Mine ecological environment restoration and management assessment method and device
Through the features extraction neural network and autoencoder, the problems of insufficient data and poor diversity in the restoration of the mine ecological environment are solved, the evaluation accuracy and efficiency are improved, and the evaluation is achieved efficient assessment of the mine ecological environment restoration.
Patent Information
- Application Number
- CN202411036118.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-07-31
AI Technical Summary
In the recovery of mining ecological environment, the existing technology has problems such as insufficient data volume, poor diversity and weak model generalization capabilities in data collection, processing and analysis, resulting in low accuracy, long cycle and high cost of restoration effect evaluation.
Feature extraction neural network is used to extract ecological parameters, and the neural network parameters are initialized in combination with complexity constraint methods. The preset training sample set data is expanded, and the dimensionality reduction process is used to use the autoencoder, and the probability neural network is used for classification recognition. Evaluate the model to determine the mine restoration situation.
Even if limited data is collected, key features can be extracted, evaluation accuracy can be improved, feature extraction models can be overfitted, data diversity can be enhanced, feature representation and dimensionality reduction processes can be optimized, and evaluation accuracy and efficiency can be improved.
Smart Images

Figure CN118569506B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to a method and device for assessing the restoration and management of a mine ecological environment. Background Art
[0002] In mining activities, multiple ecosystems such as soil, water quality, vegetation, and atmosphere are often damaged, and traditional environmental restoration methods usually face problems such as difficult to evaluate the effect, long cycle, and high cost. In order to more effectively monitor and evaluate the restoration process of the mine ecological environment, it is necessary to collect and analyze a large amount of ecological data.
[0003] However, existing technologies have several key limitations in data collection, processing and analysis, such as insufficient data volume, poor data diversity, and weak model generalization ability. These problems reduce the accuracy of repair effect evaluation, extend the repair cycle, and increase governance costs. Summary of the invention
[0004] In view of this, the purpose of the present invention is to provide a method and device for evaluating the restoration and management of the ecological environment of a mine, which can extract the corresponding key features and improve the evaluation accuracy even if limited data is collected.
[0005] In the first aspect, an embodiment of the present invention provides a mine ecological environment restoration and governance assessment method, wherein the method includes: monitoring the ecological parameters of the mine; extracting features of the ecological parameters using a pre-constructed feature extraction model, and determining key features in the ecological parameters; wherein the feature extraction model uses a feature extraction neural network for feature extraction, and the feature extraction neural network uses a complexity constraint method to initialize the neural network parameters; the feature extraction model is constructed based on a preset training sample set, and the training sample set is constructed after data expansion based on the data state of the initial sample set; key features are classified and identified through a pre-constructed assessment model, and the restoration status of the mine is determined based on the results of the classification and identification.
[0006] In combination with the first aspect, an embodiment of the present invention also provides a first implementation method of the first aspect, wherein the above method also includes: performing dimensionality reduction processing on the key features through a preset autoencoder algorithm to generate a low-dimensional feature representation corresponding to the key features; in the feature space where the low-dimensional feature representation is located, defining the low-dimensional feature representation as a target node, and adjusting the low-dimensional feature representation based on the adjacent nodes corresponding to the target node and their corresponding structural relationships; reconstructing the low-dimensional feature representation through a preset decoder to obtain the reconstructed key features.
[0007] In combination with the first aspect, an embodiment of the present invention also provides a second implementation of the first aspect, wherein the step of adjusting the low-dimensional feature representation based on the adjacent nodes corresponding to the target node and their corresponding structural relationships includes: calculating the node connection strength based on the adjacent nodes of the target node and their corresponding node similarities; summing the target node according to the node connection strength, adjacent nodes, and the normally distributed noise corresponding to the target node to adjust the target node; wherein the normally distributed noise includes the noise injection amount corresponding to the target node, and the noise injection amount is determined based on the variability corresponding to the target node.
[0008] In combination with the first aspect, an embodiment of the present invention also provides a third implementation of the first aspect, wherein a method for constructing a feature extraction model includes: obtaining a preset feature extraction neural network, initializing the neural network parameters of the feature extraction neural network based on a normally distributed random number from the input layer dimension to the output layer dimension of the feature extraction neural network; using the neural network parameters as fireflies; and, based on the correction factor learning rate and regularization factor corresponding to the feature extraction neural network, and the gradient of the loss function with respect to the neural network parameters, calculating the parameter weights corresponding to the neural network parameters, and obtaining the position vector corresponding to the firefly; inputting a preset training sample set into the feature extraction neural network, and using the loss function to evaluate the network performance of the feature extraction neural network to obtain the loss function value; calculating the relative attraction between fireflies based on the loss function value, and optimizing the position vector according to the relative attraction; until the feature extraction neural network meets the preset iteration conditions, constructing a feature extraction model based on the feature extraction neural network.
[0009] In combination with the first aspect, an embodiment of the present invention also provides a fourth implementation of the first aspect, wherein a method for constructing an evaluation model includes: obtaining a preset probabilistic neural network, inputting a pre-constructed training sample set into the probabilistic neural network for forward propagation, and obtaining an output probability of the training sample set corresponding to each preset category; weighting the output probability based on the inverse frequency of the number of category samples of the preset category; calculating the gradient of the network parameters of the probabilistic neural network based on the output probability, and adjusting the learning rate of the probabilistic neural network according to a preset fine-tuning strategy; and updating the network parameters based on a preset processing function; wherein the processing function is used to characterize the ability to process data of different categories; detecting the category balance corresponding to the current category probability, and adjusting the inverse frequency based on the detection result; until the probabilistic neural network reaches the preset iteration requirement and the category probability meets the preset category balance requirement, an evaluation model is constructed based on the probabilistic neural network.
[0010] In combination with the first aspect, an embodiment of the present invention also provides a fifth implementation of the first aspect, wherein the step of weighting the output probability based on the inverse frequency of the number of category samples of a preset category includes: performing ratio processing on the maximum number of samples in all categories and the number of category samples of the preset category to obtain the inverse frequency weight; weighting the output probability based on a preset weighted loss function and the inverse frequency weight so that the output probability meets the preset category balance requirement.
