Comprehensive intelligent early warning method for Internet of Things signals and cross information thereof

Abstract

The invention discloses a comprehensive intelligent early warning method for Internet of Things signals and cross information thereof, relates to the technical field of intelligent supervision and early warning, and solves the technical problems that equipment signals and cross information cannot be comprehensively integrated, and the fault processing efficiency and the system recovery speed are affected. The method comprises the following steps of: establishing a fault knowledge base, screening a pre-selection scheme through multi-dimensional matching, performing quantitative evaluation from three aspects of technical feasibility, resource feasibility and time feasibility, and performing weighted calculation on feasibility indexes, so that the method can be applied to complex data better; the final scheme is determined by comprehensively considering the historical repair effect and the same-type recurrence rate, the solution can be evaluated comprehensively and objectively, it is ensured that the selected scheme is feasible in technology, reasonable in resource and efficient in time, the fault processing efficiency is improved, the operation and maintenance cost is reduced, and stable operation of the Internet of Things system is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent supervision and early warning, and specifically to a comprehensive intelligent early warning method for Internet of Things signals and their cross information. Background Art

[0002] The purpose of alarm is to discover changes and abnormalities in on-site operation data and status through online monitoring, notify relevant users of the detailed information of the abnormalities, help users understand the abnormalities in operation and status, and further take rectification measures.

[0003] According to the patent application with the publication number CN114609462A, an intelligent early warning diagnosis model for comprehensive multi-feature parameters of electrical equipment status is disclosed. The diagnosis model includes a basic feature preset module, an operation feature acquisition module, and a data diagnosis module; the basic feature preset module is used to set the basic fault features of the electrical equipment to be detected; the basic feature preset module includes an insulation parameter preset unit, a wear parameter preset unit, and a temperature resistance parameter preset unit.

[0004] With the wide application of Internet of Things technology, various Internet of Things devices play key roles in different fields. However, the main problems faced in the current Internet of Things early warning field include: First, the data utilization is insufficient, and it is impossible to comprehensively integrate the signals of the device itself and cross information, making it difficult to accurately reflect the true operation status of the system; Second, the early warning model lacks scientificity. The data preprocessing is rough, the feature extraction is incomplete, and the model construction and optimization methods are unreasonable, resulting in poor timeliness and accuracy of early warning, and easy to have false alarms and missed alarms; Third, the selection of fault solution lacks an effective evaluation mechanism, and it is impossible to quickly determine the optimal solution among many solutions, affecting the fault handling efficiency and the system recovery speed, and increasing the operation and maintenance cost. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a comprehensive intelligent early warning method for Internet of Things signals and their cross information, which solves the problems of inability to comprehensively integrate the signals of the device itself and cross information, affecting the fault handling efficiency and the system recovery speed.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A comprehensive intelligent early warning method for Internet of Things signals and their cross information, including the following steps: Obtain the signal data and cross information of the Internet of Things device, and perform denoising, duplicate removal, and missing value processing to obtain preprocessed data; Extract the feature vectors corresponding to the Internet of Things signals and cross information of the preprocessed data, and fuse them to obtain a fusion vector; Divide the fusion vector and select the early warning model framework to establish an intelligent analysis early warning model, and generate normal or abnormal signals in combination with real-time data; Analyze the abnormal state signals, determine the early warning level and establish a fault knowledge base, and match and screen the fault characteristics with the fault knowledge base to obtain a preselected solution; Evaluate and obtain the technical feasibility index, resource feasibility index and time feasibility index of the analyzed preselected solutions, and at the same time perform weighted processing to obtain the feasibility index; Compare the feasibility index with the preset value to screen out the solutions to be selected, determine the selected solution in combination with its corresponding repair effect, and generate solution information.

[0007] As a further solution of the present invention, the specific method for obtaining the fusion vector is as follows: Extract the time-domain characteristics, frequency-domain characteristics and time-frequency domain characteristics of the Internet of Things signals to obtain the Internet of Things signal feature vectors, and at the same time extract the classification characteristics and numerical characteristics of the cross information to obtain the cross information feature vectors, and splice the two to obtain the fusion vector.

