Processing method and device for predicting limit strain characteristics of concrete

Abstract

The embodiment of the invention relates to a processing method and device for predicting limit strain characteristics of concrete. The method comprises the following steps: constructing a concrete limit strain characteristic prediction model as a first prediction model based on a BP network; performing data acquisition on the plurality of concrete manufacturer / building industry literature data sources to construct a first data set; using a GA algorithm to carry out a first round of training on the first prediction model according to the first data set; performing a second round of training on each first round of model according to the first data set by using a model parameter optimizer; performing performance evaluation on each second round model based on the first data set; performing model parameter solidification on the first prediction model based on the model parameter of the second round model with the highest first evaluation value; and processing batch prediction tasks based on the first prediction model after completing model parameter solidification. According to the invention, the processing efficiency of single / batch limit strain value acquisition tasks can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a processing method and device for predicting ultimate strain characteristics of concrete. Background Art

[0002] There is a limit stress point on the stress-strain curve of reinforced concrete components (referred to as concrete components) during the compression test. The strain value corresponding to this limit stress point is called the limit strain value ε ccu . From the stress model of concrete components, it can be known that the ultimate strain value is related to a series of design parameters of concrete components. These design parameters are commonly the minor axis size of stirrups, the major axis size of stirrups, the cross-sectional area of ​​stirrups, the moment of inertia of stirrup sections, the shape code of stirrups, the spacing of stirrups, the reinforcement ratio of stirrups, the strength of stirrups, the number of stirrups, the strength of concrete, the thickness of concrete cover, the longitudinal size of components, the aspect ratio of components, the cross-sectional shape of components, the strength of longitudinal reinforcement, the reinforcement ratio of longitudinal reinforcement, the characteristic value of stirrups and the characteristic value of reinforcement. Under normal circumstances, to obtain the ultimate strain value of a concrete component, either a mechanical simulation software with modeling and simulation functions is used to perform three-dimensional modeling of the component and a compression simulation analysis based on the three-dimensional model to obtain the corresponding ultimate strain value, or the ultimate strain value is obtained through the actual measurement information of the compression experiment. Through practice, we found that these two conventional processing methods consume a lot of simulation calculation time and experimental time to obtain the ultimate strain value of each concrete component, and there is a problem of low efficiency of single processing. Based on these two conventional processing methods, it is obviously impossible to quickly obtain the ultimate strain value of batch concrete components. Summary of the invention

[0003] The purpose of the present invention is to provide a processing method, device, electronic device and computer-readable storage medium for predicting the ultimate strain characteristics of concrete in view of the defects of the prior art. The present invention first constructs a prediction model of the ultimate strain characteristics of concrete based on a back propagation (BP) network, and constructs a data set by collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources; then uses a genetic algorithm (GA) algorithm to perform a first round of training on the prediction model according to the data set, and uses a model parameter optimizer to perform a second round of training on each first round model according to the data set, and performs performance evaluation on each second round model based on the data set, and solidifies the model parameters of the first prediction model based on the model parameters of the second round model with the highest evaluation value; finally, after completing the model parameter solidification, the batch prediction task is processed based on the first prediction model. The processing efficiency of single / batch tasks can be improved by processing the data acquisition task of the ultimate strain value through the prediction model provided by the present invention.

[0004] To achieve the above object, a first aspect of an embodiment of the present invention provides a method for predicting the ultimate strain characteristics of concrete, the method comprising:

[0005] Based on BP network, a concrete ultimate strain characteristic prediction model is constructed as the corresponding first prediction model;

[0006] Constructing a corresponding first data set by collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources;

[0007] Using the GA algorithm to perform a first round of training on the first prediction model according to the first data set to obtain multiple first round models; and using a preset model parameter optimizer to perform a second round of training on each of the first round models according to the first data set to obtain a corresponding second round model; and based on the first data set, a performance evaluation is performed on each of the second round models to obtain a corresponding first evaluation value; and based on the model parameters of the second round model with the highest first evaluation value, the model parameters of the first prediction model are solidified;

[0008] After the model parameters are solidified, batch prediction tasks are processed based on the first prediction model.

[0009] Preferably, the first prediction model is used to predict the ultimate strain characteristics of the concrete component according to the normalized parameter vector x input by the model and output the corresponding normalized prediction value y;

[0010] The vector length of the normalized parameter vector x is 19, and it is composed of 19 types of normalized parameters; the 19 types of normalized parameters include normalized stirrup minor axis size, normalized stirrup major axis size, normalized stirrup cross-sectional area, normalized stirrup cross-sectional inertia moment, normalized stirrup shape code, normalized stirrup spacing, normalized stirrup reinforcement ratio, normalized stirrup strength, normalized stirrup limb number, normalized concrete strength, normalized cross-sectional size, normalized concrete cover thickness, normalized component longitudinal size, normalized component aspect ratio, normalized component cross-sectional shape, normalized longitudinal reinforcement strength, normalized longitudinal reinforcement ratio, normalized stirrup eigenvalue and normalized reinforcement eigenvalue;

[0011] The normalized predicted value y is a normalized limit strain value;

[0012] The first prediction model includes a first input layer, a first hidden layer and a first output layer; the neuron transfer functions used in the first hidden layer and the first output layer in the first prediction model are both sigmoid activation functions;

[0013] The first input layer is composed of nineteen parallel input layer neurons; the input layer neurons correspond one to one to the normalized parameters in the normalized parameter vector x;

[0014] Each of the input layer neurons of the first input layer is used to receive the corresponding normalized parameter in the normalized parameter vector x as a corresponding neuron input value;

[0015] The first hidden layer is composed of a plurality of parallel hidden layer neurons, and each of the hidden layer neurons is fully connected to all the input layer neurons;

[0016] Each of the hidden layer neurons of the first hidden layer is used to perform full connection calculation on the neuron input values ​​of all the input layer neurons connected to itself to generate a corresponding first fully connected vector, and perform activation operation on the first fully connected vector based on a sigmoid activation function to obtain a corresponding first activation vector;

[0017] The first output layer has only one output layer neuron, and the output layer neuron is fully connected to all the hidden layer neurons;

[0018] The output layer neurons of the first output layer are used to perform a full connection calculation on the first activation vector of all the hidden layer neurons connected to themselves to generate a corresponding second fully connected vector, and to perform an activation operation on the second fully connected vector based on the sigmoid activation function and output the operation result as the corresponding normalized prediction value y.

[0019] Preferably, the construction industry literature data source includes journal articles, books, research reports, patents and standard specifications of the construction industry;

[0020] The first data set includes a plurality of first data records; the first data record includes a first training parameter vector and a first label prediction value; the first training parameter vector has the same data structure as the normalized parameter vector x and is also composed of nineteen normalized parameters; the first label prediction value has the same data meaning as the normalized prediction value y and is also a normalized limit strain value;

[0021] The model parameter optimizer includes an ADAM optimizer.

