Anomaly traffic detection system and method for mixture model based on ELM and deep forest
A mixed model using ELM and Deep Forests addresses the inefficiencies of existing systems by deploying these models in wireless sensors, reducing energy consumption and enhancing detection accuracy for anomalous traffic.
Patent Information
- Application Number
- JP2023528213
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2022-06-28
- Filing Date
- 2022-10-24
- Publication Date
- 2024-11-06
- Estimated Expiration
- 2042-10-24
AI Technical Summary
Existing intrusion detection systems in resource-limited wireless sensor networks face challenges with high computational overhead, dependency on large training data, slow convergence, and limited applicability to tasks with small datasets, especially in detecting anomalous traffic.
A mixed model approach combining Extreme Learning Machines (ELM) and Deep Forests for anomaly detection, involving data cleaning, feature extraction, and deploying ELM and Deep Forest models on different nodes, with retraining using accuracy and AUC as evaluation indicators.
The method significantly reduces energy consumption and improves detection accuracy for anomalous traffic in resource-limited nodes by leveraging ELM's efficiency and Deep Forest's robustness, achieving high detection efficiency and accuracy.
Smart Images

Figure 0007578331000010 
Figure 0007578331000011 
Figure 0007578331000012
Abstract
Description
[Technical field]
[0001] The present invention relates to the technical field of network security detection, and particularly, but not limited to, a mixed model anomaly traffic detection system and method based on ELM and deep forest. [Background technology]
[0002] According to the Internet Security Threat Report (ISTR), at present, some encryption hijacking criminals in the Internet environment encrypt scripts, causing some devices to become unusable due to high CPU occupancy, and network attacks occur frequently and are difficult to prevent. In addition, as the Internet of Things systems become more popular, many Internet of Things systems consisting of resource-limited terminals such as wireless sensors are used on a large scale and the environment is open for long periods of time, which makes them more likely to become targets of malicious third parties' attacks, and there is an increasing need for rapid, low-power abnormal traffic detection.
[0003] Existing intrusion detection generally uses traditional machine learning and deep learning algorithms to treat anomaly detection as a binary classification problem, with supervision based on normal and abnormal data, and establishes semi-supervised or unsupervised learning models with good results. Deep learning algorithms have been particularly successful in visual tasks, and in recent years, many attempts have been made by relevant research institutes both in China and abroad to apply them to the field of abnormal traffic.
[0004] Deep learning adopts training multiple single-layer nonlinear networks to discover and describe the complex structural features inside the problem, so it can express the essential features of the data, has strong generalization ability, and the accuracy of abnormal traffic detection is increasingly higher. However, it requires increasingly larger computation overhead, has too many hyperparameters, the learning performance is highly dependent on careful parameter tuning, there are too many training interference factors, it requires a large amount of training data, the convergence speed is slower, the time taken is longer, and it is difficult to apply to tasks with only small-scale training data.
[0005] For the above problem, deep forest has fewer hyperparameters than deep neural networks and has the advantage of being highly robust to hyperparameter settings. Deep forest is a deep model established based on non-differentiable modules, and generates deep forests with three characteristics: layer-by-layer processing, feature transformation within the model, and sufficient model complexity. In many cases, it can also achieve excellent performance with the same default settings for different data in different fields.
