CN114757096B - Bridge temperature prediction method, device, equipment and medium based on NARX neural network - Google Patents

Abstract

The invention discloses a bridge temperature prediction method, a device, computer equipment and a storage medium based on NARX neural network, wherein the method comprises the following steps: performing sequential interpolation or extraction on the obtained bridge temperature data and the corresponding meteorological data, adjusting the bridge temperature data and the corresponding meteorological data to time sequence data with the same sampling rate, and taking the adjusted meteorological data as an alternative parameter; selecting weather parameters with high correlation with bridge temperature from the alternative parameters through the maximum information coefficient, and preprocessing the selected weather parameters and the corresponding bridge temperature data to obtain sample data; the sample data is divided into training data and test data; training the open-loop architecture model by using training data, converting the open-loop architecture in the trained model into a closed-loop structure, and evaluating the prediction performance of the closed-loop structure model by using test data; and if the prediction performance is good, predicting the data value of the bridge temperature by using the closed-loop structure model. The method provided by the invention has high prediction precision and good engineering practicability.

Description

Bridge temperature prediction method, device, equipment and medium based on NARX neural network

Technical field

The invention relates to the technical field of bridge structure health monitoring, in particular to a bridge temperature prediction method, device, computer equipment and storage medium based on NARX neural network.

Background technique

Bridge structures are important infrastructure and key nodes on transportation arteries. Millions of bridges have made significant contributions to the development of our country's society and economy. Temperature load is an important form of load. Temperature changes can cause changes in the material properties of the bridge and thermal expansion and contraction of the volume. Under the connection constraints between the boundaries and components, it will lead to structural deformation, stress, strain, support reaction force, etc. Temperature effect causes the occurrence of bridge temperature diseases. In order to ensure the safety of bridge structure construction and operation, it is very necessary to obtain bridge temperature field data.

At present, the acquisition of bridge temperature field data relies heavily on on-site measurements, and real-time monitoring of bridge temperature based on structural health monitoring systems has the characteristics of high cost, long time consumption, and low analysis efficiency, and the sensors may malfunction, be damaged, or be lost during their service life. situation, it cannot meet the engineering analysis needs of most bridges and the requirements for long-term stable data acquisition. In addition, in the past, the finite element thermal analysis method was used to obtain the bridge temperature field, which also had the disadvantages of complex thermal boundary condition calculation methods, difficulty in software writing, time-consuming operation, large memory usage, and many limitations in use.

In recent years, neural network technology has been widely used in the engineering field due to its superior nonlinear fitting capabilities. Nonlinear Auto-Regressive with ExogenousInputs Model (NARX) is a recursive dynamic neural network. It is based on the BP neural network and introduces the output of the neural network into the neural network as external feedback. input layer and add s units of delay to the input. Compared with static neural networks, its output also considers the information from the time step from t-s to time t. It can be regarded as a multi-layer perceptron with input delay and short-term memory capabilities, which can better learn complex time series data. The relationship between input and output of a dynamic system. Applying this model can consider the highly nonlinear and time series characteristics of bridge temperature and external meteorological factors, and output the predicted value of the temperature of the bridge measuring point corresponding to the meteorological time series data.

Contents of the invention

In order to solve the above-mentioned deficiencies in the prior art, the present invention provides a bridge temperature prediction method, device, computer equipment and storage medium based on NARX neural network. This method utilizes the advantages of NARX neural network in processing non-linear data relationships and short-term memory capabilities. By inputting the site meteorological parameters, the corresponding bridge temperature data can be quickly obtained. This is the first application of deep learning technology in the direction of bridge temperature prediction. This method has high prediction accuracy, good engineering practicability, and can shorten calculation time.

The first object of the present invention is to provide a bridge temperature prediction method based on NARX neural network.

The second object of the present invention is to provide a bridge temperature prediction device based on NARX neural network.

A third object of the present invention is to provide a computer device.

The fourth object of the present invention is to provide a storage medium.

The first object of the present invention can be achieved by adopting the following technical solutions:

A bridge temperature prediction method based on NARX neural network, the method includes:

Perform sequence interpolation or extraction on the obtained bridge temperature data and corresponding meteorological data, adjust it to time series data with the same sampling rate, and use the adjusted meteorological data as alternative parameters; wherein the meteorological data includes multiple meteorological parameters;

Meteorological parameters with high correlation with bridge temperature are selected from the alternative parameters through the maximum information coefficient as input features of the model; the selected meteorological parameters and the corresponding bridge temperature data are preprocessed respectively to obtain sample data; where , the sample data is divided into training data and test data;

Use the training data to train the model to obtain a trained model; the model is a NARX neural network model with an open-loop architecture with external input;

Convert the open-loop architecture in the trained model to a closed-loop structure, use the test data, and use the absolute error average index to evaluate the prediction performance of the closed-loop structure model;

If the prediction result is within the preset range, the closed-loop structural model obtains the predicted value of the bridge temperature based on the input meteorological data.