[0011] In combination with the first aspect, an embodiment of the present invention also provides a sixth implementation of the first aspect, wherein a method for constructing a training sample set includes: obtaining pre-collected initial mine samples; the initial mine samples include ecological parameters of the mine; labeling the initial mine samples to construct an initial sample set; the labeled categories are used to indicate the restoration status of the mine; extracting the data status of the initial sample set, and performing data expansion on the initial sample set based on the data status through a preset generative adversarial network to generate expanded samples; merging the expanded samples with the initial sample set to generate a training sample set.
[0012] In combination with the first aspect, an embodiment of the present invention also provides a seventh implementation method of the first aspect, wherein the above method also includes: calculating the average error rate corresponding to the expanded samples; adjusting the discriminator output of the generative adversarial network based on the average error rate; and updating the discriminator parameters based on the adjusted discriminator output.
[0013] In combination with the first aspect, an embodiment of the present invention also provides an eighth implementation of the first aspect, wherein the above method also includes: evaluating the adversarial balance between the generator and the discriminator of the generative adversarial network, and updating the parameters of the generator and the parameters of the discriminator based on the evaluation results.
[0014] In the second aspect, an embodiment of the present invention also provides a mine ecological environment restoration and management assessment device, wherein the device includes: a data monitoring module for monitoring the ecological parameters of the mine; a feature extraction module for using a pre-constructed feature extraction model to extract features of the ecological parameters and determine the key features in the ecological parameters; wherein the feature extraction model uses a feature extraction neural network for feature extraction, and the feature extraction neural network uses a complexity constraint method to initialize the neural network parameters; the feature extraction model is constructed based on a preset training sample set, and the training sample set is constructed after data expansion based on the data state of the initial sample set; an execution module is used to classify and identify key features through a pre-constructed evaluation model, and determine the restoration status of the mine based on the results of the classification and identification.
[0015] The embodiment of the present invention provides a mine ecological environment restoration and management assessment method and device. After collecting the ecological parameters of the mine, the feature extraction model performs feature extraction, and the assessment model performs classification prediction, wherein the feature extraction model uses a feature extraction neural network for feature extraction, and the feature extraction neural network uses a complexity constraint method to initialize the neural network parameters, which can avoid overfitting of the feature extraction neural network. In addition, the feature extraction model is constructed based on a preset training sample set, and the training sample set is constructed after data expansion based on the data state of the initial sample set, wherein the data expansion is performed based on the data state of the reference sample set, which can reduce the generation of duplicate data and enhance the diversity of data. Based on this, even if limited data is collected, the corresponding key features can be extracted to improve the assessment accuracy.
[0016] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.
[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0019] Figure 1 A flowchart of a mine ecological environment restoration and management assessment method provided by an embodiment of the present invention;
[0020] Figure 2 A flowchart of another mine ecological environment restoration and management assessment method provided by an embodiment of the present invention;
[0021] Figure 3 A flow chart of a method for constructing a training sample set provided by an embodiment of the present invention;
[0022] Figure 4 A schematic diagram of the structure of a mine ecological environment restoration and management assessment device provided by an embodiment of the present invention;
[0023] Figure 5 A schematic diagram of the structure of another mine ecological environment restoration and management assessment device provided by an embodiment of the present invention;
[0024] Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described in combination with the embodiments below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0026] The present invention provides a method and device for evaluating the restoration and management of the ecological environment in a mine. Even if limited data is collected, the corresponding key features can be extracted to improve the evaluation accuracy.
[0027] To facilitate understanding of this embodiment, a mine ecological environment restoration and management evaluation method disclosed in an embodiment of the present invention is first described in detail, see Figure 1 A schematic diagram of a method for assessing the restoration and management of a mine ecological environment is shown, and the method may include the following steps:
[0028] Step S102, monitoring the ecological parameters of the mine.
[0029] In a specific implementation, the data of the present invention is collected from on-site sensors in the mine, and ecological parameters such as soil moisture, surface temperature, and atmospheric composition are collected in real time. In one embodiment, the attributes of the data include: ax: surface temperature, in degrees Celsius; ay: soil moisture, expressed in percentage; az: atmospheric CO2 concentration, in ppm; aw: vegetation coverage, expressed in percentage; av: water quality index, expressed in numerical value; au: surface mineral distribution, expressed in numerical value; at: soil heavy metal content, in mg / kg; as: surrounding environmental noise, expressed in numerical value; ar: spatial location information, expressed in longitude and latitude; aq: timestamp, in the format of YYYY-MM-DD, HH:MM:SS, used to indicate the year, month, and day of data collection, and the corresponding collection time.
[0030] It should be emphasized that the above data embodiment is only used to illustrate a data format and type of the present invention. In actual applications, the attributes of data are usually more than 10 attributes, and the number of data attributes may reach dozens or even hundreds.
[0031] Step S104, extracting features from ecological parameters using a pre-built feature extraction model to determine key features in the ecological parameters.
[0032] Step S106, classifying and identifying key features through a pre-built evaluation model, and determining the restoration status of the mine based on the classification and identification results.
[0033] After obtaining the ecological parameters, the data is subjected to feature extraction through a pre-built feature extraction model to determine the key features in the ecological parameters. Furthermore, the key features are classified and identified by the evaluation model to determine the classification category to which the key features belong, and then the restoration status of the mine is determined. Among them, the evaluation model classifies the key features and outputs the predicted probability corresponding to each category, thereby determining the restoration status corresponding to the mine. In an embodiment of the present invention, the classification categories include: unrepaired restoration status, repairing restoration status, and repaired restoration status, a total of 3 categories.
[0034] In specific implementation, traditional neural networks may overfit in complex or small data situations and lack an effective complexity control mechanism. The feature extraction model of the embodiment of the present invention uses a feature extraction neural network for feature extraction, and the feature extraction neural network uses a complexity constraint method to initialize the neural network parameters, which can avoid overfitting of the feature extraction neural network.
[0035] In addition, the feature extraction model is constructed based on a preset training sample set, and the training sample set is constructed after data expansion based on the data state of the initial sample set. Among them, data expansion is performed based on the data state of the reference sample set, which can reduce the generation of duplicate data and enhance data diversity. Based on this, even if limited data is collected, the corresponding key features can be extracted, thereby improving the evaluation accuracy.