[0022] Preferably, the step of collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources to construct a corresponding first data set specifically includes:

[0023] Obtain a plurality of first concrete component compression test reports from a plurality of concrete manufacturer data sources; and obtain a plurality of first industry documents from a plurality of construction industry document data sources;

[0024] And all the first concrete component compression test reports and all the first industry documents are retrieved for concrete component design parameters and corresponding component experimental measurement limit strain values ​​to obtain corresponding multiple first search records; the first search record includes a first component type, a first component design parameter group and a first experimental measurement limit strain value; the first component type includes first, second, third and fourth types; the concrete component type corresponding to the first type is a concrete component with both stirrups and longitudinal reinforcement; the concrete component type corresponding to the second type is a concrete component with only stirrups but no longitudinal reinforcement; the concrete component type corresponding to the third type is a concrete component with only longitudinal reinforcement but no stirrups; the concrete component type corresponding to the fourth type is a concrete component with neither stirrups nor longitudinal reinforcement; the first component type is a record of the first search record of the first type The number of records of the first search record of the first component type being the second type is the largest, the number of records of the first search record of the first component type being the third type is the third, and the number of records of the first search record of the first component type being the fourth type is the smallest; the first component design parameter group is composed of 19 types of original parameters that have not been normalized; the 19 types of original parameters that have not been normalized include stirrup minor axis size, stirrup major axis size, stirrup cross-sectional area, stirrup cross-sectional moment of inertia, stirrup shape code, stirrup spacing, stirrup reinforcement ratio, stirrup strength, number of stirrup limbs, concrete strength, cross-sectional size, concrete cover thickness, component longitudinal size, component aspect ratio, component cross-sectional shape, longitudinal reinforcement strength, longitudinal reinforcement ratio, stirrup characteristic value and reinforcement characteristic value; the first experimental measurement limit strain value is an unnormalized limit strain value;

[0025] Traverse all the first retrieval records; and during the traversal, take the first retrieval record currently traversed as the corresponding current retrieval record; and normalize the nineteen categories of unnormalized original parameters of the current retrieval record respectively to obtain a group of corresponding nineteen categories of normalized parameters to form a corresponding first training parameter vector; and normalize the first experimental measurement limit strain value of the current retrieval record to obtain the corresponding first label prediction value; and form a corresponding first data record with the first training parameter vector and the first label prediction value corresponding to the current retrieval record; and at the end of the traversal, form the corresponding first data set with all the first data records obtained.

[0026] Preferably, the using the GA algorithm to perform a first round of training on the first prediction model according to the first data set to obtain multiple first round models specifically includes:

[0027] Step 501: The total number of neurons in the input layer of the first prediction model n inSet to 19, the total number of neurons in the output layer n out Set to 1, and based on Get the total number of neurons n in multiple hidden layers hid The total number of neurons in the hidden layer.

[0028] Among them, F INT () is a rounding function, the value range of △n is [0, n0-1], and the preset number n0 is an integer greater than 1; the total number of hidden layer neurons in the total number of hidden layer neurons n hid The total number of matches the preset number n0;

[0029] Step 502: randomly extracting a plurality of first data records from the first data set based on a preset first extraction ratio to form a corresponding second data set;

[0030] wherein the first extraction ratio is a percentage less than 1; the total number of records in the first data set matches the product of the first extraction ratio and the total number of records in the second data set;

[0031] Step 503, based on the total number of neurons n in each hidden layer hid The shape of the first hidden layer of the first prediction model is set to obtain a corresponding first body model; and the model parameter set of each first body model is initialized; and a real number vector is converted based on the current model parameter set of each first body model to obtain a corresponding first body parameter vector; and a corresponding first body error is set for each first body model, and each first body error is initialized to 1; and a round of iteration counter is initialized to 1;

[0032] Wherein, the total number of the first individual models matches the preset number n0;

[0033] Step 504, taking each of the first individual models as the corresponding current individual model; and traversing all the first data records of the second data set; and taking the first data record currently traversed as the corresponding current data record during the traversal; and inputting the first training parameter vector of the current data record into the current individual model for prediction processing to obtain the corresponding first training prediction value; and forming a corresponding first prediction label pair from the first training prediction value and the first label prediction value of the current data record; and at the end of this traversal, bringing all the first prediction label pairs corresponding to the current individual model into a preset first error function for error calculation to obtain the corresponding first error value, and resetting the first individual error corresponding to the current individual model to the corresponding first error value;

[0034] Wherein, the first error function is a mean square error function;

[0035] Step 505: taking the inverse of the first individual error corresponding to each first individual model as the corresponding first individual fitness; summing up all the first individual fitnesses obtained to obtain the corresponding first fitness sum; taking the ratio of each first individual fitness to the first fitness sum as the corresponding first individual selection probability; and sorting all the first individual models in descending order of the first individual selection probability to obtain the corresponding first individual model sequence;

[0036] Step 506, identifying whether the preset number n0 is an odd number; if so, go to step 507; if not, go to step 508;

[0037] Step 507, taking the first first individual model in the first individual model sequence as the elite model of the time; and forming the second and third first individual models of the first individual model sequence into a corresponding first paired individual, and forming the fourth and fifth first individual models into a corresponding first paired individual, and so on, until the n0-1th and n0th first individual models form a corresponding first paired individual; and directly taking the first individual parameter vector of the elite model of the time as the corresponding first individual crossover parameter vector; and performing vector parameter crossover processing on the first individual parameter vectors of the two first individual models of each first paired individual based on a preset individual crossover algorithm and a preset first crossover probability to obtain two corresponding first individual crossover parameter vectors; and going to step 509;

[0038] The individual crossover algorithm is a type of chromosome crossover algorithm in the GA algorithm, and the chromosome crossover algorithm includes at least a single-point crossover algorithm, a multi-point crossover algorithm, and an arithmetic crossover algorithm;

[0039] Step 508: The first and second first individual models of the first individual model sequence are combined into a corresponding first paired individual, the third and fourth first individual models are combined into a corresponding first paired individual, and so on, until the n0-1th and n0th first individual models are combined into a corresponding first paired individual; and based on the individual crossover algorithm and the first crossover probability, the first individual parameter vectors of the two first individual models of each first paired individual are subjected to vector parameter crossover processing to obtain two corresponding first individual crossover parameter vectors;

[0040] Step 509, performing vector parameter mutation processing on the first individual crossover parameter vector of each of the first individual models based on a preset individual mutation algorithm and a preset first mutation probability to obtain a corresponding first individual mutation parameter vector;

[0041] The individual mutation algorithm is a type of chromosome mutation algorithm in the GA algorithm, and the chromosome mutation algorithm at least includes a basic position mutation algorithm, a uniform mutation algorithm, a non-uniform mutation algorithm and a Gaussian approximate mutation algorithm;

[0042] Step 510, resetting the model parameter set of the corresponding first individual model based on each of the first individual variation parameter vectors;

[0043] Step 511, identifying whether the one-round iteration counter exceeds a preset one-round iteration number threshold; if it exceeds, go to step 512; if it does not exceed, add 1 to the one-round iteration counter and return to step 504;

[0044] Step 512, solidifying the model parameters of each of the first individual models, and using each of the first individual models after the solidification as a corresponding first round model.

[0045] Preferably, the using a preset model parameter optimizer to perform a second round of training on each of the first round models according to the first data set to obtain a corresponding second round model specifically includes:

[0046] Step 601, initializing the second round iteration counter to 1; and using the first round model currently being trained as the corresponding current model;

[0047] Step 602: randomly select a preset first number M of the first data records from the first data set to form a corresponding first batch record set; and traverse all the first data records in the first batch record set; and during the traversal, use the first data record currently traversed as the corresponding current data record; and input the first training parameter vector of the current data record into the current model for prediction processing to obtain the corresponding second training prediction value y i ; and use the first tag prediction value of the current data record as the corresponding second tag prediction value And the second training prediction value y i and the second label prediction value Form a corresponding second prediction label pair;

[0048] Wherein, the first number M is an integer greater than 0; 1≤batch record index i≤M;

[0049] Step 603, bringing all the obtained second prediction label pairs into the preset first model loss function LOSS to calculate the corresponding first batch loss values;

[0050] Among them, the first model loss function LOSS is:

[0051]

[0052] λ is the preset first penalty factor, p is the preset first probability threshold; ε is the step function, Greater than or equal to y i The corresponding step function The function value of is 1. Less than y i The step function corresponding to The function value of is 0;

[0053] Step 604: predict the value of the second tag is greater than the second training prediction value y i The total number of the second predicted label pairs is counted to obtain the corresponding total number of first label pairs m; and the corresponding first batch probability is calculated based on the total number of the first label pairs m and the first quantity M;

[0054] in,

[0055] For is the probability function of a probabilistic event;

