A multi-source heterogeneous data fusion communication system and method

Abstract

The application discloses a kind of multi-source heterogeneous data fusion communication systems and methods, it is related to big data processing technical field, the system runtime, by integrating multiple heterogeneous data sources, real-time data acquisition and preprocessing, effectively solve the limitation of traditional single data source monitoring mode. Through data fusion and feature extraction module, the system can comprehensively capture the multidimensional state characteristics of equipment, obtain characteristic data set Fset(t), to provide accurate data support for subsequent anomaly detection and prediction. Under the joint action of anomaly detection model and prediction model, the system can timely discover potential failure and give early warning in advance, while generating adaptive control strategy to prevent further deterioration of abnormality. In addition, through feedback and adaptive optimization module, improve the accuracy of fault prediction and the effectiveness of control strategy, reduce maintenance cost, significantly improve efficiency and safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data processing, in particular to a multi-source heterogeneous data fusion communication system and method. BACKGROUND

[0002] In the widespread application of modern communication technology, communication systems, as the core of information transmission, cover a variety of forms from wireless communication to wired networks. With the diversification of data types and the development of sensor technology, multi-source heterogeneous data fusion has gradually become one of the key technologies in communication systems. This technology integrates and processes multi-source data from different sensors, network nodes, user devices, etc., achieving more efficient communication quality and more accurate data analysis.

[0003] The monitoring of equipment status usually relies on a single type of data, such as temperature sensor or vibration sensor data. However, this single data source monitoring method has significant shortcomings, making it difficult to fully reflect the comprehensive status of the equipment. Since equipment failures are usually the result of multiple factors, a single data source may not capture these complex correlations. In addition, existing monitoring systems often fail to achieve real-time anomaly detection, making it impossible to provide timely warnings before a failure occurs, resulting in uncontrollable downtime or equipment damage. These shortcomings greatly limit operational efficiency and equipment reliability. SUMMARY

[0004] To address the shortcomings of the prior art, the present application provides a multi-source heterogeneous data fusion communication system and method, which solves the problems mentioned in the background art.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: a multi-source heterogeneous data fusion communication system, comprising a data acquisition and preprocessing module, a data fusion and feature extraction module, an anomaly detection model module, a prediction model construction module, a warning and decision module, and a feedback and adaptive optimization module;

[0006] The data acquisition and preprocessing module acquires data from multiple heterogeneous data sources in real time to form an acquisition data set Draw, and preprocesses the acquisition data set Draw to obtain a preprocessed data set Dpre;

[0007] The data fusion and feature extraction module performs data fusion on the preprocessed data set Dpre to obtain temperature vibration comprehensive features Ftv(t), power consumption interference comprehensive features Fpe(t), and micro-vibration noise features Fmvn(t) at time t, then performs feature extraction to form a fused feature data set Fset(t), and stores it after labeling;

[0008] The anomaly detection model module is constructed through the fused feature data set Fset(t), and an anomaly score set Ascore(t) of the current communication system is obtained through training and analysis of the anomaly detection model;

[0009] The prediction model construction module constructs a prediction model through the anomaly score set Ascore(t) and the stored historical data set, and an abnormal situation factor Pfuture(t+n) is obtained through prediction of the prediction model;

[0010] The early warning and decision module matches the abnormal situation factor Pfuture(t+n) obtained with a preset anomaly evaluation threshold YZ, obtains an abnormal matching result M(t+n), and generates an adaptive regulation strategy scheme according to the abnormal matching result;

[0011] The feedback and adaptive optimization module matches the abnormal matching result M(t+n) with an actual feedback result R(t+n), obtains an error term, and obtains a model adjustment index Tzs for adjusting the anomaly detection model and the prediction model.

[0012] Preferably, the data acquisition and preprocessing module includes a data acquisition unit and a data preprocessing unit.

[0013] The data acquisition unit acquires data in real time from a plurality of heterogeneous data sources, and the plurality of heterogeneous data sources include temperature data acquired by a temperature sensor, vibration data acquired by a vibration sensor, electromagnetic interference data acquired by an electromagnetic interference sensor, power consumption data acquired by a device running power consumption sensor, micro-vibration frequency data acquired by an acceleration sensor, and noise data acquired by a noise sensor, to form an acquisition data group Draw, which specifically includes temperature T(t), vibration frequency V(t), electromagnetic interference intensity E(t), power consumption P(t), micro-vibration frequency MV(t), and noise frequency spectrum NS(t).

[0014] The data preprocessing unit preprocesses the acquisition data group Draw, including data standardization preprocessing, denoising preprocessing, and missing value preprocessing, to form a preprocessed data set Dpre.

[0015] The data standardization preprocessing includes using Zscore standardization preprocessing to process parameters of different dimensions on the same scale; the denoising preprocessing includes using a sliding average filtering method for denoising preprocessing to adjust the influence of high-frequency noise; and the missing value preprocessing includes using a linear interpolation method for missing value processing to retain data trends while adjusting analysis errors caused by missing values.

[0016] Preferably, the data fusion and feature extraction module includes a data fusion unit and a feature extraction unit.

[0017] The data fusion unit fuses the pre-processed data set Dpre with time t as a variable to obtain temperature vibration comprehensive feature Ftv(t), power consumption interference comprehensive feature Fpe(t) and micro-vibration noise feature Fmvn(t);

[0018] The feature extraction unit extracts the obtained temperature vibration comprehensive feature Ftv(t), power consumption interference comprehensive feature Fpe(t) and micro-vibration noise feature Fmvn(t) to form a fused feature data set Fset(t), and stores it after marking, the marking including time t and data source identification Fid;

[0019] Wherein, the feature data set Fset(t) is specifically Fset(t)={Ftv(t), Fpe(t), Fmvn(t), t, Fid}.

