Long continuous building life prediction method and system and storage medium

Abstract

The invention discloses a long continuous building life prediction method and system and a storage medium. The invention relates to the technical field of constructional engineering. Constructing an initial durability index matrix A, and taking the mass a, the compressive strength b and the splitting tensile strength c as columns of the matrix; performing standardization processing on each index in the matrix A to form an initial evaluation value xij, and eliminating dimensional difference; comprising an input layer, a bidirectional Bi-GRU layer, a full connection layer and an output layer which are connected in sequence. Wherein the input layer receives time sequence data alpha (t) of a comprehensive evaluation factor alpha; by fusing the entropy weight method and the bidirectional Bi-GRU recurrent neural network model, the technical scheme of the invention can more accurately capture the durability change rule of the long continuous building under the salt invasion-dry-wet cycle coupling effect, thereby improving the accuracy and reliability of life prediction. And engineers and decision makers are helped to more accurately evaluate the service life condition of the building and formulate more reasonable maintenance and management strategies.

Description

Technical Field

[0001] The present invention relates to the field of building engineering technology, and specifically to the use of an entropy weight method combined with a neural network technology to evaluate and predict the life of a building. The target direction is long continuous buildings affected by "salt invasion-dry-wet cycle coupling", and in particular to a life prediction method, system and storage medium for long continuous buildings. Background Art

[0002] In the northwest of my country, sulfate environment is more common due to specific geographical environment and climatic conditions. This environment poses a significant threat of salt erosion to long continuous buildings, such as large infrastructure such as bridges, due to their wide and long contact area. Salt erosion not only affects the appearance of the building, but also leads to the gradual deterioration of its structural performance, which in turn causes huge economic losses, waste of resources and potential safety issues. Therefore, the durability evaluation and prediction of such buildings in salt erosion environment has become a technical problem that needs to be solved urgently.

[0003] In order to meet this challenge, current research and practice have begun to explore the application of GM (Grey Model) neural network to estimate the life of long continuous buildings affected by salt corrosion, such as the following existing technologies:

[0004] [1]BI Y,CHEN C,HUANG X,et al.Discrimination method of biomassslagging tendency based on particle swarm optimization deep neural network(DNN)[J].Energy,2022:125368.

[0005] [2]ANJALI R,VENKATESAN G.Optimization of mechanical properties and composition of M-sand and pet particle added concrete using hybrid deepneural network-horse herd optimization algorithm[J].Constr Build Mater,2022,347:128334.

[0006] [3]WEN Z,ZHOU R,SU H.MR and stacked GRUs neural network combinedmodel and its application for deformation prediction of concrete dam[J].Expert Systems Appl,2022,201:117272.

[0007] The above traditional technology uses GM neural network as a data processing and prediction tool. When the information is incomplete or the data is limited, it is applied to the prediction of building durability in salt corrosion environment by mining the intrinsic relationship between the data.

[0008] However, traditional technical forms often adopt the method of simply superimposing various influencing indicators when dealing with this problem. This method ignores the reality that after concrete is subjected to the "salt invasion-dry-wet cycle coupling" in the actual environment, its multiple performance indicators will undergo complex changes. Therefore, in order to comprehensively and scientifically evaluate the durability of concrete, it is necessary to comprehensively consider the comprehensive impact of multiple indicators on it. But it is worth noting that the degree of influence of these indicators on the durability of concrete is not the same. Some indicators play a leading role, while others are only secondary factors.

[0009] This is equivalent to simply superimposing various indicators for evaluation, which not only fails to accurately reflect the actual durability of concrete in complex environments, but also leads to deviations in the evaluation results.

[0010] To this end, the present invention proposes a life prediction method, system and storage medium for long continuous buildings. Summary of the invention

[0011] In view of this, the present invention hopes to provide a life prediction method, system and storage medium for long continuous buildings to solve or alleviate the technical problems existing in the prior art, namely, how to more intelligently assign weights to the "salt invasion-dry-wet cycle coupling" effect, and use a neural network model to predict the life of long continuous buildings, and at least provide a beneficial option for this; the technical solution of the present invention is implemented as follows:

[0012] First, the life prediction method of long continuous buildings:

[0013] 1. Overview:

[0014] The present invention aims to integrate the entropy weight method with the bidirectional Bi-GRU recurrent neural network model to realize the intelligent prediction of the life of long continuous buildings under the coupling of salt invasion and dry-wet cycles. Specifically, firstly, the initial durability index matrix is ​​constructed and standardized to eliminate the dimensional difference and form the initial data that can be used for evaluation. Then, the entropy weight method is applied to objectively determine the weight of each durability index, so as to extract the comprehensive evaluation factor, which reflects the durability of concrete. Finally, the comprehensive evaluation factor is used as input and introduced into the bidirectional Bi-GRU recurrent neural network model to consider the time series information and perform life prediction. By training and optimizing the model, the life prediction sequence of the building is finally output, providing a scientific basis and technical support for engineering practice. This scheme combines the advantages of objective weighting methods and advanced neural network models to improve the accuracy and reliability of life prediction.