Further, using the training data to train the model to obtain a trained model specifically includes:

Using the training data to train the model, including hyperparameter learning and optimization of the model;

For the trained model, the training data is used to quantify the relationship strength between the predicted value and the target value using fitting optimization. If the relationship strength is not high enough, the model is retrained or the topology of the model is optimized to obtain a well-trained model. model;

The training data includes meteorological data _xi and corresponding bridge temperature data _yi , with _yi as the target value; the value output after the meteorological data _xi is input into the model is the predicted value

Using the training data, fitting optimization is used to quantify the relationship strength between the predicted value and the target value, specifically:

The goodness of fit formula is as follows:

In the formula, R ² represents the strength of the relationship between the predicted value and the target value, is the mean of the target value, and n is the number of samples input to the model.

Further, using the training data to train the model specifically includes:

Divide the training data into three subsets, and use the three subsets for cross-validation, where:

The first subset is the training set, which is used to calculate gradients and update network weights and biases to minimize the network loss function;

The second subset is the validation set, which monitors validation errors during the training process so that network weights and biases are saved with the minimum validation set error;

The third subset is the test set, which is used for final testing after training and verification, and outputs the performance indicators after the model training is completed;

Choose Bayesian regularization algorithm or quantified conjugate gradient algorithm to optimize network weights and biases, and introduce Dropout technology to avoid under-fitting and over-fitting;

During the training process of the model, the trainscg quantified conjugate gradient algorithm is selected, and the training data is input into the model for parameter learning and optimization of the neural network. If the verification error does not decrease, the iteration is stopped, otherwise the number of iterations is set. number of times.

Further, the test data includes meteorological data and corresponding bridge temperature;

The use of the test data and the use of the absolute error average index to evaluate the prediction performance of the closed-loop structural model include:

Input any meteorological data x _j in the test data into the closed-loop structural model to obtain the value output by the closed-loop structural model. After de-normalizing the output value, the predicted value of the bridge temperature is obtained.

The bridge temperature corresponding to the meteorological data x _j in the test data is de-standardized and used as the target value y _j of the bridge temperature;

The absolute error average index is used to reflect the error between the bridge temperature prediction value and the target value. The absolute error average is calculated as follows:

Among them, n is the number of samples input to the model;

If the average absolute error is near the preset value, it means that the prediction performance is good and can meet the engineering accuracy requirements.

Further, let the bridge temperature be x, and any corresponding meteorological parameter among the alternative parameters be y;

The meteorological parameters that are highly correlated with the bridge temperature are selected from the alternative parameters through the maximum information coefficient as the input features of the model, specifically including:

The calculation formula using the maximum information coefficient is as follows:

Among them, I(x;y) is the mutual information coefficient, p(x,y) is the joint probability between variables x and y, a and b represent the number of grids divided into two-dimensional space in the x and y directions, B The size setting satisfies a*b<B;

The larger the value of MIC(x;y), the higher the correlation between x and y; if the correlation between x and y is high, then y is selected as the input feature of the model.

Further, the selected meteorological parameters and corresponding bridge temperature data are preprocessed respectively to obtain sample data, which specifically includes:

Perform Z-Score standardization processing on the selected meteorological parameters respectively to obtain processed meteorological data as input data for the model;

Perform Z-Score normalization processing on the bridge temperature data to obtain the processed bridge temperature data as the model output value;

The processed meteorological data and the corresponding processed bridge temperature data constitute sample data.

Further, the NARX neural network model adopts a network structure with output delayed feedback, including an input layer, a hidden layer and an output layer;

Initialize the number of nodes and the number of feedback delays of the NARX neural network model. The input-output relationship of the NARX neural network model is:

y(t)=f(x(t),x(t-1),…,x(t-d),y(t-1))

Among them, t represents the current moment, d represents the number of feedback delays, y(t) and y(t-1) represent the output of the network model at the current moment and the previous moment, x(t) and x(t-d) represent the current moment and the previous moment. The input data of the network model at time d.

Further, the meteorological data obtained include meteorological parameters such as zenith angle, temperature, relative humidity, wind speed, rainfall, and total cloud coverage.

The second object of the present invention can be achieved by adopting the following technical solutions:

A bridge temperature prediction device based on NARX neural network, the device includes:

The alternative parameter acquisition module is used to perform sequence interpolation or extraction on the acquired bridge temperature data and corresponding meteorological data, adjust it to time series data with the same sampling rate, and use the adjusted meteorological data as alternative parameters; wherein, the meteorological data The data include multiple meteorological parameters;

The sample data acquisition module is used to select meteorological parameters that are highly correlated with the bridge temperature from the alternative parameters through the maximum information coefficient as the input features of the model; and predict the selected meteorological parameters and the corresponding bridge temperature data respectively. Process to obtain sample data; wherein the sample data is divided into training data and test data;

A model training module, used to train the model using the training data to obtain a trained model; the model is a NARX neural network model with an open-loop architecture with external input;

The model prediction performance evaluation module is used to convert the open-loop architecture in the trained model into a closed-loop structure, and uses the test data to evaluate the prediction performance of the closed-loop structure model using the absolute error average index;

The bridge temperature prediction module is used to use the closed-loop structural model to obtain the predicted value of the bridge temperature based on the input meteorological data if the prediction result is within the preset range.