[0008] As a further solution of the present invention, the specific method for generating normal or abnormal signals is as follows: Divide the fusion vector dataset into a training set, a validation set and a test set according to a certain ratio, and at the same time normalize each feature in the fusion vector. Then select the early warning model framework according to the fusion vector, select the loss function according to the task type of the early warning model, select the Adam algorithm to adjust the parameters of the early warning model, and input the training set into the early warning model framework for training and optimization to obtain an intelligent analysis early warning model; Then obtain the real-time data and substitute it into the intelligent analysis early warning model, and obtain normal or abnormal signals after being processed by the intelligent analysis early warning model.

[0009] As a further solution of the present invention, the specific method for obtaining the preselected solution is as follows: Set multi-level early warnings, determine the current early warning level according to the abnormal state signals, establish a fault knowledge base based on historical data, extract the fault characteristics corresponding to the early warning level, and match them with the knowledge base; Screen out the same type of solutions according to the fault characteristics and number them as i, and i = 1, 2,..., j, where j represents the number of the same type of solutions. Calculate the similarity between the fault characteristics and each solution, compare the preset threshold, and screen out the preselected solutions with qualified similarity and record them as a, and a = 1, 2,..., b, where b represents the number of preselected solutions.

[0010] As a further solution of the present invention, the specific method for establishing a fault knowledge base based on historical data is as follows: Comprehensively collect various types of faults, phenomena, causes, and corresponding solutions that may occur in devices and components in the Internet of Things system, organize and classify the collected fault information, and organize it according to dimensions such as device type, fault category, and severity to establish a fault knowledge base.

[0011] As a further solution of the present invention, the specific method for evaluating and obtaining the technical feasibility index, resource feasibility index, and time feasibility index of the analyzed preliminary solutions is as follows: Evaluate the technical feasibility, score the preliminary solution and the fault characteristics from three aspects of technical adaptability, maturity, and complexity, sum them after normalization, and obtain the technical feasibility index; Evaluate the resource feasibility, score the preliminary solution and the fault characteristics from three aspects of technical adaptability, maturity, and complexity, sum them after normalization, and obtain the technical feasibility index; Evaluate the time feasibility, score the repair duration, emergency response duration, and time window limit, complete the normalization and summation, and determine the time feasibility index.

[0012] As a further solution of the present invention, the specific method for screening and obtaining the solutions to be selected is as follows: Set the weight coefficients k1, k2, and k3 corresponding to the technical feasibility index Ja, resource feasibility index Za, and time feasibility index Ta, calculate the feasibility index Qa corresponding to the preliminary solution a according to the formula Qa = Ja×k1 + Za×k2 + Ta×k3, and screen out the solutions greater than the preset value Qy and record them as the solutions to be selected.

[0013] As a further solution of the present invention, the specific method for generating solution information is as follows: Obtain the historical data corresponding to the solutions to be selected, obtain the repair effect corresponding to the solutions to be selected according to the historical data, analyze the repair effect at the same time, calculate the recurrence rate of the same type according to the formula recurrence rate of the same type = number of recurrences of the same type / total number of recurrences × 100%, then select the solution to be selected with the smallest recurrence rate of the same type as the standard, determine the selected solution, and generate solution information.

[0014] The present invention provides a comprehensive intelligent early warning method for Internet of Things signals and their cross information. Compared with the prior art, it has the following beneficial effects: The present invention collects the self-signals and cross-information of Internet of Things devices, extracts the corresponding feature vectors, and fuses them to obtain a fused vector. The fused vector dataset is scientifically divided into a training set, a validation set, and a test set. The features are normalized, and an early warning model is constructed based on a decision tree model, and the Adam algorithm is used to optimize the parameters. This refined model construction process can better adapt to complex data compared with traditional early warning models, improving the generalization ability and early warning accuracy of the model.

[0015] The present invention establishes a fault knowledge base, screens preselected solutions through multi-dimensional matching, and quantitatively evaluates them from three aspects: technical feasibility, resource feasibility, and time feasibility. The feasibility index is calculated by weighting, and the final solution is determined by comprehensively considering the historical repair effect and the recurrence rate of the same type. This method can comprehensively and objectively evaluate the solution, ensure that the selected solution is technically feasible, reasonable in resources, and efficient in time, improve the fault handling efficiency, reduce the operation and maintenance cost, and ensure the stable operation of the Internet of Things system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a diagram of the method steps of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment 1 Please refer to Figure 1 , this application provides a comprehensive intelligent early warning method for Internet of Things signals and their cross-information, including the following steps: Step S1: Collect the self-signal data such as the operating status and environmental parameters of Internet of Things devices through various sensors, and at the same time collect other cross-information related to the Internet of Things system, such as weather conditions, geographical information, user behavior data, etc., and perform preprocessing operations such as denoising, removing duplicate data, and filling missing values on the collected data to obtain preprocessed data. This operation in this step is to improve the data quality.

