Power Internet of Things terminal identification method based on improved ReliefF-RF algorithm

Abstract

The invention discloses a power Internet of Things terminal identification method based on an improved ReliefF-RF algorithm, and the method comprises the steps: firstly collecting the data of a power Internet of Things terminal device, carrying out the preprocessing, carrying out the feature extraction and important feature screening through employing the improved ReliefF-RF algorithm in combination with a fuzzy theory, carrying out the clustering and crossing of the screened features, and carrying out the recognition of a power Internet of Things terminal. And finally, inputting the important features and the features obtained by crossing into a random forest classification model to obtain a predicted classification result. According to the scheme, the improved ReliefF algorithm and the random forest model are combined, mutual information, mutual information items and noise suppression items are introduced, and the fuzzy theory is combined, so that accurate classification and efficient management of the power terminal equipment are realized in a complex network environment and diversified protocol characteristics.

Description

Technical Field

[0001] The present invention belongs to the field of electric power Internet of Things, and specifically relates to an electric power Internet of Things terminal identification method based on an improved ReliefF-RF algorithm. Background Art

[0002] With the rapid development of the Power Internet of Things (PoT), the variety and number of power terminal devices have exploded. From small smart meters to complex energy storage systems, these devices together form a vast and complex power network. The communication protocols between these devices have also become increasingly complex and diverse, encompassing multiple protocol versions, from traditional Modbus and DNP3 to modern IEC60870-5-104. However, the lack of a unified protocol identification mechanism in traditional power grid management models leads to poor compatibility between devices and inefficient data processing, making it difficult to meet the requirements of efficient and intelligent management in modern power systems. This limitation is particularly evident when faced with large-scale distributed energy resource integration and dynamic energy management. Furthermore, the diversity and complexity of protocols creates a vast feature space, making protocol identification and data processing extremely cumbersome and inefficient. Edge IoT terminals, as key nodes in the Power Internet of Things (PoT), must achieve efficient data collection, processing, and transmission in a resource-constrained environment. However, the complexity of protocol features poses significant challenges in the lightweight design and deployment of these terminal devices. This not only increases computing and storage resource consumption but also compromises the system's real-time responsiveness and stability. For example, traditional protocol identification methods often rely on manual configuration or simple rule matching, which makes it difficult to cope with dynamically changing network environments and complex protocol characteristics, resulting in inefficient data transmission and even the inability of devices to connect to each other.

[0003] Therefore, there is an urgent need for an efficient feature selection algorithm combined with a classification algorithm to accurately classify edge IoT terminals. Summary of the Invention

[0004] In response to the above problems, the purpose of the present invention is to provide a power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm, so as to achieve accurate classification and efficient management of power terminal equipment under complex network environments and diverse protocol characteristics.

[0005] The specific technical solutions for achieving the purpose of the present invention are as follows:

[0006] A method for identifying power Internet of Things terminals based on an improved ReliefF-RF algorithm includes the following steps:

[0007] Step 1: Collect data from the power Internet of Things terminal equipment and pre-process the collected data;

[0008] Step 2: For the preprocessed data, the improved ReliefF-RF algorithm combined with fuzzy theory is used to extract features and screen important features;

[0009] Step 3: Cluster and cross-classify the selected features;

[0010] Step 4: Input the important features and the features obtained by cross-pollination in step 3 into the random forest classification model to obtain the predicted classification results.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] (1) The solution of the present invention is based on the improved ReliefF algorithm for feature selection. By adding mutual information and interactive information terms, it can not only quantify the direct correlation between features and target variables, but also deeply explore the complex relationships and potential interaction effects between features. This multi-level feature analysis method can avoid the limitations of traditional methods that only focus on a single feature, significantly improve the overall expression ability of the feature set, and provide important support for the performance optimization of the classification model; at the same time, by introducing noise suppression terms, it can effectively identify and eliminate noise features that have little contribution to the classification task or may introduce interference. This method improves the accuracy and robustness of feature selection, ensures that more pure and efficient input features are provided for subsequent models, thereby enhancing the classification accuracy and stability of the model;

[0013] (2) The solution of the present invention uses fuzzy theory to fuzzify the feature importance score, which can effectively deal with the ambiguity and uncertainty of the feature importance score. This flexible processing method improves the refinement of feature screening and enhances the adaptability and robustness of the algorithm in complex environments.