The third object of the present invention can be achieved by adopting the following technical solutions:

A computer device includes a processor and a memory for storing an executable program of the processor. When the processor executes the program stored in the memory, the above bridge temperature prediction method is implemented.

The fourth object of the present invention can be achieved by adopting the following technical solutions:

A storage medium stores a program. When the program is executed by a processor, the above bridge temperature prediction method is implemented.

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

1. The present invention provides a bridge temperature prediction method based on NARX neural network, which is the first application of deep learning technology in the direction of bridge temperature prediction. This method constructs meteorological characteristics that are strongly related to the bridge temperature as the input of the model, and considers the periodic characteristics of the time dimension of meteorological data, the nonlinear combination relationship with the structural temperature, and the short-term temporal dependence through the neural network; and through the neural network The recursive method updates the network neuron weights and thresholds in real time, which can complete multi-step prediction of the temperature of multiple measuring points on the bridge within the required time step, and the prediction accuracy is high.

2. The present invention uses a neural network to calculate the temperature of the bridge. The more traditional method of obtaining the temperature of the measuring point based on the SHM monitoring system can make up for the shortcomings of health monitoring equipment that has a certain service life limit and possible lack of data, and achieve long-term measurement of the temperature of the bridge measuring point. Stable acquisition.

3. Compared with the traditional method of calculating the temperature field based on finite elements and programming software interactive programs, the method provided by the present invention can shorten the calculation time, reduce the memory occupied space, improve the calculation efficiency, and has good engineering practicality.

Description of the drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the structures shown in these drawings without exerting creative efforts.

Figure 1 is a flow chart of a bridge temperature prediction method based on NARX neural network in Embodiment 1 of the present invention.

Figure 2 is a schematic diagram of the open-loop topology of the NARX neural network model in Embodiment 1 of the present invention.

Figure 3 is a schematic diagram of the NARX model fitting goodness (R ² ) index in Embodiment 1 of the present invention.

Figure 4 is a schematic diagram of the closed-loop topology structure of the NARX neural network model in Embodiment 1 of the present invention.

Figure 5 is a schematic diagram of the box-shaped bridge section and measuring point locations in Embodiment 1 of the present invention.

Figure 6 is a schematic diagram of the mean absolute error (MAE) index of the NARX model prediction results in Embodiment 1 of the present invention.

Figure 7 is a schematic diagram comparing the predicted value and the target value of the temperature at the middle point of the roof in Embodiment 1 of the present invention.

Figure 8 is a schematic diagram comparing the predicted value and the target value of the temperature at the middle point of the base plate in Embodiment 1 of the present invention.

Figure 9 is a schematic diagram comparing the predicted value and the target value of the temperature at the middle point of the east web in Embodiment 1 of the present invention.

Figure 10 is a schematic diagram comparing the predicted value and the target value of the temperature at the middle point of the west web in Embodiment 1 of the present invention.

Figure 11 is a structural block diagram of a bridge temperature prediction device based on NARX neural network according to Embodiment 2 of the present invention.

Figure 12 is a structural block diagram of a computer device according to Embodiment 3 of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments These are some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present invention. . It should be understood that the specific embodiments described are only used to explain the present application and are not used to limit the present application.

Example 1:

As shown in Figure 1, this embodiment provides a bridge temperature prediction method based on NARX neural network, including the following steps:

S101. Perform sequence interpolation or extraction on the acquired bridge temperature data and corresponding meteorological data, adjust them to time series data with the same sampling rate, and use the adjusted meteorological data as alternative parameters.

Among them, meteorological data includes multiple meteorological parameters.

(1) Obtain bridge temperature data and corresponding meteorological data.

The bridge temperature data used for model training comes from actual observations from the health monitoring system or finite element calculations, and the meteorological data comes from shared data downloads from meteorological websites or actual observations at the bridge site.

In this embodiment, the bridge temperature data used in pre-training is obtained by arranging temperature measurement points on the physical bridge structure and using temperature acquisition instruments.

In this embodiment, the meteorological data includes six parameters: zenith angle, temperature, relative humidity, wind speed, rainfall, and total cloud coverage. Among them, the data on temperature, relative humidity, wind speed, rainfall, and total cloud coverage come from weather station data provided by the Swiss meteorological platform Meteoblue and downloaded from the www.meteoblue.com website.

Since the meteorological station is only 500m away from the bridge site, the weather station environment can be approximately used to replace the bridge site meteorological environment. According to the relevant data of the bridge site, such as solar altitude angle, solar declination, solar hour angle, local time and location Point latitude, calculate the zenith angle γ through the following formula:

sinδ＝0.39795cos[0.98563(N-173)/180*pi]

ω=15×(ST-12)

γ=90°-arcsin(sinh _s )

Among them, h _s is the solar altitude angle; δ is the solar declination, indicating the angle between the sun's rays and the earth's equatorial plane; N is the day number, that is, the number of days from January 1 of that year; ω is the solar hour angle; ST is the local time; is the latitude of the observation location.