[0019] Step S2: Extract the corresponding feature vectors of the Internet of Things signals and cross-information according to the obtained preprocessed data, and at the same time perform feature-level fusion on the obtained feature vectors to obtain a fused vector, and the specific processing method is as follows: Feature extraction for Internet of Things signals Time-domain feature extraction Mean: Calculate the average value of a signal over a period of time, which reflects the overall level of the signal. For example, for the temperature signal collected by a temperature sensor, the mean can represent the average temperature during that period.

[0020] Frequency domain feature extraction Frequency peak: Find the peak frequencies in the power spectral density, which are usually related to the main components of the signal. For example, in an audio signal, the frequency peaks can correspond to different audio features.

[0021] Time-frequency domain feature extraction Short-time Fourier transform: Divide the signal into several short time segments, perform Fourier transform on each short segment, and obtain a time-frequency diagram, which can simultaneously observe the changes of the signal at different times and frequencies. When analyzing motor faults, the frequency changes of the current signal at different moments can be observed through the short-time Fourier transform, so as to judge the operating state of the motor.

[0022] Feature extraction for cross information Classification feature extraction Bag-of-words model: For text-based cross information, such as user comments, device description information, etc., the bag-of-words model can be used to convert it into a feature vector. First, tokenize the text, and then count the frequency of each word to form a vector, and the dimension of the vector is the size of the vocabulary.

[0023] Numerical feature extraction Directly use the original numerical values: If the cross information itself is numerical, such as longitude and latitude in geographical information, the number of clicks in user behavior data, etc., it can be directly used as an element of the feature vector.

[0024] The obtained feature vectors are fused to obtain a fused vector, and the specific processing methods are as follows: Simple concatenation: Concatenate two feature vectors in sequence to form a new feature vector. For example, if the feature vector of the Internet of Things signal is [1, 2, 3], and the feature vector of the cross information is [4, 5, 6], then the fused feature vector is [1, 2, 3, 4, 5, 6].

[0025] Weighted fusion: Assign different weights to different feature vectors according to the importance of the features, and then perform weighted summation. For example, for the feature vector [1, 2, 3] of the Internet of Things signal and the feature vector [4, 5, 6] of the cross information, if it is considered that the features of the Internet of Things signal are more important, assign a weight of 0.6, and the weight of the cross information feature is 0.4, then the fused feature vector is [0.6×1 + 0.4×4, 0.6×2 + 0.4×5, 0.6×3 + 0.4×6].

[0026] The above two methods are selected according to different requirements.

[0027] Step S3: Establish an intelligent analysis and early warning model based on the obtained fusion vector. At the same time, set a dynamic early warning threshold based on historical data, and substitute the real-time data into the model to analyze the overall operation status and generate a status analysis signal. The status analysis signal includes a normal status signal and an abnormal status signal, and the specific generation method is as follows: Divide the fusion vector dataset into a training set, a validation set, and a test set according to a certain ratio. At the same time, normalize each feature in the fusion vector and map it to a certain interval, such as [0, 1] or [-1, 1]. Then, select an early warning model framework according to the fusion vector. The early warning model frameworks include, for example, decision trees, support vector machines, multi-layer perceptrons, and convolutional neural networks, etc. In this application, a decision tree model is selected as the standard framework for construction. Select an appropriate loss function according to the task type (classification or regression) of the early warning model, and select an optimization algorithm to adjust the parameters of the early warning model. The optimization algorithms include Stochastic Gradient Descent (SGD), Adagrad, Adadelta, RMSProp, Adam, etc. In this application, Adam is used as the standard for analyzing and adjusting the model parameters, and the training set is input into the early warning model framework for training and optimization to obtain an intelligent analysis and early warning model; Then, obtain the real-time data and substitute it into the intelligent analysis and early warning model. After being processed by the intelligent analysis and early warning model, the corresponding status analysis signal is obtained. For the normal status signal, no processing is performed, and for the abnormal status signal, subsequent analysis is performed.

