Tree growth prediction method and system based on Elman dynamic recurrent neural network

Abstract

The invention discloses a tree growth prediction method and system based on an Elman dynamic recurrent neural network, and relates to the technical field of tree growth prediction, and the method comprises the steps: carrying out the collection of tree growth data, and carrying out the preprocessing of the collected data; performing feature extraction on the preprocessed data; constructing an Elman dynamic recurrent neural network model, and training the model; the trained model is evaluated and optimized, and the analysis result is accurate. According to the method, the dynamic change of time sequence data can be captured, high-quality input data and deep feature mining can be provided, the robustness and adaptability of the model are improved, accurate prediction of tree growth is realized, and finally, power transmission line managers are helped to take measures in time, and safe and stable operation of the power transmission line is ensured.

Description

Technical Field

[0001] The invention relates to the technical field of tree growth prediction, in particular to a tree growth prediction method and system based on an Elman dynamic recursive neural network. Background Art

[0002] Tree growth may cause the clearance width of the transmission line corridor to exceed the standard, thus affecting the normal operation of the transmission line and even causing safety hazards. In addition, traditional tree growth prediction methods are usually based on empirical models and statistical analysis, which makes it difficult to accurately predict future tree growth, especially when considering time series data and dynamic changes.

[0003] Using the Elman neural network, a tree growth prediction model can be established based on the historical data of tree growth and related environmental factors. The model can capture the dynamic changes and trends of time series data. The trained Elman neural network model can be used to predict the growth of trees in the future, helping power transmission line managers to take appropriate measures to manage them in a timely manner, and ensuring that the power transmission lines can operate safely and stably.

[0004] The tree growth prediction technology based on Elman dynamic recursive neural network has important application prospects in transmission line corridor management. It can help transmission line managers better understand the future growth trends of trees, formulate reasonable management strategies, and ensure the safe and stable operation of transmission lines. Summary of the invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is: how to solve the problem that it is difficult to accurately predict the future growth of trees using existing methods.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: a tree growth prediction method based on Elman dynamic recursive neural network, comprising: collecting tree growth data and preprocessing the collected data; extracting features from the preprocessed data; constructing an Elman dynamic recursive neural network model and training the model; evaluating and optimizing the trained model and analyzing the accuracy of the results.

[0008] As a preferred solution of the tree growth prediction method based on Elman dynamic recursive neural network described in the present invention, wherein: the tree growth data includes single tree height, breast diameter, crown width morphological growth record indicators and temperature, precipitation, soil nutrient environment data influencing factors; the preprocessing includes cleaning, missing value processing, normalization and time-space alignment; the time-space alignment includes time alignment and space alignment; the time alignment includes consistency of timestamps of all time series data; the space alignment includes if the data contains spatial information, it is necessary to ensure consistency of the spatial reference system; the data is divided into a training set, a validation set and a test set.

[0009] As a preferred solution of the tree growth prediction method based on Elman dynamic recurrent neural network of the present invention, the feature extraction includes time series decomposition and feature learning; the time series decomposition includes decomposing the growth record of each tree into trend, seasonality and random components, which are expressed as:

[0010] Y t =T t +R t +S t

[0011] Among them, Y t is the original time series value at time t, T t is the trend characteristic, R t is a seasonal characteristic, S t is a random feature; the feature learning includes using the automatic encoder in deep learning to perform nonlinear feature learning to obtain deep learning features.

[0012] As a preferred solution of the tree growth prediction method based on the Elman dynamic recursive neural network of the present invention, the Elman dynamic recursive neural network model includes an input layer, a hidden layer, an attention mechanism layer and an output layer; the input layer includes combining the original features, trend features, seasonal features and deep learning features in series as the input of the Elman dynamic recursive neural network, which is expressed as:

[0013] x t =concat(Y t , T t , R t , S t )

[0014] Among them, x t is the input of the Elman dynamic recurrent neural network, concat is the serial merging operation, Y t is the original feature, S t is a deep learning feature; the calculation of reset gate and update gate is expressed as:

[0015] r t =σ(W ir x t +U hr h t-1 +b r )

[0016] z t =σ(W iz x t +U hz h t-1 +b z )