[0006] Extreme learning machine (ELM) has the advantages of fewer parameters, fewer resources, faster training speed and higher learning efficiency. ELM is a single hidden layer feedforward neural network (SLFNs) learning algorithm, which has the advantages of strong learning ability and approximating complex nonlinear functions, which can be directly solved, and the final solving process is the problem of solving the Moore-Penrose generalized inverse of a matrix, and the model can be trained by setting the number of hidden layer neurons. In terms of application, ELM can reduce energy consumption while ensuring a relatively high detection rate, which is more suitable for resource-limited wireless sensor networks. Summary of the Invention [Problem to be solved by the invention]
[0007] In response to one or more problems in the prior art, the present invention provides a mixed model anomaly traffic detection system and method based on ELM and deep forest, which completes anomaly traffic detection in resource-limited nodes such as wireless sensors, and has relatively high detection efficiency and accuracy. [Means for solving the problem]
[0008] The technical solutions to achieve the objectives of the present invention are as follows:
[0009] A method for detecting abnormal traffic using a mixture model based on ELM and deep forest, comprising: Step 1: performing data cleaning, feature extraction and data dimensionality reduction on real-time traffic data collected by a lowest-level member node of a wireless sensor resource-limited node; Step 2: deploying an extreme learning machine (ELM) model and a deep forest model to different nodes in the wireless sensor network to detect mixed abnormal traffic, and outputting an abnormal traffic detection result; The method is characterized by including step 3, in which the management node in the wireless sensor retrains the deep forest model on the updated dataset using accuracy and AUC as evaluation indicators, where AUC is the area under the ROC curve.
[0010] In addition, in the method for detecting abnormal traffic using a mixture model based on ELM and deep forest according to the present invention, step 1 specifically includes: S1-1 removes dirty reads such as non-numeric characters (NAN) and infinity characters (Infinity) present in real-time traffic data; From the captured complete network information, 30-dimensional data that has a significant impact on detecting abnormal traffic is selected by referring to the Botnet Dataset for the Internet of Things (BOT-IOT Dataset) published by the Canberra Network Range Laboratory at the University of New South Wales. Based on the 30-dimensional features, S1-2 adds a new 14-dimensional feature, which is mainly statistical data such as the total number of packets for each source / destination IP. After processing, normalization and standardization were performed on the data to calculate the 4 and S1-3 forming pre-processed traffic data with four-dimensional features.
[0011] In addition, in the abnormal traffic detection method of the mixture model based on ELM and deep forest according to the present invention, step 2 specifically includes: Step 2-1 of deploying an ELM model to member nodes by a management node in the wireless sensor; Step 2-2 deploying the deep forest model to a sink node by a management node in the wireless sensor; Step 2-3: performing ELM rapid detection on the dimension-reduced real-time traffic data in the member node, adding data features to the traffic data based on the detection result, and if the detection result is normal traffic, labeling the data features as normal values and allowing the traffic data to pass, and if the detection result is abnormal traffic, labeling the data features as abnormal values; The method includes steps 2-4 of removing the data feature labels of the abnormal traffic from the sink node, performing secondary detection on the abnormal traffic using a deep forest model, adding the data features to the traffic data as its final feature values based on the secondary detection result, and fusing the data together in a management node.
[0012] Furthermore, in the method for detecting abnormal traffic using a mixture model based on ELM and deep forest according to the present invention, the pre-training process of the ELM model deployed in step 2-1 includes: Training sample X train Input the sample size as [row0,col0]. Divide the BOT-IOT dataset after preprocessing in step 1 into a training set and a test set in a ratio of 8:2. X train Step 2-1-1: Let row0 be the training set, row1 be the number of training samples, col0 be the feature dimension of the training samples, and the number of hidden layer units be n, where n is a positive integer; Step 2-1-2: Initialize the weights w and deviations b of the ELM model, and set the value range of w to [col0,n] and the value range of b to [row0,n]; Formula h=g(w·X train +b), calculate the nonlinear mapping of the ELM model based on the activation function g(x), h is the nonlinear mapping of the ELM model, and X train Step 2-1-3: set h as a training sample, obtain the output of the hidden layer by finding the inverse matrix H of h, and set the size of H to [n,row0]; Step 2-1-4, one-hot coding is performed on the data labels in the dataset, and β=H·T is calculated and obtained, where the data label indicates whether the data is abnormal traffic, T is the value after one-hot coding of the label data, and β indicates the output weight, the size of which is [n, 2]; Input real-time traffic data X test For h1=g(w X test +b), where h1 is the nonlinear mapping in the ELM model of the real-time traffic data to be tested, and multiplied by β to obtain the traffic detection result, the calculation formula is result=h1·β; Step 2-1-6 of classifying the acquired traffic detection results to acquire abnormal traffic and normal traffic; and step 2-1-7 of saving the post-training weights w, deviations b, and inverse matrix H as parameters required for deploying the ELM model.