[0036] Furthermore, based on the above embodiments, the embodiments of the present invention also provide another method for evaluating the ecological environment restoration and management of mines. In order to improve the model processing efficiency, the embodiments of the present invention also perform data dimensionality reduction on key features after feature extraction of ecological parameters, and further optimize and simplify the feature representation through the self-encoding process of the data dimensionality reduction model, thereby reducing the dimension of the data and improving processing efficiency. Figure 2 A flowchart of another method for evaluating the restoration and management of a mine ecological environment provided by an embodiment of the present invention is shown. Figure 2 As shown, the method comprises the following steps:
[0037] Step S202, monitoring the ecological parameters of the mine.
[0038] Step S204, extracting features from ecological parameters using a pre-built feature extraction model to determine key features in the ecological parameters.
[0039] In specific implementation, the feature extraction model of the embodiment of the present invention is constructed by the following method to avoid overfitting of the feature extraction neural network:
[0040] 1) Obtain a preset feature extraction neural network, and initialize the neural network parameters of the feature extraction neural network based on the normally distributed random numbers from the input layer dimension to the output layer dimension of the feature extraction neural network.
[0041] Specifically, the present invention uses a 3-layer fully connected neural network for feature extraction. In the prior art, some solutions use neural networks for feature extraction. In some neural network structures, gradient vanishing, gradient explosion or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. The present invention uses a neural network optimized based on the complexity constrained firefly algorithm, and uses a complexity evaluation index to constrain the complexity of the neural network to avoid overfitting.
[0042] In one embodiment, the parameters of the neural network The initialization of adopts a complexity constraint-based method to avoid overfitting, which can be expressed as:
[0043]
[0044] In the formula, is the regularization parameter; The generated dimensions are (input layer dimension) to Normally distributed random numbers (of the output layer dimension); is the He initialization method for a neural network with a ReLU activation function. Preferably, Set to 0.3.
[0045] 2) The neural network parameters are regarded as fireflies; and based on the correction factor learning rate and regularization factor corresponding to the feature extraction neural network, as well as the gradient of the loss function with respect to the neural network parameters, the parameter weights corresponding to the neural network parameters are calculated to obtain the position vector corresponding to the firefly.
[0046] In the specific implementation, based on the firefly optimization algorithm, a position vector is defined for each firefly (i.e., an instance of the network parameter) , the position represents a set of weights of the network, which can be expressed as:
[0047]
[0048] In the formula, Indicates The position of each firefly, i.e., its neural network parameters; Depends on the current number of iterations and current parameters The regularization factor of ; is the correction factor learning rate; is the loss function about The gradient of is the disturbance factor; is the updated learning rate. Preferably, Set to 0.3, Set to 0.03.
[0049] Further, The calculation method can be expressed as:
[0050]
[0051] In the formula, is the baseline regularization strength, Controls the rate at which the regularization strength changes over time. Preferably, Set to 3, Set to 5.
[0052] Furthermore, the correction factor learning rate As the training progresses, it is dynamically adjusted to adapt to the complexity and generalization requirements of the model at different stages, which can be expressed as:
[0053]
[0054] In the formula, is the initial regularization weight, is the attenuation coefficient. Preferably, Set to 3, Set to 5.
[0055] 3) Input the preset training sample set into the feature extraction neural network, use the loss function to evaluate the network performance of the feature extraction neural network, and obtain the loss function value.
[0056] Based on the preset training sample set, the neural network represented by each firefly is forward propagated and the loss function is used. To evaluate network performance, it can be expressed as:
[0057]
[0058] In the formula, is the average loss over all training samples; is the cross entropy loss function; It is the network output, which is obtained by the preset Softmax function; and They are Input and labels of training samples; is the sample size.
[0059] 4) Calculate the relative attraction between fireflies based on the loss function value, and optimize the position vector according to the relative attraction.
[0060] 5) Until the feature extraction neural network meets the preset iteration conditions, a feature extraction model is constructed based on the feature extraction neural network.
[0061] After the evaluation results are obtained, the feature extraction neural network is optimized to construct a feature extraction model. In this embodiment of the present invention, the relative attraction between fireflies is calculated according to the loss function value, and the fireflies with better performance will attract other fireflies to move to their position.
[0062] Specifically, Firefly to the Fireflies move, if The brightness (performance) of the first firefly is better than the Firefly direction. Then execute the following update method:
[0063]
[0064] In the formula, is the updated position; It is the attraction factor; is the light absorption coefficient; is the step length; is to generate random disturbances in the interval (0,1); is the square of the L2 norm. Preferably, Set to 3, Set to 5.
[0065] Furthermore, the light absorption coefficient The calculation of takes into account the dynamic changes of the current iteration to improve the adaptability of the algorithm. The calculation method is:
[0066]
[0067] In the formula, is the environment complexity evaluation function, For the The feature vector of the input of the iteration, is the initial light absorption coefficient, is the adjustment factor, is the current iteration number, is the total number of iterations. Preferably, Set to 3, Set to 4.
[0068] Furthermore, the environmental complexity evaluation function The calculation method can be expressed as:
[0069]
[0070] In the formula, is the number of features, It is The variance of a feature.
[0071] Further, the above steps are repeated until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.
[0072] Step S206, performing dimensionality reduction processing on the key features through a preset autoencoder algorithm to generate a low-dimensional feature representation corresponding to the key features.
[0073] Step S208, in the feature space where the low-dimensional feature representation is located, the low-dimensional feature representation is defined as a target node, and the low-dimensional feature representation is adjusted based on the adjacent nodes corresponding to the target node and their corresponding structural relationships.
[0074] Among them, the embodiment of the present invention performs dimensionality reduction processing on key features through a preset autoencoder algorithm. The autoencoder algorithm adopted by the present invention is used for data compression and dimensionality reduction. The autoencoder consists of two parts: an encoder and a decoder. The encoder is responsible for compressing the input data into a smaller code, and the decoder attempts to reconstruct the original input data from the code.
[0075] The prior art may fail to fully consider the complexity of the data structure when processing feature dimensionality reduction, resulting in information loss. The present invention adopts a dynamically adjusted constraint mechanism to automatically adjust the compression rate of the coding layer according to the characteristics of the input data, thereby optimizing the effect of feature dimensionality reduction.
[0076] Specifically, by inputting the key features into the autoencoder model and performing forward propagation through the encoder, the corresponding low-dimensional feature representation is generated. The forward propagation process can be expressed as:
[0077]
[0078] In the formula, It is The output of hidden nodes; is the ReLU activation function; is the input layer nodes; It is The bias of the hidden nodes.