[0020] Preferably, the temperature vibration comprehensive feature Ftv(t) is obtained by the following calculation formula:

[0021]

[0022] In the formula, Ftv(t) represents the temperature vibration comprehensive feature of time t, and specifically represents the dynamic relationship between the temperature T(t) and the vibration frequency V(t) of time t, T(t) represents the temperature, and specifically represents the temperature T of time t, V(t) represents the vibration frequency, and reflects the thermal state of the power grid equipment, and specifically represents the vibration frequency of time t, represents the derivative of the vibration frequency V(t) with respect to time t, and d represents the derivative;

[0023] The power consumption interference comprehensive feature Fpe(t) is obtained by the following calculation formula:

[0024]

[0025] In the formula, Fpe(t) represents the power consumption interference comprehensive feature of time t, and specifically represents the dynamic relationship between the electromagnetic interference intensity E(t) and the power consumption P(t) of time t, E(t) represents the electromagnetic interference intensity of the equipment of time t, and specifically represents the electromagnetic interference degree of the equipment operation area, P(t) represents the equipment operation power consumption of time t, and specifically represents the energy consumption amount during operation; ln represents the natural logarithm function, and is specifically used to adjust the range of the dynamic relationship between the electromagnetic interference intensity E(t) and the power consumption P(t);

[0026] The micro-vibration noise feature Fmvn(t) is obtained by the following calculation formula:

[0027]

[0028] In the formula, Fmvn(t) represents the micro-vibration noise characteristics at time t, specifically representing the dynamic relationship between the micro-vibration frequency MV(t) at time t and the noise spectrum NS(t), MV(t) represents the micro-vibration frequency at time t, specifically representing the number of micro-vibrations occurring when the device is running at time t, reflecting the stability of the device running state, NS(t) represents the noise spectrum at time t, specifically representing the environmental noise state of the device running area at time t, represents the first-order derivative of the micro-vibration frequency MV(t) with respect to time t, specifically representing the rate of change of the micro-vibration frequency MV(t) with time, used to capture transient fluctuations.

[0029] Preferably, the anomaly detection model module inputs the fused feature data set Fset(t) as model input data to build an anomaly detection model, wherein building the anomaly detection model includes using the Isolation Forest, Principal Component Analysis and Support Vector Machine algorithms to build, through training and analysis of the anomaly detection model, to reflect the running state of the communication system and obtain the current communication system anomaly score set Ascore(t);

[0030] The anomaly score set Ascore(t) is obtained by the following calculation formula:

[0031]

[0032] In the formula, Fset,i(t) represents the i-th feature in the feature data set Fset(t) at time t, μF,i represents the i-th feature mean, σF,i represents the i-th feature standard deviation, N represents the number of features, λ represents the adjustment coefficient, specifically used to balance the dynamic relationship between the feature and the distance measure, Ddist(t) represents the distance measure of the feature in the anomaly detection model, specifically representing the difference degree of the current feature and the model qualified state feature.

[0033] Preferably, the prediction model construction module includes a data preparation unit and a data calculation unit;

[0034] The data preparation unit extracts the stored historical data set synchronously through the anomaly score set Ascore(t), to form a historical anomaly score set Hscore(t), and performs feature selection and time series processing based on the anomaly score set Ascore(t) and the historical anomaly score set Hscore(t), including time series smoothing processing and difference processing, to obtain the smoothed anomaly score value S(t) and the anomaly score difference value D(t) at time t, to reflect the change trend of the anomaly score, and synchronously cover and replace the anomaly score set Ascore(t) with the smoothed anomaly score value S(t) to form the smoothed anomaly score set Ascore(t);

[0035] The data estimation unit uses a time series model to construct a prediction model, inputs the smoothed abnormal score set Ascore(t), the abnormal score difference value D(t) and the historical abnormal score set Hscore(t) as input data of the prediction model, performs prediction training through the prediction model, and obtains an abnormal situation factor Pfuture(t+n);

[0036] The abnormal situation factor Pfuture(t+n) is obtained by the following calculation formula:

[0037] Pfuture(t+n) = β0+ β1·Ascore(t) + β2·D(t) + β3·Hscore(t);

[0038] In the formula, β0 represents the bias term of the prediction model function, specifically a constant baseline, β1, β2 and β3 represent the preset weight values of the smoothed abnormal score set Ascore(t), the abnormal score difference value D(t) and the historical abnormal score set Hscore(t) respectively.

[0039] Preferably, the abnormal score value S(t) is obtained by the following calculation formula:

[0040] S(t) = α·Ascore(t) + (1-α)·Hscore(t-1);

[0041] In the formula, α represents a smoothing coefficient, which is specifically used to adjust the preset weight value distribution of the abnormal score set Ascore(t) and the historical abnormal score set Hscore(t);

[0042] The abnormal score difference value D(t) is obtained by the following calculation formula:

[0043]

[0044] Preferably, the early warning and decision module matches the abnormal situation factor Pfuture(t+n) obtained with a preset abnormal evaluation threshold YZ, and obtains an abnormal matching result M(t+n) through The matching formula, and generates an adaptive control strategy scheme according to the abnormal matching result;

[0045] When the abnormal matching result M(t+n) is 1, a predicted abnormal situation is obtained, and an adaptive control strategy scheme is generated according to the abnormal matching result, including adjusting system parameters, switching backup systems, reducing power consumption and notifying operating personnel to take action;

[0046] When the abnormal matching result M(t+n) is 0, a predicted non-abnormal situation is obtained, and no adaptive control strategy scheme is generated according to the abnormal matching result.