[0013] Furthermore, in the method for detecting abnormal traffic using a mixture model based on ELM and deep forest according to the present invention, the pre-training process of the deep forest model deployed in step 2-2 includes: Step 2-2-1, in which the XGBoost classifier, the random forest classifier and the extreme random tree classifier are selected as deep forest classifiers, the maximum depth of the XGBoost classifier is set as 5, the goal is multi-classification, the learning rate is set as 0.1, and the random forest classifier and the extreme random tree classifier are set using default parameters; Step 2-2-2 includes performing deep forest training using the processed BOT-IOT model and storing the deep forest training results.
[0014] In addition, in the method for detecting abnormal traffic using a mixture model based on ELM and deep forest according to the present invention, step 3 specifically includes: Step 3-1: Deploy the optimized deep forest model on the user-facing management node and configure the deep forest parameters; Step 3-2: pre-storing the labeled traffic data from each sink node in a temporary database and setting an upper limit for data storage; Step 3-3: After the number of pre-stored data reaches the upper limit of storage, the data features are extracted as labels, mixed with the raw data set, and then divided into a training set and a test set. The deep forest model is tuned using accuracy and AUC as evaluation indicators. Step 3-4 includes the management node loading the tuned model into the sink node.
[0015] Furthermore, in the method for detecting abnormal traffic using a mixture model based on ELM and deep forest according to the present invention, the tuning of the deep forest model in step 3-3 is performed using a K-fold cross validation method, and specifically, Step 3-3-1 to back up the raw data set, Step 3-3-2, in which the raw data set is divided into a training set and a test set, the data in the temporary database is extracted as a label set based on the data features, and the label set is combined with the training set in the raw data set to form a new data set; Step 3-3-3, in which the new dataset is divided evenly into k sets, one of which is selected as a validation set, and the remaining k-1 sets are selected as a training set; Step 3-3-4 of constructing a deep forest algorithm model using a deep forest consisting of a random forest classifier, an XGBoost classifier, an extreme random trees classifier and a logistic regression classifier; Step 3-3-5: find the model with the smallest error in the validation set for each fold in the k-1 fold, and if the number of training times is less than k, return to step 3-3-3; Step 3-3-6: take the model with the smallest error in the validation set for each fold, put it into the test set to test the error, and calculate and obtain the average performance for each round on the test set; The evaluation values include step 3-3-7, which selects the accuracy and AUC values, compares the average value of the new model with the performance of the original model, and if the performance improves, updates the deep forest model to update the dataset, otherwise continues to use the original deep forest model and deletes the data stored this time.
[0016] A mixed model anomaly traffic detection system based on ELM and deep forest using any one of the above detection methods, a real-time traffic data feature extraction and dimension reduction module for performing data cleaning, feature extraction, and data dimension reduction on real-time traffic data collected by a lowest-level member node in the resource-limited node of the wireless sensor; an anomaly traffic detection module for deploying the ELM model and the deep forest model on different nodes in the wireless sensor network, detecting mixed anomaly traffic, and outputting an anomaly traffic detection result; The management node in the wireless sensor is configured to retrain the deep forest model on the updated dataset using accuracy and AUC as evaluation indices, and the deep forest model training module is configured so that AUC is the area under the ROC curve. Effect of the Invention
[0017] Compared with the prior art, the above technical solution used in the present invention has the following technical advantages:
[0018] The mixed model anomaly traffic detection method based on ELM and deep forest of the present invention is used for resource-limited nodes such as wireless sensors, and performs mixed detection based on ELM and deep forest model on pre-processed traffic features. In wireless sensor networks, different nodes can be deployed with ELM and deep forest models according to their different roles in the network, so that multiple layers work together to greatly reduce the energy consumption for anomaly detection at each node and also improve the accuracy of anomaly detection results.