[0017] Among them, r t is the reset gate output, W ir To reset the gate weight matrix, U hr h is the weight matrix that resets the gate to connect the hidden states at different time steps. t-1 is the hidden layer state of the previous time step, b r is the bias term for the reset gate, z t is the update gate output, W iz is the weight matrix of the update gate, U hz is the weight matrix that updates the gate to connect the hidden states at different time steps, b z is the bias term of the update gate; the hidden layer includes calculating the candidate hidden state, which is expressed as:

[0018]

[0019] in, is the candidate hidden state, W ih is the hidden layer weight matrix, U hh is the weight matrix of the hidden layer connecting the hidden states at different time steps, ⊙ is the Hadamard product, b h is the bias term of the hidden layer; the updated hidden layer output is expressed as:

[0020]

[0021] Among them, h t is the hidden layer state at the current time step.

[0022] As a preferred solution of the tree growth prediction method based on Elman dynamic recurrent neural network described in the present invention, the attention mechanism layer includes encoding environmental factors, calculating correlation scores, calculating attention weights, and obtaining attention-weighted input features; the environmental factor encoding is expressed as:

[0023] c t =B(En t )

[0024] Among them, ct is the encoded data, B is the encoding operation, En t is environmental data; the calculated correlation score is expressed as:

[0025] e t，i =v T tanh(W a h t +U a x t，i +V a c t +b a )

[0026] Among them, e t，i To consider the correlation between the input features and hidden states of environmental factors, v is the parameter vector used to calculate the attention score, W a is the weight matrix for calculating the relevance score, U a is the weight matrix connecting the hidden states at different time steps in the attention mechanism layer, V a is the weight matrix from environmental factors to attention scores, b α is the bias term for calculating the relevance score, x t，i is the i-th element of the input vector at time step t; the calculated attention weight is expressed as:

[0027]

[0028] Among them, α t，i is the attention weight; the attention-weighted input feature is expressed as:

[0029]

[0030] in, is the input feature after attention weighting; the output layer calculates the output prediction as:

[0031]

[0032] Among them, y t is the output prediction, W ho is the weight matrix of the output layer, b o is the bias of the output layer, and g is the activation function.

[0033] As a preferred solution of the tree growth prediction method based on the Elman dynamic recurrent neural network of the present invention, the model training includes initializing parameters, using conventional data and adversarial data for alternating training, and updating parameters until the convergence condition is met or the predetermined number of iterations is reached; the loss function of conventional data training is calculated, which is expressed as:

[0034]

[0035] Among them, L is the loss function of regular data training, N is the length of the time period, that is, the total number of data points, and Y t is the true output at time t; a dynamic learning rate based on error sensitivity is introduced, and the dynamic learning rate update strategy is expressed as:

[0036] η t =η t-1 ·exp(-γ·ΔL t )

[0037] Among them, η t-1 is the learning rate of the previous time step, η t is the learning rate of the current time step, γ is the scaling parameter, ΔL t is the loss difference between two consecutive iterations; the model parameters are updated using dynamic learning rate and gradient, expressed as:

[0038]

[0039] Among them, θ new1 are the model parameters after regular data update, θ old are the model parameters before updating, is the loss function with respect to θ old The gradient of ; generating adversarial samples is expressed as:

[0040]

[0041] in, is an adversarial sample, ∈ is a small constant, sign is the sign function of the gradient, is the loss function L with respect to the input x t Gradient of ; Use the model to generate predictions based on adversarial examples Calculate the adversarial loss, expressed as:

[0042]

[0043] Among them, L adv is the loss function of the adversarial sample, is the prediction generated by the model based on the adversarial sample; updates the dynamic learning rate and updates the model parameters, expressed as:

[0044]

[0045] in, is the difference in loss of adversarial examples in two consecutive iterations, is the adversarial sample loss function with respect to θ new1 The gradient of θ new2To counter the updated model parameters.

