Communication power supply remote monitoring method and system based on Internet of Things

Abstract

The invention provides a communication power supply remote monitoring method and system based on the Internet of Things, and the method comprises the steps: collecting multi-dimensional operation data of a communication power supply, and obtaining original data; calculating the time sequence similarity between the original data and the historical normal operation data by using a DTW algorithm to obtain a similarity matrix; performing quantitative analysis on the causal relationship of the operation data among the dimensions in the original data based on Granger causal test to obtain a causal incidence matrix; fusing the similarity matrix with a causal incidence matrix to obtain a feature enhancement matrix; inputting the feature enhancement matrix into a deep neural network, and outputting an abnormal state judgment result; and the abnormal state judgment result is encrypted through the Internet of Things communication module and then is sent to a remote monitoring center, so that remote monitoring of the communication power supply is realized. According to the method, the abnormal state judgment result can be accurately output, and the defect that the abnormal state of the communication power supply cannot be accurately analyzed at present is overcome.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a method and system for remote monitoring of communication power supplies based on the Internet of Things. Background Art

[0002] Once a communication power supply fails, the communication system will be paralyzed and data transmission will be interrupted, which not only seriously affects the quality of communication services and brings a bad experience to users, but may also cause incalculable economic losses.

[0003] Currently, traditional remote monitoring systems monitor the operating parameters of communication power supplies remotely and then perform data analysis. However, in data processing, they lack effective algorithms to mine the information behind the data, have poor diagnostic capabilities for complex faults, and cannot determine the type and cause of the faults.

[0004] Therefore, the existing communication power supply monitoring methods have defects in data processing, analysis and other aspects, and cannot accurately analyze the abnormal state of the communication power supply. Summary of the Invention

[0005] The main purpose of the present invention is to provide a method and system for remote monitoring of communication power supplies based on the Internet of Things, aiming to overcome the defect that the abnormal state of the communication power supply cannot be accurately analyzed at present.

[0006] To achieve the above purpose, the present invention provides a method for remote monitoring of communication power supplies based on the Internet of Things, including the following steps:

[0007] Collect multi-dimensional operating data of the communication power supply to obtain raw data;

[0008] Use the DTW algorithm to calculate the time series similarity between the raw data and the historical normal operating data to obtain a similarity matrix;

[0009] Quantitatively analyze the causal relationship between the operating data of each dimension in the raw data based on Granger causality test to obtain a causal association matrix;

[0010] Fuse the similarity matrix and the causal association matrix to obtain a feature enhancement matrix; input the feature enhancement matrix into a deep neural network and output an abnormal state judgment result;

[0011] Encrypt the abnormal state judgment result through an Internet of Things communication module and send it to a remote monitoring center to realize remote monitoring of the communication power supply.

[0012] Further, the multi-dimensional operating data includes voltage, current, temperature, power, humidity, internal resistance, state of charge of the battery, electromagnetic interference intensity, and ripple coefficient.

[0013] Further, before calculating the time series similarity between the original data and the historical normal operation data using the DTW algorithm, the following steps are included:

[0014] The median filtering algorithm is used to remove the impulse noise in the original data, and the missing values are filled by linear interpolation.

[0015] Further, the quantification analysis of the causal relationship between the operation data of each dimension in the original data based on the Granger causality test to obtain the causal association matrix includes:

[0016] Based on the local outlier factor algorithm, the outliers in the original data are identified and removed, and smoothing processing is performed to obtain preprocessed data; the information entropy of the operation data of each dimension in the preprocessed data is calculated, and the dimensions with information entropy values higher than the preset threshold are selected as key dimensions;

[0017] The rescaled range analysis method is used to calculate the Hurst exponent of the operation data of each key dimension, and the time scale suitable for the data of each key dimension is determined according to the Hurst exponent;

[0018] For the operation data of each key dimension, the Granger causality test is performed at the time scale; the Granger causality test includes judging whether there is a Granger causal relationship between each key dimension by constructing an autoregressive model at the time scale;

[0019] For the key dimension pairs with Granger causal relationships, the cumulative effect value of their impulse response functions is calculated as an index of the causal relationship strength;

[0020] Based on the existence situation of the causal relationship between each key dimension and the causal relationship strength value, they are arranged in matrix form according to the order of the key dimensions to obtain the causal association matrix.

[0021] Further, the quantification analysis of the causal relationship between the operation data of each dimension in the original data based on the Granger causality test to obtain the causal association matrix includes:

[0022] The wavelet packet decomposition method is used to decompose the operation data of each dimension into different frequency bands, and according to the energy distribution of the signals in each frequency band, the main frequency band components are screened out for reconstruction to obtain preliminarily purified data; the manifold learning algorithm based on locally linear embedding is used to reduce the dimension of the preliminarily purified data to obtain reduced-dimensional data;

[0023] The autocorrelation function and mutual information entropy between every two dimensions in the reduced-dimensional data are calculated to determine the autocorrelation time scale, and the time interval corresponding to the maximum mutual information entropy is found; the autocorrelation time scale and the time interval are weighted and averaged to obtain the optimal time lag for the Granger causality test between two dimensions;

[0024] Divide the dimension-reduced data into multiple data blocks of equal length in chronological order; perform Granger causality test on each data block with the optimal time lag to obtain a local causal relationship matrix corresponding to each data block;

[0025] All local causal relationship matrices are fused to obtain a causal association matrix that can reflect the dynamic changes of data over time.

[0026] Furthermore, during the Granger causality test, the Bayesian information criterion is used to automatically select the optimal model order;

[0027] After Granger causality test, for each causal relationship in the local causal relationship matrix, the initial causal strength quantitative index is calculated;

[0028] The concept of conditional entropy in information theory is introduced to correct the initial causal strength.

[0029] Furthermore, encrypting the abnormal state judgment result and sending it to the remote monitoring center includes:

[0030] Generate initial quantum key;

[0031] Based on Logistic chaotic mapping, a chaotic sequence is generated with initial parameters, an initial quantum key is XOR-ed with the initial parameters of the chaotic sequence to obtain the initial parameters of the chaotic sequence; and an associated chaotic sequence associated with the initial quantum key is generated with the initial parameters of the chaotic sequence;

[0032] Binary-encode the abnormal state judgment result to obtain a binary data string;

[0033] The generated associated chaotic sequence is intercepted according to the same length as the binary data string, and the intercepted chaotic sequence and the binary data string are subjected to bit-by-bit XOR operation to obtain encrypted data;

[0034] The encrypted data is sent to the remote monitoring center through the IoT communication module.

[0035] Furthermore, encrypting the abnormal state judgment result and sending it to the remote monitoring center includes:

[0036] Analyze the original data based on the deep learning model, explore the correlation between dimensions, and construct a feature map;

[0037] Extracting key feature parameters from the feature map, inputting the key feature parameters into a preset cryptographic algorithm, and dynamically generating a set of data replacement rules;

[0038] Convert the abnormal state judgment result into a specific coding sequence, and rearrange and combine the specific coding sequence based on the generated data replacement rule to complete encryption and obtain encrypted data;

[0039] Send the encrypted data to the remote monitoring center through the Internet of Things communication module.