[0019] The drawings are intended to provide a further understanding of the invention and, together with the explanatory description, are intended to illustrate embodiments of the invention and are not intended to limit the invention. [Brief description of the drawings]
[0020] [Figure 1] FIG. 1 is a flowchart of a mixture model anomaly traffic detection method based on ELM and deep forest according to the present invention. [Diagram 2] FIG. 2 is an architecture diagram of the mixed model anomaly traffic detection system based on ELM and deep forest according to the present invention. [Diagram 3] FIG. 3 is a schematic diagram of an ELM model in the method for detecting abnormal traffic using a mixed model based on ELM and deep forest according to the present invention. [Figure 4]FIG. 4 is a schematic diagram of a deep forest model in the method for detecting abnormal traffic using a mixed model based on ELM and deep forest according to the present invention. [Diagram 5] FIG. 5 is a k-fold cross-validation diagram of the mixture model anomaly traffic detection method based on ELM and deep forest according to the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0021] In order to further understand the present invention, the following preferred embodiments are described in conjunction with examples, but it should be understood that these descriptions are merely intended to further illustrate the features and advantages of the present invention and are not intended to limit the scope of the claims of the present invention.
[0022] The description in this section is merely for some representative embodiments, and the present invention is not limited to the scope described in the embodiments. Combinations of different embodiments, substitutions of some technical features in different embodiments, and substitutions of some technical features in the embodiments with similar or similar existing technical means are also included in the description and protection scope of the present invention.
[0023] The abnormal traffic detection method of a mixed model based on ELM and deep forest is realized by three modules, including a real-time traffic data feature extraction and dimensionality reduction module, an abnormal traffic detection module, and a deep forest model training module, as shown in Figure 1, and specifically includes the following S1 to S3.
[0024] S1 real-time traffic data feature extraction and dimension reduction module first performs data cleaning on the real-time traffic data collected by the lowest-level member node in the resource-limited node such as wireless sensor, and then performs feature extraction and data dimension reduction. Specifically, it includes the following steps S11 to S13.
[0025] S11: Perform processing to remove non-numeric characters (NAN) and infinity-numeric characters (Infinity) that exist in real-time traffic.
[0026] S12 From the captured complete network information, select 30-dimensional data that has a significant impact on detecting abnormal traffic by referring to the BOT-IOT dataset, such as the text display of transaction protocols in network traffic, source IP addresses, destination IP addresses, etc., and based on the 30-dimensional features, add new 14-dimensional new features, which are mainly statistical data such as the total number of packets for each source / destination IP, in order to improve the detection ability of the classifier.
[0027] S13 Normalize and standardize the processed data to calculate the 4 The pre-processed traffic data with four-dimensional features is formed.
[0028] S2 The abnormal traffic detection module detects mixed abnormal traffic by deploying the ELM and the deep forest model to different nodes in the wireless sensor network, and outputs the abnormal traffic detection result. Specifically, it includes the following S21 to S24.
[0029] S21 The ELM model is deployed to member nodes by the management node in the wireless sensor.
[0030] Preferably, the pre-training process of the deployed ELM model includes the following steps a1 to a7.
[0031] Step a1 Training sample X train Input the above, set the sample size to [row0,col0], and set the number of hidden layer units to n. Here, the sample size is set to [4000000,54], and the number of hidden layer units n is selected as 109 through testing.
[0032] Step a2: Initialize the weights w and deviations b of the ELM model, where w has size [col0,n] and b has size [row0,n]. The contents of w are random numbers in the range, and b is initialized to an array of all 0s; Step a3 Formula h=g(w X train +b) to obtain the nonlinear mapping of ELM, where g(x) is the activation function, and the output of the hidden layer can be obtained by finding the inverse matrix H of h using activation functions such as sigmoid, Gaussian, and multiquadric, where the size of H is [n,row0]; The multiquadric activation function is also a radial basis function, which is a real-valued function whose value depends only on the distance from the far point.
number
[0033] Step a4: Perform one-hot coding on the data labels in the dataset to obtain T with size [row0,2], and use H in step a3 to obtain β=H·T based on the formula, where β has size [n,2]; Step a5 Enter real-time traffic X test For h1=g(w X test +b) and multiply it by β to obtain the traffic detection result result = h1·β.