[0046] As a preferred solution of the tree growth prediction method based on Elman dynamic recursive neural network described in the present invention, wherein: the evaluation and optimization of the trained model includes time series prediction evaluation and space series prediction evaluation; the time series prediction evaluation includes dividing the test set into multiple continuous prediction windows in chronological order, using the trained model to predict each prediction window, calculating the error between the predicted value and the true value, summarizing the error indicators of all prediction windows, and analyzing the performance of the model on the entire time series; the space series prediction evaluation includes dividing the spatial distribution data of the test set into different regional grids, using the model to predict each region or grid, calculating the error between the prediction result and the true distribution, summarizing the error indicators of all regions or grids, so as to evaluate the performance of the model in the overall spatial distribution prediction.

[0047] Another object of the present invention is to provide a system for tree growth prediction method based on Elman dynamic recursive neural network, which can solve the problem of tree growth prediction based on Elman dynamic recursive neural network by constructing a tree growth prediction system.

[0048] In order to solve the above technical problems, the present invention provides the following technical solutions: a tree growth prediction system based on Elman dynamic recursive neural network, including a data acquisition and processing module, a feature extraction module, a model construction module and an evaluation and optimization module; the data acquisition and processing module is used to collect tree growth data and preprocess the collected data; the feature extraction module is used to extract features from the preprocessed data; the model construction module is used to construct an Elman dynamic recursive neural network model and train the model; the evaluation and optimization module is used to evaluate and optimize the trained model and analyze the accuracy of the results.

[0049] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the tree growth prediction method based on the Elman dynamic recursive neural network are implemented.

[0050] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the tree growth prediction method based on the Elman dynamic recursive neural network as described above.

[0051] The beneficial effects of the present invention are as follows: the tree growth prediction method based on the Elman dynamic recursive neural network provided by the present invention can capture the dynamic changes of time series data, provide high-quality input data and deep feature mining, improve the robustness and adaptability of the model, and realize accurate prediction of tree growth, ultimately helping power transmission line managers to take timely measures to ensure the safe and stable operation of the transmission lines, and effectively solving the problem that traditional methods are difficult to accurately predict tree growth. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0053] Figure 1 A flowchart of a tree growth prediction method based on an Elman dynamic recurrent neural network is provided for the first embodiment of the present invention.

[0054] Figure 2 A structural diagram of a tree growth prediction system based on an Elman dynamic recursive neural network provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the above-mentioned objects, 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.

[0056] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0057] Example 1

[0058] Reference Figure 1 , which is the first embodiment of the present invention, provides a tree growth prediction method based on the Elman dynamic recursive neural network, including: collecting tree growth data and preprocessing the collected data; extracting features from the preprocessed data; building an Elman dynamic recursive neural network model and training the model; evaluating and optimizing the trained model and analyzing the accuracy of the results.

[0059] S1. Collect tree growth data and pre-process the collected data.

[0060] Tree growth data include individual tree height, breast diameter, crown width morphological growth record indicators as well as temperature, precipitation, and soil nutrient environmental data influencing factors.

[0061] Data preprocessing checks the collected multi-source heterogeneous data to see if they are complete and accurate, whether there are missing values, outliers or duplicate data, and performs corresponding cleaning, missing value processing, normalization and spatiotemporal alignment to form a unified input data format. According to the growth law of trees, time series data is constructed to reflect the dynamic changes of influencing factors over time.

[0062] Normalization: When using the Elman dynamic recurrent neural network for transmission line corridor tree growth prediction, it is very important to normalize the data. By standardizing the data, the model can learn more effectively, reach a stable state faster, and improve its predictive ability for unknown data.

[0063] Minimum-maximum normalization: Minimum-maximum normalization is a data preprocessing technique that adjusts each value in the original data set to the range of 0 to 1 through linear transformation. The specific processing formula is:

[0064] x=[x-min(x)] / min(x)-min(x)]

[0065] Among them, x' is the normalized data, and x is the original collected data.

[0066] Z-score standardization: Score standardization is a method of processing data. The difference between each data point and the mean of the data set is calculated, and then the difference is divided by the standard deviation to convert the original data into a new distribution with a mean of 0 and a standard deviation of 1. The specific processing formula is:

[0067] x'=(x-μ) / σ

[0068] Among them, x' is the data after standardization, x is the original collected data, σ is the standard deviation of the original data, and μ is the mean of the original data. This transformation means that the data point x' represents the standard deviation of the point relative to the mean in the original data distribution, that is, how much it deviates from the mean.