[0028] Step S4: Analyze the obtained abnormal status signal, set different levels of early warnings, and issue different levels of alarms according to the degree of abnormality. For example, when there is a slight abnormality in the Internet of Things signal and the cross-information display system, a yellow early warning is issued; when the abnormal situation is relatively serious, a red early warning is issued and corresponding emergency measures are taken. Determine the early warning level according to the obtained abnormal status signal, and at the same time, obtain historical data and establish a corresponding fault knowledge base based on the historical data. Then, obtain the fault characteristics corresponding to the early warning level and match them with the fault knowledge base to obtain a preliminary solution. The specific matching method is as follows: First, it is necessary to systematically construct a fault knowledge base. Collect data through multiple channels, and comprehensively collect the possible fault types, phenomena, causes, and corresponding solutions of various devices and components in the Internet of Things system. These data sources are extensive. Among them, device manuals and technical documents provide official standard fault descriptions and processing methods; historical maintenance records record actual fault cases and solution processes, which have great practical reference value; expert experience integrates professional knowledge and industry insights, enriching the dimension of fault information. After collecting the data, the fault information is structured and multi-dimensionally classified. It is classified according to device types, such as categories like sensors, communication modules, controllers, etc.; according to fault categories, it can be further divided into hardware faults, software faults, network faults, etc.; at the same time, combined with the severity of the faults, it is divided into emergency faults, serious faults, general faults, and minor faults. Through this multi-dimensional organization method, a fault knowledge base with clear levels and convenient for retrieval is established; When a fault warning is generated by the system, it is necessary to accurately extract the fault characteristics corresponding to the warning level. Fault characteristics cover multiple key dimensions. The fault phenomenon is the abnormal manifestation that can be directly perceived by users or monitored by the system, such as abnormal vibration of the device, interruption of data transmission, etc.; relevant parameters include quantitative data such as the values collected by sensors and system operation indicators; the device status describes information such as the current working mode and connection status of the device. Based on the currently extracted fault characteristics, retrieve and filter in the fault knowledge base to find all solutions of the same type, and number them in sequence as i, and i = 1, 2, …, j, where j represents the number of solutions of the same type; Methods such as cosine similarity and edit distance are used to calculate the semantic similarity for the fault phenomenon and solutions described in the text; for numerical relevant parameters, the similarity degree is measured by calculating the absolute value of the difference or relative error, etc. Compare the calculated similarity with the preset value. The preset value is comprehensively set by operators with professional knowledge and practical experience according to factors such as system characteristics and fault risks, and filter out the solutions of the same type with similarity greater than the threshold, marked as preselected solutions, numbered as a, and a = 1, 2, …, b, where b represents the number of preselected solutions.

[0029] Step S5: Obtain the preselected solutions, and evaluate the preselected solutions from three aspects: technical feasibility, resource feasibility, and time feasibility to obtain the corresponding evaluation indicators, and the specific evaluation methods are as follows: Evaluate the technical feasibility, score the technical adaptability, technical maturity, and technical complexity of the preselected solutions and the current fault characteristics respectively, and normalize the obtained scores. At the same time, sum the normalized scores to obtain the technical feasibility indicator; For the three evaluation indicators of technical adaptability, technical maturity, and technical complexity, formulate detailed scoring criteria. Adopt a 5-point system, with 1 point being the lowest and 5 points being the highest. Technical adaptability mainly considers the compatibility between the solution technology and the existing system; technical maturity is judged based on factors such as the application years and the number of users of the technology; technical complexity is evaluated according to the technical difficulty required for implementation and training costs, etc.; Suppose there is an abnormal sensor data transmission fault in an Internet of Things system, and 3 preselected solutions are obtained. For technical adaptability, the communication protocol adopted by Solution 1 is fully compatible with the existing system, and the expert gives it 5 points; Solution 2 requires partial transformation of the existing system to be compatible, and gives it 3 points; Solution 3 is completely incompatible and gives it 1 point. In terms of technical maturity, Solution 1 is a mature technology that has been applied for 5 years and gives it 4 points; Solution 2 is a newly developed technology and gives it 2 points; Solution 3 is a cutting-edge technology in the industry with few application cases and gives it 1 point. In terms of technical complexity, Solution 1 is easy to operate and technicians can easily get started, and gives it 1 point; Solution 2 requires certain technical training and gives it 3 points; Solution 3 involves complex algorithms and requires a professional team to operate and gives it 5 points; After normalizing the scoring results, the normalized scores of technical adaptability are obtained: For Solution 1, \(\frac{5 - 1}{5 - 1}=1\); for Solution 2, \(\frac{5 - 1}{3 - 1}=0.5\); for Solution 3, \(\frac{5 - 1}{1 - 1}=0\); The normalized scores of technical maturity are: For Solution 1, \(\frac{4 - 1}{4 - 1}=1\); for Solution 2, \(\frac{4 - 1}{2 - 1}\approx0.33\); for Solution 3, \(\frac{4 - 1}{1 - 1}=0\) The normalized scores of technical complexity are: For Solution 1, \(\frac{5 - 1}{1 - 1}=0\); for Solution 2, \(\frac{5 - 1}{3 - 1}=0.5\); for Solution 3, \(\frac{5 - 1}{5 - 1}=1\) Sum up to get the technical feasibility index: For Solution 1, \(1 + 1+0 = 2\); for Solution 2, \(0.5 + 0.33+0.5 = 1.33\); for Solution 3, \(0 + 0+1 = 1\).