[0014] (3) The feature clustering and interactive mining method of the main features proposed in the present invention aims to generate more expressive enhanced features through cluster analysis and feature reconstruction. The method first performs cluster analysis on the feature set based on the K-means clustering algorithm to identify the "data item" feature category. Subsequently, these "data item" features are intelligently reconstructed within the same cluster to generate new enhanced features. The generated enhanced features provide the classifier with more powerful expressive power and significantly improve the model's performance on the protocol classification task.

[0015] The present invention will be further described below with reference to specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm of the present invention.

[0017] Figure 2Schematic diagram of a data acquisition device in an embodiment of the present invention.

[0018] Figure 3 This is a schematic diagram of the feature importance results of the improved ReliefF feature importance evaluation algorithm for original data in an embodiment of the present invention.

[0019] Figure 4 Schematic diagram of the effect of the selection of the k value on the accuracy in an embodiment of the present invention.

[0020] Figure 5 Schematic diagram of the clustering effect of features after feature selection in an embodiment of the present invention.

[0021] Figure 6 Schematic diagram of the classification confusion matrix for the original data set and the data set after feature selection in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] Example

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. The described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts are within the scope of protection of the present invention.

[0024] As used in this application and the claims, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural unless the context clearly indicates otherwise. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0025] Unless otherwise specifically stated, the relative arrangement of the parts and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to actual proportional relationships. The techniques, methods and equipment known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods and equipment should be considered as part of the authorization specification. In all examples shown and discussed here, any specific values should be interpreted as being merely exemplary and not as limitations. Therefore, other examples of the exemplary embodiments may have different values. It should be noted that similar numbers and letters represent similar items in the following figures, and therefore, once an item is defined in one figure, it does not need to be further discussed in subsequent figures.

[0026] Combine Figure 1 , a power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm, comprising the following steps:

[0027] Step 1: Collect data from the power Internet of Things terminal devices and preprocess the collected data. The collected data includes actual communication data of various protocol communication devices in the power Internet of Things (such as smart meters, distributed power generation equipment, energy storage systems, etc.), simulate complex network environments and diverse protocol characteristics, and obtain data samples containing different protocol types.

[0028] In this embodiment, the data acquisition device is as follows: Figure 2 As shown, the longest characteristic bit of the collected data is 61 columns;

[0029] The communication data of the power IoT terminal device is monitored through the RS485 bus and transmitted to the serial port server. The serial port server forwards the protocol message to the Ethernet and accesses the edge IoT gateway for processing and storage through the TCP / UDP protocol.

[0030] Step 1-1, parse the actual communication data of various protocol communication devices and extract the original message information of different protocols. According to the message structure, data format and transmission characteristics of the communication protocol, pre-process the original data, including denoising, data format conversion and standardization, to provide a high-quality data basis for subsequent feature extraction, and perform data cleaning on the collected data, that is, remove irrelevant items. The processing logic in this embodiment is to traverse each character of the message and determine whether the character is a valid hexadecimal character (i.e. [0-9A-Fa-f]). If so, retain it and check whether the subsequent characters are also hexadecimal characters. If they are consecutive hexadecimal characters, retain them as a whole; if not, stop the current retention operation and continue to check the next character;

[0031] Step 1-2: Segment the cleaned continuous hexadecimal data into groups of several characters and normalize them into a standard byte-level representation (e.g., 0000 00 00 00). This operation is to structure the data and facilitate the extraction of protocol features, such as specific byte sequences, flags, or length fields, paving the way for subsequent feature selection algorithms, thereby improving the accuracy of the classification algorithm.

[0032] Step 1-3: convert the data to a unified base, such as converting a hexadecimal number to a decimal number;

[0033] Steps 1-4: Normalize the data and map it to the range [0, 1] to eliminate dimensional differences, unify data standards, and improve model performance:

[0034]

[0035] Among them, X′ is the normalized data, X is the original data, and X min is the minimum value in the data set, X max is the maximum value in the data set.

[0036] Step 2: For the preprocessed data, use the improved ReliefF algorithm combined with fuzzy theory to extract features and screen important features:

[0037] Step 2-1, extract features from preprocessed data;

[0038] Step 2-2, use the improved ReliefF-RF algorithm to determine the weight W(f i );

[0039] ReliefF is a feature selection algorithm whose core idea is to evaluate the importance of features by measuring the changes in feature values to neighbor relationships. i , its importance is estimated by accumulating the difference between Hit (nearest sample of the same type) and Miss (nearest sample of a different type). The algorithm works by randomly selecting a sample R from the training sample set each time, then finding R's k nearest neighbor samples (near Hits) from sample sets of the same type as R, and finding k nearest neighbor samples (near Misses) from sample sets of different types of R, and then updating the weight of each feature, as shown in the following formula:

[0040]

[0041] Among them, M j (C) indicates class The jth nearest neighbor sample in , m represents the number of iterations, k is the number of selected neighbors, P(c) is the proportion of category C, diff(A,R j ,H j ) represents sample R j and H j The difference in feature A.