[0047] Preferably, the feedback is matched with the actual feedback result R(t+n) according to the abnormal matching result M(t+n) by the self-adaptive optimization module, wherein the actual feedback result R(t+n) is obtained by the operator feedback, an error term is obtained, and a model adjustment index Tzs for adjusting the abnormal detection model and the prediction model is obtained, and the model adjustment index Tzs is taken as an adjustment factor to indirectly adjust the abnormal detection performance and the prediction performance of the model by affecting the parameter update in the training process of the abnormal detection model and the prediction model;

[0048] The model adjustment index Tzs is obtained by the following calculation formula:

[0049] Tzs(t+n)=Tzs(t)-η·(M(t+n)-R(t+n));

[0050] In the formula, Tzs(t) represents the model adjustment index at time t, specifically including the abnormal detection model and the prediction model, and η represents the learning rate, specifically representing the control adjustment step.

[0051] A communication method for multi-source heterogeneous data fusion, comprising the following steps:

[0052] Step one: the data acquisition and preprocessing module acquires data from multiple heterogeneous data sources in real time to form an acquisition data set Draw, and pre-processes the acquisition data set Draw to obtain a pre-processed data set Dpre;

[0053] Step two: the data fusion and feature extraction module performs data fusion on the pre-processed data set Dpre to obtain temperature vibration comprehensive features Ftv(t), power consumption interference comprehensive features Fpe(t), and micro-vibration noise features Fmvn(t) at time t, then performs feature extraction to form a fused feature data set Fset(t), and stores it after labeling;

[0054] Step three: the abnormal detection model module constructs an abnormal detection model through the fused feature data set Fset(t), and obtains an abnormal score set Ascore(t) of the current communication system through training and analysis of the abnormal detection model;

[0055] Step four: the prediction model construction module constructs a prediction model through the abnormal score set Ascore(t) and the stored historical data set, and obtains an abnormal situation factor Pfuture(t+n) through prediction by the prediction model;

[0056] Step five: the early warning and decision module matches the abnormal situation factor Pfuture(t+n) with a preset abnormal evaluation threshold YZ to obtain an abnormal matching result M(t+n), and generates a self-adaptive regulation strategy scheme according to the abnormal matching result.

[0057] Step six: The feedback and adaptive optimization module matches the abnormal matching result M(t+n) with the actual feedback result R(t+n), obtains the error term and obtains the model adjustment index Tzs for adjusting the abnormal detection model and the prediction model.

[0058] The application provides a multi-source heterogeneous data fusion communication system and method, which has the following beneficial effects:

[0059] (1) When the system is running, by integrating multiple heterogeneous data sources, real-time data collection and preprocessing are performed, effectively solving the limitations of traditional single data source monitoring methods. Through the data fusion and feature extraction module, the system can comprehensively capture the multi-dimensional state characteristics of the equipment and obtain the feature data set Fset(t), thereby providing accurate data support for subsequent anomaly detection and prediction. Under the joint action of the anomaly detection model and the prediction model, the system can timely discover potential faults and issue early warnings in advance, and generate adaptive control strategies to prevent further deterioration of the anomaly. In addition, through the feedback and adaptive optimization module, the system can continuously adjust and optimize itself according to the actual feedback, improving the accuracy of fault prediction and the effectiveness of the control strategy. Ultimately, this system effectively compensates for the shortcomings of traditional systems in fault warning and real-time response, reduces maintenance costs, and significantly improves overall efficiency and safety.

[0060] (2) By using smoothing processing and difference analysis, the abnormal score data is deeply optimized to ensure that the model has higher robustness and accuracy when processing data. In addition, the system can adaptively adjust the parameters of the prediction model according to the dynamic relationship between the real-time abnormal score set Ascore(t) and the historical abnormal score set Hscore(t), making the system's prediction of future abnormal situations more accurate, thereby effectively avoiding the risk of excessive alarms and missed alarms. This mechanism significantly improves the system's ability to respond to potential problems in complex environments, reduces unnecessary maintenance operations and resource consumption, and further optimizes the system's operating efficiency.

[0061] (3) According to the abnormal situation factor Pfuture(t+n), the potential abnormal situation is judged and predicted, and the targeted regulation strategy can be quickly generated, including adjusting system parameters, switching backup systems, reducing power consumption and notifying operators, and the actual feedback result R(t+n) can be verified after implementation, through the manual inspection of the operator and the automatic feedback loop, the system can accurately identify the accuracy of the prediction, and the model parameters are updated in real time based on the error. This closed-loop feedback mechanism ensures that the system always maintains efficient abnormal detection and response capabilities in the process of constantly adapting to environmental changes. Finally, through the continuous optimization of the model adjustment index Tzs, the abnormal detection model and the prediction model have stronger self-adaptation ability, greatly improving the response effect to complex faults, reducing false positives and false negatives, and thus ensuring the long-term stable operation of the system and effectively controlling the maintenance cost. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 A communication system block diagram of the multi-source heterogeneous data fusion of the application is shown in the figure.

