Dynamically updated bridge pile foundation progress prediction method

Abstract

The invention discloses a dynamically updated bridge pile foundation progress prediction method. The method comprises the steps that S1, initial bridge pile foundation linear meter progress data are obtained; s2, data cleaning; s3, judging and sorting the data; s4, training and obtaining a prediction model; s5, performing prediction; and S6, dynamically updating the LSTM prediction model. According to the method, aiming at the characteristic that the construction progress process has strong time sequence, a machine learning algorithm based on the time sequence is adopted to carry out progress prediction, through feature extraction of continuous multi-time-point data, progress prediction of a next time point is carried out, preorder data features are effectively utilized, and through construction of state variables, progress prediction of the next time point is carried out. The characteristics of far-end time point data are reserved, the long-time-period data characteristics are effectively reserved, and the progress prediction accuracy of the bridge pile foundation is improved.

Description

Technical Field

[0001] The invention belongs to the field of rail transportation, and in particular relates to a dynamically updated bridge pile foundation progress prediction method. Background Art

[0002] Bridge pile foundation construction is a key link in bridge engineering. Pile foundation construction is greatly affected by geological conditions. There are great differences in pile foundation construction methods and technical requirements under different geological conditions. At the same time, high construction accuracy is required. The position, verticality, and depth of the pile foundation directly affect the stress state of the bridge structure. Due to the concealment of the pile foundation, quality inspection and monitoring are difficult. Pile foundation construction is usually irreversible. Once the pile is built, if problems are found, the difficulty and cost of repair are extremely high. At the same time, pile foundation construction usually faces complex environments such as high groundwater levels, unstable soil layers, and sensitive surrounding environments, making construction difficult. Therefore, the progress of bridge pile foundation construction is affected by many complex factors such as geological conditions, construction technology, safety control, environmental protection, and quality inspection. It is not only the core of bridge engineering progress control, but also the focus of the entire project progress management.

[0003] The existing progress management of bridge pile foundations basically relies on spreadsheet statistics, which is not only labor-intensive but also prone to data errors. Especially for rail transit linear projects, there are a large number of complex and large bridges. This method is difficult to handle a large amount of data and complex relationships, and it is impossible to predict and analyze the progress of pile foundation construction, resulting in project managers being unable to grasp the project progress in real time, which may cause project risks to get out of control.

[0004] In summary, a dynamically updated bridge pile foundation progress prediction method is urgently needed. Summary of the invention

[0005] In order to solve the problem in the background technology, the present invention provides a bridge pile foundation progress prediction method with accurate prediction and high data utilization and dynamic update.

[0006] To this end, the technical solution of the present invention is as follows:

[0007] A dynamically updated bridge pile foundation progress prediction method comprises the following steps:

[0008] S1, obtaining the initial bridge pile foundation extension meter progress data;

[0009] S2, cleaning the initial bridge pile foundation meter progress data to obtain a bridge pile foundation meter progress data set;

[0010] S3, judging and sorting the data; judging whether it is the first time to use the machine learning prediction model, if so, sorting the data in the bridge pile foundation extended meter progress data set obtained in S2 in chronological order, and then constructing and obtaining the training data set, and then executing S4; if not, executing step S5;

[0011] S4, training and obtaining the prediction model, includes the following steps:

[0012] S41, importing the training data set into the LSTM prediction model to be trained;

[0013] S42, calculate the forget gate using the following formula:

[0014] f t =σ(w f ·[h t-1 ,x t ]+b f );

[0015] Among them, x t is the data of pile foundation length in meters, h t-1 is the predicted data of pile foundation length, x t and h t-1 All belong to the training data set, t represents the time step number; w f is the fully connected network weight of the forget gate, b f is the bias of the forget gate, σ(·) is the sigmoid function; output forget weight f t , used to filter out useless information in the data of bridge pile foundation length;

[0016] S43, calculate the input gate, first calculate the retention weight i of the deep bridge extension data state by the following formula t :

[0017] i t =σ(w i ·[h t-1 ,x t ]+b i );

[0018] Among them, w i is the fully connected network weight of the input gate, b i is the bias of the input gate;

[0019] Then, calculate the data status of deep bridge length The calculation formula is as follows:

[0020]

[0021] Among them, w c The data status of deep bridge length The fully connected network weights are used to Mapping, b c The data status of deep bridge length The bias of , tanh(·) is the tanh activation function;

[0022] S44, Update status:

[0023] The bridge length data state C at time step t-1 t-1 The bridge length data state C at time step t is updated by the following formula: t :

[0024]

[0025] Among them, ° represents the product;

[0026] S45, calculate the output gate by the following formula:

[0027] o t =σ(w o ·[h t-1 ,x t ]+b o );

[0028] Among them, t is the information loss, w o is the fully connected network weight of the output gate, b o is the bias of the output gate;

[0029] S46, calculate the hidden state by the following formula to obtain the predicted value h of the pile foundation extension data at time step t t :

[0030] h t =o t °tanh(C t );

[0031] S47, the predicted value h of the pile foundation length data obtained in S46 t Compare the values ​​in the corresponding training data set in S3. When the difference is less than 0.01 or the number of iterations reaches 1000 steps, the training is stopped and the trained LSTM prediction model is output.