[0069] Generally speaking, minimum-maximum normalization is suitable for situations where the data distribution is relatively uniform, Z-score normalization is suitable for situations where the data distribution is relatively normal, and score correction normalization is suitable for situations where the data distribution characteristics need to be retained. For data with different characteristics, different normalization methods may be required to ensure that the data of each feature is within an appropriate range. For each numerical variable, use the selected normalization method to convert it to a standard format.

[0070] Spatiotemporal alignment includes time alignment and space alignment; time alignment includes consistency of timestamps of all time series data; space alignment includes ensuring consistency of spatial reference systems if the data contains spatial information, and obtaining preprocessed data Y t ; Divide the data into training set, validation set and test set.

[0071] S2. Extract features from the preprocessed data.

[0072] Feature extraction includes time series decomposition and feature learning.

[0073] Time series decomposition involves decomposing each tree's growth record into trend, seasonal, and random components, expressed as:

[0074] Y t =T t +R t +S t

[0075] Among them, Y t is the original time series value at time t, that is, the preprocessed data, T t is a trend feature, indicating the long-term trend of data, R t is a seasonal feature, indicating a periodically repeated seasonal pattern, S t is a random feature, representing the remainder that cannot be explained by trend or seasonality.

[0076] Feature learning uses the autoencoder in deep learning to perform nonlinear feature learning to obtain deep learning features.

[0077] Trend features: reflect the overall trend of tree growth in the long term, such as an increase or decrease in growth rate. Seasonal features: depict the cyclical changes in tree growth in different seasons, such as faster growth in spring than in autumn. Random / residual features: capture abnormal changes and unpredictable fluctuations in the data, which may be caused by external environmental factors such as weather conditions or pests and diseases. Deep learning features: advanced features extracted by autoencoders that can represent nonlinear complex relationships and inherent patterns in tree growth data.

[0078] S3. Build the Elman dynamic recursive neural network model and train the model.

[0079] The Elman dynamic recurrent neural network model includes an input layer, a hidden layer, an attention mechanism layer and an output layer.

[0080] Input Layer:

[0081] The original features, trend features, seasonal features and deep learning features are combined in series as the input of the Elman dynamic recurrent neural network, expressed as:

[0082] x t =concat(Y t , T t , R t , S t )

[0083] Among them, x t is the input of the Elman dynamic recurrent neural network, is a comprehensive input vector containing various features, concat is a serial merging operation, and Y t is the original feature, S t For deep learning features.

[0084] The calculation of reset gate and update gate is expressed as:

[0085] r t =σ(W ir x t +U hr h t-1 +b r )

[0086] z t =σ(W iz x t +U hz h t-1 +b z )

[0087] Among them, r t The reset gate output controls the reset of the past state information. ir To reset the gate weight matrix, U hr h is the weight matrix that resets the gate to connect the hidden states at different time steps. t-1 is the hidden layer state of the previous time step, b r is the bias term for the reset gate, z t To update the gate output, decide how much old state to keep, W iz is the weight matrix of the update gate, U hz is the weight matrix that connects the hidden states at different time steps for the update gate, b z is the bias term for the update gate.

[0088] Hidden Layer:

[0089] Calculate the candidate hidden state, expressed as:

[0090]

[0091] in, is the candidate hidden state, W ih is the hidden layer weight matrix, U hh is the weight matrix of the hidden layer connecting the hidden states at different time steps, ⊙ is the Hadamard product, b h is the bias term of the hidden layer.

[0092] Update the hidden layer output as:

[0093]

[0094] Among them, h t is the hidden layer state at the current time step.

[0095] Attention mechanism layer:

[0096] Encode environmental factors, calculate relevance scores, calculate attention weights, and obtain attention-weighted input features;

[0097] Environmental factors are coded as follows:

[0098] c t =B(En t )

[0099] Among them, c t is the encoded data, B is the encoding operation, En t for environmental data;

[0100] The calculation of the relevance score is expressed as:

[0101] e t，i =v T tanh(W a h t +U a x t，i +V a ct+b a )

[0102] Among them, e t，i To consider the correlation between the input features and hidden states of environmental factors, v is the parameter vector used to calculate the attention score, W a is the weight matrix for calculating the relevance score, U a The weight matrix that connects the hidden states of different time steps for the attention mechanism layer is used to convert the hidden layer state h of the previous time step t-1 The relevance score e of information incorporated into the attention mechanism t，i In the calculation of V a is the weight matrix from environmental factors to attention scores, b a is the bias term for calculating the relevance score, x t，i is the i-th element of the input vector at time step t.