[0056] Step 605, identifying the first batch loss value and the first batch probability; if the first batch loss value meets the preset first loss value convergence range and the first batch probability is greater than or equal to the first probability threshold p, go to step 609; if the first batch loss value does not meet the first loss value convergence range or the first batch probability is less than the first probability threshold p, go to step 606;

[0057] Step 606, setting the learning rate of the model parameter optimizer, specifically: identifying the second-round iteration counter; if the second-round iteration counter is 1, setting the learning rate to a preset first learning rate; if the second-round iteration counter is greater than 1, identifying whether the remainder of the second-round iteration counter divided by the preset first number of iterations is 1, and if so, taking the product of the current learning rate and the preset first attenuation factor as the new learning rate;

[0058] Step 607, setting the optimization objective function O of the model parameter optimizer and its corresponding constraint conditions;

[0059] Wherein, the optimization objective function O is:

[0060] The constraints of the optimization objective function O are:

[0061] Step 608, performing a round of optimization processing on the current model parameter set of the current model based on the model parameter optimizer;

[0062] Step 609, identifying whether the second-round iteration counter exceeds a preset second-round iteration number threshold; if so, going to step 610; if not, adding 1 to the second-round iteration counter and returning to step 602;

[0063] Step 610, solidify the model parameters of the current model, and use the current model after the parameters are solidified as a corresponding second-round model.

[0064] Preferably, the performing performance evaluation on each of the second-round models based on the first data set to obtain a corresponding first evaluation value specifically includes:

[0065] Step 71: randomly extract multiple first data records from the first data set based on a preset second extraction ratio to form a corresponding first evaluation data set; and count the total number of records in the first evaluation data set to obtain a corresponding second number K;

[0066] wherein the second extraction ratio is a percentage less than 1; the total number of records of the first data set multiplied by the second extraction ratio and the second number K match;

[0067] Step 72, taking each of the second round models as the corresponding current model one by one;

[0068] Step 73, traverse all the first data records of the first evaluation data set; and during the traversal, use the first data record currently traversed as the corresponding current data record; and input the first training parameter vector of the current data record into the current model for prediction processing to obtain the corresponding third training prediction value y j ; and use the first tag prediction value of the current data record as the corresponding third tag prediction value And the third training prediction value y j and the third label prediction value Form a corresponding third prediction label pair;

[0069] Among them, 1≤evaluation record index j≤K;

[0070] Step 74: Calculate the corresponding first relative error e based on each of the third predicted labels. j ; and for all the first relative errors e obtained j Perform mean calculation to obtain the corresponding first relative error mean e aver ; and according to all the first relative errors e j and the first relative error mean e aver Calculate the corresponding first relative error standard deviation s;

[0071] in,

[0072] Step 75: according to the first relative error mean value e aver and the first relative error standard deviation s to perform evaluation value calculation to obtain the corresponding first evaluation value;

[0073] in, a1 and a2 are two preset weighting parameters.

[0074] Preferably, processing batch prediction tasks based on the first prediction model after completing the solidification of the model parameters specifically includes:

[0075] After the model parameters are solidified, a first batch prediction list is received; the first batch prediction list includes one or more first concrete component parameter groups; the first concrete component parameter group consists of 19 types of original parameters that have not been normalized; the 19 types of original parameters that have not been normalized include stirrup minor axis size, stirrup major axis size, stirrup cross-sectional area, stirrup cross-sectional moment of inertia, stirrup shape code, stirrup spacing, stirrup reinforcement ratio, stirrup strength, number of stirrup limbs, concrete strength, cross-sectional size, concrete cover thickness, component longitudinal size, component aspect ratio, component cross-sectional shape, longitudinal reinforcement strength, longitudinal reinforcement ratio, stirrup characteristic value and reinforcement characteristic value;

[0076] All the first concrete component parameter groups are traversed; and during the traversal, the first concrete component parameter group currently traversed is used as the corresponding current parameter group; and the 19 categories of unnormalized original parameters of the current parameter group are respectively normalized to obtain a group of corresponding 19 categories of normalized parameters to form a corresponding first model input vector; and the first model input vector is input into the first prediction model for prediction processing to obtain the corresponding first model prediction value; and the first model prediction value is denormalized to obtain the corresponding first limit strain prediction value; and the current parameter group and the first limit strain prediction value form a corresponding first prediction record; and at the end of the traversal, all the first prediction records obtained form a corresponding first batch prediction report and save it.

[0077] A second aspect of an embodiment of the present invention provides a device for implementing the processing method for predicting the ultimate strain characteristics of concrete as described in the first aspect, the device comprising: a model building module, a data set building module, a model training module and a model application module;

[0078] The model building module is used to build a concrete ultimate strain characteristic prediction model based on a BP network as a corresponding first prediction model;

[0079] The data set construction module is used to construct a corresponding first data set by collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources;

[0080] The model training module is used to use the GA algorithm to perform a first round of training on the first prediction model according to the first data set to obtain multiple first round models; and use a preset model parameter optimizer to perform a second round of training on each of the first round models according to the first data set to obtain a corresponding second round model; and based on the first data set, the performance of each of the second round models is evaluated to obtain a corresponding first evaluation value; and based on the model parameters of the second round model with the highest first evaluation value, the model parameters of the first prediction model are solidified;

[0081] The model application module is used to process batch prediction tasks based on the first prediction model after completing the solidification of model parameters.

[0082] A third aspect of an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0083] The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement the method steps described in the first aspect above;

[0084] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0085] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions. When the computer instructions are executed by a computer, the computer executes the instructions of the method described in the first aspect above.

[0086] The embodiment of the present invention provides a processing method, device, electronic device and computer-readable storage medium for predicting the ultimate strain characteristics of concrete. As can be seen from the above content, the embodiment of the present invention first constructs a prediction model of the ultimate strain characteristics of concrete by using a BP network, and constructs a data set by collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources; then uses a GA algorithm to perform a first round of training on the prediction model according to the data set, and uses a model parameter optimizer to perform a second round of training on each first round model according to the data set, and performs a performance evaluation on each second round model based on the data set, and solidifies the model parameters of the first prediction model based on the model parameters of the second round model with the highest evaluation value; finally, after completing the model parameter solidification, the batch prediction task is processed based on the first prediction model. The prediction model provided by the embodiment of the present invention is used to process the data acquisition task of the ultimate strain value, which not only improves the processing efficiency of single tasks, but also improves the processing efficiency of batch tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 A schematic diagram of a processing method for predicting the ultimate strain characteristics of concrete provided in the first embodiment of the present invention;

[0088] Figure 2 A module structure diagram of a processing device for predicting the ultimate strain characteristics of concrete provided in the second embodiment of the present invention;

[0089] Figure 3 A schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION

[0090] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0091] Embodiment 1 of the present invention provides a method for predicting the ultimate strain characteristics of concrete, such as Figure 1 As shown in the schematic diagram of a processing method for predicting the ultimate strain characteristics of concrete provided in the first embodiment of the present invention, the method mainly includes the following steps:

[0092] Step 1: construct a concrete ultimate strain characteristic prediction model based on BP network as the corresponding first prediction model.

[0093] Here, the first prediction model of the embodiment of the present invention is used to predict the ultimate strain characteristics of the concrete component according to the normalized parameter vector x input by the model and output the corresponding normalized prediction value y; wherein, the vector length of the normalized parameter vector x is 19, and it is composed of nineteen types of normalized parameters; the nineteen types of normalized parameters include normalized stirrup minor axis size, normalized stirrup major axis size, normalized stirrup cross-sectional area, normalized stirrup cross-sectional moment of inertia, normalized stirrup shape coding, normalized stirrup spacing, normalized stirrup reinforcement ratio, normalized stirrup strength, normalized number of stirrup limbs, normalized concrete strength, normalized cross-sectional size, normalized concrete cover thickness, normalized member longitudinal size, normalized member aspect ratio, normalized member cross-sectional shape, normalized longitudinal reinforcement strength, normalized longitudinal reinforcement ratio, normalized stirrup characteristic value and normalized reinforcement characteristic value; the normalized prediction value y is a normalized ultimate strain value.