[0042] The traditional ReliefF algorithm has limitations and does not consider the interaction between features and the influence of noise. Therefore, the improved method introduces the following items:

[0043] (1) Mutual information term

[0044] Mutual information measures the value of a single feature f i The mutual information term is used to help screen features that directly contribute to sub-classification. Its formula is:

[0045] I(f i ;y)=H(f i )-H(f i |y)

[0046] Where H(f i ) is the feature f i The information entropy of f i uncertainty,

[0047] H(f i |y) is the conditional information entropy under the condition provided by the target variable y, that is, when y is known, f i the remaining uncertainty;

[0048]

[0049] A larger mutual information value indicates a more important feature.

[0050] (2) Interaction Information Items

[0051] Mutual information is a measure of the correlation between the joint effect of two features and the target variable.

[0052] I(f i ,f j ;y)=H(f i ,f j )-H(f i ,f j |y)

[0053]

[0054]

[0055] Where H(f i ,f j ) is the feature f i With f j The joint information entropy of these two features, H(f i ,f j |y) is the conditional information entropy under the condition provided by the target variable y, that is, when y is known, f i With f j The joint residual uncertainty.

[0056] The larger the mutual information, the stronger the synergy and correlation between the two features.

[0057] (3) Noise term

[0058] The noise term added by the present invention is directly calculated based on the variance of the feature, and its suppression strength is controlled by the weight parameter λ. The noise suppression formula is:

[0059]

[0060] Finally, the improved importance scoring formula can be obtained:

[0061] W(f i )=α·W′(f i )+β·I(f i ;y)+γ·I(f i ,f j ;y)-λ·Noise i

[0062] Among them, W′(f i ) represents the feature f determined by the ReliefF algorithm i The original weight, I(f i ; y) represents feature f i The mutual information term, H(f i ) is the feature f i The information entropy of f i The uncertainty of H(f i |y) is the conditional information entropy under the condition provided by the target variable y, that is, when y is known, f i The remaining uncertainty of I(f i ,f j ; y) represents feature f i The mutual information term, H(f i ,f j ) is the feature f i With f j The joint information entropy of i ,f j |y) is the conditional information entropy under the condition provided by the target variable y, that is, when y is known, f i With f j The joint residual uncertainty, Noise i Represents feature f i The noise term, n is the number of samples, f ij represents the j-th observation value of the i-th feature, Represents feature f i The average value of , α, β, γ and λ are weight values.

[0063] For ease of comparison, the scores are normalized to the interval [0,1]:

[0064]

[0065] Step 2-3: Next, the normalized weights are graded using fuzzy logic to screen out features with higher importance, and the fuzzy membership function is used to determine the important features:

[0066] Score(x) = trimf(x; [a, b, c])

[0067] Among them, trimf is the triangular membership function, c>b>a, and the input x is the weight of each feature after normalization;

[0068] When x is less than a or greater than c, Score is 0. When x∈b, When x∈[b,c],

[0069] In this embodiment, the specific parameters are selected as follows:

[0070] Score(x)=trimf(x;[0.5,0.6,0.7])

[0071] When x is less than 0.5 or greater than 0.7, Score is 0. When x∈[0.5, 0.6], When x∈[0.6, 0.7], The features with a score greater than 0.6 are selected as important features;

[0072] In this embodiment, the grid search method is used to search the parameters involved. The search range of each parameter is set to 'lambda':[0.01,0.05,0.1], 'beta':[0.1,0.5,1.0], 'alpha':[0.5,1.0,1.5], 'gamma':[0.1,0.2,0.3]. The parameter combination finally searched is the optimal parameter: {'lambda':0.05, 'beta':0.5, 'alpha':1.0, 'gamma':0.2}. Combining fuzzy logic and optimal k value, the final feature is selected as [0,1,2,3,4,5,7,8,9,10,11,12,13,14,15,16,18,19,20,22,24,26,28,34,35,51,52]. Its feature importance is as follows Figure 3 As shown, the k value is selected as Figure 4 shown.