[0103] The calculation of attention weight is expressed as:

[0104]

[0105] Among them, α t，i is the attention weight.

[0106] The attention-weighted input feature is represented as:

[0107]

[0108] in, It is the input feature after attention weighting.

[0109] The output layer calculates the output prediction as:

[0110]

[0111] Among them, y t is the output prediction, W ho is the weight matrix of the output layer, b o is the bias of the output layer, and g is the activation function.

[0112] Train the model:

[0113] Initialize the parameters, use regular data and adversarial data for alternating training, and update the parameters until the convergence condition is met or the predetermined number of iterations is reached. Calculate the loss function of regular data training, expressed as:

[0114]

[0115] Among them, L is the loss function of regular data training, N is the length of the time period, that is, the total number of data points, and Y t is the actual output at time t.

[0116] A dynamic learning rate based on error sensitivity is introduced, and the dynamic learning rate update strategy is expressed as:

[0117] η t =η t-1 ·exp(-γ·ΔL t )

[0118] Among them, η t-1 is the learning rate of the previous time step, η t is the learning rate of the current time step, γ is a positive scaling parameter that controls the sensitivity of the learning rate change, ΔL t is the loss difference between two consecutive iterations, defined as L t-1 -L t .

[0119] Update the model parameters using dynamic learning rate and gradient, expressed as:

[0120]

[0121] Among them, θ new1 are the model parameters after regular data update, θ old are the model parameters before updating, is the loss function with respect to θ old gradient.

[0122] Generate adversarial samples, expressed as:

[0123]

[0124] in, is an adversarial sample, ∈ is a small constant used to control the amplitude of the perturbation, sign is the sign function of the gradient, is the loss function L with respect to the input x t gradient.

[0125] Use the model to generate predictions based on adversarial examples Calculate the adversarial loss, expressed as:

[0126]

[0127] Among them, L adv is the loss function of the adversarial sample, is the prediction generated by the model based on the adversarial example.

[0128] Update the dynamic learning rate and model parameters, expressed as:

[0129]

[0130] in, is the difference in loss of adversarial examples in two consecutive iterations, is the adversarial sample loss function with respect to θ new1 The gradient of θ new2 To counter the updated model parameters.

[0131] S4. Evaluate and optimize the trained model and analyze the accuracy of the results.

[0132] The evaluation and optimization of the trained model includes time series prediction evaluation and spatial series prediction evaluation.

[0133] Time series forecasting evaluation divides the test set into multiple continuous forecast windows in chronological order, uses the trained model to predict each forecast window, and calculates the error between the predicted value and the true value. Commonly used error indicators include mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE). The error indicators of all forecast windows are summarized to analyze the performance of the model on the entire time series.

[0134] Spatial sequence prediction evaluation divides the spatial distribution data of the test set into different regional grids, uses the model to predict each region or grid, and calculates the error between the prediction result and the true distribution. Spatial correlation metrics (such as Moran's I), spatial error or similarity metrics (such as spatial cross-validation metrics) can be used to summarize the error indicators of all regions or grids to evaluate the performance of the model in overall spatial distribution prediction.

[0135] Example 2

[0136] Reference Figure 2 , which is the second embodiment of the present invention, is different from the previous embodiment in that it provides a tree growth prediction system based on the Elman dynamic recursive neural network, including: a data acquisition and processing module, a feature extraction module, a model building module and an evaluation and optimization module.

[0137] The data collection and processing module is used to collect tree growth data and pre-process the collected data.

[0138] The feature extraction module is used to extract features from the preprocessed data.

[0139] The model building module is used to build the Elman dynamic recurrent neural network model and train the model.

[0140] The evaluation and optimization module is used to evaluate and optimize the trained model and analyze the accuracy of the results.