[0079] Furthermore, graph theory is used in the feature space to enhance the association learning between data points by defining the relationship between nodes and edges, thereby capturing complex data structures. In specific implementation, considering the graph structure relationship between data points, the embodiment of the present invention calculates the node connection strength based on the adjacent nodes of the target node and the similarity between their corresponding nodes; and sums the target node according to the node connection strength, adjacent nodes, and the normal distribution noise corresponding to the target node to adjust the target node.
[0080] Specifically, the normal distribution noise includes the noise injection amount corresponding to the target node, and the noise injection amount is determined based on the variability corresponding to the target node. The node update method can be expressed as:
[0081]
[0082] In the formula, Considering adjacent nodes After The new state of the node; It is The output of hidden nodes; Representation Node The set of neighbor nodes of Inject vector for noise; Is a node and The strength of the connection between The mean is 0 and the variance is Based on this, the low-dimensional feature representation is adjusted.
[0083] Furthermore, the amount of noise injected Automatically adjust according to the variability of input data, the calculation method can be expressed as:
[0084]
[0085] In the formula, is the standard deviation of the input data; is a regulating factor. Preferably, Set to 0.3.
[0086] In one embodiment, the connection strength According to the dynamic calculation of similarity between nodes, it can be expressed as:
[0087]
[0088] In the formula, is an adjustable scale parameter. Preferably, Set to 0.1.
[0089] Step S210, reconstructing the low-dimensional feature representation through a preset decoder to obtain reconstructed key features.
[0090] The decoder reconstructs the compressed features and tries to restore the original number, which can be expressed as:
[0091]
[0092] In the formula, is the reconstructed input node ; It is from Hidden nodes to The decoding weights of the input nodes; is the bias in the decoding stage; is the Sigmoid activation function.
[0093] The autoencoder algorithm and decoder used in the embodiment of the present invention are pre-trained, wherein for each layer of the autoencoder, the initialization method of the weight and bias parameters of the autoencoder can be expressed as:
[0094]
[0095] In the formula, Indicates that from Input nodes to The weights of the output nodes; is the regularization factor; represents a normal distribution with a mean of 0 and a standard deviation of 1; is the number of nodes in the input layer. Preferably, Set to 0.3.
[0096] In addition, the embodiment of the present invention uses a composite loss function to optimize the training of the autoencoder. The composite loss function includes a reconstruction error and a graph structured error, which can be expressed as:
[0097]
[0098] In the formula, is the total loss function; is a parameter that controls the importance of the chart structure error; is the reconstruction error, is the graph structure error, is the total number of samples. Preferably, Set to 0.8.
[0099] Furthermore, the composite loss function is used to update the parameters of the autoencoder, and the updating method is the gradient descent method.
[0100] Further, the above steps are repeated until the preset stop iteration condition is met, indicating that the data dimension reduction model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.
[0101] Step S212, classify and identify key features through a pre-built evaluation model, and determine the restoration status of the mine based on the classification and identification results.
[0102] After the key features are reduced in dimension through the above steps, the pre-built evaluation model is used to classify and identify the reduced data. Among them, traditional classification methods may not be able to effectively deal with the problem of class imbalance, affecting the accuracy and reliability of the classification results.
[0103] The evaluation model of the embodiment of the present invention adopts a probabilistic neural network as a classifier algorithm. Based on the traditional probabilistic neural network, a fine-tuning strategy is adopted to allow the network to dynamically adjust its parameters according to the distribution of data during the training process. By adopting a fine-tuning module, the network of the present invention can adjust parameters according to the gradient of the loss function in each training cycle, thereby enhancing the model's adaptability to specific data features.
[0104] In the specific implementation, the construction method of the evaluation model is as follows:
[0105] 1) Obtain a preset probabilistic neural network, input the pre-constructed training sample set into the probabilistic neural network for forward propagation, and obtain the output probability of the training sample set corresponding to each preset category.
[0106] Specifically, a preset probabilistic neural network is first obtained, and its model parameters, including weights and biases, are initialized. In one embodiment, the present invention adopts a random initialization strategy and defines the initial values of the weight matrix and the bias vector as:
[0107]
[0108]
[0109] In the formula, To initialize the weight matrix, is the standard deviation, through normal distribution generate; To initialize the bias vector, the initial value is zero vector.
[0110] Furthermore, the data is forward propagated through the network to calculate the output probability of each category, which can be expressed as:
[0111]
[0112]
[0113] In the formula, is the input vector; and Respectively Layer weights and biases; is the weighted input; is the output after activation; is the activation function.
[0114] In one embodiment, the activation function is the ReLU function, and the calculation method can be expressed as:
[0115]
[0116] In the formula, To obtain The maximum value function in an interval.
[0117] 2) Based on the inverse frequency of the number of category samples of the preset category, the output probability is weighted.
[0118] In order to solve the problem of category imbalance, the embodiment of the present invention also uses cross entropy loss to perform weighted processing on the output probability. Specifically, the inverse frequency weight is obtained by performing ratio processing on the maximum number of samples in all categories and the number of category samples in the preset category; the output probability is weighted based on the preset weighted loss function and the inverse frequency weight so that the output probability meets the preset category balance requirement. The method can be expressed as:
[0119]
[0120] In the formula, is the weighted loss function, is the total number of categories; For the The weight of the class; is the true label; is the predicted probability, which is calculated by the Softmax function and can be expressed as:
[0121]
[0122] In the formula, Output for the network.
[0123] Furthermore, the weight It is the inverse frequency weighting based on the number of category samples, which can be expressed as:
[0124]
[0125] In the formula, For the The number of samples of the class, is the maximum number of samples in all classes.
[0126] 3) Calculate the gradient of the network parameters of the probabilistic neural network based on the output probability, adjust the learning rate of the probabilistic neural network according to the preset fine-tuning strategy; and update the network parameters based on the preset processing function. The processing function is used to characterize the ability to process different types of data.