[0040] The present invention also provides an Internet of Things-based remote monitoring system for communication power supplies, including:

[0041] An acquisition module for acquiring multi-dimensional operation data of the communication power supply to obtain original data;

[0042] A calculation module for calculating the time series similarity between the original data and the historical normal operation data using the DTW algorithm to obtain a similarity matrix;

[0043] An analysis module for quantitatively analyzing the causal relationship between the operation data of each dimension in the original data based on the Granger causality test to obtain a causal association matrix;

[0044] An output module for fusing the similarity matrix and the causal association matrix to obtain a feature enhancement matrix; inputting the feature enhancement matrix into a deep neural network and outputting an abnormal state judgment result;

[0045] A monitoring module for encrypting and sending the abnormal state judgment result to the remote monitoring center through the Internet of Things communication module to achieve remote monitoring of the communication power supply.

[0046] The present invention also provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the method described in any one of the above are implemented.

[0047] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in any one of the above are implemented.

[0048] The communication power supply remote monitoring method and system based on the Internet of Things provided by the present invention include: collecting multi-dimensional operation data of the communication power supply to obtain original data; using the DTW algorithm to calculate the time series similarity between the original data and the historical normal operation data to obtain a similarity matrix; quantitatively analyzing the causal relationship between the operation data of each dimension in the original data based on the Granger causality test to obtain a causal association matrix; fusing the similarity matrix and the causal association matrix to obtain a feature enhancement matrix; inputting the feature enhancement matrix into a deep neural network to output an abnormal state judgment result; encrypting the abnormal state judgment result through an Internet of Things communication module and sending it to a remote monitoring center to achieve remote monitoring of the communication power supply. In the present invention, by combining the DTW algorithm and the Granger causality test to obtain the similarity matrix and the causal association matrix, and fusing them and inputting them into the deep neural network, the abnormal state judgment result can be accurately output, overcoming the defect that the abnormal state of the communication power supply cannot be accurately analyzed at present. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a schematic diagram of the steps of the communication power supply remote monitoring method based on the Internet of Things in an embodiment of the present invention;

[0050] Figure 2 is a block diagram of the structure of the communication power supply remote monitoring system based on the Internet of Things in an embodiment of the present invention;

[0051] Figure 3 is a schematic block diagram of the structure of a computer device in an embodiment of the present invention.

[0052] The implementation, functional features and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0054] Refer to Figure 1 , an embodiment of the present invention provides a communication power supply remote monitoring method based on the Internet of Things, including the following steps:

[0055] Step S1, collect multi-dimensional operation data of the communication power supply to obtain original data;

[0056] Step S2, use the DTW algorithm to calculate the time series similarity between the original data and the historical normal operation data to obtain a similarity matrix;

[0057] Step S3: Quantitatively analyze the causal relationship between the operation data of each dimension in the original data based on Granger causality test to obtain a causal association matrix;

[0058] Step S4: Fuse the similarity matrix and the causal association matrix to obtain a feature enhancement matrix; input the feature enhancement matrix into a deep neural network to output an abnormal state judgment result;

[0059] Step S5: Encrypt the abnormal state judgment result through the IoT communication module and send it to the remote monitoring center to achieve remote monitoring of the communication power supply.

[0060] In this embodiment, as described in step S1 above, the communication power supply is the core power source of the communication system, and the stability of its operation state is directly related to the normal operation of the communication network. The multi-dimensional operation data can comprehensively reflect the working conditions of the communication power supply from different angles, providing a basic basis for accurately judging whether it is in an abnormal state subsequently.

[0061] In this embodiment, data collection is realized by means of various types of sensors distributed at key parts of the communication power supply. For example, a voltage sensor can monitor the value of the power supply output voltage in real time, a current sensor can obtain the magnitude of the current, a power sensor is used to measure the power consumption of the power supply, a temperature sensor can sense the heat generation during the operation of the power supply, a humidity sensor can detect the humidity level of the environment where the power supply is located, and there may also be sensors for monitoring the electromagnetic interference intensity, etc. The original data collected has the characteristics of multi-dimension and dynamic change. The data of different dimensions are interrelated and each contains different information, and will change in real time with the operation of the communication power supply and the change of the external environment.

[0062] As described in step S2 above, the dynamic time warping (DTW) algorithm is mainly used to solve the problems of stretching and offset that may exist in time series data on the time axis. In the actual operation process of the communication power supply, due to the influence of various factors, even when it is in a normal state, the data collected at different time periods may not be completely aligned on the time axis, but they may essentially have similar change patterns. The DTW algorithm finds the optimal matching path between two time series to minimize the sum of the distances between corresponding points, thereby accurately measuring the similarity between them.

[0063] Specifically, the currently collected raw data and the pre-stored historical normal operation data are regarded as two time series respectively. The algorithm constructs a two-dimensional matrix, where each element in the matrix represents the distance between a point in the raw data sequence and a point in the historical normal data sequence. Then, an optimal path is searched in this matrix through the method of dynamic programming. The sum of the elements on this path is the DTW distance between the two time series. The smaller the distance, the higher the similarity. The above calculations are performed on each dimension of the raw data with the historical normal data respectively, and finally a similarity matrix is obtained. Each element in the matrix represents the similarity between the corresponding dimensions. The similarity matrix can intuitively show the similarity degree of the currently collected raw data and the historical normal data in each dimension, providing an important reference basis for subsequent judgment of whether the communication power supply has abnormalities. If the similarity of a certain dimension is low, it may mean that there are abnormal changes in the operating parameters corresponding to this dimension.

[0064] As described in step S3 above, Granger causality test is a statistical method used to judge the causal relationship between time series variables. In the multi-dimensional operation data of the communication power supply, there may be causal relationships of mutual influence between data in different dimensions. For example, an increase in the temperature of the power supply will cause a change in its output voltage, or a sudden increase in current may cause an increase in power. The basic idea of the Granger causality test is that if the past values of a variable (such as A) can help predict the future values of another variable (such as B), then A can be considered the Granger cause of B.

[0065] Specifically, Granger causality tests are performed on the time series data of any two dimensions in the raw data. First, the lag order of the test needs to be determined, that is, how many past time points of data are considered to affect the future value. Then, by establishing a regression model, the prediction effects of the current variable are respectively tested with and without the past values of another variable. If the difference is significant, it indicates that there is a causal relationship, and a quantitative index of the causal relationship, such as the F statistic, is calculated according to the test results. The above tests are performed on all pairwise combinations between dimensions, and finally a causal association matrix is obtained. The elements in the matrix represent the strength of the causal relationship between the corresponding dimensions. The causal association matrix helps to deeply understand the internal connections between the operating parameters of the communication power supply, clarify which parameter changes are caused by other parameters, and which parameters are the causes of other parameter changes. This is of great significance for accurately diagnosing the root cause of communication power supply failures and effective fault prediction.