[0015] (II) Technical solution:

[0016] In order to achieve the above technical objectives, after collecting experimental data of long continuous buildings (such as bridges) under the coupled effects of salt invasion and dry-wet cycles, including the mass a, compressive strength b and splitting tensile strength c of concrete, the present invention selects the following implementation scheme:

[0017] 2.1 Step S1, data collection and preprocessing:

[0018] Construct the initial durability index matrix A, and use the mass a, compressive strength b and splitting tensile strength c as the columns of the matrix. Standardize each index in the initial durability index matrix A to form the initial evaluation value x ij , eliminating dimensional differences.

[0019] 2.1.1 Step S100, constructing an initial durability index matrix A:

[0020]

[0021] Among them, a i represents the quality of the i-th sample, b i represents the compressive strength of the i-th sample, c i represents the splitting tensile strength of the i-th sample, and n is the number of samples.

[0022] 2.1.2 Step S101, perform Min-Max normalization processing:

[0023] Perform Min-Max standardization on each indicator in the initial durability indicator matrix A to form the initial evaluation value x ij ,Right now:

[0024] Among them, A ijrepresents the element in the i-th row and j-th column of matrix A, min(A j ) and max(A j ) represent the minimum and maximum values ​​of the j-th column in the initial durability index matrix A, respectively.

[0025] After standardization, the initial durability index matrix A is transformed into the initial evaluation value matrix X:

[0026]

[0027] 2.2 Step S2, execute entropy weight method:

[0028] For each indicator in the initial durability indicator matrix A, calculate the proportion of each sample value to the total value of the indicator to form a proportion matrix P, and calculate the information entropy e of each indicator based on the proportion matrix P j And import it into the weight matrix W, and calculate the weight w of each indicator based on the constraints of the weight matrix W j And extract the comprehensive evaluation factor α from the weight matrix W (α is the sum of the standardized value of each indicator multiplied by its weight).

[0029] 2.2.1 Step S200, calculate the weight matrix P:

[0030] For each indicator in the initial durability indicator matrix A, calculate the proportion of each sample value to the total value of the indicator to form the proportion matrix P:

[0031]

[0032] in, It indicates the proportion of the i-th sample value under the j-th indicator to the total value of the indicator.

[0033] 2.2.2 Step S201, calculate information entropy e j And form the weight matrix W:

[0034] For each indicator in the weight matrix P, calculate its information entropy e j :

[0035] Among them, j = 1, 2, 3 correspond to the three indicators of mass, compressive strength and splitting tensile strength respectively;

[0036] The information entropy e j Import it into the weight matrix W, and calculate the weight of each indicator based on the constraints of the weight matrix W (the sum of the weights of each indicator is 1):

[0037] 2.2.3 Step S202, calculate the comprehensive evaluation factor α:

[0038] According to the weight matrix W and the initial evaluation value matrix X (obtained in step S101), the comprehensive evaluation factor α is calculated:

[0039] Among them, x ij Represents the initial evaluation value of the i-th sample under the j-th indicator.

[0040] 2.3 Step S3, activate the bidirectional Bi-GRU recurrent neural network model:

[0041] It includes an input layer, a bidirectional Bi-GRU layer, a fully connected layer, and an output layer connected in sequence; the input layer receives the time series data α(t) of the comprehensive evaluation factor α. The bidirectional Bi-GRU layer is used to capture the positive and negative dependencies in the time series data. The fully connected layer maps the output of the Bi-GRU layer to the life prediction value s. The output layer outputs the life prediction sequence S of the building.

[0042] In the time series data α(t), t=1,2,...,T represents the index of the time series, and T is the total time step.

[0043] 2.3.1 Step S300, input layer:

[0044] Receive the time series data α(t) and use it as the input of the bidirectional Bi-GRU layer.

[0045] 2.3.2 Step S301, Bidirectional Bi-GRU layer:

[0046] It includes forward GRU and backward GRU, which capture the positive and reverse dependencies in time series data respectively.

[0047] (1) The hidden state of the forward GRU is OR{h t};

[0048] (2) The hidden state of the backward GRU is OL{h t};

[0049] (3) The output h of the bidirectional Bi-GRU layer t For: h t =[OR{h t},OL{h t}]; where [·,·] represents the concatenation operation of vectors.

[0050] The specific execution process is as follows:

[0051] S3010, calculate the hidden state of the forward GRU

[0052] Among them, GRU frepresents the function of the forward GRU, α(t) is the input of the current time step (i.e., the time series data of the comprehensive evaluation factor α), is the hidden state at the previous time step.