[0127] Furthermore, the performance of the probabilistic neural network is evaluated based on the output probability, and the model is optimized based on the evaluation results. In the specific implementation, the gradient is calculated by the back propagation algorithm, and the network parameters are updated according to the micro-adjustment strategy. The parameter update method can be expressed as:
[0128]
[0129]
[0130] In the formula, and For the Layer Wheel weights and biases; and For the Layer Wheel weights and biases; is the learning rate of the current round; and are the gradients of the loss function with respect to weights and biases respectively; For processing functions of different data categories, Represents the data for each category.
[0131] Furthermore, define the function It is used to characterize the ability to process different types of data. The calculation method can be expressed as:
[0132]
[0133] In the formula, is the first adaptive adjustment parameter; is the second adaptive adjustment parameter.
[0134] Furthermore, an adaptive learning mechanism is used to dynamically adjust and , the adjustment method can be expressed as:
[0135]
[0136]
[0137] In the formula, and is the loss function For parameters and The partial derivative of and Respectively a first adaptive adjustment parameter and a second adaptive adjustment parameter of an iteration; and Respectively The first adaptive adjustment parameter and the second adaptive adjustment parameter of the iteration.
[0138] Further, The calculation method can be expressed as:
[0139]
[0140] In the formula, is the batch size, It is The network output of samples is is the corresponding true label, is the input vector.
[0141] Before each training cycle of the model starts, the embodiment of the present invention also activates the fine-tuning module to adjust the learning rate according to the performance evaluation result of the previous training cycle, which can be expressed as:
[0142]
[0143] In the formula, Indicates The learning rate of the round; is the initial learning rate; is the decay rate; is the current training round number. Preferably, Set to 0.03.
[0144] Furthermore, the attenuation rate The calculation method can be expressed as:
[0145]
[0146] In the formula, is the predetermined minimum learning rate, is a predetermined decay period length. Preferably, Set to 0.0001, Set to 1000.
[0147] 4) Check the class balance corresponding to the current class probability, and adjust the inverse frequency based on the test results.
[0148] 5) Until the probabilistic neural network reaches the preset iteration requirement and the category probability meets the preset category balance requirement, an evaluation model is constructed based on the probabilistic neural network.
[0149] After each iteration, the balance between classes is checked. If class imbalance is detected, , check and adjust class weights To achieve class balance, it can be expressed as:
[0150]
[0151] In the formula, is the total number of samples; For the The number of samples of the class.
[0152] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.
[0153] Another mine ecological environment restoration and management assessment method proposed in an embodiment of the present invention adopts a firefly algorithm based on complexity constraints to optimize neural network parameters, and combines it with a dynamic adjustment strategy, which can effectively control the complexity of the model, prevent overfitting of the feature extraction model, and improve the stability of model training.
[0154] In addition, the embodiment of the present invention also performs dimensionality reduction processing on key features. During the dimensionality reduction process, by combining the autoencoder and graph theory, graph theory is used to enhance the correlation learning between data points, which can capture more complex data structures, dynamically adjust the compression rate of the coding layer, and optimize the dimensionality reduction effect according to the characteristics of the input data. This optimizes the feature representation and dimensionality reduction process, and improves the efficiency and quality of data processing.
[0155] Furthermore, the embodiment of the present invention also sets a fine-tuning strategy for the probabilistic neural network. The fine-tuning strategy allows the model to dynamically adjust parameters according to the specific distribution of the data during the training process, thereby solving the problem of category imbalance, enhancing the model's recognition and classification accuracy of specific data features, and improving the model's classification performance and adaptability.
[0156] Furthermore, based on the above embodiment, the embodiment of the present invention also provides another method for evaluating the restoration and management of the ecological environment of mines, which mainly describes the method for constructing a training sample set, wherein the feature extraction model, the data dimension reduction model (that is, the above autoencoder algorithm), and the evaluation model can be trained by the training sample set, wherein the data dimension reduction model can be trained using the data extracted by the feature extraction model, and the evaluation model can be trained using the data processed by the data dimension reduction model. Specifically, Figure 3 A flow chart of a method for constructing a training sample set provided by an embodiment of the present invention is shown. Figure 3 , the method comprises the following steps:
[0157] Step S11, obtaining the pre-collected initial mine sample.
[0158] Step S12: labeling the initial mine samples and constructing an initial sample set.
[0159] In a specific implementation, the initial sample of the mine includes the ecological parameters of the mine, and the data of the above embodiment can be specifically referred to. Furthermore, the collected data is labeled to construct a sample set. The categories labeled in the embodiment of the present invention are used to indicate the restoration status of the mine. Among them, the labeling method of the present invention is manual labeling. In one embodiment, the labeled categories include: unrepaired restoration status, restoration in progress, and restoration completed restoration status, a total of 3 categories.
[0160] Step S13, extracting the data state of the initial sample set, and performing data expansion on the initial sample set based on the data state through a preset generative adversarial network to generate expanded samples.
[0161] It is understandable that in the task of the present invention, the collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model and affect the accuracy of the model. The existing technology may not effectively solve the quality control problem during data expansion and easily cause mode collapse. The embodiments of the present invention can use techniques such as rotation, flipping, scaling, and noise injection to perform data expansion.
[0162] In order to enhance data diversity and reduce the problem of mode collapse in data expansion, the present invention adopts a generative adversarial network based on perturbative robust enhancement for sample generation. Among them, the generative adversarial network based on perturbative robust enhancement includes two components: a generator (G) and a discriminator (D). The purpose of the generator is to create seemingly real data to deceive the discriminator; the task of the discriminator is to distinguish the generated data from the real data. In view of the problem that traditional generative adversarial networks are prone to mode collapse in data expansion, the present invention adopts a state storage mechanism so that the generator can refer to the state of data generated in the past when generating new data, reduce the generation of duplicate data, and enhance data diversity.
[0163] In specific implementation, the embodiment of the present invention assists in data expansion by extracting the state of data, which can be expressed as:
[0164]
[0165] In the formula, Indicates from memory A function that extracts states according to a certain strategy; For historical status.
[0166] In this embodiment of the present invention, data expansion is performed by generating adversarial networks, and the above data state is used to represent the historical generation state at each iteration. The historical generation status Extraction is performed to assist in this generation.
[0167] The generative adversarial network of the embodiment of the present invention is trained through the following steps:
[0168] a-pair generator and the discriminator The parameters are randomly initialized and the memoized state library is set to empty. The initialization method can be expressed as:
[0169]
[0170]
[0171] In the formula, is the parameter of the generator; Indicates compliance with a specific distribution; are the parameters of the discriminator; represents a normal distribution with mean 0 and covariance the identity matrix.