[0034] Step a6: Execute result = [item.tolist().index(max(item.tolist()))for item in result] on the obtained results to classify them into abnormal traffic and normal traffic.
[0035] Step a7: Save w, b, and H after training, which are the parameters required to deploy the ELM model.
[0036] S22 Deploy the deep forest model to the sink node by the management node in the wireless sensor.
[0037] Preferably, the pre-training process of the deployed deep forest model includes the following steps b1 and b2.
[0038] Step b1: Select the XGBoost classifier, the random forest classifier and the extreme random tree classifier as deep forest classifiers, respectively, set the maximum depth of the XGBoost classifier as 5, the goal as multi-classification, and the learning rate as 0.1, and set the random forest classifier and the extreme random tree classifier using default parameters.
[0039] Step b2: Perform deep forest training using the processed BOT-IOT model, and save the deep forest training result.
[0040] S23 For the traffic data after dimensionality reduction, first perform ELM rapid detection in the member node, and add one data feature detection_result to the traffic data based on the detection result. If the detection result is normal traffic, label the detection_result as a normal value '0' and pass it through. If abnormal traffic is detected, label it as an abnormal value '1'.
[0041] S24 After the data labeled as abnormal traffic is identified in the sink node, the detection_result feature value is first deleted, a secondary detection is performed using a deep forest model, the detection_result value is determined based on the result, and the data is collectively fused and sent to the management node for processing.
[0042] The S3 Deep Forest Model Training module retrains the Deep Forest model on the updated dataset by the management node in the wireless sensor based on evaluation metrics such as accuracy and AUC.
[0043] Preferably, the deep forest model training module specifically executes the following steps S31 to S33.
[0044] Deploy the optimized deep forest algorithm and configure deep forest parameters on the management node for S31 users.
[0045] S32: Prestore the labeled traffic data from each sink node in a temporary database, and set the upper limit of data storage to 10,000.
[0046] S33: After the number of pre-stored data reaches a set threshold, the detection_result features are extracted as labels and mixed with the raw dataset, then the dataset is divided into a training set and a test set, and the deep forest model is tuned based on evaluation indices such as accuracy and AUC.
[0047] Preferably, the tuning algorithm used in this example is K-fold cross-validation, as shown in FIG. a step c1 of backing up the raw data set; Step c2, in which the raw data set is divided into a training set and a test set, and the data in the temporary database is extracted as a label set with detection_result features, and then merged with the training set in the raw data set as a new data set; Step c3 of equally dividing the new dataset into k sets, selecting one of the sets as a validation set and the remaining k-1 sets as a training set, where k is 10; Step c4 of constructing a deep forest algorithm model using a deep forest consisting of a random forest classifier, an XGBoost classifier, an extreme random tree classifier and a logistic regression classifier; Step c5 of searching for a model with the smallest error in the validation set for each fold in the k-1 fold, and returning to step c3 if the number of training times is less than k; Step c6 of taking the model with the smallest error in the validation set for each fold and putting it into the test set to test the error, and calculating and obtaining the average performance of each time on the test set; The evaluation values include step c7 of selecting the accuracy and AUC values, comparing the average value of the new model with the performance of the original model, and if the performance improves, updating the deep forest model to update the dataset, otherwise continuing to use the original deep forest model and deleting the new data stored this time.
[0048] The evaluation index is: accuracy
number
number
number
number
number
number
number
number
[0049] S34: The manager node loads the tuned model to the sink node.
[0050] The present invention provides an anomaly traffic detection system and method for resource-limited nodes such as wireless sensors, and performs mixed detection based on ELM and deep forest models on pre-processed traffic features. In a wireless sensor network, ELM and deep forest models are deployed respectively according to the roles of different nodes in the network, and multiple layers work together to significantly reduce the energy consumption for anomaly detection at each node while also improving the accuracy of anomaly detection results.