[0053] S3011, calculate the hidden state of the backward GRU

[0054] Among them, GRU b represents the function of the backward GRU, α(t) is the input of the current time step, is the hidden state at the next time step (note that in backward propagation, the time steps are reversed).

[0055] S3012, calculate the output h of the bidirectional Bi-GRU layer t :

[0056] 2.3.3 Step S302, fully connected layer:

[0057] The output h of the bidirectional Bi-GRU layer t Mapped to life expectancy prediction value s t :s t =W f ·h t +b f ;

[0058] Among them, W f and b f are the weight matrix and bias vector of the fully connected layer respectively.

[0059] 2.3.4 Step S303, output layer:

[0060] Output the building life prediction sequence S: S = {s1, s2, ..., s T}.

[0061] 2.4 Training method of bidirectional Bi-GRU recurrent neural network model:

[0062] Although step S3 uses a bidirectional Bi-GRU recurrent neural network model, the model still needs to be trained before deployment. The training process includes:

[0063] P1. Collect the historical concrete mass a', historical compressive strength b' and historical splitting tensile strength c' of the same target building to form a data set D, and divide it into training set D train and the test set D test ;

[0064] P2, using training set D train Train the model;

[0065] P3, optimize model parameters through mean square error MSE loss function and Adam optimizer;

[0066] P4. Continuously adjust the model parameters through iterative training until the model is in the test set D test Achieve satisfactory performance.

[0067] (III) Mechanism for solving technical problems:

[0068] 3.1 Entropy weight method realizes intelligent weight assignment:

[0069] The information entropy of each durability index is calculated using the principle of entropy weight method. Information entropy is a measure of information uncertainty. The smaller the information entropy, the greater the discreteness of the index value, the more information it provides, and the greater the role it plays in the comprehensive evaluation. Therefore, it should be given a larger weight. Conversely, the larger the information entropy, the smaller the weight it is given. Through this method, the weight of each durability index in the comprehensive evaluation can be objectively determined, avoiding the interference of subjective factors.

[0070] According to the calculated weights, the standardized values ​​of each durability index are weighted and summed to extract the comprehensive evaluation factor. This factor comprehensively reflects the durability of concrete under the coupling of salt invasion and dry-wet cycles, and provides key input data for the subsequent prediction of the neural network model.

[0071] 3.2 Neural network model realizes life prediction:

[0072] A bidirectional Bi-GRU recurrent neural network model is used for life prediction. This model combines the advantages of bidirectional structure and GRU (gated recurrent unit), can capture the positive and negative dependencies in time series data at the same time, and is suitable for processing data with time series characteristics. The model structure includes an input layer, a bidirectional Bi-GRU layer, a fully connected layer, and an output layer. After training, the comprehensive evaluation factor is input as input data into the trained neural network model, and the model will output the corresponding building life prediction sequence. This sequence reflects the expected life of the building at different time points, providing an important reference for engineering practice.

[0073] Second, the life prediction system of long continuous buildings:

[0074] like Figure 2 As shown, the system includes a processor and a memory connected to the processor, wherein the memory stores program instructions, and when the program instructions are executed by the processor, the processor executes the life prediction method as described above. The processor is connected to,

[0075] (1) The data preprocessing module is responsible for collecting the experimental data of long continuous buildings: including the mass a, compressive strength b and splitting tensile strength c of concrete. The collected data are cleaned and denoised to ensure the accuracy and completeness of the data. At the same time, it is also responsible for constructing the initial durability index matrix and standardizing each index in the matrix to eliminate dimensional differences and form initial evaluation values ​​that can be used for subsequent analysis.

[0076] (2) Calculate the information entropy of each durability index and measure the uncertainty of information. Weight assignment module of entropy weight method: According to the size of information entropy, objectively determine the weight of each durability index in the comprehensive evaluation. Perform weighted summation on the standardized values ​​of each durability index and extract the comprehensive evaluation factor to reflect the durability of concrete.

[0077] (3) Implement the neural network model module of the bidirectional Bi-GRU recurrent neural network: including input layer, bidirectional Bi-GRU layer, fully connected layer and output layer.

[0078] At the same time, the model is trained using historical data, and the model parameters are adjusted through forward propagation and back propagation algorithms to minimize the prediction error. The mean square error (MSE) is used as the loss function to evaluate the model performance, and optimization algorithms such as the Adam optimizer are used to accelerate the training process.

[0079] (4) Input the comprehensive evaluation factors as input data into the trained neural network model to perform life prediction and result output module:

[0080] The model outputs the corresponding building life prediction sequence, which reflects the expected life of the building at different time points.

[0081] Compared with the prior art, the present invention has the following beneficial effects:

[0082] 1. Improved prediction accuracy and reliability: By integrating the entropy weight method with the bidirectional Bi-GRU recurrent neural network model, the technical solution of the present invention can more accurately capture the durability change law of long continuous buildings under the coupling of salt invasion and dry-wet cycles, thereby improving the accuracy and reliability of life prediction. It helps engineers and decision makers to more accurately evaluate the life status of buildings and formulate more reasonable maintenance and management strategies.