[0172] b-Generator combined with the real data of the current batch and random interference signals , and the state extracted from the memory bank , generate new data samples , which can be expressed as:
[0173]
[0174] In the formula, The data output by the generator; is the interference signal; is a generator function.
[0175] Furthermore, the random interference signal The calculation method can be expressed as:
[0176]
[0177] in, represents the standard deviation of the interference signal, represents the mean value of the interference signal. Preferably, is set to 0.01, is set to 0.
[0178] c-Discriminator Also receives fake data from the generator And real data samples , perform classification training and output the classification results, which can be expressed as:
[0179]
[0180] In the formula, represents the output of the discriminator, where ; 0 represents false data, 1 represents true data.
[0181] d- According to the feedback from the discriminator (i.e., the discriminator scores the authenticity of the generated data), adjust the parameters of the generator to optimize the quality of the generated data, which can be expressed as:
[0182]
[0183] In the formula, is the updated generator parameter; is the learning rate; represents the gradient of the generator parameters. Preferably, It is set to dynamically adjust the mode, and the initial value is set to 0.01.
[0184] Furthermore, the gradient The calculation method can be expressed as:
[0185]
[0186] In the formula, is the gradient of the discriminator with respect to the generator output.
[0187] Furthermore, in order to optimize the training process of the generator based on the performance feedback of the discriminator, the embodiment of the present invention uses an adaptive parameter adjustment mechanism to dynamically adjust the generator learning rate , thereby improving the quality and diversity of generated data, in one embodiment, define It represents the performance index of the discriminator in distinguishing real data from generated data, and is calculated as:
[0188]
[0189] In the formula, is the label (0 for generated data, 1 for real data), is the input data.
[0190] Furthermore, the learning rate The adjustment is based on The way to adaptively adjust the value of can be expressed as:
[0191]
[0192] In the formula, is the learning rate of the adjusted generator, is a positive adjustment factor. Preferably, Set to 0.4.
[0193] e-Furthermore, the parameters of the discriminator are updated.
[0194] Different from the parameter updating method of the generator, when updating the discriminator parameters, the present invention adopts a probabilistic discriminator adjustment mechanism controlled by dynamic threshold to improve the stability of the generative adversarial network and the authenticity of the generated data. In the specific implementation, the average error rate corresponding to the expanded samples is first calculated, and then the discriminator output of the generative adversarial network is adjusted based on the average error rate. Furthermore, based on the adjusted discriminator output, the discriminator parameters are updated.
[0195] Specifically, for the discriminator output For data is the probability of true data, using a dynamic threshold The updated and adjusted discriminator output can be expressed as:
[0196]
[0197] In the formula, is the threshold of the tth iteration, is a positive parameter that controls the output sensitivity. Preferably, Set to 0.3.
[0198] Furthermore, the dynamic threshold According to the discriminator in front The average error rate in the iterations Adjustment, the adjustment method can be expressed as:
[0199]
[0200] In the formula, For the The threshold for the iterations, is the learning rate used to adjust the speed at which the threshold changes; represents the deviation of the discriminator from the ideal error rate of 0.5. Preferably, Set to 0.03.
[0201] Further, using the adjusted discriminator output Update the parameters of the discriminator , which can be expressed as:
[0202]
[0203] In the formula, is the updated discriminator parameter, is the learning rate of the discriminator. Preferably, Set to 0.05.
[0204] Furthermore, some of the high-quality data states generated this time are stored in the memory bank , used for future data generation process, can be expressed as:
[0205]
[0206] In the formula, the update operation generates data by Add to existing memory is implemented in .
[0207] f-Furthermore, the adversarial balance between the generator and the discriminator of the generative adversarial network is evaluated, and based on the evaluation results, the parameters of the generator and the parameters of the discriminator are updated.
[0208] In specific implementation, the embodiment of the present invention also evaluates the confrontation balance between the generator and the discriminator to ensure that one of them does not completely suppress the other during the training process, which can be expressed as:
[0209]
[0210] In the formula, represents a balance measure; The number of samples for evaluation; Indicates A generated sample.
[0211] Further, the above steps are repeated until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.
[0212] Step S14: merge the expanded sample with the initial sample set to generate a training sample set.
[0213] Another mine ecological environment restoration and management assessment method provided by an embodiment of the present invention improves the generative adversarial network, and uses the generative adversarial network based on interference robust enhancement for data expansion. It can effectively avoid the common mode collapse problem in traditional generative adversarial networks, avoid the repeated generation of similar or low-quality data, and can generate more high-quality and diversified data samples, thereby improving the training effect and generalization ability of the model.
[0214] Furthermore, based on the above embodiments, the embodiments of the present invention also provide a mine ecological environment restoration and management assessment device, Figure 4 The structure diagram of a mine ecological environment restoration and management evaluation device provided by an embodiment of the present invention is shown as follows: Figure 4 As shown, the device includes: a data monitoring module 100, which is used to monitor the ecological parameters of the mine; a feature extraction module 200, which is used to extract features of the ecological parameters using a pre-constructed feature extraction model, and determine the key features in the ecological parameters; wherein the feature extraction model uses a feature extraction neural network to extract features, and a complexity constraint algorithm is used to constrain the neural network parameters of the feature extraction neural network; the feature extraction model is constructed based on a preset training sample set, and the training sample set is constructed after data expansion based on the data state of the initial sample set; an execution module 300, which is used to classify and identify key features through a pre-constructed evaluation model, and determine the restoration status of the mine based on the results of the classification and identification.
[0215] The embodiment of the present invention provides a mine ecological environment restoration and management assessment device, whose implementation principle and technical effects are the same as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference can be made to the corresponding contents in the aforementioned method embodiment.
[0216] Furthermore, based on the above embodiment, the embodiment of the present invention also provides another mine ecological environment restoration and management assessment device, Figure 5A schematic structural diagram of another device for evaluating the ecological environment restoration and management of a mine provided by an embodiment of the present invention is shown, wherein the device further includes a data processing module 400, which is used to perform dimensionality reduction processing on key features through a preset autoencoder algorithm to generate a low-dimensional feature representation corresponding to the key features; in the feature space where the low-dimensional feature representation is located, the low-dimensional feature representation is defined as a target node, and the low-dimensional feature representation is adjusted based on the adjacent nodes corresponding to the target node and their corresponding structural relationships; the low-dimensional feature representation is reconstructed through a preset decoder to obtain the reconstructed key features.