[0066] As described in step S4 above, the similarity matrix mainly reflects the similarity between the current data and the historical normal data, while the causal association matrix reflects the causal relationship between the data in each dimension. By fusing these two matrices, the information contained in each of them can be fully utilized to obtain a feature enhancement matrix containing more key features. The fusion method can be simple weighted summation or a more complex non-linear fusion method. Select an appropriate fusion strategy according to the actual situation to highlight the role of important information in different matrices.

[0067] The above deep neural network is a machine learning model with powerful learning ability. It can automatically learn complex patterns and features from a large amount of data. In this step, the feature enhancement matrix is passed as input to a pre-trained deep neural network. This network usually consists of multiple hidden layers, each hidden layer contains multiple neurons, and the neurons are connected by weights. During the training process, the network continuously adjusts these weights so that for known normal and abnormal samples, the corresponding classification results can be accurately output.

[0068] After the feature enhancement matrix is input into the trained deep neural network, the above deep neural network will analyze and judge the input data according to the learned patterns and output a result indicating whether the communication power supply is in an abnormal state. This result can be a binary judgment (normal or abnormal) or a numerical value representing the probability of abnormality, which is convenient for the operation and maintenance personnel to take corresponding measures according to the actual situation.

[0069] As described in step S5 above, the above Internet of Things communication module is a bridge for realizing data transmission between the communication power supply and the remote monitoring center. Common Internet of Things communication technologies include Wi-Fi, Bluetooth, ZigBee, LoRa, etc. Different communication technologies have different characteristics and application scenarios, and can be selected according to actual needs. These modules can send the abnormal state judgment result in the form of digital signals through the corresponding communication network.

[0070] To ensure the security and privacy of the abnormal state judgment results during transmission, it is necessary to encrypt the data. The encryption algorithm can use a symmetric encryption algorithm (such as AES) or an asymmetric encryption algorithm (such as RSA), or the two algorithms can be used in combination to improve the security of encryption. At the sending end, the encryption key is used to encrypt the abnormal state judgment results and convert them into ciphertext form; at the receiving end, the remote monitoring center uses the corresponding decryption key to decrypt the ciphertext and restore the original abnormal state judgment results. After the remote monitoring center receives the decrypted abnormal state judgment results, the operation and maintenance personnel can understand the operating status of the communication power supply in real time. If the judgment results show that the communication power supply is in an abnormal state, the monitoring center can issue an alarm in a timely manner and provide relevant abnormal information, such as possible fault causes, affected ranges, etc., so that the operation and maintenance personnel can quickly take measures to troubleshoot and repair the faults, thereby realizing the remote real-time monitoring and management of the communication power supply.

[0071] In this embodiment, by combining the DTW algorithm and Granger causality test to obtain the similarity matrix and the causal association matrix, and fusing and inputting them into the deep neural network, the abnormal state judgment results can be accurately output, overcoming the defect that the abnormal state of the communication power supply cannot be accurately analyzed currently.

[0072] In one embodiment, the multi-dimensional operation data includes voltage, current, temperature, power, humidity, internal resistance, state of charge of the battery, electromagnetic interference intensity, and ripple coefficient.

[0073] In one embodiment, before calculating the time series similarity between the original data and the historical normal operation data using the DTW algorithm, it includes:

[0074] The median filtering algorithm is used to remove the impulse noise in the original data, and the missing values are filled by linear interpolation.

[0075] In this embodiment, when collecting the multi-dimensional operation data of the communication power supply, external interferences (such as electromagnetic interference, sudden sensor failures, etc.) will cause impulse noise to be mixed into the data. The impulse noise appears as isolated outliers in the data, with a large deviation from the normal data, which will seriously affect the subsequent data processing and analysis results, such as increasing the similarity calculation error and inaccurate causal relationship analysis. Therefore, removing the impulse noise can improve the data quality and provide a reliable basis for subsequent analysis.

[0076] Median filtering is a non - linear filtering method. For each data point in the original data sequence, a set of data within its neighborhood is selected (the neighborhood size can be set according to the actual situation, such as including the point and N data points before and after it). This set of data is sorted by numerical value, and the median value of the sorted sequence is taken as the result of filtering this point. Since impulse noise is usually an isolated extremely large or extremely small value, it is easily excluded from the median value after sorting, thus achieving the purpose of removing noise.

[0077] Operation process: First, determine the size of the filtering window (i.e., the neighborhood range), and then calculate the filtered value of each data point in the original data in turn according to the above principle, and finally obtain a new data sequence after removing impulse noise.

[0078] During the data acquisition process, some data may be missing due to reasons such as sensor failures and communication interruptions. Missing values will damage the integrity and continuity of the data, making subsequent analysis unable to proceed normally or resulting in deviations. Therefore, it is necessary to fill in the missing values to ensure the usability of the data and the accuracy of the analysis results. The linear interpolation method is based on the assumption that between two known data points, the change of data is linear. For a data point with a missing value, if there are known data points before and after it, the missing value can be estimated through a linear relationship based on the numerical values of these two known data points and the distance between them.

[0079] In one embodiment, the quantification analysis of the causal relationship between the running data of each dimension in the original data is performed based on Granger causality test to obtain a causal association matrix, including:

[0080] Identify and remove the outliers in the original data based on the local outlier factor algorithm and perform smoothing processing to obtain pre - processed data; calculate the information entropy of the running data of each dimension in the pre - processed data, and select the dimensions with information entropy values higher than the preset threshold as key dimensions;

[0081] Use the rescaled range analysis method to calculate the Hurst exponent of the running data of each key dimension, and determine the time scale suitable for the data of each key dimension according to the Hurst exponent;

[0082] Perform Granger causality test on the running data of each of the key dimensions at the time scale; the Granger causality test includes, at the time scale, by constructing an autoregressive model, determining whether there is a Granger causal relationship between each key dimension;

[0083] For the key dimension pairs with Granger causal relationships, calculate the cumulative effect value of their impulse response functions as an index of causal relationship strength;

[0084] Based on the existence and intensity values of causal relationships among key dimensions, arrange them in matrix form according to the order of key dimensions to obtain a causal association matrix.

[0085] In this embodiment, there may be outliers in the original data. These outliers may be caused by measurement errors, sudden interferences, etc. They will seriously interfere with the subsequent causal relationship analysis and cause deviations in the analysis results. Smoothing processing is to reduce the random fluctuations in the data, highlight the trend characteristics of the data, and make the data more stable and easy to analyze.

[0086] The Local Outlier Factor (LOF) algorithm determines whether a data point is an outlier by calculating the local density of each data point relative to other data points in its neighborhood. If the local density of a data point is much lower than that of other data points in its neighborhood, it is considered an outlier. Smoothing processing can use methods such as moving average and exponential smoothing to eliminate random noise by weighted averaging the data.