[0083] 2. Intelligent assignment of weights is realized: Traditional methods often rely on expert experience or subjective judgment to determine the weights of various durability indicators, while the technical solution of the present invention adopts the entropy weight method, which can objectively determine the weights according to the discrete degree and amount of information of the data, avoiding the interference of subjective factors and making the weight assignment more scientific and reasonable.

[0084] 3. Enhanced generalization ability of the model: By introducing the bidirectional Bi-GRU recurrent neural network model, the technical solution of the present invention can capture the positive and negative dependencies in the time series data and better process the data with time series characteristics. This makes the model more generalizable and adaptable when facing life prediction under different buildings and different environmental conditions.

[0085] 4. Provides a comprehensive evaluation basis: The technical solution of the present invention not only considers a single durability index, but also integrates the information of multiple indexes to extract a comprehensive evaluation factor as the input of the neural network model. This provides a more comprehensive evaluation basis, making the prediction results more comprehensive and accurate.

[0086] 5. Supports intelligent decision-making: The technical solution of the present invention provides a scientific basis and technical support for engineering practice, enabling engineers and decision makers to formulate more scientific and reasonable maintenance, reinforcement or demolition strategies based on accurate prediction results, thereby extending the service life of the building, reducing maintenance costs, and improving the overall economic and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0088] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0089] Figure 2 It is a schematic diagram of the system composition of the present invention;

[0090] Figure 3 It is a visualization schematic diagram of the life prediction sequence S of the application example of the present invention; wherein the horizontal axis is the life (in months), and the vertical axis is the value of the comprehensive evaluation factor α;

[0091] Figure 4 It is a scatter point comparison schematic diagram of the test examples of the present invention; wherein the horizontal axis is the life span (in months), and the vertical axis is the comprehensive evaluation factor α value of the experimental group;

[0092] Figure 5 This is a schematic diagram of the Python execution program of step S2 of the first embodiment of the present invention;

[0093] Figure 6 A schematic diagram of a Python architecture program of a bidirectional Bi-GRU recurrent neural network model in step S3 of the first embodiment of the present invention;

[0094] Figure 7 This is a schematic diagram of the model training Python architecture program of the second embodiment of the present invention. DETAILED DESCRIPTION

[0095] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below;

[0096] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0097] Explanation of relevant terms:

[0098] (1) Concrete quality a: a parameter that measures the quality of concrete materials and directly affects the durability of the building;

[0099] (2) Compressive strength b: The ability of concrete to resist damage when under compression, which is an important indicator for evaluating its mechanical properties;

[0100] (3) Splitting tensile strength c: The ability of concrete to resist splitting tensile failure, reflecting the strength of its tensile properties;

[0101] (4) Initial durability index matrix A: a matrix consisting of multiple durability indexes (such as mass, compressive strength, splitting tensile strength), used to comprehensively evaluate the durability of concrete;

[0102] (5) Index: a parameter used to measure a particular aspect of performance or characteristic during the evaluation process, such as the quality standard of quality a, the compressive strength standard of compressive strength b, etc.;

[0103] (6) Total index: For a specific index (such as compressive strength), the sum of all sample values ​​is used to calculate the weight matrix;

[0104] (7) Sample value: a specific measurement or observation of a certain indicator (such as quality) in a data set;

[0105] (8) Weight matrix P: a matrix consisting of the weight of each sample value in its corresponding total index value, used to reflect the relative importance of each sample in the whole;

[0106] (9) Information entropy e j : A measure of the discreteness and information content of the indicator value. The smaller the information entropy, the greater the amount of information provided by the indicator, and the more important it is in the comprehensive evaluation;

[0107] (10) Comprehensive evaluation factor α: a factor obtained by multiplying the standardized value of each index by its weight and summing them up, which is used to reflect the overall durability of concrete;

[0108] (11) Bidirectional Bi-GRU recurrent neural network: a neural network model that can simultaneously capture the positive and negative dependencies in time series data and is suitable for processing data with time series characteristics;

[0109] (12) Bi-GRU layer: The core layer in the bidirectional Bi-GRU recurrent neural network, which consists of two GRUs (gated recurrent units) in opposite directions and can process forward and reverse timing information at the same time;

[0110] (13) Life prediction value s: a single prediction result output by the neural network model, indicating the expected life of the building at a specific point in time;

[0111] (14) Life prediction sequence S: A sequence consisting of multiple life prediction values ​​s, reflecting the changes in the expected life of a building at different time points.

[0112] Embodiment 1: Figure 1 As shown, this embodiment will provide an implementation scheme of a life prediction method for a long continuous building.