[0217] The above-mentioned data processing module 400 is also used for the steps of adjusting the low-dimensional feature representation based on the adjacent nodes corresponding to the target node and their corresponding structural relationships, including: calculating the node connection strength based on the adjacent nodes of the target node and the corresponding node similarities; summing the target node according to the node connection strength, adjacent nodes, and the normally distributed noise corresponding to the target node to adjust the target node; wherein the normally distributed noise includes the noise injection amount corresponding to the target node, and the noise injection amount is determined based on the variability corresponding to the target node.
[0218] The feature extraction module 200 is also used to obtain a preset feature extraction neural network, initialize the neural network parameters of the feature extraction neural network based on the normally distributed random numbers from the input layer dimension to the output layer dimension of the feature extraction neural network; use the neural network parameters as fireflies; and, based on the correction factor learning rate and regularization factor corresponding to the feature extraction neural network, and the gradient of the loss function with respect to the neural network parameters, calculate the parameter weights corresponding to the neural network parameters to obtain the position vector corresponding to the firefly; input the preset training sample set into the feature extraction neural network, use the loss function to evaluate the network performance of the feature extraction neural network, and obtain the loss function value; calculate the relative attraction between fireflies based on the loss function value, and optimize the position vector according to the relative attraction; until the feature extraction neural network meets the preset iteration conditions, a feature extraction model is constructed based on the feature extraction neural network.
[0219] The above-mentioned execution module 300 is also used to obtain a preset probabilistic neural network, input a pre-constructed training sample set into the probabilistic neural network for forward propagation, and obtain the output probability of the training sample set corresponding to each preset category; based on the inverse frequency of the number of category samples of the preset category, the output probability is weighted; based on the output probability, the gradient of the network parameters of the probabilistic neural network is calculated, and according to the preset fine-tuning strategy, the learning rate of the probabilistic neural network is adjusted; and, based on a preset processing function, the network parameters are updated; wherein the processing function is used to characterize the ability to process data of different categories; the category balance corresponding to the current category probability is detected, and the inverse frequency is adjusted based on the detection result; until the probabilistic neural network reaches the preset iteration requirement, and the category probability meets the preset category balance requirement, an evaluation model is constructed based on the probabilistic neural network.
[0220] The above-mentioned execution module 300 is also used to perform ratio processing on the maximum number of samples in all categories and the number of category samples in the preset category to obtain the inverse frequency weight; and weight the output probability based on the preset weighted loss function and the inverse frequency weight so that the output probability meets the preset category balance requirement.
[0221] The device also includes a construction module 500, which is used to obtain pre-collected initial mine samples; the initial mine samples include ecological parameters of the mine; the initial mine samples are labeled to construct an initial sample set; the labeled categories are used to indicate the restoration status of the mine; the data status of the initial sample set is extracted, and the initial sample set is expanded based on the data status through a preset generative adversarial network to generate expanded samples; the expanded samples are merged with the initial sample set to generate a training sample set.
[0222] The above-mentioned construction module 500 is also used to calculate the average error rate corresponding to the expanded samples; based on the average error rate, adjust the discriminator output of the generative adversarial network; based on the adjusted discriminator output, update the parameters of the discriminator.
[0223] The above-mentioned building module 500 is also used to evaluate the adversarial balance between the generator and the discriminator of the generative adversarial network, and based on the evaluation results, update the parameters of the generator and the parameters of the discriminator.
[0224] An embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above Figures 1 to 3 The steps of the method are shown.
[0225] The embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. The computer program is executed by a processor. Figures 1 to 3The embodiment of the present invention also provides a structural diagram of an electronic device, such as Figure 6 FIG. 1 is a schematic diagram of the structure of the electronic device, wherein the electronic device includes a processor 61 and a memory 60, the memory 60 stores computer executable instructions that can be executed by the processor 61, and the processor 61 executes the computer executable instructions to implement the above Figures 1 to 3 The method shown. Figure 6 In the illustrated embodiment, the electronic device further includes a bus 62 and a communication interface 63 , wherein the processor 61 , the communication interface 63 and the memory 60 are connected via the bus 62 .
[0226] Among them, the memory 60 may include a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 63 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 62 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc., and can also be an AMBA (Advanced Microcontroller Bus Architecture, on-chip bus standard) bus, wherein AMBA defines three types of buses, including APB (Advanced Peripheral Bus) bus, AHB (Advanced High-performance Bus) bus and AXI (Advanced eXtensible Interface) bus. The bus 62 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.
[0227] The processor 61 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 61. The above processor 61 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor 61 reads the information in the memory and combines its hardware to complete the above Figures 1 to 3 Any of the methods shown.
[0228] A computer program product of a mine ecological environment restoration and management assessment method and device provided in an embodiment of the present invention includes a computer-readable storage medium storing a program code, and the instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can refer to the method embodiment, which will not be repeated here. A person skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here. In addition, in the description of the embodiment of the present invention, unless otherwise clearly specified and limited, the terms "installation", "connection" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in the present invention can be understood in specific circumstances. If the function is implemented in the form of a software functional unit and sold or used as an independent product, it 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, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method 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., various media that can store program codes.