[0063] Figure 2 A communication method step diagram of the multi-source heterogeneous data fusion of the application is shown in the figure. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0065] Embodiment 1

[0066] The application provides a multi-source heterogeneous data fusion communication system, please refer to Figure 1 , including a data acquisition and preprocessing module, a data fusion and feature extraction module, an abnormal detection model module, a prediction model construction module, a warning and decision module, and a feedback and adaptive optimization module.

[0067] The data acquisition and preprocessing module acquires data from multiple heterogeneous data sources in real time to form an acquisition data set Draw, and preprocesses the acquisition data set Draw to obtain a preprocessed data set Dpre.

[0068] The data fusion and feature extraction module performs data fusion on the preprocessed data set Dpre, obtains temperature vibration comprehensive features Ftv(t), power consumption interference comprehensive features Fpe(t) and micro-vibration noise features Fmvn(t) at time t, then performs feature extraction, forms a fused feature data set Fset(t), and stores after labeling;

[0069] The anomaly detection model module constructs an anomaly detection model through the fused feature data set Fset(t), obtains an anomaly score set Ascore(t) of the current communication system through training and analysis of the anomaly detection model;

[0070] The prediction model construction module constructs a prediction model through the anomaly score set Ascore(t) and the stored historical data set, and obtains an abnormal situation factor Pfuture(t+n) through prediction of the prediction model;

[0071] The early warning and decision module matches the obtained abnormal situation factor Pfuture(t+n) with a preset anomaly evaluation threshold YZ, obtains an abnormal matching result M(t+n), and generates an adaptive regulation strategy scheme according to the abnormal matching result;

[0072] The feedback and adaptive optimization module matches the abnormal matching result M(t+n) with an actual feedback result R(t+n), obtains an error term and a model adjustment index Tzs for adjusting the anomaly detection model and the prediction model.

[0073] In this embodiment, by integrating multiple heterogeneous data sources, real-time data collection and preprocessing are performed, effectively solving the limitations of traditional single data source monitoring methods. Through the data fusion and feature extraction module, the system can comprehensively capture the multi-dimensional state features of the equipment, obtain the feature data set Fset(t), and thus provide accurate data support for subsequent anomaly detection and prediction. Under the joint action of the anomaly detection model and the prediction model, the system can timely discover potential faults and issue early warnings in advance, and generate adaptive regulation strategies to prevent further deterioration of the anomaly. In addition, through the feedback and adaptive optimization module, the system can continuously adjust and optimize itself according to actual feedback, improving the accuracy of fault prediction and the effectiveness of regulation strategies. Ultimately, this system effectively compensates for the shortcomings of traditional systems in fault warning and real-time response, reduces equipment downtime, reduces maintenance costs, and significantly improves overall efficiency and safety.

[0074] Embodiment 2

[0075] This embodiment is an explanation and description in Embodiment 1, please refer to Figure 1 , specifically: the data acquisition and preprocessing module includes a data acquisition unit and a data preprocessing unit;

[0076] The data acquisition unit acquires data in real time from a plurality of heterogeneous data sources, including temperature data acquired by a temperature sensor, vibration data acquired by a vibration sensor, electromagnetic interference data acquired by an electromagnetic interference sensor, power consumption data acquired by a device running power consumption sensor, micro-vibration frequency data acquired by an acceleration sensor, and noise data acquired by a noise sensor, to form an acquisition data set Draw, specifically including temperature T(t), vibration frequency V(t), electromagnetic interference intensity E(t), power consumption P(t), micro-vibration frequency MV(t), and noise spectrum NS(t);

[0077] The data preprocessing unit pre-processes the acquisition data set Draw, including data standardization preprocessing, denoising preprocessing, and missing value preprocessing, to form a pre-processed data set Dpre;

[0078] The data standardization preprocessing includes using Zscore standardization preprocessing to process parameters of different dimensions on the same scale; the denoising preprocessing includes using a sliding average filtering method for denoising preprocessing to adjust the influence of high-frequency noise; the missing value preprocessing includes using a linear interpolation method for missing value processing to retain data trends while adjusting analysis errors caused by missing values.

[0079] The data fusion and feature extraction module includes a data fusion unit and a feature extraction unit;

[0080] The data fusion unit performs data fusion on the pre-processed data set Dpre with time t as the variable, to obtain temperature vibration comprehensive features Ftv(t), power consumption interference comprehensive features Fpe(t), and micro-vibration noise features Fmvn(t);

[0081] The feature extraction unit performs feature extraction on the obtained temperature vibration comprehensive features Ftv(t), power consumption interference comprehensive features Fpe(t), and micro-vibration noise features Fmvn(t) to form a fused feature data set Fset(t), and stores it after labeling, the labeling including time t and data source identification Fid;

[0082] The feature data set Fset(t) is specifically Fset(t)={Ftv(t),Fpe(t),Fmvn(t),t,Fid}.

[0083] The temperature vibration comprehensive features Ftv(t) are obtained by the following calculation formula:

[0084]

[0085] In the formula, Ftv(t) represents a temperature vibration comprehensive feature at time t, and specifically represents a dynamic relationship between a temperature T(t) at time t and a vibration frequency V(t) at time t, T(t) represents a temperature, and specifically represents a temperature T at time t, V(t) represents a vibration frequency, and reflects a thermal state of the power grid equipment, and specifically represents a vibration frequency at time t, represents a derivative of the vibration frequency V(t) with respect to time t, and d represents a derivative;

[0086] The power consumption interference comprehensive feature Fpe(t) is obtained by the following calculation formula:

[0087]