[0030] Evaluate the resource feasibility, score the preselected solutions for the manpower requirements, material resources, and data resources of the current fault characteristics respectively, and perform normalization in the same way, and sum up at the same time to obtain the resource feasibility index; Still taking the above sensor fault as an example, in terms of manpower requirements, Solution 1 only requires 1 ordinary technician to implement and gives it 5 points; Solution 2 requires 2 technicians with specific skills and gives it 3 points; Solution 3 requires a 5-person professional team and gives it 1 point. In terms of material resources, the required parts of Solution 1 are in sufficient stock and give it 5 points; for Solution 2, some parts need to be purchased with a long cycle and give it 2 points; for Solution 3, it is difficult to purchase the required imported equipment and give it 1 point. In terms of data resources, the system data relied on by Solution 1 is complete and accurate and gives it 5 points; Solution 2 needs to supplement some data and gives it 3 points; for Solution 3, the required special data is difficult to obtain and gives it 1 point.

[0031] After normalization, the normalized scores of manpower requirements are: For Solution 1, \(\frac{5 - 1}{5 - 1}=1\); for Solution 2, \(\frac{5 - 1}{3 - 1}=0.5\); for Solution 3, \(\frac{5 - 1}{1 - 1}=0\); The normalized scores of material resources are: For Solution 1, \(\frac{5 - 1}{5 - 1}=1\); for Solution 2, \(\frac{5 - 1}{2 - 1}=0.25\); for Solution 3, \(\frac{5 - 1}{1 - 1}=0\); Data resource normalization score: Option 1: 5−15−1 = 1; Option 2: 5−13−1 = 0.5; Option 3: 5−11−1 = 0; Resource feasibility index: Option 1: 1+1+1 = 3; Option 2: 0.5+0.25+0.5 = 1.25; Option 3: 0+0+0 = 0.

[0032] Evaluate the time feasibility, score the preselected options and the repair duration, emergency response duration, and time window limit of the current fault characteristics respectively, perform normalization processing, and sum them up to obtain the time feasibility index; Still the above sensor fault. In terms of the repair duration, for Option 1, it is expected to be repaired in 2 hours and is scored 5 points; for Option 2, it takes 6 hours and is scored 3 points; for Option 3, it takes 12 hours and is scored 1 point. In terms of the emergency response duration, Option 1 can be started within 10 minutes and is scored 5 points; Option 2 takes 30 minutes and is scored 3 points; Option 3 takes 1 hour and is scored 1 point. In terms of the time window limit, Option 1 can be implemented at any time and is scored 5 points; Option 2 needs to be carried out during the low peak period of the system at night and is scored 3 points; Option 3 can only be implemented during specific periods on weekends and is scored 1 point.