[0032] S5, extracting the bridge pile foundation extension meter progress data within the time period to be predicted from the bridge pile foundation extension meter progress data set obtained in S2, and inputting it into the trained machine learning prediction model LSTM, and then predicting the bridge pile foundation extension meter progress in the subsequent time of the time period.

[0033] Preferably, it also includes S6, dynamically updating the LSTM prediction model: as the construction of the bridge pile foundation progresses, the LSTM prediction model is continuously trained using newly acquired data, the parameters in the LSTM prediction model are corrected, and the prediction accuracy of the LSTM prediction model is improved.

[0034] In S3, a training data set is constructed by the following method: taking the continuous K+1 values ​​in the bridge pile foundation extension meter progress data set as a data pair, and taking the first K values ​​in the data pair as the input value part, and the K+1th value as the true value part, dividing the bridge pile foundation extension meter progress data set into several data pairs, and finally merging all the data pairs into a training data set.

[0035] In S47, before training to obtain the LSTM prediction model, the data involved in the training is completely extracted from the training data set, and the prediction results generated during the training process do not participate in the training of the prediction model.

[0036] The initial bridge pile foundation extended meter progress data includes the bridge construction site name, pier number, pile foundation number, pile foundation extended meter data, pile foundation progress date and pile foundation completion date.

[0037] In S2, the following methods are used for data cleaning: bilinear interpolation is used to handle missing values; the one-svm method is used to handle outliers; duplicate value detection at the same time point is used to handle duplicate data; and a unified method of standard units is used to achieve data standardization.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. The method of the present invention utilizes the data reported in real time in the project, based on a machine learning prediction model, to achieve the purpose of predicting the progress of the bridge pile foundation project, and has practical application value for ensuring the progress of railway project construction.

[0040] 2. The traditional method of bridge progress prediction relies on the filling in of construction team reports and manual prediction, and is highly dependent on the experience of engineering management personnel. However, the method of the present invention adopts a machine learning algorithm based on time series to predict the progress of the construction process, which has a strong time series characteristic. It extracts features from continuous multi-time point data to predict the progress of the next time point, effectively utilizing the features of the previous data, and retains the characteristics of the remote time point data through the construction of state variables, effectively retaining the data characteristics of a long period of time, thereby improving the accuracy of the bridge pile foundation progress prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of the method of the present invention;

[0042] Figure 2 A schematic diagram of the construction progress of a certain extra-large bridge in the embodiment;

[0043] Figure 3 It is a schematic diagram comparing the predicted result and the actual progress of a certain extra-large bridge in the embodiment. DETAILED DESCRIPTION

[0044] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0045] like Figure 1 As shown, the dynamically updated bridge pile foundation progress prediction method of the present invention comprises the following steps:

[0046] S1, obtain the initial bridge pile foundation extension meter progress data:

[0047] The project construction progress management personnel fill in the bridge pile foundation construction data through the information management platform, upload the data to the dedicated table space of the database through the HTTP / HTTPS protocol, and realize efficient decoupled storage of the data. At the same time, the data is stored and indexed through the dedicated table space structure of the database to optimize the data query performance and obtain the initial bridge pile foundation extension progress data.

[0048] The initial bridge pile foundation length progress data includes the bridge construction site name, pier number, pile foundation number, pile foundation length data, pile foundation progress date and pile foundation completion date.

[0049] S2, data cleaning:

[0050] The initial bridge pile foundation meter progress data was cleaned, specifically:

[0051] Bilinear interpolation is used to handle missing values;

[0052] One-SVM method is used to handle outliers;

[0053] Duplicate data were processed by using duplicate value detection at the same time point;

[0054] A unified method of standard unit system is adopted to realize data standardization; after data cleaning, a data set of bridge pile foundation extension progress with high accuracy, strong consistency, no duplication and good integrity is obtained.

[0055] S3, judge and organize the data:

[0056] Determine whether it is the first time to use the machine learning prediction model. If not, jump to step S5; if yes, sort the data in the bridge pile foundation extension progress data set obtained in S2 in chronological order to construct a training data set for machine learning of the prediction model;

[0057] The specific steps of constructing the training data set are as follows: taking K+1 consecutive values ​​as a data pair, taking the first K values ​​in the data pair as the input value part, and the K+1th value as the true value part, dividing the bridge pile foundation extension progress data set into several data pairs, and finally merging all the data pairs into a training data set.