[0113] In this embodiment, regarding step S1, data collection and preprocessing: constructing an initial durability index matrix A. Performing standardization on each index in the initial durability index matrix A to form an initial evaluation value x ij , eliminating dimensional differences.

[0114] Specifically, the construction of matrix A is based on the collection of quality a, compressive strength b, and splitting tensile strength c of historical concrete samples of the same target building. These indicators are key factors in evaluating the durability of concrete and are therefore selected as input features of the model. Each row of the matrix represents a sample and each column represents a feature (quality, compressive strength, splitting tensile strength).

[0115] Further, in step S100, an initial durability index matrix A is constructed:

[0116]

[0117] Among them, a i represents the quality of the i-th sample, b i represents the compressive strength of the i-th sample, ci represents the splitting tensile strength of the i-th sample, and n is the number of samples.

[0118] Further, in step S101, Min-Max standardization is performed: Min-Max standardization is a method of scaling the data range to [0, 1]. Min-Max standardization is performed on each indicator in the initial durability indicator matrix A to form an initial evaluation value x ij ,Right now:

[0119]

[0120] Among them, A ij represents the element in the i-th row and j-th column of matrix A, min(A j ) and max(A j ) represent the minimum and maximum values ​​of the j-th column in the initial durability index matrix A, respectively.

[0121] It can be understood that this method is implemented by subtracting the minimum value of the column and dividing it by the difference between the maximum and minimum values ​​of the column. Standardization eliminates the dimensional differences between different features, so that all features have the same weight and influence in the model. At the same time, after standardization, the initial durability index matrix A is transformed into the initial evaluation value matrix X:

[0122] It is understandable that by eliminating dimensional differences, the model is able to learn the relationship between features and targets more accurately. Standardization helps speed up model training and improves prediction accuracy. Standardized feature values ​​are in the same range, which makes the model's results easier to interpret and understand. Standardization also helps in comparison and analysis between features.

[0123] In this embodiment, regarding step S2, the entropy weight method is performed: for each indicator in the initial durability indicator matrix A, the proportion of each sample value to the total value of the indicator is calculated to form a weight matrix P, and the information entropy e of each indicator is calculated based on the weight matrix P. j And import it into the weight matrix W, and calculate the weight w of each indicator based on the constraints of the weight matrix W j And extract the comprehensive evaluation factor α from the weight matrix W (α is the sum of the standardized value of each indicator multiplied by its weight).

[0124] Specifically, step S200 calculates the weight matrix P: assume each indicator in the initial durability indicator matrix A, calculate the weight of each sample value in the total value of the indicator, and form the weight matrix P:

[0125]

[0126] in, It indicates the proportion of the i-th sample value under the j-th indicator to the total value of the indicator.

[0127] It can be understood that the construction of the weight matrix P is based on the initial durability index matrix A, and is obtained by calculating the proportion of each sample value to the total value of the index. This calculation method can reflect the relative importance of each sample on each index.

[0128] Specifically, in step S201, the information entropy e is calculated. j And form the weight matrix W: For each indicator in the weight matrix P, calculate its information entropy e j :

[0129] Among them, j = 1, 2, 3 correspond to the three indicators of mass, compressive strength and splitting tensile strength respectively;

[0130] The information entropy e j Import it into the weight matrix W, and calculate the weight of each indicator based on the constraints of the weight matrix W (the sum of the weights of each indicator is 1):

[0131] It should be pointed out that the smaller the information entropy, the greater the degree of variation of the indicator and the more information it provides, so it should be given a greater weight in the comprehensive evaluation.

[0132] Specifically, step S202, calculate the comprehensive evaluation factor α: calculate the comprehensive evaluation factor α according to the weight matrix W and the initial evaluation value matrix X (obtained in step S101): The comprehensive evaluation factor α reflects the overall performance of each sample after comprehensively considering all indicators.

[0133] Among them, x ij Represents the initial evaluation value of the i-th sample under the j-th indicator.

[0134] It can be understood that the entropy weight method can more accurately reflect the importance of each indicator in the comprehensive evaluation by calculating the information entropy and weight. Taking into account the information of all indicators, the durability of concrete can be evaluated more comprehensively. By calculating the comprehensive evaluation factor α, the overall performance of each sample can be more accurately reflected.

[0135] Furthermore, the Python execution program of step S2 is as follows Figure 5As shown, A / A.sum(axis=0) is used to calculate the proportion of each sample value in the total value of the indicator to form the proportion matrix P. A.sum(axis=0) calculates the sum of each column of the A matrix, and then divides it by the A matrix through the broadcast mechanism to obtain the proportion matrix P. -1 / np.log(n)*np.sum(P*np.log(P),axis=0) is used to calculate the information entropy ej of each indicator. np.log(P) calculates the natural logarithm of the proportion matrix P, P*np.log(P) calculates the product of each element, and np.sum(...,axis=0) calculates the sum of each column to obtain the information entropy ej. (1-ej) / (1-ej).sum() is used to calculate the weight wj of each indicator and form the weight matrix W. np.sum(wj*X,axis=1) is used to calculate the comprehensive evaluation factor α of each sample. wj*X calculates the product of the weight matrix W and the initial evaluation value matrix X, and np.sum(...,axis=1) calculates the sum of each row to obtain the comprehensive evaluation factor α.