[0094] The model structure of the first prediction model in the embodiment of the present invention includes a first input layer, a first hidden layer and a first output layer; and the neuron transfer functions used in the first hidden layer and the first output layer in the first prediction model are both sigmoid activation functions.

[0095] 1) First input layer:

[0096] The first input layer of the first prediction model of the embodiment of the present invention is composed of nineteen parallel input layer neurons; here, the input layer neurons correspond one-to-one to the normalized parameters in the normalized parameter vector x.

[0097] Each input layer neuron of the first input layer is used to receive the corresponding normalized parameters in the normalized parameter vector x as the corresponding neuron input value.

[0098] 2) First hidden layer:

[0099] The first hidden layer of the first prediction model of the embodiment of the present invention is composed of a plurality of parallel hidden layer neurons, each hidden layer neuron is fully connected to all input layer neurons. The optimal value of the total number of hidden layer neurons needs to be confirmed by subsequent steps.

[0100] Each hidden layer neuron of the first hidden layer is used to perform a full connection calculation on the neuron input values ​​of all input layer neurons connected to itself to generate a corresponding first fully connected vector, and perform an activation operation on the first fully connected vector based on the sigmoid activation function to obtain a corresponding first activation vector.

[0101] 3) First output layer:

[0102] The first output layer of the first prediction model in the embodiment of the present invention has only one output layer neuron, and the output layer neuron is fully connected to all hidden layer neurons.

[0103] The output layer neurons of the first output layer are used to perform a full connection calculation on the first activation vectors of all hidden layer neurons connected to themselves to generate a corresponding second fully connected vector, and perform an activation operation on the second fully connected vector based on the sigmoid activation function and output the operation result as the corresponding normalized prediction value y.

[0104] Step 2, constructing a corresponding first data set by collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources;

[0105] Among them, the literature data sources of the construction industry include journal articles, books, research reports, patents and standard specifications of the construction industry;

[0106] The first data set includes a plurality of first data records; the first data record includes a first training parameter vector and a first label prediction value; the first training parameter vector has the same data structure as the normalized parameter vector x and is also composed of nineteen normalized parameters; the first label prediction value has the same data meaning as the normalized prediction value y and is also a normalized limit strain value;

[0107] Specifically, it includes: step 21, obtaining a plurality of first concrete component compression test reports from a plurality of concrete manufacturer data sources; and obtaining a plurality of first industry documents from a plurality of construction industry document data sources;

[0108] Step 22, searching all first concrete component compression test reports and all first industry documents for concrete component design parameters and corresponding component experimental measurement limit strain values ​​to obtain corresponding multiple first search records;

[0109] Here, the first search record of the embodiment of the present invention includes a first component type, a first component design parameter group, and a first experimental measurement limit strain value;

[0110] Among them, the first component type includes the first, second, third and fourth types; the concrete component type corresponding to the first type is a concrete component with both stirrups and longitudinal reinforcements; the concrete component type corresponding to the second type is a concrete component with only stirrups but no longitudinal reinforcements; the concrete component type corresponding to the third type is a concrete component with only longitudinal reinforcements but no stirrups; the concrete component type corresponding to the fourth type is a concrete component with neither stirrups nor longitudinal reinforcements; it should be noted that the number of records of the first search record whose first component type is the first type is the largest, the number of records of the first search record whose first component type is the second type is the second largest, the number of records of the first search record whose first component type is the third type is the third largest, and the number of records of the first search record whose first component type is the fourth type is the smallest;

[0111] The first component design parameter group consists of 19 types of original parameters that have not been normalized; the 19 types of original parameters that have not been normalized include stirrup minor axis size, stirrup major axis size, stirrup cross-sectional area, stirrup cross-sectional inertia moment, stirrup shape code, stirrup spacing, stirrup reinforcement ratio, stirrup strength, number of stirrup legs, concrete strength, cross-sectional size, concrete cover thickness, component longitudinal size, component aspect ratio, component cross-sectional shape, longitudinal reinforcement strength, longitudinal reinforcement ratio, stirrup characteristic value and reinforcement characteristic value;

[0112] The ultimate strain value measured in the first experiment is an unnormalized ultimate strain value;

[0113] Step 23, traverse all first retrieval records; and during the traversal, take the currently traversed first retrieval record as the corresponding current retrieval record; and normalize the 19 categories of unnormalized original parameters of the current retrieval record respectively to obtain a set of corresponding 19 categories of normalized parameters to form a corresponding first training parameter vector; and normalize the first experimental measurement limit strain value of the current retrieval record to obtain the corresponding first label prediction value; and form a corresponding first data record with the first training parameter vector and the first label prediction value corresponding to the current retrieval record; and at the end of the traversal, all the obtained first data records form a corresponding first data set.

[0114] Step 3, using the GA algorithm to perform a first round of training on the first prediction model according to the first data set to obtain multiple first round models; and using a preset model parameter optimizer to perform a second round of training on each first round model according to the first data set to obtain a corresponding second round model; and based on the first data set, a performance evaluation is performed on each second round model to obtain a corresponding first evaluation value; and based on the model parameters of the second round model with the highest first evaluation value, the model parameters of the first prediction model are solidified;

[0115] Specifically, it includes: step 31, using the GA algorithm to perform a first round of training on the first prediction model according to the first data set to obtain multiple first round models;

[0116] Specifically, step 3101: the total number of neurons in the input layer of the first prediction model n in Set to 19, the total number of neurons in the output layer n out Set to 1, and based on Get the total number of neurons n in multiple hidden layers hid The total number of neurons in the hidden layer.

[0117] Among them, F INT () is a rounding function, the value range of △n is [0,n0-1], the preset number n0 is an integer greater than 1; the total number of hidden layer neurons in the total number of hidden layer neurons n hidThe total number matches the preset number n0;

[0118] For example, the total number of neurons in the input layer is n in =19, total number of neurons in the output layer n out =1, the value range of △n is [0,10], n0=11, then we get the total number of neurons n in the 11 hidden layers. hid The total number of neurons in the hidden layer is n. hid They are 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24 respectively;

[0119] Step 3102: randomly extracting a plurality of first data records from the first data set based on a preset first extraction ratio to form a corresponding second data set;

[0120] wherein the first extraction ratio is a percentage less than 1; the total number of records in the first data set multiplied by the first extraction ratio matches the total number of records in the second data set;

[0121] Step 3103, based on the total number of neurons n in each hidden layer hid The shape of the first hidden layer of the first prediction model is set to obtain the corresponding first volume model; and the model parameter set of each first volume model is initialized; and the real vector conversion is performed based on the current model parameter set of each first volume model to obtain the corresponding first volume parameter vector; and a corresponding first volume error is set for each first volume model, and each first volume error is initialized to 1; and a round of iteration counter is initialized to 1;

[0122] Wherein, the total number of the first body models matches the preset number n0;

[0123] Step 3104, taking each first body model as the corresponding current individual model; and traversing all the first data records of the second data set; and taking the currently traversed first data record as the corresponding current data record during the traversal; and inputting the first training parameter vector of the current data record into the current individual model for prediction processing to obtain the corresponding first training prediction value; and forming a corresponding first prediction label pair from the first training prediction value and the first label prediction value of the current data record; and at the end of this traversal, bringing all the first prediction label pairs corresponding to the current individual model into the preset first error function for error calculation to obtain the corresponding first error value, and resetting the first body error corresponding to the current individual model to the corresponding first error value;

[0124] Wherein, the first error function is a mean square error function;

[0125] Step 3105, taking the inverse of the first body error corresponding to each first body model as the corresponding first body fitness; and summing up all the obtained first body fitnesses to obtain the corresponding first body fitness sum; and taking the ratio of each first body fitness to the first body fitness sum as the corresponding first body selection probability; and sorting all the first body models in descending order of the first body selection probability to obtain the corresponding first body model sequence;