[0033] After normalization, the normalization score of the repair duration: Option 1: 5−15−1 = 1; Option 2: 5−13−1 = 0.5; Option 3: 5−11−1 = 0; The normalization score of the emergency response duration: Option 1: 5−15−1 = 1; Option 2: 5−13−1 = 0.5; Option 3: 5−11−1 = 0; The normalization score of the time window limit: Option 1: 5−15−1 = 1; Option 2: 5−13−1 = 0.5; Option 3: 5−11−1 = 0. The time feasibility index: Option 1: 1+1+1 = 3; Option 2: 0.5+0.5+0.5 = 1.5; Option 3: 0+0+0 = 0. Perform a weighted sum of the obtained technical feasibility index Ja, resource feasibility index Za, and time feasibility index Ta, and calculate the feasibility index Qa corresponding to the preselected option a according to the formula Qa = Ja×k1 + Za×k2 + Ta×k3, where k1, k2, and k3 are the corresponding weight coefficients, and k1 + k2 + k3 = 1. The specific values are set by the operator himself. At the same time, sort the obtained feasibility index Qa from large to small, then screen the feasibility index Qa of the preselected option a with the preset value Qy as the standard, and select the preselected option with the feasibility index Qa greater than the preset value Qy, denoted as the option to be selected; Next, obtain the historical data corresponding to the alternative solutions to be selected, and based on the historical data, obtain the repair effect corresponding to the alternative solutions to be selected. At the same time, analyze the repair effect. By calculating the recurrence rate of the same type, and the specific calculation formula is recurrence rate of the same type = number of recurrences of the same type / total number of recurrences × 100%. Then, select the alternative solution with the smallest recurrence rate of the same type as the standard to determine the selected solution and generate solution information.

[0034] Embodiment 2 As Embodiment 2 of the present invention, it is implemented on the basis of Embodiment 1, and the difference from Embodiment 1 is that the method for extracting the feature vectors of the Internet of Things signals and cross-information in step S2 is different, and the specific differences are as follows: Time-domain feature extraction Peak value: The maximum and minimum values of a signal within a certain time period, which can reflect the extreme situation of the signal. For example, in the signal collected by a pressure sensor, the peak value can represent the maximum bearing value of the pressure.

[0035] Frequency-domain feature extraction Power spectral density: Convert the time-domain signal to the frequency domain through Fourier transform and calculate the power spectral density, which can reflect the distribution of the energy of the signal at different frequencies. For example, in the signal collected by a vibration sensor, the power spectral density can help analyze the vibration frequency characteristics of the device and determine whether there is abnormal vibration.

[0036] Time-frequency domain feature extraction Wavelet transform: Decompose the signal using wavelet functions to obtain wavelet coefficients at different scales and frequencies, which has good time-frequency localization characteristics and can more accurately analyze the transient characteristics of the signal. For example, in image signal processing, wavelet transform can be used to extract features such as edges and textures of images.

[0037] Feature extraction for cross-information Classification feature extraction One-hot encoding: For categorical cross-information, such as weather conditions (sunny, cloudy, rainy, etc.), device types (sensor A, sensor B, etc.), one-hot encoding can be used to convert it into a vector form. For example, represent sunny as [1, 0, 0], cloudy as [0, 1, 0], and rainy as [0, 0, 1].

[0038] Numerical feature extraction Statistical features: Calculate some statistical features for numerical cross-information, such as mean, variance, maximum value, minimum value, etc. For example, for the number of operations of a user on an Internet of Things device within a certain period of time, the mean and variance can be calculated as features of the user's behavior.

[0039] Embodiment 3 As the third embodiment of the present invention, the key lies in combining the implementation processes of the first embodiment and the second embodiment.

[0040] For some data in the above formula, only their numerical values are taken for calculation, and the parameter units are not substituted for calculation. At the same time, the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0041] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An integrated intelligent early warning method for Internet of Things signals and their cross information, characterized in that, It includes the following steps: Obtain the signal data and cross information of the Internet of Things devices, and perform denoising, duplicate removal, and missing value processing to obtain preprocessed data; Extract the feature vectors of the Internet of Things signals and cross information corresponding to the preprocessed data, and fuse them to obtain a fused vector; Divide the fused vector and select an early warning model framework to establish an intelligent analysis and early warning model, and generate normal or abnormal signals in combination with real-time data; Analyze the abnormal signals, determine the early warning level, establish a fault knowledge base, and match and screen the fault features with the fault knowledge base to obtain a preselected solution; Evaluate and obtain the technical feasibility index, resource feasibility index, and time feasibility index of the preselected solution analysis, and perform weighted processing to obtain the feasibility index; Compare the feasibility index with the preset value, screen to obtain the solution to be selected, determine the selected solution in combination with its corresponding repair effect, and generate solution information.