[0088] In the formula, Fpe(t) represents a power consumption interference comprehensive feature at time t, and specifically represents a dynamic relationship between an electromagnetic interference intensity E(t) at time t and a power consumption P(t) at time t, E(t) represents an electromagnetic interference intensity of the equipment at time t, and specifically represents an electromagnetic interference degree of an equipment operation region, P(t) represents a power consumption of the equipment at time t, and specifically represents an energy consumption amount during operation, and ln represents a natural logarithm function, and is specifically used to adjust a range of the dynamic relationship between the electromagnetic interference intensity E(t) and the power consumption P(t);

[0089] The micro-vibration noise feature Fmvn(t) is obtained by the following calculation formula:

[0090]

[0091] In the formula, Fmvn(t) represents a micro-vibration noise feature at time t, and specifically represents a dynamic relationship between a micro-vibration frequency MV(t) at time t and a noise spectrum NS(t) at time t, MV(t) represents a micro-vibration frequency at time t, and specifically represents a number of micro-vibrations occurring when the equipment operates at time t, and reflects stability of an equipment operation state, and NS(t) represents a noise spectrum at time t, and specifically represents an environmental noise state of an equipment operation region at time t, represents a first-order derivative of the micro-vibration frequency MV(t) with respect to time t, and specifically represents a rate of change of the micro-vibration frequency MV(t) with time, and is used to capture instantaneous fluctuations.

[0092] Embodiment 3

[0093] This embodiment is an explanation and description made in Embodiment 2, and please refer to Figure 1Specifically, the abnormality detection model module inputs the fused feature data set Fset(t) as model input data into the constructed abnormality detection model, wherein the abnormality detection model is constructed by using an isolation forest, principal component analysis, and a support vector machine algorithm, and the running state of the communication system is reflected through training and analysis of the abnormality detection model, and the abnormality score set Ascore(t) of the current communication system is obtained.

[0094] The abnormality score set Ascore(t) is obtained by the following calculation formula:

[0095]

[0096] In the formula, Fset,i(t) represents the i-th feature in the feature data set Fset(t) at time t, μF,i represents the mean of the i-th feature, σF,i represents the standard deviation of the i-th feature, N represents the number of features, λ represents an adjustment coefficient, specifically used to balance the dynamic relationship between the feature and the distance measure, and Ddist(t) represents the distance measure of the feature in the abnormality detection model, specifically representing the difference degree of the current feature and the model qualified state feature.

[0097] The prediction model construction module includes a data preparation unit and a data calculation unit.

[0098] The data preparation unit extracts the stored historical data set synchronously based on the abnormality score set Ascore(t) to form a historical abnormality score set Hscore(t), and performs feature selection and time series processing based on the abnormality score set Ascore(t) and the historical abnormality score set Hscore(t), including time series smoothing processing and difference processing, to obtain a smoothed abnormality score value S(t) and an abnormality score difference value D(t) at time t, to reflect the change trend of the abnormality score, and to synchronously cover and replace the abnormality score set Ascore(t) with the smoothed abnormality score value S(t) to form a smoothed abnormality score set Ascore(t);

[0099] The data calculation unit constructs a prediction model by using a time series model, inputs the smoothed abnormality score set Ascore(t), the abnormality score difference value D(t), and the historical abnormality score set Hscore(t) as input data of the prediction model, and performs prediction training through the prediction model to obtain an abnormality condition factor Pfuture(t+n).

[0100] The abnormality condition factor Pfuture(t+n) is obtained by the following calculation formula:

[0101] Pfuture(t+n) = β0+ β1·Ascore(t) + β2·D(t) + β3·Hscore(t);

[0102] In the formula, β0represents the bias term of the prediction model function, specifically representing a constant baseline, β1, β2and β3respectively represent the preset weight values of the smoothed abnormal score set Ascore(t), the abnormal score difference value D(t) and the historical abnormal score set Hscore(t).

[0103] The abnormal score value S(t) is obtained by the following calculation formula:

[0104] S(t) = α·Ascore(t) + (1-α)·Hscore(t-1);

[0105] In the formula, α represents the smoothing coefficient, which is specifically used to regulate the preset weight value distribution of the abnormal score set Ascore(t) and the historical abnormal score set Hscore(t).

[0106] The abnormal score difference value D(t) is obtained by the following calculation formula:

[0107]

[0108] In this embodiment, the abnormal score data is deeply optimized by using smoothing processing and difference analysis, ensuring that the model has higher robustness and accuracy when processing data. In addition, the system can adaptively adjust the parameters of the prediction model according to the dynamic relationship between the real-time abnormal score set Ascore(t) and the historical abnormal score set Hscore(t), making the system's prediction of future abnormal conditions more accurate, thereby effectively avoiding the risk of excessive alarm and missed alarm. This mechanism significantly improves the system's ability to respond to potential problems in complex environments, reduces unnecessary maintenance operations and resource consumption, and further optimizes the system's running efficiency.

[0109] Embodiment 4

[0110] This embodiment is an explanation and description in embodiment 3, please refer to Figure 1 Specifically, the early warning and decision module matches the abnormal situation factor Pfuture(t+n) obtained with the preset abnormal evaluation threshold YZ, obtains the abnormal matching result M(t+n) through the matching formula, and generates an adaptive control strategy scheme according to the abnormal matching result;

[0111] When the abnormal matching result M(t+n) is 1, it is obtained that the abnormal situation is predicted to occur, and an adaptive control strategy scheme is generated according to the abnormal matching result, including adjusting system parameters, switching backup systems, reducing power consumption and notifying operating personnel to take action.