[0126] Here, the smaller the first individual error is, the greater the corresponding first individual fitness is, and then the corresponding first individual selection probability is also greater, that is, the first individual model sequence is sorted in descending order of model performance;

[0127] Step 3106, identifying whether the preset number n0 is an odd number; if so, go to step 3107; if not, go to step 3108;

[0128] Step 3107, taking the first body model in the first body model sequence as the elite model of the time; and forming the second and third first body models of the first body model sequence into a corresponding first paired individual, and forming the fourth and fifth first body models into a corresponding first paired individual, and so on, until the n0-1th and n0th first body models form a corresponding first paired individual; and directly taking the first body parameter vector of the elite model of the time as the corresponding first body crossover parameter vector; and performing vector parameter crossover processing on the first body parameter vectors of the two first body models of each first paired individual based on a preset individual crossover algorithm and a preset first crossover probability to obtain two corresponding first body crossover parameter vectors; and then going to step 3109;

[0129] The individual crossover algorithm is a type of chromosome crossover algorithm in the GA algorithm, and the chromosome crossover algorithm includes at least a single-point crossover algorithm, a multi-point crossover algorithm, and an arithmetic crossover algorithm; the first crossover probability is a preset probability value, such as 0.4;

[0130] Step 3108, the 1st and 2nd first body models of the first body model sequence are combined into a corresponding first paired individual, the 3rd and 4th first body models are combined into a corresponding first paired individual, and so on, until the n0-1th and n0th first body models are combined into a corresponding first paired individual; and based on the individual crossover algorithm and the first crossover probability, the first body parameter vectors of the two first body models of each first paired individual are subjected to vector parameter crossover processing to obtain two corresponding first body crossover parameter vectors;

[0131] Step 3109, performing vector parameter mutation processing on the first body cross parameter vector of each first body model based on a preset individual mutation algorithm and a preset first mutation probability to obtain a corresponding first body mutation parameter vector;

[0132] The individual mutation algorithm is a type of chromosome mutation algorithm in the GA algorithm, and the chromosome mutation algorithm includes at least a basic position mutation algorithm, a uniform mutation algorithm, a non-uniform mutation algorithm, and a Gaussian approximate mutation algorithm; the first mutation probability is a preset probability value, such as 0.01;

[0133] Step 3110, resetting the model parameter set of the corresponding first body model based on each first body variation parameter vector;

[0134] Step 3111, identifying whether a round of iteration counter exceeds a preset round of iteration number threshold; if exceeded, go to step 3112; if not exceeded, add 1 to the round of iteration counter and return to step 3104;

[0135] Here, the threshold of the number of iterations is a preset positive integer;

[0136] Step 3112, solidify the model parameters of each first body model, and use each first body model after the parameters are solidified as a corresponding first round model;

[0137] Step 32, and use a preset model parameter optimizer to perform a second round of training on each first round model according to the first data set to obtain a corresponding second round model;

[0138] Wherein, the model parameter optimizer includes an ADAM optimizer;

[0139] Specifically, it includes: step 3201, initializing the second round iteration counter to 1; and taking the first round model currently trained as the corresponding current model;

[0140] Step 3202: randomly select a preset first number M of first data records from the first data set to form a corresponding first batch record set; and traverse all first data records in the first batch record set; and during the traversal, use the currently traversed first data record as the corresponding current data record; and input the first training parameter vector of the current data record into the current model for prediction processing to obtain the corresponding second training prediction value y i ; and use the first label prediction value of the current data record as the corresponding second label prediction value And the second training prediction value y i and the second label prediction value Form a corresponding second prediction label pair;

[0141] Wherein, the first number M is an integer greater than 0; 1≤batch record index i≤M;

[0142] Step 3203, bringing all the obtained second prediction label pairs into the preset first model loss function LOSS to calculate the corresponding first batch loss values;

[0143] Among them, the first model loss function LOSS is:

[0144]

[0145] λ is the preset first penalty factor, p is the preset first probability threshold; ε is the step function, Greater than or equal to y i The corresponding step function The function value of is 1. Less than y i The corresponding step function The function value of is 0;

[0146] Step 3204: predict the value of the second tag Greater than the second training prediction value y i The total number of the second predicted label pairs is counted to obtain the corresponding total number of first label pairs m; and the corresponding first batch probability is calculated based on the total number of first label pairs m and the first quantity M;

[0147] in,

[0148] For is the probability function of a probabilistic event;

[0149] Step 3205, identify the first batch loss value and the first batch probability; if the first batch loss value meets the preset first loss value convergence range and the first batch probability is greater than or equal to the first probability threshold p, go to step 3209; if the first batch loss value does not meet the first loss value convergence range or the first batch probability is less than the first probability threshold p, go to step 3206;

[0150] Here, the first loss value convergence range is a preset loss value range;

[0151] Step 3206, setting the learning rate of the model parameter optimizer;

[0152] Specifically, the second-round iteration counter is identified; if the second-round iteration counter is 1, the learning rate is set to the preset first learning rate; if the second-round iteration counter is greater than 1, whether the remainder of the second-round iteration counter divided by the preset first number of iterations is 1 is identified, and if so, the product of the current learning rate and the preset first attenuation factor is used as the new learning rate;

[0153] Here, the first learning rate is a preset learning rate; the first number of iterations is a preset positive integer; the first attenuation factor is a preset percentage less than 1; the model parameter optimizer of the embodiment of the present invention performs a decay process on the learning rate every time iterates the first number of iterations, and the processing method is new learning rate = old learning rate × first decay factor;

[0154] For example, it is known that the first learning rate is 0.01, the first number of iterations is 100, and the first attenuation factor is 0.5; then, when the second iteration counter = 1, the learning rate of the model parameter optimizer is 0.01; when the second iteration counter = 101, the remainder of the second iteration counter divided by the first number of iterations is 1, and the new learning rate of the model parameter optimizer = the old learning rate × the first attenuation factor = 0.01 × 0.5 = 0.005; when the second iteration counter = 201, the remainder of the second iteration counter divided by the first number of iterations is 1, and the new learning rate of the model parameter optimizer = the old learning rate × the first attenuation factor = 0.005 × 0.5 = 0.0025;

[0155] Step 3207, setting the optimization objective function O of the model parameter optimizer and its corresponding constraint conditions;

[0156] Among them, the optimization objective function O is:

[0157] The constraints for optimizing the objective function O are:

[0158] Step 3208, performing a round of optimization processing on the current model parameter set of the current model based on the model parameter optimizer;

[0159] Step 3209, identifying whether the second-round iteration counter exceeds the preset second-round iteration number threshold; if it exceeds, go to step 3210; if it does not exceed, add 1 to the second-round iteration counter and return to step 3202;

[0160] Here, the second-round iteration number threshold is a preset integer;

[0161] Step 3210, solidify the model parameters of the current model, and use the current model after the parameters are solidified as a corresponding second-round model;

[0162] Step 33, performing performance evaluation on each second-round model based on the first data set to obtain a corresponding first evaluation value;

[0163] Specifically, it includes: step 331, randomly extracting a plurality of first data records from the first data set based on a preset second extraction ratio to form a corresponding first evaluation data set; and counting the total number of records of the first evaluation data set to obtain a corresponding second number K;

[0164] The second extraction ratio is a percentage less than 1; the product of the total number of records of the first data set and the second extraction ratio matches the second quantity K;

[0165] Step 332, taking each second round model as the corresponding current model one by one;

[0166] Step 333, traverse all first data records of the first evaluation data set; and during the traversal, use the first data record currently traversed as the corresponding current data record; and input the first training parameter vector of the current data record into the current model for prediction processing to obtain the corresponding third training prediction value y j ; and use the first label prediction value of the current data record as the corresponding third label prediction value And the third training prediction value y j And the third label prediction value Form a corresponding third prediction label pair;

[0167] Among them, 1≤evaluation record index j≤K;

[0168] Step 334: Calculate the corresponding first relative error e based on each third predicted label j ; and for all the first relative errors e j Perform mean calculation to obtain the corresponding first relative error mean e aver ; and according to all the first relative errors e j and the first relative error mean e aver Calculate the corresponding first relative error standard deviation s;

[0169] in,

[0170] Step 335: according to the first relative error mean value e aver The first relative error standard deviation s is used to calculate the evaluation value to obtain the corresponding first evaluation value;

[0171] in, a1 and a2 are two preset weighting parameters;

[0172] Step 34: solidify the model parameters of the first prediction model based on the model parameters of the second round model with the highest first evaluation value.