(2) Perform sequence interpolation or extraction on the bridge temperature data and corresponding meteorological data, adjust them to time series data with the same sampling rate, and use the adjusted meteorological data as alternative parameters.

Perform linear interpolation or extraction on the temperature data of measuring points obtained from actual observations and the corresponding meteorological data, and adjust them to time series data with the same sampling rate. The sample time interval is 1 hour. The adjusted meteorological data will be used as alternative parameters for the training model. .

S102. Select the meteorological parameters that are highly correlated with the bridge temperature from the candidate parameters through the maximum information coefficient as the input features of the model.

In this embodiment, the candidate parameters include six parameters in the meteorological data: zenith angle, temperature, relative humidity, wind speed, rainfall, and total cloud coverage.

Implement dimensionality reduction for six alternative parameters, and quantitatively select meteorological factors that have a strong correlation with bridge temperature. The maximum information coefficient (MIC) is used to measure the correlation between each parameter in the candidate parameters and the bridge temperature.

MIC is a quantification of the relative entropy between the joint distribution and the marginal distribution of two random variables. It is used to measure the degree of linear or nonlinear correlation between two variables x and y. The basic idea is to discretize two variables in a two-dimensional space and use a scatter diagram to represent it, then divide the two-dimensional space into a certain number of squares in the x and y directions, and observe the placement of the scatter points in each square. In the case of scattered points, the joint probability is calculated to simplify the calculation of the mutual information coefficient, which solves the problem that the joint probability is difficult to solve in mutual information; finally, normalization is performed to obtain the value of MIC. There are many different ways to divide it into certain squares in the x and y directions, and the maximum MIC value is calculated as the final result. The MIC calculation formula is as follows:

In the formula, I(x,y) is the mutual information coefficient, p(x,y) is the joint probability between variables x and y, a and b represent the number of grids divided into two-dimensional space in the x and y directions, Set a*b<B, and the size of B is set to about 0.6 power of the number of data samples.

Meteorological parameters with high correlation with bridge temperature are selected as input features of the model through the maximum information coefficient (MIC). That is, the larger the value of MIC, the higher the correlation between x and y.

In this embodiment, the meteorological parameters selected through MIC include five parameters: zenith angle, temperature, relative humidity, wind speed, and total cloud coverage, which are used as input features of the model.

S103. Preprocess the selected meteorological parameters and the bridge temperature data under corresponding environmental conditions respectively to obtain sample data, and divide the sample data into training data and test data.

(1) Perform Z-Score standardization processing on the selected meteorological parameters to obtain processed meteorological data.

The selected meteorological parameters are used as input features of the model. Before inputting the data of the input features into the model, Z-Score standardization processing can be performed to eliminate the impact of data dimensions and data range differences of different features, which greatly improves neural network training. Data performance. The processed meteorological parameter data is called processed meteorological data and is used as the input data of the model.

After processing, the data is transformed into dimensionless data with a mean of 0 and a variance of 1, and conforms to the standard normal distribution. This method can uniformly transform data of different magnitudes into the same magnitude to ensure comparability between data and improve the stability of the neural network. The Z-Score normalization formula is as follows:

In the formula: x is the pre-processed data, x’ is the post-processed data, μ is the mean of all pre-processed data, and σ is the standard deviation of all pre-processed data.

(2) Perform Z-Score standardization processing on the bridge temperature data to obtain the processed bridge temperature data.

According to the selected meteorological parameters, bridge temperature data under corresponding environmental conditions are selected from the bridge temperature data. Z-Score standardization is also performed on the selected bridge temperature data to obtain the processed bridge temperature data as the value of the model output. .

(3) Obtain sample data based on the processed meteorological data and processed bridge temperature data.

The processed meteorological data and the processed bridge temperature data constitute the sample data.

(4) Divide the sample data into training data and test data.

Under the condition that the overall time series data samples are not out of order, 80% of the sample data is used as training data and 20% is used as test data.

S104. Use the training data to train the NARX neural network model.

(1) Build an open-loop NARX neural network model with external input.

As shown in Figure 2, the open-loop NARX neural network model provided in this example includes an input layer, a hidden layer and an output layer.

Initialize the number of nodes and the number of feedback delays of the NARX neural network model. The input-output relationship of the NARX neural network model is:

y(t)=f(x(t),x(t-1),…,x(t-d),y(t-1))

Where t represents the current moment and d represents the number of feedback delays.

f(x,y) is a nonlinear mapping function about x,y, y(t), y(t-1) represents the network output at the current moment and the previous moment, x(t), x(t-d) represents the current moment and network input at the first d moments. The hidden layer contains h neurons, and the optimal number of neurons in the hidden layer can be selected based on the empirical formula:

Among them, h is the number of hidden layer neurons, m is the number of input neurons, n is the number of output neurons, and a is any constant between 1 and 10.