[0112] When the abnormal matching result M(t+n) is 0, it is obtained that the abnormal situation is predicted not to occur, and no adaptive control strategy scheme is generated according to the abnormal matching result.

[0113] The feedback and adaptive optimization module matches the abnormal matching result M(t+n) with the actual feedback result R(t+n), wherein the actual feedback result R(t+n) is obtained through operator feedback, an error term is obtained, and a model adjustment index Tzs for adjusting the abnormal detection model and the prediction model is obtained, and the model adjustment index Tzs is used as a regulation factor to indirectly adjust the abnormal detection performance and the prediction performance of the model by affecting the parameter update in the training process of the abnormal detection model and the prediction model.

[0114] The actual feedback result R(t+n) is obtained through operator feedback and is collected and obtained in an automatic feedback loop mode.

[0115] The operator feedback is obtained through manual inspection and confirmation by the operator, for example, the operator judges the actual state of the system by inspecting the equipment, checking the system alarm and inspecting the physical environment, and at the same time, the operator fills out a feedback report or directly inputs the system as part of R(t+n) to provide a more comprehensive system state.

[0116] The automatic feedback loop mode is collected and obtained as follows: when the system successfully avoids the abnormality after executing the control strategy, the feedback result R(t+n) will be displayed as normal; if it fails to avoid, the feedback result will record the actual fault or abnormal state of the system;

[0117] When the actual feedback result R(t+n) is 1, it means that the predicted abnormal situation of the system has not been successfully solved or the abnormality has indeed occurred, which means that although the system has made a warning and control, the problem still exists, or the system has indeed experienced an unforeseen abnormality;

[0118] When the actual feedback result R(t+n) is 0, it means that the predicted abnormal situation of the system has been successfully solved or the abnormality has not occurred. This means that the prediction of the system may have errors, or the control strategy successfully prevented potential abnormalities;

[0119] The model adjustment index Tzs is obtained by the following calculation formula:

[0120] Tzs(t+n) = Tzs(t) - η·(M(t+n) - R(t+n));

[0121] In the formula, Tzs(t) represents the model adjustment index at time t, which specifically includes an anomaly detection model and a prediction model, and η represents a learning rate, specifically indicating a control adjustment step size.

[0122] In this embodiment, potential abnormal situations are judged and predicted according to the abnormal situation factor Pfuture(t+n), and targeted control strategies can be quickly generated, including adjusting system parameters, switching backup systems, reducing power consumption, and notifying operators. It can also be verified by actual feedback results R(t+n) after implementation, through manual inspection by operators and an automated feedback loop, the system can accurately identify the accuracy of the prediction, and update the model parameters in real time based on the error. This closed-loop feedback mechanism ensures that the system always maintains efficient anomaly detection and response capabilities in the process of constantly adapting to environmental changes. Ultimately, through the continuous optimization of the model adjustment index Tzs, the anomaly detection model and the prediction model have stronger adaptive ability, greatly improving the response effect to complex faults and reducing false positives and false negatives, thereby ensuring the long-term stable operation of the system and effectively controlling the maintenance cost.

[0123] Embodiment 5

[0124] A communication method for multi-source heterogeneous data fusion, please refer to Figure 2 , specifically: comprising the following steps:

[0125] Step one: the data acquisition and preprocessing module acquires data from multiple heterogeneous data sources in real time to form an acquisition data set Draw, and pre-processes the acquisition data set Draw to obtain a pre-processed data set Dpre;

[0126] Step two: the data fusion and feature extraction module performs data fusion on the pre-processed data set Dpre to obtain temperature vibration comprehensive features Ftv(t), power consumption interference comprehensive features Fpe(t), and micro-vibration noise features Fmvn(t) at time t, then performs feature extraction to form a fused feature data set Fset(t), and stores it after labeling;

[0127] Step three: the anomaly detection model module constructs an anomaly detection model through the fused feature data set Fset(t), and obtains an abnormal score set Ascore(t) of the current communication system through training and analysis of the anomaly detection model;

[0128] Step four: the prediction model construction module constructs a prediction model through the abnormal score set Ascore(t) and the stored historical data set, and obtains an abnormal situation factor Pfuture(t+n) through the prediction model;

[0129] Step five: the early warning and decision module matches the acquired abnormal situation factor Pfuture(t+n) with the preset abnormal evaluation threshold YZ, acquires an abnormal matching result M(t+n), and generates an adaptive regulation strategy scheme according to the abnormal matching result;

[0130] Step six: the feedback and adaptive optimization module matches the abnormal matching result M(t+n) with the actual feedback result R(t+n), acquires an error term, and acquires a model adjustment index Tzs for adjusting the abnormal detection model and the prediction model.

[0131] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A communication system for multi-source heterogeneous data fusion, characterized in that: It includes a data acquisition and preprocessing module, a data fusion and feature extraction module, an anomaly detection model module, a prediction model construction module, an early warning and decision-making module, and a feedback and adaptive optimization module; The data acquisition and preprocessing module acquires data in real time from several heterogeneous data sources, forms a data acquisition group Draw, and preprocesses the data acquisition group Draw to obtain a preprocessed dataset Dpre. The data fusion and feature extraction module performs data fusion on the preprocessed dataset Dpre to obtain the comprehensive temperature vibration feature Ftv(t), comprehensive power consumption interference feature Fpe(t), and micro-vibration noise feature Fmvn(t) at time t. Then, it performs feature extraction to form the fused feature data set Fset(t), which is then labeled and stored. The anomaly detection model module constructs an anomaly detection model using the fused feature data set Fset(t), and obtains the anomaly score set Ascore(t) of the current communication system through the training and analysis of the anomaly detection model. The prediction model building module constructs a prediction model using the anomaly score set Ascore(t) and the stored historical dataset, and uses the prediction model to predict and obtain the anomaly factor Pfuture(t+n). The early warning and decision-making module matches the acquired abnormal situation factor Pfuture(t+n) with the preset abnormal assessment threshold YZ to obtain the abnormal matching result M(t+n), and generates an adaptive control strategy scheme based on the abnormal matching result. The feedback and adaptive optimization module matches the anomaly matching result M(t+n) with the actual feedback result R(t+n) to obtain the error term and the model adjustment index Tzs used to adjust the anomaly detection model and the prediction model.