[0173] Step 4: After the model parameters are solidified, batch prediction tasks are processed based on the first prediction model;

[0174] Specifically, it includes: step 41, receiving the first batch prediction list after completing the solidification of model parameters;

[0175] The first batch prediction list includes one or more first concrete component parameter groups; the first concrete component parameter group consists of 19 types of unnormalized original parameters; the 19 types of unnormalized original parameters include stirrup minor axis size, stirrup major axis size, stirrup cross-sectional area, stirrup cross-sectional inertia moment, stirrup shape code, stirrup spacing, stirrup reinforcement ratio, stirrup strength, number of stirrup legs, concrete strength, cross-sectional size, concrete cover thickness, component longitudinal size, component aspect ratio, component cross-sectional shape, longitudinal reinforcement strength, longitudinal reinforcement ratio, stirrup characteristic value and reinforcement characteristic value;

[0176] Step 42, traverse all first concrete component parameter groups; and during the traversal, take the currently traversed first concrete component parameter group as the corresponding current parameter group; and respectively normalize the 19 categories of unnormalized original parameters of the current parameter group to obtain a group of corresponding 19 categories of normalized parameters to form a corresponding first model input vector; and input the first model input vector into the first prediction model for prediction processing to obtain the corresponding first model prediction value; and perform denormalization processing on the first model prediction value to obtain the corresponding first limit strain prediction value; and form a corresponding first prediction record by the current parameter group and the first limit strain prediction value; and at the end of the traversal, form a corresponding first batch prediction report by all the obtained first prediction records and save it.

[0177] Figure 2 This is a module structure diagram of a processing device for predicting the ultimate strain characteristics of concrete provided in the second embodiment of the present invention. The device is a terminal device or server that implements the aforementioned method embodiment, and can also be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiment. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 2 As shown, the device includes: a model construction module 201, a data set construction module 202, a model training module 203 and a model application module 204.

[0178] The model building module 201 is used to build a concrete ultimate strain characteristic prediction model based on a BP network as a corresponding first prediction model.

[0179] The data set construction module 202 is used to construct a corresponding first data set by collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources.

[0180] The model training module 203 is used to use the GA algorithm to perform a first round of training on the first prediction model according to the first data set to obtain multiple first-round models; and use a preset model parameter optimizer to perform a second round of training on each first-round model according to the first data set to obtain a corresponding second-round model; and based on the first data set, perform performance evaluation on each second-round model to obtain a corresponding first evaluation value; and solidify the model parameters of the first prediction model based on the model parameters of the second-round model with the highest first evaluation value.

[0181] The model application module 204 is used to process batch prediction tasks based on the first prediction model after completing the solidification of the model parameters.

[0182] An embodiment of the present invention provides a processing device for predicting the ultimate strain characteristics of concrete, which can execute the method steps in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here.

[0183] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the model building module can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the above-mentioned module is determined. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0184] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more digital signal processors (DSP), or one or more field programmable gate arrays (FPGA). For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0185] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the above method embodiments are generated. The above computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The above-mentioned computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the above-mentioned computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, Bluetooth, microwave, etc.) methods. The above-mentioned computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media integrated. The above-mentioned available medium can be a magnetic medium (such as a floppy disk, a hard disk, a tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0186] Figure 3 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. The electronic device may be a terminal device or a server for implementing the method of the aforementioned embodiment, or may be a terminal device or a server for implementing the method of the aforementioned embodiment connected to the aforementioned terminal device or server. Figure 3As shown, the electronic device may include: a processor 301 (such as a CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transceiver 303. Various instructions may be stored in the memory 302 to complete various processing functions and implement the processing steps described in the aforementioned embodiment method. Preferably, the electronic device involved in the embodiment of the present invention also includes: a power supply 304, a system bus 305 and a communication port 306. The system bus 305 is used to realize the communication connection between components. The above-mentioned communication port 306 is used for connecting and communicating between the electronic device and other peripherals.

[0187] exist Figure 3 The system bus 305 mentioned in the figure can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The communication interface is used to realize the communication between the database access device and other devices (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM), and may also include non-volatile memory (Non-Volatile Memory), such as at least one disk storage.

[0188] The above-mentioned processor can be a general-purpose processor, including a central processing unit CPU, a network processor (Network Processor, NP), a graphics processing unit (Graphics Processing Unit, GPU), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0189] It should be noted that an embodiment of the present invention further provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the method and processing process provided in the above embodiments.

[0190] An embodiment of the present invention further provides a chip for executing instructions, wherein the chip is used to execute the processing steps described in the aforementioned method embodiment.

[0191] The embodiment of the present invention provides a processing method, device, electronic device and computer-readable storage medium for predicting the ultimate strain characteristics of concrete. As can be seen from the above content, the embodiment of the present invention first constructs a prediction model of the ultimate strain characteristics of concrete by using a BP network, and constructs a data set by collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources; then uses a GA algorithm to perform a first round of training on the prediction model according to the data set, and uses a model parameter optimizer to perform a second round of training on each first round model according to the data set, and performs a performance evaluation on each second round model based on the data set, and solidifies the model parameters of the first prediction model based on the model parameters of the second round model with the highest evaluation value; finally, after completing the model parameter solidification, the batch prediction task is processed based on the first prediction model. The prediction model provided by the embodiment of the present invention is used to process the data acquisition task of the ultimate strain value, which not only improves the processing efficiency of single tasks, but also improves the processing efficiency of batch tasks.

[0192] The professionals should further realize that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0193] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0194] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting the ultimate strain characteristics of concrete, characterized in that: The method comprises: Based on BP network, a concrete ultimate strain characteristic prediction model is constructed as the corresponding first prediction model; Constructing a corresponding first data set by collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources; Using the GA algorithm to perform a first round of training on the first prediction model according to the first data set to obtain multiple first round models; and using a preset model parameter optimizer to perform a second round of training on each of the first round models according to the first data set to obtain a corresponding second round model; and based on the first data set, a performance evaluation is performed on each of the second round models to obtain a corresponding first evaluation value; and based on the model parameters of the second round model with the highest first evaluation value, the model parameters of the first prediction model are solidified; After the model parameters are solidified, batch prediction tasks are processed based on the first prediction model.