2. The comprehensive intelligent early warning method for an Internet of Things signal and its cross information according to claim 1, wherein, The specific method for obtaining the fused vector is as follows: Extract the time-domain features, frequency-domain features, and time-frequency domain features of the Internet of Things signals to obtain the Internet of Things signal feature vectors. At the same time, extract the classification features and numerical features of the cross information to obtain the cross information feature vectors, and splice the two to obtain the fused vector.

3. The comprehensive intelligent early warning method for Internet of Things signals and their cross information according to claim 2, characterized in that, The specific method for generating normal or abnormal signals is as follows: Divide the fused vector dataset into a training set, a validation set, and a test set according to a certain ratio. At the same time, normalize each feature in the fused vector. Then, select an early warning model framework according to the fused vector, select a loss function according to the task type of the early warning model, select the Adam algorithm to adjust the parameters of the early warning model, and input the training set into the early warning model framework for training and optimization to obtain an intelligent analysis and early warning model; Then obtain real-time data and substitute it into the intelligent analysis and early warning model, and obtain normal or abnormal signals after being processed by the intelligent analysis and early warning model.

4. The integrated intelligent early warning method for Internet of Things signals and their cross information according to claim 1, characterized in that, The specific method for obtaining the preselected solution is as follows: Set multi-level early warnings, determine the current early warning level according to the abnormal signals, establish a fault knowledge base based on historical data, extract the fault features corresponding to the early warning level, and match them with the knowledge base; Screen out the same type of solutions according to the fault features and number them as i, where i = 1, 2,..., j, where j represents the number of the same type of solutions. Calculate the similarity between the fault features and each solution, compare with the preset threshold, and screen out the preselected solutions with qualified similarity and record them as a, where a = 1, 2,..., b, where b represents the number of preselected solutions.

5. The integrated intelligent early warning method for Internet of Things signals and their cross information according to claim 4, characterized in that The specific method for establishing a fault knowledge base based on historical data is as follows: Comprehensively collect the possible fault types, phenomena, causes, and corresponding solutions of various devices and components in the Internet of Things system, organize and classify the collected fault information, and organize it according to dimensions such as device type, fault category, and severity to establish a fault knowledge base.

6. The integrated intelligent early warning method for Internet of Things signals and their cross information according to claim 1, characterized in that, The specific method for evaluating and obtaining the technical feasibility index, resource feasibility index, and time feasibility index of the preselected solution analysis is as follows: Evaluate the technical feasibility, score the preselected solution and the fault features from three aspects of technical adaptability, maturity, and complexity, and sum them after normalization to obtain the technical feasibility index; Evaluate the resource feasibility, score the preselected solutions and fault characteristics in terms of technical adaptability, maturity, and complexity, sum them after normalization, and obtain the technical feasibility index; Evaluate the time feasibility, score the repair duration, emergency response duration, and time window limit, complete the normalization and summation, and determine the time feasibility index.

7. The comprehensive intelligent early warning method for Internet of Things signals and their cross information according to claim 1, characterized in that, The specific method for screening and obtaining the solutions to be selected is as follows: Set the weight coefficients k1, k2, and k3 corresponding to the technical feasibility index Ja, resource feasibility index Za, and time feasibility index Ta, calculate the feasibility index Qa corresponding to the preselected solution a according to the formula Qa = Ja×k1 + Za×k2 + Ta×k3, and screen out the solutions greater than the preset value Qy as the solutions to be selected.

8. An integrated intelligent early warning method for Internet of Things signals and their cross-information according to claim 1, characterized in that, The specific method for generating the solution information is as follows: Obtain the historical data corresponding to the solutions to be selected, obtain the repair effect corresponding to the solutions to be selected according to the historical data, analyze the repair effect at the same time, calculate the recurrence rate of the same type according to the formula recurrence rate of the same type = number of recurrences of the same type / total number of recurrences × 100%, then select the solution to be selected with the smallest recurrence rate of the same type as the standard, determine the selected solution, and generate the solution information.

Images