2. The communication system for multi-source heterogeneous data fusion according to claim 1, characterized in that: The data acquisition and preprocessing module includes a data acquisition unit and a data preprocessing unit; The data acquisition unit collects data in real time from multiple heterogeneous data sources, including temperature data collected by a temperature sensor, vibration data collected by a vibration sensor, electromagnetic interference data collected by an electromagnetic interference sensor, power consumption data collected by a device operating power consumption sensor, micro-vibration frequency data collected by an acceleration sensor, and noise data collected by a noise sensor, forming a data acquisition group called Draw.

3. The communication system for multi-source heterogeneous data fusion according to claim 2, characterized in that: The data collection group Draw specifically includes temperature T(t), vibration frequency V(t), electromagnetic interference intensity E(t), power consumption P(t), micro-vibration frequency Mv(t), and noise spectrum Ns(t).

4. The communication system for multi-source heterogeneous data fusion according to claim 2, characterized in that: The data preprocessing unit preprocesses the collected data group Draw, including data standardization preprocessing, noise reduction preprocessing, and missing value preprocessing, to form the preprocessed dataset Dpre. The data standardization preprocessing includes Z-score standardization preprocessing, which processes parameters of different dimensions on the same scale; the denoising preprocessing includes moving average filtering; and the missing value preprocessing includes linear interpolation.

5. A communication system for multi-source heterogeneous data fusion according to claim 1, characterized in that: The data fusion and feature extraction module includes a data fusion unit and a feature extraction unit; The data fusion unit performs data fusion on the preprocessed dataset Dpre with time t as the variable to obtain the comprehensive temperature vibration feature Ftv(t), the comprehensive power consumption interference feature Fpe(t), and the micro-vibration noise feature Fmvn(t). The feature extraction unit extracts features from the acquired temperature vibration comprehensive feature Ftv(t), power consumption interference comprehensive feature Fpe(t), and micro vibration noise feature Fmvn(t), forming a fused feature data set Fset(t), which is then labeled and stored. The label includes time t and data source identifier Fid. The feature data set Fset(t) is specifically defined as Fset(t) = {Ftv(t), Fpe(t), Fmvn(t), t, Fid}.

6. The communication system for multi-source heterogeneous data fusion according to claim 1, characterized in that: The comprehensive characteristic of temperature vibration, Ftv(t), is obtained through the following calculation formula: In the formula, Ftv(t) represents the comprehensive temperature vibration characteristics at time t, specifically the dynamic relationship between the temperature T(t) at time t and the vibration frequency V(t). T(t) represents the temperature, specifically the temperature T at time t, and V(t) represents the vibration frequency, reflecting the thermal state of the power grid equipment, specifically the vibration frequency at time t. d represents the derivative of the vibration frequency V(t) with respect to time t. The power consumption interference comprehensive characteristic Fpe(t) is obtained by the following calculation formula: In the formula, Fpe(t) represents the comprehensive characteristics of power consumption and interference at time t, specifically the dynamic relationship between the electromagnetic interference intensity E(t) and power consumption P(t) at time t. E(t) represents the electromagnetic interference intensity of the equipment at time t, specifically the degree of electromagnetic interference in the equipment's operating area. P(t) represents the power consumption of the equipment at time t, specifically the energy consumption during operation. ln represents the natural logarithm function, specifically used to adjust the range of the dynamic relationship between the electromagnetic interference intensity E(t) and power consumption P(t). The micro-vibration noise characteristic Fmvn(t) is obtained by the following calculation formula: In the formula, Fmvn(t) represents the micro-vibration noise characteristics at time t, specifically the dynamic relationship between the micro-vibration frequency Mv(t) and the noise spectrum Ns(t) at time t. Mv(t) represents the micro-vibration frequency at time t, specifically the number of micro-vibrations that occur during the operation of the equipment at time t, reflecting the stability of the equipment's operating state. Ns(t) represents the noise spectrum at time t, specifically the environmental noise state of the equipment in the operating region at time t. This represents the first derivative of the micro-vibration frequency Mv(t) with respect to time t, specifically indicating the rate of change of the micro-vibration frequency Mv(t) with time, used to capture instantaneous fluctuations.

7. A communication system for multi-source heterogeneous data fusion according to claim 1, characterized in that: The anomaly detection model module uses the fused feature data set Fset(t) as the model input data to input the constructed anomaly detection model. The construction of the anomaly detection model includes the use of isolated forest, principal component analysis and support vector machine algorithms. Through the training and analysis of the anomaly detection model, the operating status of the communication system is reflected, and the current anomaly score set Ascore(t) of the communication system is obtained. The anomaly score set Ascore(t) is obtained through the following calculation formula: In the formula, Fset,i(t) represents the i-th feature in the feature data set Fset(t) at time t, μF,i represents the mean of the i-th feature, σF,i represents the standard deviation of the i-th feature, N represents the number of features, λ represents the adjustment coefficient, which is used to balance the dynamic relationship between features and distance metrics, and Ddist(t) represents the distance metric of the feature in the anomaly detection model, which specifically represents the degree of difference between the current feature and the model's qualified state features.