[0136] In this embodiment, regarding step S3, a bidirectional Bi-GRU recurrent neural network model is activated: it includes an input layer, a bidirectional Bi-GRU layer, a fully connected layer and an output layer connected in sequence; wherein the input layer receives the time series data α(t) of the comprehensive evaluation factor α. The bidirectional Bi-GRU layer is used to capture the positive and negative dependencies in the time series data. The fully connected layer maps the output of the Bi-GRU layer to the life prediction value s. The output layer outputs the life prediction sequence S of the building.

[0137] Among them, in the time series data α(t), t=1,2,...,T represents the index of the time series, and T is the total time step.

[0138] Specifically, step S300, input layer: The main function of the input layer is to receive time series data α(t), which is a comprehensive evaluation factor obtained by entropy weight method, which contains key information about building durability. This time series data will be used as input to the bidirectional Bi-GRU layer for subsequent processing. By receiving and transmitting time series data, the input layer provides the model with the necessary information source, so that the model can learn and predict based on this data.

[0139] Specifically, step S301, bidirectional Bi-GRU layer: consists of a forward GRU and a backward GRU, which are responsible for capturing the forward and reverse dependencies in the time series data respectively. The forward GRU processes data from front to back, while the backward GRU processes data from back to front. The output ht of the bidirectional Bi-GRU layer is the concatenation of the hidden states of the forward and backward GRUs, which can capture both past information and future information. Among them:

[0140] (1) The hidden state of the forward GRU is OR{h t};

[0141] (2) The hidden state of the backward GRU is OL{h t};

[0142] (3) The output h of the bidirectional Bi-GRU layer t For: h t =[OR{h t},OL{h t}]; where [·,·] represents the concatenation operation of vectors.

[0143] Furthermore, the specific execution process of this step is as follows:

[0144] S3010, calculate the hidden state of the forward GRU

[0145] Among them, GRU f represents the function of the forward GRU, α(t) is the input of the current time step (i.e., the time series data of the comprehensive evaluation factor α), is the hidden state at the previous time step.

[0146] S3011, calculate the hidden state of the backward GRU

[0147] Among them, GRU b represents the function of the backward GRU, α(t) is the input of the current time step, is the hidden state at the next time step (note that in backward propagation, the time steps are reversed).

[0148] S3012, calculate the output h of the bidirectional Bi-GRU layer t :

[0149] It can be understood that by capturing the forward and reverse dependencies in the time series data, the bidirectional Bi-GRU layer is able to more fully understand the dynamic changes of the data, thereby improving the prediction accuracy of the model.

[0150] Specifically, step S302, fully connected layer: The fully connected layer is responsible for converting the output h of the bidirectional Bi-GRU layer t Mapped to life expectancy prediction value s t Upper:s t =W f ·h t +b f .

[0151] This mapping process is achieved through the weight matrix Wf and the bias vector b f They are the parameters of the fully connected layer and need to be learned through training.

[0152] It can be understood that the fully connected layer, as a bridge between the bidirectional Bi-GRU layer and the output layer, can transform complex hidden states into concise lifespan prediction values, allowing the model to directly output the results that users care about.

[0153] Specifically, step S303, output layer: output building life prediction sequence S: S = {s1, s2, ..., s T}. The sequence consists of a series of life prediction values ​​st, where t represents the index of the time series. Through the output layer, users can intuitively see the model's prediction results for the building life.

[0154] Furthermore, the Python architecture of the bidirectional Bi-GRU recurrent neural network model is as follows Figure 6 As shown, the principle is:

[0155] alpha_data represents time series data, and its shape is (num_samples,T,1), which means there are num_samples data samples, each sample contains T time steps, and each time step has 1 feature. target_data is the corresponding target life data, and its shape is (num_samples,T), which represents the target life value of each sample at each time step.

[0156] Sequential model is used for sequential stacking. Bidirectional(GRU(...),input_shape=(T,1)) constructs a bidirectional Bi-GRU layer to capture the forward and reverse dependencies of time series data. Among them, units=units sets the number of hidden units of GRU, and return_sequences=True ensures that the output of each time step is returned. Dense(1) constructs a fully connected layer responsible for converting the output of the Bi-GRU layer into a life prediction value.

[0157] After training is completed, use the predict method to predict lifespan, with the input being alpha_data and the output being the predicted lifespan value.