[0087] First, use the LOF algorithm to scan the original data, calculate the LOF value of each data point, set a suitable threshold, and identify and remove the data points with LOF values higher than this threshold. Then, perform smoothing processing on the data after removing outliers. For example, use the simple moving average method to calculate the average value of each data point and several data points before and after it as the smoothed value of this point, and finally obtain the preprocessed data.

[0088] In multi-dimensional running data, the importance of data in different dimensions for causal relationship analysis may be different. Information entropy can measure the uncertainty and amount of information of data. Dimensions with higher information entropy values usually contain more variation information and may be more critical for causal relationship analysis. By selecting key dimensions, unnecessary computational effort can be reduced and the analysis efficiency can be improved.

[0089] Information entropy is a concept in information theory, which represents the degree of uncertainty of a random variable. Calculate the information entropy for each dimension of the preprocessed data respectively, set a preset threshold, and screen out the dimensions with information entropy values higher than this threshold as key dimensions for subsequent causal relationship analysis.

[0090] Data in different dimensions have different time characteristics. For example, some data show obvious change patterns on a relatively short time scale, while some data need to show their trends on a longer time scale. The Hurst exponent can reflect the long-term memory and trend of time series data. By calculating the Hurst exponent, the appropriate time scale for analyzing the data of each key dimension can be determined, making the Granger causality test more accurate.

[0091] The rescaled range analysis (R / S analysis) is a method used to analyze the long-term memory of time series. The Hurst exponent (H) ranges from 0 to 1. When H = 0.5, the time series is a random walk and there is no long-term memory. When H > 0.5, the time series has positive long-term memory, that is, the past trend will continue into the future. When H < 0.5, the time series has negative long-term memory, that is, the past trend will reverse.

[0092] Perform R / S analysis on the operation data of each key dimension and calculate its Hurst exponent. Determine the time scale suitable for the data of this dimension according to the size of the Hurst exponent. For example, for dimensions with a larger H value, a longer time scale can be selected for analysis; for dimensions with an H value close to 0.5, a moderate time scale can be selected.

[0093] Granger causality test is used to judge whether there is a causal relationship between two time series. In the multi-dimensional operation data of communication power supplies, understanding the causal relationships between key dimensions helps to deeply understand the operation mechanism of the power supply and provides a basis for fault diagnosis and prediction. The basic idea of the Granger causality test is that if the past values of a variable X can help predict the future values of another variable Y, and this prediction effect is better than only using the past values of Y, then X is considered the Granger cause of Y. The test is achieved by establishing an autoregressive model. Consider the prediction ability of two regression models for Y, one including the past values of X and the other not, and judge whether there is a causal relationship by comparing the goodness of fit of these two models (usually using the F test).

[0094] Specifically, under the determined time scale, pair up the operation data of each key dimension. For each pair of data, establish autoregressive models with and without the past values of the other party respectively. Compare the goodness of fit of the two models through the F test to judge whether there is a Granger causal relationship. For key dimension pairs with a Granger causal relationship, calculate the cumulative effect value of their impulse response function as an indicator of the strength of the causal relationship. Just judging that there is a causal relationship between two key dimensions is not enough; it is also necessary to understand the strength of this causal relationship. The impulse response function can describe the dynamic impact of a unit impulse of one variable on another variable. By calculating its cumulative effect value, the strength of the causal relationship can be quantified. The impulse response function is defined based on the vector autoregressive (VAR) model, which represents the response of other variables over time after applying a unit impulse to one variable at a certain moment. The cumulative effect value is the integral of the impulse response function within a certain time range, which reflects the overall impact degree of the impulse on another variable.

[0095] For the key dimension pairs with Granger causality, establish the corresponding VAR model and calculate its impulse response function. Then sum the impulse response function within a certain time range to obtain the cumulative effect value, which is used as the causality strength index.

[0096] Organize the existence and strength of the causal relationships between the key dimensions in the form of a matrix for subsequent analysis and processing. The causal association matrix can intuitively display the causal relationship network between the key dimensions, which helps to quickly identify which dimensions have strong causal connections. The rows and columns of the matrix correspond to the respective key dimensions, and the elements in the matrix represent the causal relationship between the corresponding two key dimensions. If there is Granger causality between two key dimensions, the element value is the corresponding causality strength index; if there is no causal relationship, the element value is 0. Fill in the existence and strength values of the causal relationships between the key dimensions into the matrix in the order of the key dimensions to finally obtain the causal association matrix.

[0097] In one embodiment, the method for quantitatively analyzing the causal relationships between the operation data of each dimension in the original data based on Granger causality test to obtain a causal association matrix includes:

[0098] Use the wavelet packet decomposition method to decompose the operation data of each dimension into different frequency bands, and filter out the main frequency band components for reconstruction according to the energy distribution of the signals in each frequency band to obtain the preliminary purified data; perform dimensionality reduction on the preliminary purified data through the manifold learning algorithm based on locally linear embedding to obtain the dimensionality-reduced data;

[0099] Calculate the autocorrelation function and mutual information entropy between every two dimensions in the dimensionality-reduced data to determine the autocorrelation time scale and find the time interval corresponding to the maximum mutual information entropy; perform weighted average calculation on the autocorrelation time scale and the time interval to obtain the optimal time lag for Granger causality test between two dimensions.

[0100] Divide the dimensionality-reduced data into multiple equally long data blocks in chronological order; perform Granger causality test on each data block with the optimal time lag to obtain the local causal relationship matrix corresponding to each data block;

[0101] Fuse all the local causal relationship matrices to obtain a causal association matrix that can reflect the dynamic changes of the data over time.

[0102] In this embodiment, the original data may contain noise and redundant information, which affects the accuracy of causal relationship analysis. Wavelet packet decomposition can decompose the operation data of each dimension into different frequency bands. By screening the main frequency band components for reconstruction, noise and secondary information can be removed, the data can be preliminarily purified, and the key features in the data can be highlighted. Wavelet packet decomposition is a more refined time-frequency analysis method than wavelet decomposition. It can perform multi-level decomposition on both the high-frequency and low-frequency parts of the signal simultaneously. Each frequency band corresponds to a different frequency range, and the energy distribution of the signal in different frequency bands reflects the composition of its frequency components. The wavelet packet decomposition algorithm is used to decompose the operation data of each dimension into different frequency bands, and the energy of the signals in each frequency band is calculated. An energy ratio threshold is set, and the main frequency band components with an energy ratio higher than this threshold are screened out and reconstructed to obtain preliminarily purified data.

[0103] The data after preliminary purification may still have a high dimension, which will increase the computational complexity and analysis difficulty. The manifold learning algorithm based on Locally Linear Embedding (LLE) can map the data to a low-dimensional space while preserving the local geometric structure of the data, reduce the data dimension, and improve the computational efficiency.