8. A communication system for multi-source heterogeneous data fusion according to claim 1, characterized in that: The prediction model construction module includes a data preparation unit and a data extrapolation unit; The data preparation unit synchronously extracts the stored historical dataset from the anomaly score set Ascore(t) to form a historical anomaly score set Hscore(t). Based on the anomaly score set Ascore(t) and the historical anomaly score set Hscore(t), feature selection and time series processing are performed, including time series smoothing and differencing. The smoothed anomaly score value S(t) and the anomaly score difference value D(t) at time t are obtained to reflect the changing trend of the anomaly score. Simultaneously, the smoothed anomaly score value S(t) is used to cover and replace the anomaly score set Ascore(t) to form a smoothed anomaly score set Ascore(t). The data extrapolation unit constructs a prediction model using a time series model. It inputs the smoothed abnormal score set Ascore(t), the abnormal score difference value D(t), and the historical abnormal score set Hscore(t) as the input data of the prediction model. The prediction model is then trained to obtain the abnormal situation factor Pfuture(t+n).

9. A communication system for multi-source heterogeneous data fusion according to claim 8, characterized in that: The anomaly factor Pfuture(t+n) is obtained through the following formula: Pfuture(t+n)=β0+β1·Ascore(t)+β2·D(t)+β3·Hscore(t); In the formula, β0 represents the bias term of the prediction model function, specifically the constant baseline, and β1, β2 and β3 represent the preset weight values ​​of the smoothed abnormal score set Ascore(t), the abnormal score difference value D(t) and the historical abnormal score set Hscore(t), respectively.

10. A communication system for multi-source heterogeneous data fusion according to claim 8, characterized in that: The anomaly score S(t) is obtained using the following formula: S(t)=α·Ascore(t)+(1-α)·Hscore(t-1); In the formula, α represents the smoothing coefficient, which is specifically used to control the preset weight values ​​of the abnormal score set Ascore(t) and the historical abnormal score set Hscore(t). The anomaly score difference value D(t) is obtained by the following formula:

11. A communication system for multi-source heterogeneous data fusion according to claim 1, characterized in that: The early warning and decision-making module matches the acquired anomaly factor Pfuture(t+n) with the preset anomaly assessment threshold YZ, and then... The matching formula obtains the abnormal matching result M(t+n), and an adaptive control strategy scheme is generated based on the abnormal matching result; When the abnormal matching result M(t+n) is 1, the abnormal situation is predicted, and an adaptive control strategy is generated based on the abnormal matching result, including adjusting system parameters, switching to backup system, reducing power consumption, and notifying operators to take action. When the abnormal matching result M(t+n) is 0, the prediction is obtained without any abnormalities, and no abnormal matching result is generated to generate an adaptive control strategy scheme.

12. A communication system for multi-source heterogeneous data fusion according to claim 1, characterized in that: The feedback and adaptive optimization module matches the anomaly matching result M(t+n) with the actual feedback result R(t+n). The actual feedback result R(t+n) includes the error term obtained through operator feedback and the model adjustment index Tzs used to adjust the anomaly detection model and the prediction model. The model adjustment index Tzs is used as an adjustment factor to indirectly adjust the anomaly detection performance and prediction performance of the model by affecting the parameter updates during the training process of the anomaly detection model and the prediction model. The model adjustment index Tzs is obtained through the following formula: Tzs(t+n)=Tzs(t)-η·(M(t+n)-R(t+n)); In the formula, Tzs(t) represents the model adjustment exponent at time t, specifically including the anomaly detection model and the prediction model, and η represents the learning rate, specifically the control adjustment step size.

13. A communication method for multi-source heterogeneous data fusion, comprising the multi-source heterogeneous data fusion communication system as described in any one of claims 1 to 12, characterized in that: Includes the following steps: Step 1: The data acquisition and preprocessing module collects data in real time from multiple heterogeneous data sources, forming a data collection group Draw, and preprocesses the data collection group Draw to obtain the preprocessed dataset Dpre. Step 2: The data fusion and feature extraction module performs data fusion on the preprocessed dataset Dpre to obtain the comprehensive temperature vibration feature Ftv(t), comprehensive power consumption interference feature Fpe(t), and micro-vibration noise feature Fmvn(t) at time t. Then, feature extraction is performed to form the fused feature data set Fset(t), which is then labeled and stored. Step 3: The anomaly detection model module constructs an anomaly detection model using the fused feature data set Fset(t). Through training and analysis of the anomaly detection model, the anomaly score set Ascore(t) of the current communication system is obtained. Step 4: The prediction model building module constructs a prediction model using the anomaly score set Ascore(t) and the stored historical dataset, and uses the prediction model to predict and obtain the anomaly factor Pfuture(t+n). Step 5: The early warning and decision-making module matches the acquired abnormal situation factor Pfuture(t+n) with the preset abnormal assessment threshold YZ to obtain the abnormal matching result M(t+n), and generates an adaptive control strategy scheme based on the abnormal matching result. Step 6: The feedback and adaptive optimization module matches the anomaly matching result M(t+n) with the actual feedback result R(t+n), obtains the error term, and obtains the model adjustment index Tzs used to adjust the anomaly detection model and the prediction model.

Images