In this embodiment, the hidden layer contains 10 neurons, the hidden layer activation function is tansig, the output layer activation function is linear linear function, and the feedback delay number d is set to 3.

(2) Use training data to perform hyperparameter learning and optimization of the open-loop NARX neural network model.

The training data was divided into three subsets for cross-validation, where:

The first subset is the training set, which is used to calculate gradients and update network weights and biases to minimize the network loss function MSE;

The second subset is the validation set, which monitors validation errors during the training process so that network weights and biases are saved with the minimum validation set error;

The third subset is the test set, which is used for final testing after training and validation, and outputs performance indicators after the model training is completed.

During the training process of the NARX model, an open-loop network structure with output delayed feedback is used for training. Bayesian regularization algorithm or quantified conjugate gradient algorithm is selected according to the performance of the neural network to optimize network weights and biases, and Dropout technology is introduced to avoid The occurrence of underfitting and overfitting.

During the model training process in this embodiment, the training set accounts for 70% of the total batch of training data, and the verification set and the test set both account for 15% of the total batch of training data. Set the number of iterations epochs=150, the learning rate lr=0.01, select the quantized conjugate gradient algorithm (trainscg) for training, and input the training data into the model for parameter learning and optimization of the neural network.

During the training process of the neural network, if the verification error does not decrease within six iterations, the iteration will be stopped, otherwise it will be iterated to the above epochs.

S105. For the trained NARX neural network model, use the training data and use fitting optimization to quantify the relationship strength between the predicted value output by the NARX neural network model and the target value, and then obtain the trained model.

After the model is trained, the training data is used to quantify the strength of the relationship between the model input and output using the goodness of fit (R ² ). Among them, the training data includes meteorological data _xi and the corresponding bridge temperature data _yi , _yi is the target value, and the value output after the meteorological data _xi is input into the model is the predicted value.

The goodness of fit formula is as follows:

in, is the mean of the target value, and n is the number of samples input to the model.

The goodness of fit of this embodiment on the training set, validation set, and test set is shown in Figure 3. The goodness of fit (R ² ) of each data set reaches above 0.975. The greater the relationship strength, the higher the accuracy of the model, which proves Input feature selection is reasonable.

After the goodness of fit reaches a satisfactory level, you need to further check its loss function L(y,f(x)) on the training set and test set to measure the difference between the real value y and the predicted value f(x) degree. It is necessary to prevent the gap between the test set loss function and the training set from being too large to avoid overfitting; otherwise, retrain the NARX neural network model with random initial weights and thresholds, or adjust the number of hidden layer neurons to optimize its topology. The network loss function is calculated using mean square error (MSE).

S106. Convert the open-loop architecture in the trained model to a closed-loop structure, and use test data to evaluate the prediction performance of the closed-loop structural model; if the prediction results are within the preset range, the closed-loop structural model obtains the bridge temperature based on the input meteorological data. predicted value.

After step S105, the open-loop structure of the NARX neural network model is converted into a closed-loop structure. The NARX neural network model of the closed-loop structure is shown in Figure 4.

In order to test the generalization ability of the NARX neural network model and evaluate the prediction performance of the NARX neural network model, the meteorological data x _j of 240 consecutive time steps (corresponding to 240 hours and 10 days) in the test data are input into the closed-loop NARX neural network model to obtain NARX The value output by the neural network model is denormalized to obtain the predicted value of the bridge temperature. The bridge temperature corresponding to the meteorological data x _j in the test data is denormalized and used as the target value y _j of the bridge temperature. The target value of the bridge temperature is the same as the original bridge temperature data in step S101; the predicted value of the bridge temperature is compared with the bridge temperature Compare with target value.

To evaluate the performance of the NARX neural network model, the mean absolute error (MAE) index is used to reflect the error between the predicted value and the target value. The mean absolute error (MAE) is calculated as follows:

Among them, n is the number of samples input to the model.

The cross-sectional form and measuring point locations of the bridge in the calculation example in this embodiment are shown in Figure 5. The structural temperature value within 10 days under summer meteorological conditions in a subtropical site is calculated. For the evaluation of the prediction performance of the neural network model, the mean absolute error (MAE) index is shown in Figure 6, and the comparison between the predicted value and the target value of the bridge temperature at each measuring point is shown in Figures 7 to 10. The results show that the mean absolute error (MAE) of the bridge temperature prediction results is around 2.0, that is, the predicted value is consistent with the target value, meets the engineering accuracy requirements, and has good applicability.

For a closed-loop structural model with good applicability, one or more predicted values are output based on one or more sets of input meteorological data, and the predicted values of the bridge temperature are obtained by de-standardizing the predicted values.

Those skilled in the art can understand that all or part of the steps in the method of implementing the above embodiments can be completed by instructing relevant hardware through a program, and the corresponding program can be stored in a computer-readable storage medium.