[0141] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0142] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0143] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0144] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0145] Example 3

[0146] The third embodiment of the present invention is different from the first two embodiments in that it is used to verify the technical effects adopted in the present invention in order to verify the real effects of the method.

[0147] Dataset: Tree growth data collected from multiple transmission line corridors, including growth records of height, diameter at breast height, and crown width, as well as data on environmental factors such as temperature, precipitation, and soil nutrients.

[0148] The ratio of training data to test data is: 70% training data, 15% validation data, and 15% test data.

[0149] Evaluation indicators: prediction accuracy (PredictionAccuracy), response time (ResponseTime), model stability (ModelStability).

[0150] Traditional methods: Tree growth prediction methods based on empirical models and statistical analysis.

[0151] This embodiment uses the traditional method and our invented method to perform detection at the same time, and the detection comparison results are shown in the following table:

[0152] Table 1 Overall comparison table

[0153]

[0154]

[0155] Through the above comparison, it can be clearly seen that the method of the present invention has significant advantages in capturing the dynamic changes of time series data and improving the robustness and adaptability of the model. It ultimately achieves accurate prediction of tree growth, helps transmission line managers take timely measures to ensure the safe and stable operation of transmission lines, and effectively solves the problem that traditional methods are difficult to accurately predict tree growth.

[0156] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A tree growth prediction method based on Elman dynamic recursive neural network, characterized by: include, Collect tree growth data and pre-process the collected data; Perform feature extraction on the preprocessed data; Construct the Elman dynamic recursive neural network model and train the model; Evaluate and optimize the trained model and analyze the accuracy of the results.

2. The tree growth prediction method based on Elman dynamic recursive neural network as claimed in claim 1, characterized in that: The tree growth data include the growth record indexes of individual tree height, breast diameter, and crown width, as well as the influencing factors of temperature, precipitation, and soil nutrient environment data; The preprocessing includes cleaning, missing value processing, normalization and spatiotemporal alignment; The spatiotemporal alignment includes time alignment and space alignment; The time alignment includes consistency of timestamps of all time series data; The spatial alignment includes ensuring that the spatial reference system is consistent if the data contains spatial information; Divide the data into training, validation, and test sets.

3. The tree growth prediction method based on Elman dynamic recursive neural network as claimed in claim 2, characterized in that: The feature extraction includes time series decomposition and feature learning; The time series decomposition includes decomposing each tree's growth record into trend, seasonal and random components, expressed as, Y t =T t +R t +S t Among them, Y t is the original time series value at time t, T t is the trend characteristic, R t is a seasonal characteristic, S t is a random feature; The feature learning includes using an autoencoder in deep learning to perform nonlinear feature learning to obtain deep learning features.

4. The tree growth prediction method based on Elman dynamic recursive neural network as claimed in claim 3, characterized in that: The Elman dynamic recursive neural network model comprises an input layer, a hidden layer, an attention mechanism layer and an output layer; The input layer includes combining the original features, trend features, seasonal features and deep learning features in series as the input of the Elman dynamic recurrent neural network, expressed as: x t =concat(Y t ,T t ,R t ,S t ) Among them, x t is the input of the Elman dynamic recurrent neural network, concat is the serial merging operation, Y t is the original feature, S t For deep learning features; The calculation of reset gate and update gate is expressed as, r t =σ(W ir x t +U hr h t-1 +b r ) z t =σ(W iz x t +U hz h t-1 +b z ) Among them, r t is the reset gate output, W ir To reset the gate weight matrix, U hr h is the weight matrix that resets the gate to connect the hidden states at different time steps. t-1 is the hidden layer state of the previous time step, b r is the bias term for the reset gate, z t is the update gate output, W iz is the weight matrix of the update gate, U hz is the weight matrix that connects the hidden states at different time steps for the update gate, b z is the bias term of the update gate; The hidden layer includes calculating candidate hidden states, expressed as, in, is the candidate hidden state, W ih is the hidden layer weight matrix, U hh is the weight matrix of the hidden layer connecting the hidden states at different time steps, ⊙ is the Hadamard product, b h is the bias term of the hidden layer; Update the hidden layer output as, Among them, h t is the hidden layer state at the current time step.