[0104] The LLE algorithm assumes that the data is distributed on a low-dimensional manifold, and each data point can be linearly represented by other data points within its neighborhood. By finding the weights of this linear representation and applying these weights in the low-dimensional space, the dimensionality reduction of the data is achieved. The preliminarily purified data is input into the LLE algorithm, parameters such as the neighborhood size are determined, and the algorithm will calculate the linear representation weights of each data point and map the data to a low-dimensional space to obtain the dimensionality-reduced data.

[0105] The autocorrelation function can reflect the correlation of the data at different time points and is used to determine the autocorrelation time scale of the data; the mutual information entropy measures the degree of information sharing between two variables. Finding the time interval corresponding to the maximum value of the mutual information entropy can reflect the maximum correlation time point between the data of two dimensions. Combining the two can more accurately determine the time lag of the Granger causality test. The autocorrelation function is a function that measures the correlation of a time series at different time points, and the larger its value, the stronger the correlation. The mutual information entropy is a concept in information theory used to measure the dependence between two random variables, and the larger the value, the higher the degree of information sharing between the two variables. For each pair of dimensions in the dimensionality-reduced data, their autocorrelation functions and mutual information entropies are calculated respectively. The autocorrelation time scale is determined through the autocorrelation function curve, and the time interval corresponding to the maximum value of the mutual information entropy is found by searching the mutual information entropy curve.

[0106] The time intervals corresponding to the maximum autocorrelation time scale and mutual information entropy are weighted and averaged. By comprehensively considering the autocorrelation of the data and the maximum correlation between dimensions, a more reasonable time lag for Granger causality test is obtained, improving the accuracy of causal relationship analysis. According to the data characteristics and experience, the weights of the autocorrelation time scale and the time interval are set, and the two are weighted and averaged to obtain the optimal time lag for Granger causality test between every two dimensions.

[0107] The dimensionality-reduced data is divided into multiple equally long data blocks in chronological order to consider the dynamic change characteristics of the data over time. The data in different time periods may have different causal relationship patterns, and block processing can capture this dynamic change more carefully. Specifically, the length of the data block is determined, and the dimensionality-reduced data is divided into multiple equally long data blocks in chronological order.

[0108] Granger causality test is performed on each data block with the optimal time lag to obtain the local causal relationship matrix corresponding to each data block, which can reflect the causal relationship between dimensions within each time period. The Granger causality test determines whether there is a causal relationship by comparing the predictive ability of the past values of one variable for the future values of another variable. An autoregressive model is established within each data block, and an F-test is performed to determine the causal relationship. Within each data block, with the optimal time lag as a parameter, Granger causality test is performed on every two dimensions. According to the test results, the existence of the causal relationship is determined, and the results are organized into a local causal relationship matrix.

[0109] Each local causal relationship matrix reflects the causal relationship in different time periods. By fusing them, a causal association matrix that can reflect the dynamic change of the data over time can be obtained, more comprehensively showing the evolution of the causal relationship between dimensions. Methods such as weighted average and moving average can be used to fuse all local causal relationship matrices. For example, higher weights are assigned to the relatively new data blocks to highlight the changes in recent causal relationships, and finally a causal association matrix is obtained.

[0110] In one embodiment, during the Granger causality test, the Bayesian Information Criterion (BIC) is used to automatically select the optimal model order. When performing the Granger causality test, an autoregressive model needs to be established. The selection of the model order is crucial. An overly high order will lead to overfitting of the model, increasing the computational complexity and potentially introducing noise. An overly low order will cause the model to fail to fully capture the characteristics and patterns of the data, resulting in inaccurate test results. The BIC can find a balance between the goodness-of-fit and complexity of the model, automatically select the optimal model order, and improve the accuracy and reliability of the Granger causality test. The basic idea of the BIC criterion is to penalize the complexity of the model while considering the model's ability to fit the data. As the model order increases, the goodness-of-fit usually improves, but the model complexity also increases, and the BIC value will first decrease and then increase. The order corresponding to the minimum BIC value is the optimal order.

[0111] Specifically, set a range of possible orders, for example, from 1 to a preset maximum order. For each order within this range, establish an autoregressive model of the corresponding order and perform the Granger causality test. Calculate the BIC value for each model. Compare all the BIC values and select the order with the minimum BIC value as the optimal model order.

[0112] After performing the Granger causality test, for each causal relationship in the local causal relationship matrix, calculate the initial causal strength quantification index;

[0113] Introduce the concept of conditional entropy in information theory to correct the initial causal strength.

[0114] In this embodiment, after completing the Granger causality test and determining that there is a causal relationship between dimensions, it is necessary to quantify the strength of this causal relationship. The initial causal strength quantification index can intuitively reflect the magnitude of the causal influence of one variable on another variable, helping to further analyze and understand the interaction relationship between dimensions.

[0115] A common method for calculating the initial causal strength quantification index is based on the F-statistic in the Granger causality test. The F-statistic measures the degree of improvement in the predictive ability of another variable after including the lag terms of one variable. The larger the value of the F-statistic, the more significant the contribution of introducing the lag terms of this variable to predicting another variable, that is, the stronger the causal relationship.

[0116] For each causal relationship (i.e., the pair of dimensions determined to have a causal relationship through the Granger causality test) in the local causal relationship matrix, extract the value of the F-statistic from the results of the Granger causality test. Use the value of the F-statistic as the initial causal strength quantification index.

[0117] Although the initial causal strength quantification index (such as the F - statistic) can reflect the strength of the causal relationship to a certain extent, it does not fully consider the information transmission between variables and the complexity of the dependence relationship. Conditional entropy is a concept in information theory, which can measure the uncertainty of another variable under the condition that one variable is known. Introducing conditional entropy to correct the initial causal strength can more comprehensively and accurately reflect the essence of the causal relationship and avoid the bias that may be brought by indicators based solely on statistical tests.

[0118] Conditional entropy represents the average uncertainty of variable B under the condition that variable A is known. If the conditional entropy of another variable is smaller when one variable is known, it indicates that the dependence relationship between the two variables is stronger, and the causal relationship may also be stronger. By combining conditional entropy with the initial causal strength, the initial causal strength can be adjusted and corrected.

[0119] In one embodiment, after encrypting the abnormal state judgment result, it is sent to the remote monitoring center, including:

[0120] Generate an initial quantum key;

[0121] Based on the Logistic chaotic map, generate a chaotic sequence with the initial parameters, perform an exclusive - OR operation on the initial quantum key and the initial parameters of the chaotic sequence to obtain the initial parameters of the chaotic sequence; generate an associated chaotic sequence associated with the initial quantum key with the initial parameters of the chaotic sequence;

[0122] Perform binary encoding on the abnormal state judgment result to obtain a binary data string;

[0123] Intercept the generated associated chaotic sequence according to the same length as the binary data string, and perform a bit - by - bit exclusive - OR operation on the intercepted chaotic sequence and the binary data string to obtain the encrypted data;

[0124] Send the encrypted data to the remote monitoring center through the Internet of Things communication module.