It should be noted that although the method operations of the above embodiments are described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all illustrated operations must be performed to achieve desired results. . Instead, the steps depicted can be executed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step for execution, and/or one step may be broken down into multiple steps for execution.

Example 2:

As shown in Figure 11, this embodiment provides a bridge temperature prediction device based on NARX neural network. The device includes an alternative parameter acquisition module 1101, a sample data acquisition module 1102, a model training module 1103, and a model prediction performance evaluation module 1104. and the Bridge Temperature Prediction Module 1105, which:

The alternative parameter acquisition module 1101 is used to perform sequence interpolation or extraction on the acquired bridge temperature data and corresponding meteorological data, adjust it to time series data with the same sampling rate, and use the adjusted meteorological data as alternative parameters; wherein, the Meteorological data includes multiple meteorological parameters;

The sample data acquisition module 1102 is used to select meteorological parameters that are highly correlated with the bridge temperature from the alternative parameters through the maximum information coefficient as the input features of the model; perform separate operations on the selected meteorological parameters and the corresponding bridge temperature data. Preprocess to obtain sample data; wherein the sample data is divided into training data and test data;

The model training module 1103 is used to train the model using the training data to obtain a trained model; the model is a NARX neural network model with an open-loop architecture with external input;

The model prediction performance evaluation module 1104 is used to convert the open-loop architecture in the trained model into a closed-loop structure, use the test data, and use the absolute error average index to evaluate the prediction performance of the closed-loop structure model;

The bridge temperature prediction module 1105 is used to obtain the predicted value of the bridge temperature based on the input meteorological data by the closed-loop structural model if the prediction result is within a preset range.

The specific implementation of each module in this embodiment can be referred to the above-mentioned Embodiment 1, which will not be described one by one here. It should be noted that the device provided in this embodiment is only exemplified by the division of the above-mentioned functional modules. In practical applications, , the above functions can be allocated to different functional modules as needed, that is, the internal structure is divided into different functional modules to complete all or part of the functions described above.

Example 3:

This embodiment provides a computer device. The computer device can be a computer. As shown in Figure 12, it has a processor 1202, a memory, an input device 1203, a display 1204 and a network interface 1205 connected through a system bus 1201. The processor uses In order to provide computing and control capabilities, the memory includes a non-volatile storage medium 1206 and an internal memory 1207. The non-volatile storage medium 1206 stores an operating system, computer programs and databases. The internal memory 1207 is a non-volatile storage medium. The operating system and computer program in the medium provide an environment for running. When the processor 1202 executes the computer program stored in the memory, the bridge temperature prediction method of the above-mentioned Embodiment 1 is implemented, as follows:

Perform sequence interpolation or extraction on the obtained bridge temperature data and corresponding meteorological data, adjust it to time series data with the same sampling rate, and use the adjusted meteorological data as alternative parameters; wherein the meteorological data includes multiple meteorological parameters;

Meteorological parameters with high correlation with bridge temperature are selected from the alternative parameters through the maximum information coefficient as input features of the model; the selected meteorological parameters and the corresponding bridge temperature data are preprocessed respectively to obtain sample data; where , the sample data is divided into training data and test data;

Use the training data to train the model to obtain a trained model; the model is a NARX neural network model with an open-loop architecture with external input;

Convert the open-loop architecture in the trained model to a closed-loop structure, use the test data, and use the absolute error average index to evaluate the prediction performance of the closed-loop structure model;

If the prediction result is within the preset range, the closed-loop structural model obtains the predicted value of the bridge temperature based on the input meteorological data.

Example 4:

This embodiment provides a storage medium, which is a computer-readable storage medium that stores a computer program. When the computer program is executed by a processor, the bridge temperature prediction method of the above-mentioned Embodiment 1 is implemented, as follows:

Perform sequence interpolation or extraction on the obtained bridge temperature data and corresponding meteorological data, adjust it to time series data with the same sampling rate, and use the adjusted meteorological data as alternative parameters; wherein the meteorological data includes multiple meteorological parameters;

Meteorological parameters with high correlation with bridge temperature are selected from the alternative parameters through the maximum information coefficient as input features of the model; the selected meteorological parameters and the corresponding bridge temperature data are preprocessed respectively to obtain sample data; where , the sample data is divided into training data and test data;

Use the training data to train the model to obtain a trained model; the model is a NARX neural network model with an open-loop architecture with external input;

Convert the open-loop architecture in the trained model to a closed-loop structure, use the test data, and use the absolute error average index to evaluate the prediction performance of the closed-loop structure model;

If the prediction result is within the preset range, the closed-loop structural model obtains the predicted value of the bridge temperature based on the input meteorological data.

It should be noted that the computer-readable storage medium in this embodiment may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of computer readable storage media may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard drive, random access memory (RAM), read only memory (ROM), removable Programmed read-only memory (EPROM or flash memory), fiber optics, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above.

