion formula is: ； in: Represents the semantic vector dimension; represents the weight matrix of the encoder network; represents the bias vector of the encoder network; represents the output fusion pallet parameter vector; S22: Calculate the attention score of the semantic vector. The calculation formula is: ； in: represents the query vector; represents the key vector; Represents the semantic vector dimension; S23: Normalize the attention scores to obtain normalized attention weights, so that the sum of all attention weights is 1. The normalization calculation formula is: ； in: represents the attention score of the i-th dimension; represents the normalized attention weight of the i-th dimension; S24: Use the normalized attention weights to perform weighted summation on the semantic value vector to obtain a simplified pallet parameter vector. The calculation formula is: ； in: represents the i-th semantic value vector; represents the i-th value weight matrix; S25: Decode the simplified pallet parameter vector through the decoder network to reconstruct the fused pallet parameter vector. The decoding and reconstruction calculation formula is: ； in: represents the weight matrix of the decoder network; represents the bias vector of the decoder network; S26: Construct a target loss function based on the reconstructed fusion pallet parameter vector and the original fusion pallet parameter vector. The target loss function calculation formula is: ； in: Represents the number of pallet parameter vectors; represents the parameter vector of the i-th fusion pallet; represents the parameter vector of the fusion pallet after the i-th reconstruction; S27: Optimizing and solving the target loss function to obtain neural coding model parameters and performing vector compression processing, wherein mixed conjugate gradient is an implementation method of the optimization solution; In the step S27, the objective loss function is optimized and solved to obtain the neural coding model parameters, including: S27.1: Initialize the neural encoding model parameters and calculate the target loss function gradient under the current model parameters. The calculation formula is: ； in: Represents the target loss function gradient of the kth iteration; represents the model parameters of the kth iteration; Indicates that the parameters The objective loss function for optimizing variables; S27.2: Calculate the conjugate gradient direction, and perform weighted processing on the conjugate gradient direction and the gradient direction to obtain the hybrid search direction of the current round. The hybrid search direction calculation formula is: ； in: Represents the dynamic weight parameter calculated in the kth round; Represents the gradient direction of the target loss function; Represents the conjugate gradient direction obtained by the k-th round of iteration; Indicates mixed search direction; The conjugate gradient calculation formula is: ； in: represents the adjustment factor, and the calculation formula of the adjustment factor is: ； S27.3: Determine the update step size by line search method, the calculation formula is: ； in: represents the step size parameter variable; Represents the update step size calculated in the kth round; S27.4: Update the neural encoding model parameters according to the calculated update step size. The parameter update calculation formula is: ； in: represents the model parameters obtained in the k+1th iteration; S27.5: iteratively execute steps S27.1 to S27.4 until the maximum number of iterations is reached, and output the final model parameters as the optimal neural coding model parameters; The dynamic weight parameters include: S51: performing incremental calculation based on the conjugate gradient and gradient of the current round to obtain the conjugate gradient increment and gradient increment, and the calculation formula is: ； ； in: and They represent the conjugate gradient increment and gradient increment of the kth round respectively; S52: Perform weight calculation based on the calculated conjugate gradient increment and gradient increment to obtain the weight parameter of the current round. The weight calculation formula is: ； in: Represents a small positive number, and its value is , used to prevent the denominator from being 0; In the step S3, a composite material pallet performance prediction model is constructed. The model takes a simplified pallet parameter vector as input and takes a pallet comprehensive performance evaluation result as output, including: S31: Randomly initialize the gradient boosting decision tree model and calculate the initial deviation. The initial deviation calculation formula is: ； in: represents the initial residual of the i-th reduced pallet parameter vector sample data; Represents the model prediction value With the true value The loss function between represents the sample data of the i-th simplified pallet parameter vector; Represents the predicted value of the model on the i-th streamlined pallet parameter vector sample data; represents the true value of the i-th simplified pallet parameter vector sample data; represents the predicted value of the model at the initial moment on the i-th streamlined pallet parameter vector sample data; represents the partial derivative operation; S32: Use the calculated deviation as the target variable to perform decision tree fitting; S33: Perform comprehensive performance evaluation of the pallet based on the calculated final model; The step S32 uses the calculated deviation as the target variable to perform decision tree fitting, including: S32.1: Calculate the gradient of all sample data. The calculation formula is: ； in: represents the gradient of the sample data of the i-th reduced pallet parameter vector; represents the predicted value of the t-1 round model on the i-th streamlined pallet parameter vector sample data; S32.2: Sort the sample data according to the absolute value of the calculated gradient, and select the one with the largest absolute value of gradient. samples as the important sample set A, and randomly sample from the remaining samples samples as the non-important sample set B; in: Represents the proportion of important sample sets, with a value range of (0,1); Represents the proportion of non-important sample sets, with a value range of (0,1); Represents the total amount of sample data of simplified pallet parameter vector; S32.3: The gradient of non-important samples is amplified, and important samples are not processed. The gradient processing calculation formula is: ； in: represents the gradient of the sample data of the i-th streamlined pallet parameter vector after adjustment; S32.4: Use the adjusted gradients of all samples as classification targets for the sample set Building a decision tree ; in: represents the decision tree constructed in round t; x represents the reduced pallet parameter vector; S32.5: Update the model based on the decision tree constructed in the current round. The calculation formula is: ； in: represents the learning rate; Represents the model after the tth iteration; Represents the model after the t-1th iteration; S32.6: Calculate the deviation value based on the updated model, and iterate steps S32.1 to S32.5 until the maximum number of iterations is reached.

Images