5. The tree growth prediction method based on Elman dynamic recursive neural network as claimed in claim 4, characterized in that: The attention mechanism layer includes encoding environmental factors, calculating relevance scores, calculating attention weights, and obtaining attention-weighted input features; The environmental factor coding is expressed as: c t =B(A t ) Among them, c t is the encoded data, B is the encoding operation, En t for environmental data; The calculated relevance score is expressed as, e t，i =v T tanh(W a h t +U a x t，i +V a c t +b a ) Among them, e t，i To consider the correlation between the input features and hidden states of environmental factors, v is the parameter vector used to calculate the attention score, W a is the weight matrix for calculating the relevance score, U a is the weight matrix connecting the hidden states at different time steps in the attention mechanism layer, V a is the weight matrix from environmental factors to attention scores, b a is the bias term for calculating the relevance score, x t，i is the i-th element of the input vector at time step t; The calculated attention weight is expressed as, Among them, α t，i is the attention weight; The attention-weighted input feature is expressed as, in, The input features after attention weighting; The output layer computes the output prediction as, Among them, y t is the output prediction, W ho is the weight matrix of the output layer, b o is the bias of the output layer, and g is the activation function.

6. The tree growth prediction method based on Elman dynamic recursive neural network as claimed in claim 5, characterized in that: The training of the model includes initializing parameters, using conventional data and adversarial data for alternating training, and updating parameters until a convergence condition is met or a predetermined number of iterations is reached; Calculate the loss function for regular data training, expressed as, Among them, L is the loss function of regular data training, N is the length of the time period, that is, the total number of data points, and Y t is the real output at time t; A dynamic learning rate based on error sensitivity is introduced, and the dynamic learning rate update strategy is expressed as: or t =the t-1 ·exp(-γ·ΔL t ) Among them, η t-1 is the learning rate of the previous time step, η t is the learning rate of the current time step, γ is the scaling parameter, ΔL t is the loss difference between two consecutive iterations; Update the model parameters using dynamic learning rate and gradient, expressed as, Among them, θ new1 are the model parameters after regular data update, θ old are the model parameters before updating, is the loss function with respect to θ old The gradient of Generate adversarial samples, expressed as, in, is an adversarial sample, ∈ is a small constant, sign is the sign function of the gradient, is the loss function L with respect to the input x t The gradient of Use the model to generate predictions based on adversarial examples Calculate the adversarial loss, expressed as, Among them, L adv is the loss function of the adversarial sample, The predictions generated by the model based on the adversarial examples; Update the dynamic learning rate and model parameters, expressed as, in, is the difference in loss of adversarial examples in two consecutive iterations, is the adversarial sample loss function with respect to θ new1 The gradient of θ new2 To counter the updated model parameters.

7. The tree growth prediction method based on Elman dynamic recursive neural network as claimed in claim 6, characterized in that: The evaluation and optimization of the trained model includes time series prediction evaluation and spatial series prediction evaluation; The time series prediction evaluation includes dividing the test set into multiple continuous prediction windows in chronological order, using the trained model to predict each prediction window, calculating the error between the predicted value and the true value, summarizing the error indicators of all prediction windows, and analyzing the performance of the model on the entire time series; The spatial sequence prediction evaluation includes dividing the test set spatial distribution data into different regional grids, using the model to predict each region or grid, calculating the error between the prediction result and the true distribution, and summarizing the error indicators of all regions or grids to evaluate the performance of the model in overall spatial distribution prediction.

8. A system using the tree growth prediction method based on Elman dynamic recursive neural network as claimed in any one of claims 1 to 7, characterized in that: Including data acquisition and processing module, feature extraction module, model building module and evaluation and optimization module; The data collection and processing module is used to collect tree growth data and pre-process the collected data; The feature extraction module is used to extract features from the preprocessed data; The model building module is used to build an Elman dynamic recursive neural network model and train the model; The evaluation and optimization module is used to evaluate and optimize the trained model and analyze the accuracy of the results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the tree growth prediction method based on the Elman dynamic recurrent neural network described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the tree growth prediction method based on the Elman dynamic recurrent neural network described in any one of claims 1 to 7 are implemented.

Images