To sum up, the present invention first performs sequence interpolation or extraction on the bridge measuring point temperature data and the corresponding meteorological data, and adjusts them to time series data with the same sampling rate; then, the maximum information coefficient calculation is used to select the temperature data with high correlation with the bridge temperature. Meteorological parameters are used as input features of the model; an open-loop NARX neural network model with external input is constructed, and the training data is used to learn and optimize the hyperparameters of the NARX neural network model; finally, the open-loop NARX neural network model in the trained NARX neural network model is The architecture is converted into a closed-loop structure, the network neuron weights and thresholds are updated in real time, and the prediction of the bridge temperature within the required time step is recursively completed. Through the method provided by the invention, the data of the site meteorological parameters can be input into the model of the closed-loop structure, the bridge temperature data can be quickly obtained, and the prediction accuracy is high, which has strong engineering practicability.

The above are only preferred embodiments of the patent of the present invention, but the scope of protection of the patent of the present invention is not limited thereto. Any person familiar with the technical field can, within the scope disclosed by the patent of the present invention, proceed according to the patent of the present invention. Any equivalent substitution or change of the technical solution and its inventive concept shall fall within the scope of protection of the patent of the present invention.

Claims (6)

1. A bridge temperature prediction method based on NARX neural network, characterized in that the method includes: Perform sequence interpolation or extraction on the obtained bridge temperature data and corresponding meteorological data, adjust it to time series data with the same sampling rate, and use the adjusted meteorological data as alternative parameters; wherein the meteorological data includes multiple meteorological parameters; Meteorological parameters with high correlation with bridge temperature are selected from the alternative parameters through the maximum information coefficient as input features of the model; the selected meteorological parameters and the corresponding bridge temperature data are preprocessed respectively to obtain sample data; where , the sample data is divided into training data and test data; Use the training data to train the model to obtain a trained model; the model is a NARX neural network model with an open-loop architecture with external input; Convert the open-loop architecture in the trained model to a closed-loop structure, use the test data, and use the absolute error average index to evaluate the prediction performance of the closed-loop structure model; If the prediction result is within the preset range, the closed-loop structural model will obtain the predicted value of the bridge temperature based on the input meteorological data; Among them, the meteorological parameters with high correlation with the bridge temperature are selected from the alternative parameters through the maximum information coefficient as the input features of the model, specifically including: Assume the bridge temperature is x, and any corresponding meteorological parameter among the alternative parameters is y; The calculation formula using the maximum information coefficient is as follows: In the formula, I(x;y) is the mutual information coefficient, p(x,y) is the joint probability between variables x and y, a and b represent the number of grids divided into two-dimensional space in the x and y directions, The size setting of B satisfies a*b<B; The larger the value of MIC(x;y), the higher the correlation between x and y; if the correlation between x and y is high, then y is selected as the input feature of the model; The method of using the training data to train the model to obtain a trained model specifically includes: Divide the training data into three subsets, and use the three subsets for cross-validation. The first subset is the training set, which is used to calculate gradients and update network weights and biases to minimize the network loss function; the second subset is The set is the validation set, which monitors validation errors during the training process so that the network weights and biases are saved with the minimum validation set error; the third subset is the test set, which is used for final testing after training and validation, and outputs the performance of the model after completion of training Indicators; select Bayesian regularization algorithm or quantified conjugate gradient algorithm to optimize network weights and biases, and introduce Dropout technology to avoid under-fitting and over-fitting; During the training process of the model, the trainscg quantified conjugate gradient algorithm is selected, and the training data is input into the model for parameter learning and optimization of the neural network. If the verification error does not decrease, the iteration is stopped, otherwise the number of iterations is set. number of times; For the trained model, the training data is used to quantify the relationship strength between the predicted value and the target value using fitting optimization. If the relationship strength is not high enough, the model is retrained or the topology of the model is optimized to obtain a well-trained model. model; among them, the goodness-of-fit formula is as follows: In the formula, _yi is the bridge temperature data corresponding to the meteorological data _xi in the training data, as the target value; The predicted value output after inputting the meteorological data x _i into the model; R ² represents the strength of the relationship between the predicted value and the target value,/> is the mean of the target value, n is the number of samples input to the model; The use of the test data and the use of the absolute error average index to evaluate the prediction performance of the closed-loop structural model include: Input any meteorological data x _j in the test data into the closed-loop structural model to obtain the value output by the closed-loop structural model. After de-normalizing the output value, the predicted value of the bridge temperature is obtained. The bridge temperature corresponding to the meteorological data x _j in the test data is de-standardized and used as the target value y _j of the bridge temperature; The absolute error average index is used to reflect the error between the bridge temperature prediction value and the target value. The absolute error average is calculated as follows: If the average absolute error is near the preset value, it means that the prediction performance is good and can meet the engineering accuracy requirements. 