[0058] S4, train and obtain the prediction model:

[0059] The training data obtained in step S3 is grouped according to time series and input into the input interface of the LSTM prediction model used for predicting the progress data of pile foundation extension meters, the machine learning network is trained, and the training is repeated by adjusting the hyperparameters including the learning rate, the optimization method, and the optimization method parameters to obtain a LSTM prediction model with high accuracy, which specifically includes the following steps:

[0060] S41, importing the training data set into the LSTM prediction model to be trained;

[0061] S42, the forget gate is calculated by the following formula, and the forget gate is used to determine the bridge length data state C t-1 Weight:

[0062] f t =σ(w f ·[h t-1 ,x t ]+b f );

[0063] Among them, x t is the data of pile foundation length in meters, h t-1 is the predicted data of pile foundation length, x t and h t-1 All belong to the training data set, t represents the time step number;

[0064] w f is the fully connected network weight of the forget gate, b f is the bias of the forget gate, σ(·) is the sigmoid function, and outputs the forget weight f t , f t It is the proportion of forgotten information, which is used to filter out useless information in the progress data of bridge pile foundation.

[0065] S43, calculate the input gate: the input gate is divided into two parts: first, determine the retention weight i of the deep bridge length data state t , the calculation formula is as follows:

[0066] i t =σ(w i ·[h t-1 ,x t ]+b i );

[0067] Among them, w i is the fully connected network weight of the input gate, which is used to extract the pile foundation length data x t Data status of deep bridge length The relationship between i is the bias of the input gate.

[0068] Then, calculate the data status of deep bridge length The calculation formula is as follows:

[0069]

[0070] Among them, w c The data status of deep bridge length The fully connected network weights are used to Mapping, b c The data status of deep bridge length , and tanh(·) is the tanh activation function.

[0071] S44, Update status:

[0072] The bridge length data state C at time step t-1 t-1 The bridge length data state C at time step t is updated by the following formula: t :

[0073]

[0074] Here, ° represents the product.

[0075] S45, calculate the output gate by the following formula:

[0076] o t =σ(w o ·[h t-1 ,x t ]+b o );

[0077] Among them, t is the information loss, w o is the fully connected network weight of the output gate, b o is the bias of the output gate.

[0078] The output gate is used to control the state weight at the time of output, and the state weight is used to represent the pile foundation extension meter data x t Information loss during prediction propagation.

[0079] S46, the hidden state is calculated by the following formula, through the information loss o t and the bridge length data state C at time step t t Get the predicted value h of the pile foundation extension data at time step t t , as shown below:

[0080]

[0081] S47, the predicted value h of the pile foundation length data obtained in S46t Compared with the true value in S3 at the corresponding time step, when h t When the difference between the corresponding true value is less than 0.01 or the number of iterations of the prediction model reaches 1000 steps, the training is stopped and the trained LSTM prediction model is output. It should be noted that before the LSTM prediction model is trained, the data involved in the training is completely extracted from the training data set. The prediction results generated during the training process do not participate in the training of the prediction model. The prediction results are only used as the standard for the end of training.

[0082] S5, make predictions:

[0083] From the bridge pile foundation extension meter progress data set obtained in S2, the bridge pile foundation extension meter progress data within the time period to be predicted is extracted and input into the LSTM prediction model obtained in S4, so as to predict the bridge pile foundation extension meter progress in the subsequent time period.

[0084] In one embodiment of the present invention, for Figure 2 The software operation of a certain extra-large bridge shown in the figure is based on Python platform and is developed using the pytorch library. First, the length of the data time step is determined. Due to the long construction period of railway bridges, the data interval is selected as the length of the time step in weeks; then, the pile foundation extension meter data is sampled according to the week as the time step, and the latest data is extracted from the bridge pile foundation extension meter progress data set and 8 groups are continuously sampled. These 8 groups of data are input into the pile foundation extension meter progress data prediction network obtained by S4. The network extracts features from the data at time W-7, where W is the time step number corresponding to the time to be predicted, and the extracted pile foundation extension meter progress features are iteratively fused in chronological order. Through 8 iterations of fusion forward conduction, the pile foundation extension meter progress data at time W+1 is predicted. Then, based on the pile foundation extension meter progress data at time W+1, 8 groups of data are sampled from time W-6 to predict time W+2, and so on, until the data at time W+12 is predicted. That is, after each update of the bridge pile foundation progress data, the progress is predicted for the next 3 months. The prediction results are shown as follows Figure 3 As shown, it can be seen that the gap between the predicted value and the true value is small.