[0229] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. Finally, it should be noted that the above embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions recorded in the above embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A method for assessing the restoration and management of the ecological environment in mines. It is characterized in that include: Monitoring of ecological parameters of the mine; The ecological parameters are extracted by using a pre-built feature extraction model to determine key features in the ecological parameters; wherein the feature extraction model uses a feature extraction neural network to extract features, and the feature extraction neural network uses a complexity constraint method to initialize neural network parameters; the feature extraction model is constructed based on a preset training sample set, and the training sample set is constructed after data expansion based on the data state of the initial sample set; Classifying and identifying the key features through a pre-built assessment model, and determining the restoration status of the mine based on the classification and identification results; The method for constructing the feature extraction model comprises: Obtain a preset feature extraction neural network, and initialize neural network parameters of the feature extraction neural network based on a normally distributed random number from an input layer dimension to an output layer dimension of the feature extraction neural network; The neural network parameters are used as fireflies; and based on the correction factor learning rate and regularization factor corresponding to the feature extraction neural network, and the gradient of the loss function with respect to the neural network parameters, the parameter weights corresponding to the neural network parameters are calculated to obtain the position vector corresponding to the firefly; Inputting a preset training sample set into the feature extraction neural network, using a loss function to evaluate the network performance of the feature extraction neural network, and obtaining a loss function value; Calculating the relative attraction between the fireflies based on the loss function value, and optimizing the position vector according to the relative attraction; Until the feature extraction neural network meets a preset iteration condition, a feature extraction model is constructed based on the feature extraction neural network; Wherein, the position vector is optimized based on the following formula: In the formula, is the updated position; It is the attraction factor; is the light absorption coefficient; is the step length; is to generate random disturbances in the interval (0,1); is the square of the L2 norm; The light absorption coefficient The calculation method is: In the formula, is the environment complexity evaluation function corresponding to the training sample set, For the The feature vector of the training sample set input in the iteration, is the initial light absorption coefficient, is the adjustment factor, is the current iteration number, is the total number of iterations.
2. The method according to claim 1, It is characterized in that The method further comprises: Performing dimensionality reduction processing on the key features through a preset autoencoder algorithm to generate a low-dimensional feature representation corresponding to the key features; In the feature space where the low-dimensional feature representation is located, the low-dimensional feature representation is defined as a target node, and the low-dimensional feature representation is adjusted based on adjacent nodes corresponding to the target node and their corresponding structural relationships; The low-dimensional feature representation is reconstructed through a preset decoder to obtain reconstructed key features.
3. The method according to claim 2, It is characterized in that The step of adjusting the low-dimensional feature representation based on the adjacent nodes corresponding to the target node and the corresponding structural relationships thereof includes: Calculate the node connection strength based on the adjacent nodes of the target node and the similarities between their corresponding nodes; The target node is summed according to the node connection strength, the adjacent nodes, and the normally distributed noise corresponding to the target node to adjust the target node; wherein the normally distributed noise includes the noise injection amount corresponding to the target node, and the noise injection amount is determined based on the variability corresponding to the target node.
4. The method according to claim 1, It is characterized in that The method for constructing the evaluation model comprises: Obtain a preset probabilistic neural network, input a pre-constructed training sample set into the probabilistic neural network for forward propagation, and obtain an output probability of the training sample set corresponding to each preset category; Based on the inverse frequency of the number of category samples of the preset category, weighting the output probability; Calculating the gradient of the network parameters of the probabilistic neural network based on the output probability, adjusting the learning rate of the probabilistic neural network according to a preset fine-tuning strategy; and updating the network parameters based on a preset processing function; wherein the processing function is used to characterize the ability to process different categories of data; Testing the class balance corresponding to the current class probability, and adjusting the inverse frequency based on the test result; Until the probabilistic neural network reaches a preset iteration requirement and the category probability meets a preset category balance requirement, an evaluation model is constructed based on the probabilistic neural network.
5. The method according to claim 4, It is characterized in that The step of weighting the output probability based on the inverse frequency of the number of class samples of the preset class includes: Performing a ratio processing on the maximum number of samples in all categories and the number of category samples of the preset category to obtain an inverse frequency weight; The output probability is weighted based on a preset weighted loss function and the inverse frequency weight so that the output probability meets a preset category balance requirement.
6. The method according to claim 1, It is characterized in that The method for constructing the training sample set includes: Obtaining a pre-collected initial mine sample; the initial mine sample includes ecological parameters of the mine; The initial mine samples are labeled to construct an initial sample set; the labeled categories are used to indicate the restoration status of the mine; Extracting the data state of the initial sample set, and performing data expansion on the initial sample set based on the data state through a preset generative adversarial network to generate expanded samples; The expanded samples are combined with the initial sample set to generate a training sample set.
7. The method according to claim 6, It is characterized in that The method further comprises: Calculate the average error rate corresponding to the expanded samples; Based on the average error rate, adjusting a discriminator output of the generative adversarial network; Based on the adjusted output of the discriminator, parameters of the discriminator are updated.
8. The method according to claim 7, It is characterized in that The method further comprises: The adversarial balance between the generator and the discriminator of the generative adversarial network is evaluated, and based on the result of the evaluation, the parameters of the generator and the parameters of the discriminator are updated.
9. A mine ecological environment restoration and management assessment device, It is characterized in that include: Data monitoring module, used to monitor ecological parameters of mines; A feature extraction module, used to extract features from the ecological parameters using a pre-built feature extraction model, and determine key features in the ecological parameters; wherein the feature extraction model uses a feature extraction neural network to extract features, and the feature extraction neural network uses a complexity constraint method to initialize neural network parameters; the feature extraction model is constructed based on a preset training sample set, and the training sample set is constructed after data expansion based on the data state of the initial sample set; An execution module, used to classify and identify the key features through a pre-built evaluation model, and determine the restoration status of the mine based on the classification and identification results; The feature extraction module is also used to obtain a preset feature extraction neural network, initialize the neural network parameters of the feature extraction neural network based on the normally distributed random numbers from the input layer dimension to the output layer dimension of the feature extraction neural network; use the neural network parameters as fireflies; and, based on the correction factor learning rate and regularization factor corresponding to the feature extraction neural network, and the gradient of the loss function with respect to the neural network parameters, calculate the parameter weights corresponding to the neural network parameters to obtain the position vector corresponding to the firefly; input a preset training sample set into the feature extraction neural network, use the loss function to evaluate the network performance of the feature extraction neural network, and obtain the loss function value; calculate the relative attraction between the fireflies based on the loss function value, and optimize the position vector according to the relative attraction; until the feature extraction neural network meets the preset iteration conditions, construct a feature extraction model based on the feature extraction neural network; Wherein, the position vector is optimized based on the following formula: In the formula, is the updated position; It is the attraction factor; is the light absorption coefficient; is the step length; is to generate random disturbances in the interval (0,1); is the square of the L2 norm; The light absorption coefficient The calculation method is: In the formula, is the environment complexity evaluation function corresponding to the training sample set, For the The feature vector of the training sample set input in the iteration, is the initial light absorption coefficient, is the adjustment factor, is the current iteration number, is the total number of iterations.