2. The bridge temperature prediction method according to claim 1, characterized in that the selected meteorological parameters and corresponding bridge temperature data are separately preprocessed to obtain sample data, which specifically includes: Perform Z-Score standardization processing on the selected meteorological parameters respectively to obtain processed meteorological data as input data for the model; Perform Z-Score normalization processing on the bridge temperature data to obtain the processed bridge temperature data as the model output value; The processed meteorological data and the corresponding processed bridge temperature data constitute sample data. 3. The bridge temperature prediction method according to any one of claims 1 to 2, characterized in that the NARX neural network model adopts a network structure with output delayed feedback, including an input layer, a hidden layer and an output layer; Initialize the number of nodes and the number of feedback delays of the NARX neural network model. The input-output relationship of the NARX neural network model is: y(t)=f(x(t),x(t-1),…,x(t-d),y(t-1)) Among them, t represents the current moment, d represents the number of feedback delays, y(t) and y(t-1) represent the output of the network model at the current moment and the previous moment, x(t) and x(t-d) represent the current moment and the previous moment. The input data of the network model at time d. 4. The bridge temperature prediction method according to any one of claims 1 to 2, characterized in that the obtained meteorological data includes meteorological parameters such as zenith angle, temperature, relative humidity, wind speed, rainfall, and total cloud coverage. . 5. A bridge temperature prediction device based on NARX neural network, characterized in that the device includes: The alternative parameter acquisition module is used to perform sequence interpolation or extraction on the acquired bridge temperature data and corresponding meteorological data, adjust it to time series data with the same sampling rate, and use the adjusted meteorological data as alternative parameters; wherein, the meteorological data The data include multiple meteorological parameters; The sample data acquisition module is used to select meteorological parameters that are highly correlated with the bridge temperature from the alternative parameters through the maximum information coefficient as the input features of the model; and predict the selected meteorological parameters and the corresponding bridge temperature data respectively. Process to obtain sample data; wherein the sample data is divided into training data and test data; A model training module, used to train the model using the training data to obtain a trained model; the model is a NARX neural network model with an open-loop architecture with external input; The model prediction performance evaluation module is used to convert the open-loop architecture in the trained model into a closed-loop structure, and uses the test data to evaluate the prediction performance of the closed-loop structure model using the absolute error average index; The bridge temperature prediction module is used to use the closed-loop structural model to obtain the predicted value of the bridge temperature based on the input meteorological data if the prediction result is within the preset range; Among them, the meteorological parameters with high correlation with the bridge temperature are selected from the alternative parameters through the maximum information coefficient as the input features of the model, specifically including: Assume the bridge temperature is x, and any corresponding meteorological parameter among the alternative parameters is y; The calculation formula using the maximum information coefficient is as follows: In the formula, I(x;y) is the mutual information coefficient, p(x,y) is the joint probability between variables x and y, a and b represent the number of grids divided into two-dimensional space in the x and y directions, The size setting of B satisfies a*b<B; The larger the value of MIC(x;y), the higher the correlation between x and y; if the correlation between x and y is high, then y is selected as the input feature of the model; The method of using the training data to train the model to obtain a trained model specifically includes: Divide the training data into three subsets, and use the three subsets for cross-validation. The first subset is the training set, which is used to calculate gradients and update network weights and biases to minimize the network loss function; the second subset is The set is the validation set, which monitors validation errors during the training process so that the network weights and biases are saved with the minimum validation set error; the third subset is the test set, which is used for final testing after training and validation, and outputs the performance of the model after completion of training Indicators; select Bayesian regularization algorithm or quantified conjugate gradient algorithm to optimize network weights and biases, and introduce Dropout technology to avoid under-fitting and over-fitting; During the training process of the model, the trainscg quantified conjugate gradient algorithm is selected, and the training data is input into the model for parameter learning and optimization of the neural network. If the verification error does not decrease, the iteration is stopped, otherwise the number of iterations is set. number of times; For the trained model, the training data is used to quantify the relationship strength between the predicted value and the target value using fitting optimization. If the relationship strength is not high enough, the model is retrained or the topology of the model is optimized to obtain a well-trained model. model; among them, the goodness-of-fit formula is as follows: In the formula, _yi is the bridge temperature data corresponding to the meteorological data _xi in the training data, as the target value; The predicted value output after inputting the meteorological data x _i into the model; R ² represents the strength of the relationship between the predicted value and the target value,/> is the mean of the target value, n is the number of samples input to the model; The use of the test data and the use of the absolute error average index to evaluate the prediction performance of the closed-loop structural model include: Input any meteorological data x _j in the test data into the closed-loop structural model to obtain the value output by the closed-loop structural model. After de-normalizing the output value, the predicted value of the bridge temperature is obtained. The bridge temperature corresponding to the meteorological data x _j in the test data is de-standardized and used as the target value y _j of the bridge temperature; The absolute error average index is used to reflect the error between the bridge temperature prediction value and the target value. The absolute error average is calculated as follows: If the average absolute error is near the preset value, it means that the prediction performance is good and can meet the engineering accuracy requirements. 6. A storage medium storing a program, characterized in that when the program is executed by a processor, the bridge temperature prediction method according to any one of claims 1 to 4 is implemented.