[0158] Embodiment 2: This embodiment provides a training method for the bidirectional Bi-GRU recurrent neural network model as described in Embodiment 1, and the training process includes:

[0159] P1. Data preparation and division:

[0160] Collect data: First, it is necessary to collect the historical concrete mass a', historical compressive strength b', and historical splitting tensile strength c' data of the same target building. These data will be used as input features of the model for training and testing the model.

[0161] Furthermore, the collected data are organized into a dataset D, where each sample contains features such as the mass a', compressive strength b' and splitting tensile strength c' of the concrete, as well as the corresponding lifespan label (also obtained from historical data). The dataset D is divided into the training set D train and the test set D test .

[0162] P2. Model training:

[0163] The bidirectional Bi-GRU recurrent neural network model is trained using the divided training set Dtrain. That is, the samples in the training set are input into the model, and the output is calculated through forward propagation. During the training process, the square difference between the model prediction value and the true value is calculated through the mean square error MSE loss function.

[0164] P3. Model parameter optimization:

[0165] Use the Adam optimizer to optimize the model's parameters and minimize the loss function. Adam is an effective stochastic gradient descent algorithm that combines the advantages of the momentum and RMSprop algorithms and can adaptively adjust the learning rate. Through the Adam optimizer, the model's parameters are updated according to the gradient of the loss function, including the weight matrix W f and the bias vector b f .

[0166] P4, iterative training and performance evaluation:

[0167] Through multiple iterations of the training process, the model parameters are continuously adjusted to minimize the loss function. In each iteration, a batch of samples in the training set is used to train the model and calculate the loss.

[0168] In this embodiment, the Python execution framework of the above training method can be found in Figure 7 , where the TensorFlow library is imported and then compiled as in Example 1 ( Figure 6 ), specify the Adam optimizer and mean squared error loss function, then train the model using the training data and evaluate its performance on the test data.

[0169] Application example: This embodiment will provide a specific application method of the technical solution provided in Example 1:

[0170] Assume that the parameters of durability index matrix A are as follows:

[0171] [2.361,2.452,2.543,2.634,2.725,2.816,2.907,2.998,3.089,3.180];

[0172] [2.270,2.361,2.452,2.543,2.634,2.725,2.816,2.907,2.998,3.089];

[0173] [2.179,2.270,2.361,2.452,2.543,2.634,2.725,2.816,2.907,2.998];

[0174] [2.088,2.179,2.270,2.361,2.452,2.543,2.634,2.725,2.816,2.907];

[0175] represents the durability index of 4 different samples (or buildings) at 10 different time steps. Based on the implementation method (S1-S3) provided in Example 1, the final life prediction sequence S is output as follows: Figure 3 shown.

[0176] Experimental example: This example aims to verify the use effect of the bidirectional Bi-GRU recurrent neural network model of the entropy weight method disclosed in Example 1 compared with the traditional GM neural network model.

[0177] (I) Experimental data:

[0178] An exemplary dataset D is constructed, which contains a series of mass a, compressive strength b and splitting tensile strength c of a set of historical concrete samples, as well as the corresponding building life labels:

[0179] data = {

[0180] 'sample_id':[1,2,3,...,100],#sample number

[0181] 'quality_a':[240,245,250,...,2400],#Quality a (unit: kg / m 3 )

[0182] 'strength_b':[4,4.5,5,...,55],#Compressive strength b (unit: MPa)

[0183] 'strength_c':[4.5,4.7,4.9,...,53.5],#splitting tensile strength c (unit: MPa)

[0184] 'lifespan':[5,5.5,6,...,80]#Building lifespan label (unit: year)

[0185] }

[0186] (II) Experimental methods

[0187] Normalize the dataset D to ensure comparability between different features. Divide the dataset D into a training set and a test set.

[0188] 2.1 Experimental group (scheme disclosed in Example 1):

[0189] The entropy weight method is used to determine the weights of mass a, compressive strength b, and splitting tensile strength c. A bidirectional Bi-GRU recurrent neural network model is constructed, and the weights obtained by the entropy weight method are applied to the input features. The model is trained using the training set and predicted using the test set.

[0190] 2.2 Control group:

[0191] Build a traditional GM neural network model. Use the training set to train the model and use the test set to make predictions.

[0192] (III) Result analysis:

[0193] Randomly select 50 time points from the life span prediction sequences output by the experimental group and the control group. Draw a scatter plot to compare the degree of fit between the prediction results of the experimental group and the control group and the actual life span labels.

[0194] The expected experimental results are as follows Figure 4 As shown: The scatter plot of the experimental group is more consistent with the life label curve, indicating that the bidirectional Bi-GRU recurrent neural network model combined with the entropy weight method can more accurately predict the building life. The scatter plot of the control group has a large error with the life label curve, indicating that the prediction performance of the traditional GM neural network model is relatively poor.