[0085] In another embodiment of the present invention, S6 is also included for dynamically updating the LSTM prediction model: as the construction of the bridge pile foundation progresses, the prediction model is continuously trained using newly acquired data, and the parameters in the LSTM prediction model are continuously corrected to improve the prediction accuracy of the LSTM prediction model.

Claims

1. A dynamically updated bridge pile foundation progress prediction method, characterized in that: The following steps are involved: S1, obtaining the initial bridge pile foundation extension meter progress data; S2, cleaning the initial bridge pile foundation meter progress data to obtain a bridge pile foundation meter progress data set; S3, judging and sorting the data; judging whether it is the first time to use the machine learning prediction model, if so, sorting the data in the bridge pile foundation extended meter progress data set obtained in S2 in chronological order, and then constructing and obtaining the training data set, and then executing S4; if not, executing step S5; S4, training and obtaining the prediction model, includes the following steps: S41, importing the training data set into the LSTM prediction model to be trained; S42, calculate the forget gate using the following formula: f t =σ(w f ·[h t-1 ,x t ]+b f ); Among them, x t is the data of pile foundation length in meters, h t-1 is the predicted data of pile foundation length, x t and h t-1 All belong to the training data set, t represents the time step number; w f is the fully connected network weight of the forget gate, b f is the bias of the forget gate, σ(·) is the sigmoid function; output forget weight f t , used to filter out useless information in the data of bridge pile foundation length; S43, calculate the input gate, first calculate the retention weight i of the deep bridge extension data state by the following formula t : i t =σ(w i ·[h t-1 ,x t ]+b i ); Among them, w i is the fully connected network weight of the input gate, b i is the bias of the input gate; Then, calculate the data status of deep bridge length The calculation formula is as follows: Among them, w c The data status of deep bridge length The fully connected network weights are used to Mapping, b c The data status of deep bridge length The bias of , tanh(·) is the tanh activation function; S44, Update status: The bridge length data state C at time step t-1 t-1 The bridge length data state C at time step t is updated by the following formula: t : Among them, ° represents the product; S45, calculate the output gate by the following formula: the t =σ(w o ·[h t-1 ,x t ]+b o ); Among them, t is the information loss, w o is the fully connected network weight of the output gate, b o is the bias of the output gate; S46, calculate the hidden state by the following formula to obtain the predicted value h of the pile foundation extension data at time step t t : S47, the predicted value h of the pile foundation length data obtained in S46 t Compare the values ​​in the corresponding training data set in S3. When the difference is less than 0.01 or the number of iterations reaches 1000 steps, the training is stopped and the trained LSTM prediction model is output. S5, extracting the bridge pile foundation extension meter progress data within the time period to be predicted from the bridge pile foundation extension meter progress data set obtained in S2, and inputting it into the trained machine learning prediction model LSTM, and then predicting the bridge pile foundation extension meter progress in the subsequent time of the time period.

2. The dynamically updated bridge pile foundation progress prediction method according to claim 1 is characterized in that: It also includes S6, dynamically updating the LSTM prediction model: as the construction of the bridge pile foundation progresses, the LSTM prediction model is continuously trained using newly acquired data, the parameters in the LSTM prediction model are corrected, and the prediction accuracy of the LSTM prediction model is improved.

3. The dynamically updated bridge pile foundation progress prediction method according to claim 2 is characterized in that: In S3, a training data set is constructed by the following method: taking the continuous K+1 values ​​in the bridge pile foundation extension meter progress data set as a data pair, and taking the first K values ​​in the data pair as the input value part, and the K+1th value as the true value part, dividing the bridge pile foundation extension meter progress data set into several data pairs, and finally merging all the data pairs into a training data set.

4. The dynamically updated bridge pile foundation progress prediction method according to claim 1 is characterized in that: In S47, before training to obtain the LSTM prediction model, the data involved in the training is completely extracted from the training data set, and the prediction results generated during the training process do not participate in the training of the prediction model.

5. The dynamically updated bridge pile foundation progress prediction method according to claim 1 is characterized in that: The initial bridge pile foundation extended meter progress data includes the bridge construction site name, pier number, pile foundation number, pile foundation extended meter data, pile foundation progress date and pile foundation completion date.

6. The dynamically updated bridge pile foundation progress prediction method according to claim 1 is characterized in that: In S2, the following methods are used for data cleaning: bilinear interpolation is used to handle missing values; the one-svm method is used to handle outliers; duplicate value detection at the same time point is used to handle duplicate data; and a unified method of standard units is used to achieve data standardization.

Images