[0125] In this embodiment, first, an initial quantum key is generated. The quantum key has extremely high security and can prevent the key from being eavesdropped and copied during transmission. Generating the initial quantum key can provide a secure basis for subsequent encryption of the abnormal state judgment result, ensuring the confidentiality and integrity of data transmission. Quantum key distribution utilizes the principles of quantum mechanics, such as the common BB84 protocol. The sender (communication power supply end) represents binary information with quantum states and transmits it to the receiver (remote monitoring center) through a special quantum channel. After the receiver makes a measurement, the two parties then communicate the measurement methods through an ordinary classical channel and screen out the data corresponding to the same measurement method, and these data form the initial quantum key.

[0126] In this embodiment, the communication power supply end and the remote monitoring center first establish connections for both the quantum channel and the classical channel. The communication power supply end randomly generates quantum states representing binary information according to the BB84 protocol and sends them out. After receiving, the remote monitoring center makes measurements and feeds back the measurement method to the communication power supply end through the classical channel. The two parties select the data with consistent measurement methods to form an initial quantum key.

[0127] The above Logistic chaotic mapping can generate seemingly random and complex sequences and is very sensitive to initial conditions. Combining the initial quantum key with the original initial parameters of the chaotic sequence can make the generated chaotic sequence associated with the quantum key, enhancing the security of encryption. Even if an attacker obtains some information about the chaotic sequence, it is difficult to restore it without the quantum key.

[0128] Specifically, through a bit operation (XOR operation), the information of the initial quantum key is incorporated into the initial parameters of the chaotic sequence, making the chaotic sequence closely associated with the quantum key.

[0129] Operation process: First, select the initial parameters of the Logistic chaotic mapping. Convert the initial quantum key into binary form. Perform an XOR operation on the binary-form initial quantum key and the initial parameters to obtain new initial parameters for the chaotic sequence.

[0130] Use the new initial parameters associated with the initial quantum key to generate a chaotic sequence, which will be used to encrypt the abnormal state judgment result. Because it is associated with the quantum key, it makes the encryption process more secure. Starting from the newly obtained initial parameters of the chaotic sequence, continuously calculate according to the rules of the Logistic chaotic mapping to generate the chaotic sequence. The number of calculations is determined according to the length of the sequence required for subsequent encryption.

[0131] Computers generally process data in binary form. Compile the abnormal state judgment result into a binary data string, which is convenient for computer processing and also facilitates subsequent encryption operations with the chaotic sequence. According to the data type and format of the abnormal state judgment result, select an appropriate encoding method. For example, if it is text information, use common ASCII code or Unicode code for encoding; if it is numerical information, convert it into a binary value. Finally, turn the result into a binary data string.

[0132] Perform a bit-by-bit XOR operation on the associated chaotic sequence and the binary data string to encrypt the abnormal state judgment result. The XOR operation is simple, efficient, and reversible, and can effectively scramble the original data, making the data difficult to be cracked during transmission. Specifically, intercept the generated associated chaotic sequence to the same length as the binary data string. Perform an XOR operation on the corresponding bits of the intercepted chaotic sequence and the binary data string to obtain the encrypted binary data string, that is, the encrypted data.

[0133] Finally, the encrypted data is transmitted to the remote monitoring center through the Internet of Things communication module to achieve the remote transmission of the abnormal state judgment result. Since the data has been encrypted, even if intercepted during transmission, it is difficult for attackers to obtain sensitive information, ensuring the security of data transmission.

[0134] In one embodiment, sending the abnormal state judgment result to the remote monitoring center after encryption includes:

[0135] Analyze the original data based on the deep learning model, mine the correlation rules between dimensions, and construct them into a feature map;

[0136] Extract key feature parameters from the feature map, and input the key feature parameters into a preset cryptographic algorithm to dynamically generate a set of data permutation rules;

[0137] Convert the abnormal state judgment result into a specific coding sequence, and rearrange and combine the specific coding sequence based on the generated data permutation rules to complete encryption and obtain encrypted data;

[0138] Send the encrypted data to the remote monitoring center through the Internet of Things communication module.

[0139] In this embodiment, the original data contains the operation information of multiple dimensions of the communication power supply, but the correlations between these information are complex and not intuitive. By analyzing the original data through the deep learning model, the hidden correlation rules between dimensions can be mined. The feature map can present these correlations in an intuitive way, providing rich and valuable feature information for the subsequent encryption process, making the encryption rules closely combined with the internal characteristics of the data itself, and enhancing the security and pertinence of encryption.

[0140] The deep learning model has powerful learning and representation capabilities, and can automatically learn complex patterns and features from a large amount of data. Common deep learning models such as convolutional neural networks (CNNs), recurrent neural networks (RNNs) and their variants (such as LSTMs, GRUs), etc., can select appropriate models according to the characteristics and structures of the original data. Through the non-linear transformation of multiple layers of neurons, the model extracts and transforms the features of the original data, finally learns the correlation relationships between dimensions, and visualizes and structurally represents these relationships in the form of a map.

[0141] The feature map contains a large amount of feature information, but not all information is equally important for the encryption process. Extracting key feature parameters can focus on those features that have a greater impact on the internal structure and correlations of the data, reduce redundant information, and improve the efficiency and effectiveness of generating encryption rules in the subsequent process.

[0142] According to preset criteria (such as the importance score of features, the correlation with the abnormal state judgment result, etc.), key feature parameters are screened out from the feature map. Feature selection algorithms (such as correlation analysis-based, model-based feature selection, etc.) can be used to determine the key features.

[0143] The above-mentioned preset cryptographic algorithm has certain mathematical properties and encryption mechanisms. When the key feature parameters are input into the cryptographic algorithm, the rules and properties of the algorithm can be utilized to dynamically generate a set of data permutation rules closely related to the data itself in combination with the key features of the data. These rules will be used to encrypt the abnormal state judgment result, making the encryption process more flexible and secure.

[0144] Common cryptographic algorithms such as AES, RSA, etc. have different encryption principles and operation rules. Taking the key feature parameters as the input of the algorithm, through the operations and transformations of the algorithm, a set of rules for data permutation are generated. These rules can define operations such as the exchange order of data elements and position transformation.

[0145] The extracted key feature parameters are input into the preset cryptographic algorithm, and calculations and processing are carried out according to the regulations of the algorithm, and finally a set of data permutation rules are obtained.

[0146] To facilitate the encryption operation, the abnormal state judgment result needs to be converted into a specific coding sequence. The coding sequence usually adopts binary or other digital coding forms, which can represent the result in a unified format, facilitating subsequent rearrangement and combination according to the data permutation rules. Specifically, according to the data type and format of the abnormal state judgment result, a suitable coding method is selected. For example, if the result is text information, ASCII code or Unicode code can be used for coding; if it is numerical information, it can be converted into binary numerical representation. Finally, the abnormal state judgment result is converted into a specific coding sequence.