[0195] All the above embodiments only express the implementation methods of the relevant practical applications of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

[0196] For those skilled in the art, it can be further appreciated that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in 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 function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel 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.

[0197] At the same time, those skilled in the art can understand that all or part of the processes in all the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

Claims

1. A life prediction method for a long continuous building, comprising collecting the mass a, compressive strength b and splitting tensile strength c of the concrete of the long continuous building under the coupling of salt invasion and dry-wet cycles, characterized in that: Follow these steps: S1, construct the initial durability index matrix A, take the mass a, compressive strength b and splitting tensile strength c as the columns of the matrix; standardize each index in the initial durability index matrix A to form the initial evaluation value x ij ; S2, for each indicator in the initial durability indicator matrix A, calculate the proportion of each sample value to the total value of the indicator to form a proportion matrix P, and calculate the information entropy e of each indicator based on the proportion matrix P j And import it into the weight matrix W, and calculate the weight w of each indicator based on the constraints of the weight matrix W j And extract the comprehensive evaluation factor α from the weight matrix W; S3, activates the bidirectional Bi-GRU recurrent neural network model, including sequentially connected, An input layer for receiving the time series data α(t) of the comprehensive evaluation factor α; Bidirectional Bi-GRU layer that captures forward and backward dependencies in time series data; Map the output of the Bi-GRU layer to the fully connected layer of the life prediction value s; The output layer outputs the building life prediction sequence S.

2. The life prediction method according to claim 1, characterized in that: In S1, the initial durability index matrix A is: Among them, a i represents the quality of the i-th sample, b i represents the compressive strength of the i-th sample, c i represents the splitting tensile strength of the i-th sample, and n is the number of samples; Perform Min-Max standardization on each indicator in the initial durability indicator matrix A to form the initial evaluation value x ij : Among them, A ij represents the element in the i-th row and j-th column of matrix A, min(A j ) and max(A j ) represent the minimum and maximum values ​​of the j-th column in the initial durability index matrix A, respectively.

3. The life prediction method according to claim 1, characterized in that: In S2, the weight matrix P, information entropy e j The implementation method of the weight matrix W is: in, It indicates the proportion of the i-th sample value under the j-th indicator to the total value of the indicator; Among them, j = 1, 2, 3 correspond to the three indicators of mass, compressive strength and splitting tensile strength respectively; The information entropy e j Import it into the weight matrix W, and let the sum of the weights of each indicator be 1 as a constraint to calculate the weight of each indicator:

4. The life prediction method according to claim 3, characterized in that: The calculation method of the comprehensive evaluation factor α is: Among them, x ij Represents the initial evaluation value of the i-th sample under the j-th indicator.

5. The life prediction method according to any one of claims 1 to 3, characterized in that: In S3, the bidirectional Bi-GRU layer includes a forward GRU and a backward GRU, which respectively capture the forward and reverse dependencies in the time series data, wherein: The hidden state of the forward GRU is OR{h t }; The hidden state of the backward GRU is OL{h t }; The output h of the bidirectional Bi-GRU layer t For: h t =[OR{h t },OL{h t }]; where [·,·] represents the concatenation operation of vectors.

6. The life prediction method according to claim 5, characterized in that: The operation process of the bidirectional Bi-GRU layer is as follows: S3010, calculate the hidden state of the forward GRU Among them, GRU f represents the forward GRU function, α(t) is the input of the current time step, is the hidden state of the previous time step; S3011, calculate the hidden state of the backward GRU Among them, GRU b represents the function of the backward GRU, α(t) is the input of the current time step, is the hidden state of the next time step; S3012, calculate the output h of the bidirectional Bi-GRU layer t :

7. The life prediction method according to claim 5, characterized in that: In S3, the fully connected layer converts the output h of the bidirectional Bi-GRU layer t Mapped to life expectancy prediction value s t :s t =W f ·h t +b f ; Among them, W f and b f are the weight matrix and bias vector of the fully connected layer respectively; t=1,2,...,T represents the index of the time series, and T is the total time step.

8. The life prediction method according to claim 7, characterized in that: In S3, the output layer outputs the life prediction sequence S of the building: S = {s1, s2, ..., s T }.

9. A life prediction system for long continuous buildings, characterized in that: The system includes a processor and a memory connected to the processor, wherein the memory stores program instructions, and when the program instructions are executed by the processor, the processor executes the life prediction method according to any one of claims 1 to 8; the processor is connected to: A data preprocessing module responsible for collecting experimental data of long continuous buildings; Calculate the information entropy of each durability indicator and measure the entropy weight method weight assignment module for the uncertainty of information; A neural network model module that implements a bidirectional Bi-GRU recurrent neural network; The comprehensive evaluation factors are input as input data into the trained neural network model to perform life prediction and result output module for life prediction.

10. A storage medium, characterized in that: The storage medium stores information for implementing the life prediction method as described in any one of claims 1-8.

Images