[0073] Step 3: Cluster and cross-classify the selected features. Further use the improved ReliefF algorithm combined with fuzzy theory in step 3 to select the more important features. Perform feature combination and interactive mining on these main features to explore the potential relationship between the features and improve the expression ability of the model. Specifically:

[0074] Step 3-1: The purpose of clustering is to group highly correlated features together, use the K-means clustering algorithm to determine the cluster centers of important features, and minimize the distance between the feature cluster centers and the features:

[0075]

[0076] Among them, k is the number of clusters, C i is the i-th cluster, μ i is the i-th cluster center, f i is the transposed feature vector, and eventually all important features will be assigned to a cluster center;

[0077] In this example, the clustering effect is as follows Figure 5 shown.

[0078] Step 3-2: To better fit the actual application scenario (hexadecimal message data), the feature interaction based on register data is improved. For each feature group, the high and low bits of the register are first combined into decimal data representing the actual value. After clustering, the adjacent features in each feature group are crossed to obtain the new features after the crossover:

[0079] Register ij =f high,ij ×256×f low,ij

[0080] Among them, f high,ij With f low,ij are adjacent features in the same feature group;

[0081] The obtained features are then normalized.

[0082] Step 4: Input the important features and the features obtained by cross-pollination in step 3 into the random forest classification model to obtain the predicted classification results;

[0083] Random forests can naturally handle multi-classification tasks. They do this by constructing multiple decision trees based on random subsamples and random feature subsets, and then synthesizing the predictions of each tree (e.g., majority voting) to reach the final classification decision. In multi-classification tasks, each tree generates a single prediction based on the category distribution of its leaf nodes. Random forests effectively balance the performance of each category through an ensemble strategy, providing robustness, resistance to overfitting, and excellent adaptability to nonlinear classification boundaries. This characteristic makes them particularly effective in multi-classification tasks. The algorithm works as follows:

[0084] Assume that the training data set is D={(x1,y1),(x2,y2),...,(x N ,y N )}, the algorithm generates T subsets D1,D1,...,D by random sampling with replacement T , where D t The size of ,t∈{1,2,...,T} is the same as the size of the original training dataset D.

[0085] To enhance the diversity of the model, the random forest is used to select each subsample set D t Train a decision tree and randomly select m when splitting each node of the tree. subset (m subset ≤m, where m is the total number of features), and then select the feature that makes the splitting effect the best from this subset, that is, the feature subset with the smallest Gini index. The Gini index formula is as follows:

[0086]

[0087] where p k Indicates the proportion of samples belonging to the kth class in the dataset.

[0088] Then, on the randomly selected feature subset, the node with the smallest weighted Gini index of the left and right child nodes is selected for splitting:

[0089]

[0090] Among them, D is the corresponding data set on the current node, D L With D R Let them be the sample sets of the left and right child nodes respectively.

[0091] Random forest is to integrate multiple decision trees and fuse the guess results. In the classification task of this article, all trees have a predicted value h for the input x. t (x), the final category prediction is determined by majority voting:

[0092]

[0093] Where T is the total number of trees in the random forest during training, and II is the indicator function, which is the function when h t When (x) is equal to c, the function value is 1; otherwise it is 0. Here, h t (x) is the prediction result of the t-th base learner (each decision tree in the random forest) for the input x. c is the category of all power IoT terminals.

[0094] In this example, the number of trees selected in the random forest algorithm is set to 5. The running accuracy of the original data set, the feature set selected by the ReliefF algorithm, and the feature set selected by the improved ReliefF algorithm are shown in Table 1, and the confusion matrix is shown in Figure 6 shown.

[0095] Table 1

[0096] Original dataset Improve ReliefF Accuracy 0.9706 0.9822

[0097] This solution collects and preprocesses multi-protocol communication data, then uses an improved ReliefF algorithm combined with fuzzy theory to select efficient features. It then intelligently reconstructs key features using K-means clustering to generate enhanced features, thus avoiding the curse of dimensionality inherent in traditional high-order polynomial features. Experimental results show that the classification model accuracy increased from 0.9706 to 0.9822, significantly improving data classification performance in complex scenarios and validating the effectiveness of this approach.

[0098] In addition, the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following steps are implemented:

[0099] Step 1: Collect data from the power Internet of Things terminal equipment and pre-process the collected data;

[0100] Step 2: For the preprocessed data, the improved ReliefF algorithm combined with fuzzy theory is used to extract features and screen important features;

[0101] Step 3: Cluster and cross-classify the selected features;

[0102] Step 4: Input the important features and the features obtained by cross-pollination in step 3 into the random forest classification model to obtain the predicted classification results.