[0147] Furthermore, the generated data permutation rules are used to rearrange and combine the specific coding sequence, changing the original order and structure of the data, and realizing the encryption of the abnormal state judgment result. This permutation rule generated based on the internal features of the data makes the encrypted result have high security, and it is difficult for attackers to crack the encrypted data without understanding the rule generation mechanism.

[0148] In this embodiment, the elements in the specific coding sequence are rearranged and combined according to the data permutation rules. For example, the rule may specify the exchange of elements at certain positions, the moving order of elements, etc. Through these operations, the encrypted coding sequence, that is, the encrypted data, is obtained.

[0149] The Internet of Things communication module provides a data transmission channel between the communication power supply and the remote monitoring center. The encrypted data is sent to the remote monitoring center through the Internet of Things communication module to achieve the remote transmission of the abnormal state judgment result. Since the data has been encrypted, it can effectively prevent data from being stolen and tampered with during the transmission process, ensuring the security and integrity of the data.

[0150] In one embodiment, encrypting the abnormal state judgment result and sending it to the remote monitoring center includes:

[0151] Encoding the multi-dimensional operation data using a preset coding table to obtain encoded data;

[0152] Sequentially adding each character in the encoded data to a directed graph; the directed graph includes multiple nodes, and adjacent nodes are connected by directed edges;

[0153] Based on the abnormal state judgment result, updating the encoded characters in the preset coding table to obtain an updated coding table different from the preset coding table;

[0154] Encoding the abnormal state judgment result based on the updated coding table to obtain result encoded data;

[0155] Sequentially adding the characters in the result encoded data one by one to each directed edge of the directed graph;

[0156] Based on the characters on the directed edge and the character feature relationship between the characters on the two nodes connected by the directed edge, determining the target directed edge;

[0157] Combining the characters on each target directed edge in sequence to obtain an encryption key, encrypting the abnormal state judgment result with it, and sending it to the remote monitoring center.

[0158] In this embodiment, the multi-dimensional operation data has various forms and is not conducive to direct processing. Encoding through a preset coding table can convert these complex data into a unified character sequence that is convenient for subsequent operations, laying a foundation for subsequent encryption using a directed graph. The preset coding table defines the correspondence between multi-dimensional operation data and specific characters. For example, data of different dimensions such as voltage values and current values are mapped to corresponding characters according to set rules. Sequentially read the values of each dimension of the multi-dimensional operation data, and according to the mapping rules of the preset coding table, convert each dimension value into the corresponding character, and finally form encoded data.

[0159] Integrate the encoded data into a directed graph. By leveraging the structural characteristics of the directed graph, provide a data carrier for subsequent graph-based encryption operations, and introduce more encryption dimensions through the positional relationship of characters in the directed graph. The nodes and edges of the directed graph form a specific topological structure. By adding characters to it in an orderly manner, connections based on the graph structure can be established between characters. Starting from the starting node of the directed graph, in accordance with the order of characters in the encoded data, add characters to the nodes one by one. For each added character, move to the next node along the directed edge.

[0160] Dynamically update the encoding table according to the abnormal state judgment result, making the encryption rule closely associated with the communication power supply state, enhancing the pertinence and security of encryption, and increasing the difficulty of cracking. Through the preset update rule, adjust the character mapping relationship in the preset encoding table based on certain features in the abnormal state judgment result. Analyze the abnormal state judgment result, extract key features such as the abnormal type and severity, and modify the character mapping relationship of the corresponding data dimension in the preset encoding table according to the preset update rule to generate an updated encoding table. Use the updated encoding table to encode the abnormal state judgment result, making the encoding result consistent with the updated encryption rule to ensure the effectiveness and coherence of encryption. The updated encoding table redefines the mapping relationship between data and characters. Encoding the abnormal state judgment result based on this relationship can obtain encoded data with a different result from the original encoding.

[0161] Further integrate the result encoded data with the directed graph with added encoded data. By the distribution of characters on the directed edges, provide more information for determining the target directed edges and enrich the encryption logic. Directed edges connect nodes. Adding characters to the directed edges can construct a more complex character relationship network in the directed graph. Starting from the starting edge of the directed graph, add the characters in the result encoded data to the directed edges one by one, with each character corresponding to one directed edge.

[0162] By analyzing the character feature relationships, screen out the target directed edges with specific relationships from numerous directed edges. These target directed edges will be used to generate the encryption key, making the generation of the encryption key more logical and secure. Predetermine the character feature relationships, such as the difference in ASCII codes of characters, the parity of characters, etc. Based on these relationships, judge whether a directed edge is a target directed edge. Traverse each directed edge of the directed graph, analyze the characters on the edge and the characters on the two connected nodes, and judge whether the directed edge is a target directed edge according to the predetermined character feature relationships.

[0163] Finally, combine the characters on the target directed edge into an encryption key, and use this key to encrypt the abnormal state judgment result to ensure the security of data during transmission and prevent data from being stolen or tampered with. The encryption key is generated based on the character feature relationship in the directed graph, has randomness and complexity, and can effectively protect data security. Combine the characters on the target directed edge into an encryption key in the order of the target directed edge in the directed graph, use this key to encrypt the abnormal state judgment result, and then send the encrypted data to the remote monitoring center through the Internet of Things communication module.

[0164] Referring to Figure 2 , an embodiment of the present invention also provides an Internet of Things-based remote monitoring system for communication power supplies, including:

[0165] An acquisition module for acquiring multi-dimensional operation data of the communication power supply to obtain original data;

[0166] A calculation module for calculating the time series similarity between the original data and the historical normal operation data using the DTW algorithm to obtain a similarity matrix;

[0167] An analysis module for quantitatively analyzing the causal relationship between the operation data of each dimension in the original data based on Granger causality test to obtain a causal association matrix;

[0168] An output module for fusing the similarity matrix and the causal association matrix to obtain a feature enhancement matrix; inputting the feature enhancement matrix into a deep neural network and outputting an abnormal state judgment result;

[0169] A monitoring module for encrypting the abnormal state judgment result through the Internet of Things communication module and sending it to the remote monitoring center to achieve remote monitoring of the communication power supply.

[0170] In this embodiment, for the specific implementation of each module in the above system embodiment, please refer to that described in the above method embodiment, and details will not be repeated here.

[0171] Referring to Figure 3 , an embodiment of the present invention also provides a computer device, which can be a server, and its internal structure can be as Figure 3As shown in the figure. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements the above method.

[0172] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0173] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0174] In summary, the communication power remote monitoring method and system based on the Internet of Things provided in the embodiments of the present invention include: collecting multi-dimensional operation data of the communication power to obtain raw data; using the DTW algorithm to calculate the time series similarity between the raw data and the historical normal operation data to obtain a similarity matrix; based on Granger causality test, quantitatively analyzing the causal relationship between the operation data of each dimension in the raw data to obtain a causal association matrix; fusing the similarity matrix and the causal association matrix to obtain a feature enhancement matrix; inputting the feature enhancement matrix into a deep neural network to output an abnormal state judgment result; encrypting and sending the abnormal state judgment result to a remote monitoring center through an Internet of Things communication module to achieve remote monitoring of the communication power. In the present invention, by combining the DTW algorithm and Granger causality test to obtain a similarity matrix and a causal association matrix, and fusing them and inputting them into a deep neural network, an abnormal state judgment result can be accurately output, overcoming the defect that the abnormal state of the communication power cannot be accurately analyzed at present.