[0051] The description and application of the present invention herein are illustrative and are not intended to limit the scope of the present invention to the above-mentioned embodiments. The relevant description of the effects or advantages of the specification may not be realized in actual experimental examples due to the uncertainty of specific condition parameters or other factors, and the relevant description of the effects or advantages is not intended to limit the scope of the invention. Modifications and changes of the embodiments disclosed herein are possible, and substitutions of the embodiments and various equivalent members are known to those skilled in the art. As will be appreciated by those skilled in the art, the present invention may be realized in other forms, structures, arrangements, proportions, and other assemblies, materials, and members without departing from the spirit or essential characteristics of the present invention. Other modifications and changes can be made to the embodiments disclosed herein without departing from the spirit and spirit of the present invention.
Claims
1. A method for detecting abnormal traffic using a mixture model based on ELM and deep forest, comprising: Step 1: performing data cleaning, feature extraction and data dimensionality reduction on real-time traffic data collected by a lowest-level member node of a wireless sensor resource-limited node; 2. Deploying an extreme learning machine (ELM) model and a deep forest model to different nodes in the wireless sensor network to detect mixed abnormal traffic, and outputting an abnormal traffic detection result; Step 3: the management node in the wireless sensor retrains the deep forest model on the updated dataset using accuracy and AUC as evaluation indicators, where AUC is the area under the ROC curve; Step 1 is specifically: S1-1 removes dirty reads such as non-numeric values and infinite values present in real-time traffic data; S1-2: from the captured complete network information, select 30-dimensional data that has a significant impact on the detection of abnormal traffic by referring to the BOT-IOT dataset, and based on the 30-dimensional features, add new 14-dimensional features, which are mainly statistical data such as the total number of packets for each source / destination IP; and performing normalization and standardization operations on the processed data to form pre-processed traffic data having a total of 44 dimensional features (S1-3).
2. Step 2 is specifically: Step 2-1: deploying an ELM model to member nodes by a management node in a wireless sensor; Step 2-2 deploying the deep forest model to a sink node by a management node in the wireless sensor; Step 2-3: performing ELM rapid detection on the dimension-reduced real-time traffic data in the member node, adding data features to the traffic data based on the detection result, and if the detection result is normal traffic, labeling the data features as normal values and allowing the traffic data to pass, and if the detection result is abnormal traffic, labeling the data features as abnormal values; The method for detecting abnormal traffic using a mixed model based on ELM and deep forest as described in claim 1, further comprising steps 2-4 of removing the data feature labels of the abnormal traffic from the sink node, performing secondary detection on the abnormal traffic using a deep forest model, adding the data features to the traffic data as its final feature values based on the secondary detection result, and fusing the data together in a management node.
3. The pre-training process of the ELM model deployed in step 2-1 is as follows: Training sample X train Enter the sample size in 0 , col0], the BOT-IOT dataset after preprocessing in step 1 is divided into a training set and a test set in a ratio of 8:2, and X train Let row be the training set. 0 Let be the number of training samples, and col 0 Step 2-1-1: set the feature dimension of the training sample to n, the number of hidden layer units to n, and n to a positive integer; Initialize the weights w and deviations b of the ELM model and set the value range of w to [col 0 , n], and the value range of b is [row 0 , n] in step 2-1-2; Formula h=g(w・X train +b), where g(x) is the activation function, h is the nonlinear mapping of the ELM model, and X train is used as a training sample, and the output of the hidden layer is obtained by finding the inverse matrix H of h, and the size of H is [n, row 0 Step 2-1-3: Step 2-1-4: perform one-hot coding on the data labels in the data set to calculate and obtain β=H·T, where the data label indicates whether the data is abnormal traffic, T is the value after one-hot coding of the label data, and β indicates an output weight whose magnitude is [n, 2]; Input real-time traffic data X test For h 1 = g (w x test +b), and 1 is the nonlinear mapping in the ELM model of the real-time traffic data to be tested, and is multiplied by β to obtain the traffic detection result, and the calculation formula is result=h 1 Step 2-1-5, which is β, A step 2-1-6 of classifying the acquired traffic detection results to acquire abnormal traffic and normal traffic; The method for detecting abnormal traffic using a mixture model based on ELM and deep forest as described in claim 2, further comprising step 2-1-7 of saving the trained weights w, deviations b and inverse matrix H as parameters required for deploying the ELM model.