2. The method for predicting the ultimate strain characteristics of concrete according to claim 1, characterized in that: The first prediction model is used to predict the ultimate strain characteristics of the concrete component according to the normalized parameter vector x input by the model and output the corresponding normalized prediction value y; The vector length of the normalized parameter vector x is 19, and it is composed of 19 types of normalized parameters; the 19 types of normalized parameters include normalized stirrup minor axis size, normalized stirrup major axis size, normalized stirrup cross-sectional area, normalized stirrup cross-sectional inertia moment, normalized stirrup shape code, normalized stirrup spacing, normalized stirrup reinforcement ratio, normalized stirrup strength, normalized stirrup limb number, normalized concrete strength, normalized cross-sectional size, normalized concrete cover thickness, normalized component longitudinal size, normalized component aspect ratio, normalized component cross-sectional shape, normalized longitudinal reinforcement strength, normalized longitudinal reinforcement ratio, normalized stirrup eigenvalue and normalized reinforcement eigenvalue; The normalized predicted value y is a normalized limit strain value; The first prediction model includes a first input layer, a first hidden layer and a first output layer; the neuron transfer functions used in the first hidden layer and the first output layer in the first prediction model are both sigmoid activation functions; The first input layer is composed of nineteen parallel input layer neurons; the input layer neurons correspond one to one to the normalized parameters in the normalized parameter vector x; Each of the input layer neurons of the first input layer is used to receive the corresponding normalized parameter in the normalized parameter vector x as a corresponding neuron input value; The first hidden layer is composed of a plurality of parallel hidden layer neurons, and each of the hidden layer neurons is fully connected to all the input layer neurons; Each of the hidden layer neurons of the first hidden layer is used to perform full connection calculation on the neuron input values ​​of all the input layer neurons connected to itself to generate a corresponding first fully connected vector, and perform activation operation on the first fully connected vector based on a sigmoid activation function to obtain a corresponding first activation vector; The first output layer has only one output layer neuron, and the output layer neuron is fully connected to all the hidden layer neurons; The output layer neurons of the first output layer are used to perform a full connection calculation on the first activation vectors of all the hidden layer neurons connected to themselves to generate a corresponding second fully connected vector, and perform an activation operation on the second fully connected vector based on a sigmoid activation function and output the operation result as the corresponding normalized prediction value y.

3. The method for predicting the ultimate strain characteristics of concrete according to claim 2, characterized in that: The construction industry literature data sources include journal articles, books, research reports, patents and standard specifications of the construction industry; The first data set includes a plurality of first data records; the first data record includes a first training parameter vector and a first label prediction value; the first training parameter vector has the same data structure as the normalized parameter vector x and is also composed of nineteen normalized parameters; the first label prediction value has the same data meaning as the normalized prediction value y and is also a normalized limit strain value; The model parameter optimizer includes an ADAM optimizer.

4. The method for predicting the ultimate strain characteristics of concrete according to claim 3, characterized in that: The step of collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources to construct a corresponding first data set specifically includes: Obtain a plurality of first concrete component compression test reports from a plurality of concrete manufacturer data sources; and obtain a plurality of first industry documents from a plurality of construction industry document data sources; And all the first concrete component compression test reports and all the first industry documents are retrieved for concrete component design parameters and corresponding component experimental measurement limit strain values ​​to obtain corresponding multiple first search records; the first search record includes a first component type, a first component design parameter group and a first experimental measurement limit strain value; the first component type includes first, second, third and fourth types; the concrete component type corresponding to the first type is a concrete component with both stirrups and longitudinal reinforcement; the concrete component type corresponding to the second type is a concrete component with only stirrups but no longitudinal reinforcement; the concrete component type corresponding to the third type is a concrete component with only longitudinal reinforcement but no stirrups; the concrete component type corresponding to the fourth type is a concrete component with neither stirrups nor longitudinal reinforcement; the first component type is a record of the first search record of the first type The number of records of the first search record of the first component type being the second type is the largest, the number of records of the first search record of the first component type being the third type is the third, and the number of records of the first search record of the first component type being the fourth type is the smallest; the first component design parameter group is composed of 19 types of original parameters that have not been normalized; the 19 types of original parameters that have not been normalized include stirrup minor axis size, stirrup major axis size, stirrup cross-sectional area, stirrup cross-sectional moment of inertia, stirrup shape code, stirrup spacing, stirrup reinforcement ratio, stirrup strength, number of stirrup limbs, concrete strength, cross-sectional size, concrete cover thickness, component longitudinal size, component aspect ratio, component cross-sectional shape, longitudinal reinforcement strength, longitudinal reinforcement ratio, stirrup characteristic value and reinforcement characteristic value; the first experimental measurement limit strain value is an unnormalized limit strain value; Traverse all the first retrieval records; and during the traversal, take the first retrieval record currently traversed as the corresponding current retrieval record; and normalize the nineteen categories of unnormalized original parameters of the current retrieval record respectively to obtain a group of corresponding nineteen categories of normalized parameters to form a corresponding first training parameter vector; and normalize the first experimental measurement limit strain value of the current retrieval record to obtain the corresponding first label prediction value; and form a corresponding first data record with the first training parameter vector and the first label prediction value corresponding to the current retrieval record; and at the end of the traversal, form the corresponding first data set with all the first data records obtained.

5. The method for predicting the ultimate strain characteristics of concrete according to claim 3, characterized in that: The step of performing a first round of training on the first prediction model according to the first data set using the GA algorithm to obtain a plurality of first round models specifically includes: Step 501: The total number of neurons in the input layer of the first prediction model n in Set to 19, the total number of neurons in the output layer n out Set to 1, and based on Get the total number of neurons n in multiple hidden layers hid The total number of neurons in the hidden layer. Among them, F INT () is a rounding function, the value range of △n is [0, n0-1], and the preset number n0 is an integer greater than 1; the total number of hidden layer neurons in the total number of hidden layer neurons n hid The total number of matches the preset number n0; Step 502: randomly extracting a plurality of first data records from the first data set based on a preset first extraction ratio to form a corresponding second data set; wherein the first extraction ratio is a percentage less than 1; the total number of records in the first data set matches the product of the first extraction ratio and the total number of records in the second data set; Step 503, based on the total number of neurons n in each hidden layer hid The shape of the first hidden layer of the first prediction model is set to obtain a corresponding first body model; and the model parameter set of each first body model is initialized; and a real number vector is converted based on the current model parameter set of each first body model to obtain a corresponding first body parameter vector; and a corresponding first body error is set for each first body model, and each first body error is initialized to 1; and a round of iteration counter is initialized to 1; Wherein, the total number of the first individual models matches the preset number n0; Step 504, taking each of the first individual models as the corresponding current individual model; and traversing all the first data records of the second data set; and taking the first data record currently traversed as the corresponding current data record during the traversal; and inputting the first training parameter vector of the current data record into the current individual model for prediction processing to obtain the corresponding first training prediction value; and forming a corresponding first prediction label pair from the first training prediction value and the first label prediction value of the current data record; and at the end of this traversal, bringing all the first prediction label pairs corresponding to the current individual model into a preset first error function for error calculation to obtain the corresponding first error value, and resetting the first individual error corresponding to the current individual model to the corresponding first error value; Wherein, the first error function is a mean square error function; Step 505: taking the inverse of the first individual error corresponding to each first individual model as the corresponding first individual fitness; summing up all the first individual fitnesses obtained to obtain the corresponding first fitness sum; taking the ratio of each first individual fitness to the first fitness sum as the corresponding first individual selection probability; and sorting all the first individual models in descending order of the first individual selection probability to obtain the corresponding first individual model sequence; Step 506, identifying whether the preset number n0 is an odd number; if so, go to step 507; if not, go to step 508; Step 507, taking the first first individual model in the first individual model sequence as the elite model of the time; and forming the second and third first individual models of the first individual model sequence into a corresponding first paired individual, and forming the fourth and fifth first individual models into a corresponding first paired individual, and so on, until the n0-1th and n0th first individual models form a corresponding first paired individual; and directly taking the first individual parameter vector of the elite model of the time as the corresponding first individual crossover parameter vector; and performing vector parameter crossover processing on the first individual parameter vectors of the two first individual models of each first paired individual based on a preset individual crossover algorithm and a preset first crossover probability to obtain two corresponding first individual crossover parameter vectors; and going to step 509; The individual crossover algorithm is a type of chromosome crossover algorithm in the GA algorithm, and the chromosome crossover algorithm includes at least a single-point crossover algorithm, a multi-point crossover algorithm, and an arithmetic crossover algorithm; Step 508: The first and second first individual models of the first individual model sequence are combined into a corresponding first paired individual, the third and fourth first individual models are combined into a corresponding first paired individual, and so on, until the n0-1th and n0th first individual models are combined into a corresponding first paired individual; and based on the individual crossover algorithm and the first crossover probability, the first individual parameter vectors of the two first individual models of each first paired individual are subjected to vector parameter crossover processing to obtain two corresponding first individual crossover parameter vectors; Step 509, performing vector parameter mutation processing on the first individual crossover parameter vector of each of the first individual models based on a preset individual mutation algorithm and a preset first mutation probability to obtain a corresponding first individual mutation parameter vector; The individual mutation algorithm is a type of chromosome mutation algorithm in the GA algorithm, and the chromosome mutation algorithm at least includes a basic position mutation algorithm, a uniform mutation algorithm, a non-uniform mutation algorithm, and a Gaussian approximate mutation algorithm; Step 510, resetting the model parameter set of the corresponding first individual model based on each of the first individual variation parameter vectors; Step 511, identifying whether the one-round iteration counter exceeds a preset one-round iteration number threshold; if it exceeds, go to step 512; if it does not exceed, add 1 to the one-round iteration counter and return to step 504; Step 512, solidifying the model parameters of each of the first individual models, and using each of the first individual models after the solidification as a corresponding first round model.