[0175] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided by the present invention and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0176] It should be noted that in this article, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, apparatus, article, or method including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, apparatus, article, or method. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, apparatus, article, or method including the element.

[0177] The above are only the preferred embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent structural or equivalent process transformation made by using the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A communication power remote monitoring method based on the Internet of Things, characterized in that: The following steps are involved: Collect multi-dimensional operation data of communication power supply to obtain original data; Using the DTW algorithm, the time series similarity between the original data and the historical normal operation data is calculated to obtain a similarity matrix; Based on the Granger causality test, the causal relationship between the operating data of each dimension in the original data is quantitatively analyzed to obtain a causal correlation matrix; The similarity matrix is ​​merged with the causal association matrix to obtain a feature enhancement matrix; the feature enhancement matrix is ​​input into a deep neural network to output an abnormal state judgment result; The abnormal state judgment result is encrypted and sent to the remote monitoring center through the Internet of Things communication module to realize remote monitoring of the communication power supply.

2. The method for remote monitoring of communication power supply based on Internet of Things according to claim 1, characterized in that: The multi-dimensional operating data includes voltage, current, temperature, power, humidity, internal resistance, battery charge state, electromagnetic interference intensity, and ripple factor.

3. The method for remote monitoring of communication power supply based on Internet of Things according to claim 1, characterized in that: Before calculating the time series similarity between the original data and the historical normal operation data using the DTW algorithm, the following steps are included: The median filtering algorithm is used to remove the impulse noise in the original data, and the missing values ​​are filled by linear interpolation.

4. The method for remote monitoring of communication power supply based on Internet of Things according to claim 1, characterized in that: The causal relationship between the running data in each dimension in the original data is quantitatively analyzed based on the Granger causality test to obtain a causal correlation matrix, including: Based on the local outlier factor algorithm, outliers in the original data are identified and removed, and smoothing is performed to obtain preprocessed data; the information entropy of the running data in each dimension of the preprocessed data is calculated, and the dimension with an information entropy value higher than the preset threshold is selected as the key dimension; The Hurst index of the operating data of each key dimension is calculated using the re-standardized range analysis method, and the time scale of the data adaptation of each key dimension is determined based on the Hurst index; For the operation data of each of the key dimensions, a Granger causality test is performed at the time scale; the Granger causality test includes determining whether there is a Granger causal relationship between the key dimensions by constructing an autoregressive model at the time scale; For the key dimension pairs with Granger causality, the cumulative effect value of the impulse response function is calculated as the causal strength indicator; Based on the existence of causal relationships among key dimensions and the causal relationship strength values, the key dimensions are arranged in a matrix form in the order of the key dimensions to obtain a causal association matrix.

5. The method for remote monitoring of communication power supply based on Internet of Things according to claim 1, characterized in that: The causal relationship between the running data in each dimension in the original data is quantitatively analyzed based on the Granger causality test to obtain a causal correlation matrix, including: The wavelet packet decomposition method is used to decompose the running data of each dimension into different frequency bands. According to the energy distribution of the signals in each frequency band, the main frequency band components are screened out for reconstruction to obtain preliminary purified data. The preliminary purified data is reduced in dimension by the manifold learning algorithm based on local linear embedding to obtain reduced-dimensional data. Calculate the autocorrelation function and mutual information entropy between every two dimensions in the dimension reduction data to determine the autocorrelation time scale and find the time interval corresponding to the maximum value of the mutual information entropy; perform weighted average calculation on the autocorrelation time scale and time interval to obtain the optimal time lag for Granger causality test between two dimensions; Divide the dimension-reduced data into multiple data blocks of equal length in chronological order; perform Granger causality test on each data block with the optimal time lag to obtain a local causal relationship matrix corresponding to each data block; All local causal relationship matrices are fused to obtain a causal association matrix that can reflect the dynamic changes of data over time.

6. The method for remote monitoring of communication power supply based on Internet of Things according to claim 5, characterized in that: In the process of Granger causality test, the Bayesian information criterion is used to automatically select the optimal model order; After Granger causality test, for each causal relationship in the local causal relationship matrix, the initial causal strength quantitative index is calculated; The concept of conditional entropy in information theory is introduced to correct the initial causal strength.

7. The method for remote monitoring of communication power supply based on Internet of Things according to claim 1, characterized in that: The abnormal state judgment result is encrypted and sent to the remote monitoring center, including: Generate initial quantum key; Based on Logistic chaotic mapping, a chaotic sequence is generated with initial parameters, an initial quantum key is XOR-ed with the initial parameters of the chaotic sequence to obtain the initial parameters of the chaotic sequence; and an associated chaotic sequence associated with the initial quantum key is generated with the initial parameters of the chaotic sequence; Binary-encode the abnormal state judgment result to obtain a binary data string; The generated associated chaotic sequence is intercepted according to the same length as the binary data string, and the intercepted chaotic sequence and the binary data string are subjected to bit-by-bit XOR operation to obtain encrypted data; The encrypted data is sent to the remote monitoring center through the IoT communication module.

8. The method for remote monitoring of communication power supply based on Internet of Things according to claim 1, characterized in that: The abnormal state judgment result is encrypted and sent to the remote monitoring center, including: Analyze the original data based on the deep learning model, explore the correlation between dimensions, and construct a feature map; Extracting key feature parameters from the feature map, inputting the key feature parameters into a preset cryptographic algorithm, and dynamically generating a set of data replacement rules; Converting the abnormal state judgment result into a specific coding sequence, rearranging and combining the specific coding sequence based on the generated data replacement rule, completing encryption, and obtaining encrypted data; The encrypted data is sent to a remote monitoring center via the Internet of Things communication module.

9. A communication power supply remote monitoring system based on the Internet of Things, characterized in that: include: The acquisition module is used to collect multi-dimensional operation data of the communication power supply and obtain the original data; A calculation module, used to calculate the time series similarity between the original data and the historical normal operation data by using the DTW algorithm to obtain a similarity matrix; An analysis module, for quantitatively analyzing the causal relationship between the operating data of each dimension in the original data based on Granger causality test, to obtain a causal correlation matrix; An output module, used for fusing the similarity matrix with the causal association matrix to obtain a feature enhancement matrix; Inputting the feature enhancement matrix into a deep neural network and outputting an abnormal state judgment result; The monitoring module is used to encrypt the abnormal state judgment result and send it to the remote monitoring center through the Internet of Things communication module to realize remote monitoring of the communication power supply.

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

Images