4. The pre-training process of the deep forest model deployed in step 2-2 is as follows: The method for detecting abnormal traffic using a mixture model based on ELM and deep forest as claimed in claim 1, further comprising: step 2-2-1: selecting each of the XGBoost classifier, the random forest classifier and the extreme random tree classifier as a deep forest classifier; setting the maximum depth of the XGBoost classifier as 5, the target as multi-classification, the learning rate as 0.1, and setting the random forest classifier and the extreme random tree classifier using default parameters; and step 2-2-2: performing deep forest training using the processed BOT-IOT model and storing the deep forest training result.
5. Step 3 is specifically: Step 3-1: Deploying the optimized deep forest model on a user-facing management node and configuring deep forest parameters; Step 3-2: pre-storing the labeled traffic data from each sink node in a temporary database and setting an upper limit for data storage; Step 3-3: After the number of pre-stored data reaches the upper limit of storage, data features are extracted as labels, mixed with the raw data set, and then divided into a training set and a test set. The deep forest model is tuned using accuracy and AUC as evaluation indices. The method for detecting anomaly in a mixed model based on ELM and deep forest as claimed in claim 1, further comprising: a step 3-4 in which the management node loads the tuned model into the sink node.
6. The tuning of the deep forest model in step 3-3 uses the K-fold cross-validation method. Specifically, Step 3-3-1 to back up the raw data set; Step 3-3-2, in which the raw data set is divided into a training set and a test set, the data in the temporary database is extracted as a label set based on the data features, and the label set is combined with the training set in the raw data set to form a new data set; Step 3-3-3 of equally dividing the new dataset into k sets, selecting one set as a validation set and the remaining k-1 sets as a training set; Step 3-3-4 of constructing a deep forest algorithm model using a deep forest consisting of a random forest classifier, an XGBoost classifier, an extreme random tree classifier and a logistic regression classifier; Step 3-3-5 of searching for the model with the smallest error in the validation set of each fold in the k-1 fold, and returning to step 3-3-3 if the number of training times is less than k; Step 3-3-6: take the model with the smallest error in the validation set for each fold, put it into the test set to test the error, and calculate and obtain the average performance for each round on the test set; The method for detecting anomalies in a mixed model based on ELM and deep forest as described in claim 5, characterized in that the evaluation values include selecting the accuracy and AUC values, comparing the average value of the new model with the performance of the original model, and if the performance improves, updating the deep forest model to update the dataset, otherwise continuing to use the original deep forest model and deleting the data currently stored.
7. The ELM and deep forest based mixture model anomaly traffic detection system according to any one of claims 1 to 6, comprising: a real-time traffic data feature extraction and dimension reduction module for performing data cleaning, feature extraction, and data dimension reduction on real-time traffic data collected by a lowest-level member node in the resource-limited node of the wireless sensor; an anomaly traffic detection module for deploying the ELM model and the deep forest model on different nodes in the wireless sensor network, detecting mixed anomaly traffic, and outputting an anomaly traffic detection result; An ELM and deep forest based mixed model anomaly traffic detection system, characterized in that it comprises: a management node in the wireless sensor for retraining the deep forest model on the updated dataset using accuracy and AUC as evaluation indicators, where AUC is the area under the ROC curve; and a deep forest model training module.
Citation Information
Patent Citations
Data transmission method and device, GPU and readable storage medium
CN111124994A
Network abnormal flow detection method, model and system
CN112784881A
Random forest model training method and device, and abnormal traffic detection method and device
CN114553591A
Internet of Things botnet traffic detection and classification method based on deep forest
CN114615045A