[0103] In addition, the present invention further provides a computer storable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0104] Step 1: Collect data from the power Internet of Things terminal equipment and pre-process the collected data;

[0105] Step 2: For the preprocessed data, the improved ReliefF algorithm combined with fuzzy theory is used to extract features and screen important features;

[0106] Step 3: Cluster and cross-classify the selected features;

[0107] Step 4: Input the important features and the features obtained by cross-pollination in step 3 into the random forest classification model to obtain the predicted classification results.

[0108] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for identifying power Internet of Things terminals based on an improved ReliefF-RF algorithm, characterized in that: The following steps are involved: Step 1: Collect data from the power Internet of Things terminal equipment and pre-process the collected data; Step 2: Use the improved ReliefF algorithm to extract features and screen important features from the preprocessed data; Step 3: Cluster and cross-classify the selected features; Step 4: Input the important features and the features obtained by cross-pollination in step 3 into the random forest classification model to obtain the predicted classification results.

2. The power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm according to claim 1 is characterized in that: The data collection and preprocessing in step 1 are specifically as follows: Step 1-1: Clean the collected data to remove irrelevant items; Step 1-2: Then, the cleaned data is segmented, and the cleaned continuous data is segmented into groups of several characters, and the data is normalized into a standard byte-level representation; Steps 1-3: Convert the data to a unified base; Steps 1-4: Normalize the data: Among them, X′ is the normalized data, X is the original data, and X min is the minimum value in the data set, X max is the maximum value in the data set.

3. The power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm according to claim 2 is characterized in that: The collected data includes actual communication data of various protocol communication devices in the power Internet of Things, simulating complex network environments and diverse protocol characteristics to obtain data samples containing different protocol types.

4. The power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm according to claim 1 is characterized in that: The feature extraction and screening in step 2 are specifically as follows: Step 2-1, extract features from preprocessed data; Step 2-2, use the improved ReliefF-RF algorithm to determine the weight W(f i ); Step 2-3: After normalizing the weights, use the fuzzy membership function to determine the important features: Score(x) = trimf(x; [a, b, c]) Among them, trimf is the triangular membership function, c>b>a, and the input x is the weight of each feature after normalization; When x is less than a or greater than c, Score is 0. When x∈b, When x∈[b,c], Finally, based on the score obtained by Score, features greater than a certain threshold are selected as important features.

5. The electric power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm according to claim 4 is characterized in that: The weights of the extracted features determined in step 2-2 are specifically as follows: W(f i )=α·W′(f i )+β·I(f i ;y)+γ·I(f i ,f j ;y)-λ·Noise i I(f i ;y)=H(f i )-H(f i |y) I(f i ,f j ;y)=H(f i ,f j )-H(f i ,f j |y) Among them, W′(f i ) represents the feature f determined by the ReliefF algorithm i The original weight, I(f i ; y) represents feature f i The mutual information term, H(f i ) is the feature f i The information entropy of f i The uncertainty of H(f i |y) is the conditional information entropy under the condition provided by the target variable y, that is, when y is known, f i The remaining uncertainty of I(f i ,f j ; y) represents feature f i The mutual information term, H(f i ,f j ) is the feature f i With f j The joint information entropy of i ,f j |y) is the conditional information entropy under the condition provided by the target variable y, that is, when y is known, f i With f j The joint residual uncertainty, Noise i Represents feature f i The noise term, n is the number of samples, f ij represents the j-th observation value of the i-th feature, Represents feature f i The average value of , α, β, γ and λ are weight values.

6. The electric power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm according to claim 5 is characterized in that: The weight of the feature is normalized: Among them, min(W(f i )) is the minimum value of all feature weights, max(W(f i )) is the maximum value among all feature weights.

7. The electric power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm according to claim 1 is characterized in that: The feature clustering and crossover in step 3 are specifically as follows: Step 3-1: Use the K-means clustering algorithm to determine the cluster centers of important features: Among them, k is the number of clusters, C i is the i-th cluster, μ i is the i-th cluster center, f i is the transposed feature vector, and eventually all important features will be assigned to a cluster center; Step 3-2: Cross the adjacent features in each feature group after clustering to obtain new features after crossover: Register ij =f high,ij ×256+f low,ij Among them, f high,ij With f low,ij are adjacent features in the same feature group; The obtained features are then normalized.

8. The electric power Internet of Things terminal identification method based on the improved ReliefF-RF algorithm according to claim 1 is characterized in that: The predicted classification results obtained in step 4 are specifically: Among them, T is the total number of trees in the random forest model during training, and II is the indicator function, which is when h t When (x) is equal to c, the function value is 1; otherwise, it is 0. t (x) is the prediction result of the t-th base learner for the input x, and c is the category of all power Internet of Things terminals.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer storable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images