6. The method for predicting the ultimate strain characteristics of concrete according to claim 3, characterized in that: The using a preset model parameter optimizer to perform a second round of training on each of the first round models according to the first data set to obtain a corresponding second round model specifically includes: Step 601, initializing the second round iteration counter to 1; and using the first round model currently being trained as the corresponding current model; Step 602: randomly select a preset first number M of the first data records from the first data set to form a corresponding first batch record set; and traverse all the first data records in the first batch record set; and during the traversal, use the first data record currently traversed as the corresponding current data record; and input the first training parameter vector of the current data record into the current model for prediction processing to obtain the corresponding second training prediction value y i ; and use the first tag prediction value of the current data record as the corresponding second tag prediction value And the second training prediction value y i and the second label prediction value Form a corresponding second prediction label pair; Wherein, the first number M is an integer greater than 0; 1≤batch record index i≤M; Step 603, bringing all the obtained second prediction label pairs into the preset first model loss function LOSS to calculate the corresponding first batch loss values; Among them, the first model loss function LOSS is: λ is the preset first penalty factor, p is the preset first probability threshold; ε is the step function, y * i Greater than or equal to y i The corresponding step function The function value of is 1. Less than y i The step function corresponding to The function value of is 0; Step 604: predict the value of the second tag is greater than the second training prediction value y i The total number of the second predicted label pairs is counted to obtain the corresponding total number of first label pairs m; and the corresponding first batch probability is calculated based on the total number of the first label pairs m and the first quantity M; in, For is the probability function of a probabilistic event; Step 605, identifying the first batch loss value and the first batch probability; if the first batch loss value meets the preset first loss value convergence range and the first batch probability is greater than or equal to the first probability threshold p, go to step 609; if the first batch loss value does not meet the first loss value convergence range or the first batch probability is less than the first probability threshold p, go to step 606; Step 606, setting the learning rate of the model parameter optimizer, specifically: identifying the second-round iteration counter; if the second-round iteration counter is 1, setting the learning rate to a preset first learning rate; if the second-round iteration counter is greater than 1, identifying whether the remainder of the second-round iteration counter divided by the preset first number of iterations is 1, and if so, taking the product of the current learning rate and the preset first attenuation factor as the new learning rate; Step 607, setting the optimization objective function O of the model parameter optimizer and its corresponding constraint conditions; Wherein, the optimization objective function O is: The constraints of the optimization objective function O are: Step 608, performing a round of optimization processing on the current model parameter set of the current model based on the model parameter optimizer; Step 609, identifying whether the second-round iteration counter exceeds a preset second-round iteration number threshold; if so, going to step 610; if not, adding 1 to the second-round iteration counter and returning to step 602; Step 610, solidify the model parameters of the current model, and use the current model after the parameters are solidified as a corresponding second-round model.

7. The method for predicting the ultimate strain characteristics of concrete according to claim 3, characterized in that: The performing performance evaluation on each of the second-round models based on the first data set to obtain a corresponding first evaluation value specifically includes: Step 71: randomly extract multiple first data records from the first data set based on a preset second extraction ratio to form a corresponding first evaluation data set; and count the total number of records in the first evaluation data set to obtain a corresponding second number K; wherein the second extraction ratio is a percentage less than 1; the total number of records of the first data set multiplied by the second extraction ratio and the second number K match; Step 72, taking each of the second round models as the corresponding current model one by one; Step 73, traverse all the first data records of the first evaluation data set; and during the traversal, use the first data record currently traversed as the corresponding current data record; and input the first training parameter vector of the current data record into the current model for prediction processing to obtain the corresponding third training prediction value y j ; and use the first tag prediction value of the current data record as the corresponding third tag prediction value And the third training prediction value y j and the third label prediction value Form a corresponding third prediction label pair; Among them, 1≤evaluation record index j≤K; Step 74: Calculate the corresponding first relative error e based on each of the third predicted labels. j ; and for all the first relative errors e obtained j Perform mean calculation to obtain the corresponding first relative error mean e aver ; and according to all the first relative errors e j and the first relative error mean e aver Calculate the corresponding first relative error standard deviation s; in, Step 75: according to the first relative error mean value e aver and the first relative error standard deviation s to perform evaluation value calculation to obtain the corresponding first evaluation value; 8. The method for predicting the ultimate strain characteristics of concrete according to claim 2, characterized in that: After the model parameters are solidified, processing batch prediction tasks based on the first prediction model specifically includes: After the model parameters are solidified, a first batch prediction list is received; the first batch prediction list includes one or more first concrete component parameter groups; the first concrete component parameter group consists of 19 types of original parameters that have not been normalized; the 19 types of original parameters that have not been normalized include stirrup minor axis size, stirrup major axis size, stirrup cross-sectional area, stirrup cross-sectional moment of inertia, stirrup shape code, stirrup spacing, stirrup reinforcement ratio, stirrup strength, number of stirrup limbs, concrete strength, cross-sectional size, concrete cover thickness, component longitudinal size, component aspect ratio, component cross-sectional shape, longitudinal reinforcement strength, longitudinal reinforcement ratio, stirrup characteristic value and reinforcement characteristic value; All the first concrete component parameter groups are traversed; and during the traversal, the first concrete component parameter group currently traversed is used as the corresponding current parameter group; and the 19 categories of unnormalized original parameters of the current parameter group are respectively normalized to obtain a group of corresponding 19 categories of normalized parameters to form a corresponding first model input vector; and the first model input vector is input into the first prediction model for prediction processing to obtain the corresponding first model prediction value; and the first model prediction value is denormalized to obtain the corresponding first limit strain prediction value; and the current parameter group and the first limit strain prediction value form a corresponding first prediction record; and at the end of the traversal, all the first prediction records obtained form a corresponding first batch prediction report and save it.

9. A device for executing the processing method for predicting the ultimate strain characteristics of concrete according to any one of claims 1 to 8, characterized in that: The device comprises: a model construction module, a data set construction module, a model training module and a model application module; The model building module is used to build a concrete ultimate strain characteristic prediction model based on a BP network as a corresponding first prediction model; The data set construction module is used to construct a corresponding first data set by collecting data from multiple concrete manufacturer data sources and multiple construction industry literature data sources; The model training module is used to use the GA algorithm to perform a first round of training on the first prediction model according to the first data set to obtain multiple first round models; and use a preset model parameter optimizer to perform a second round of training on each of the first round models according to the first data set to obtain a corresponding second round model; and based on the first data set, the performance of each of the second round models is evaluated to obtain a corresponding first evaluation value; and based on the model parameters of the second round model with the highest first evaluation value, the model parameters of the first prediction model are solidified; The model application module is used to process batch prediction tasks based on the first prediction model after completing the solidification of model parameters.

10. An electronic device, characterized in that: include: memory, processors, and transceivers; The processor is used to couple with the memory, read and execute instructions in the memory, so as to implement the method according to any one of claims 1 to 8; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a computer, the computer is enabled to execute the method according to any one of claims 1 to 8.

Images