Fusion processing system based on multi-source heterogeneous data

Abstract

The invention discloses a fusion processing system based on multi-source heterogeneous data, which relates to the technical field of data fusion processing, and comprises a real-time data acquisition module, an anomaly detection module, an intelligent evaluation module and a risk analysis and response module, various parameter data generated in the operation process of various sensors are collected in real time, and subtle changes of the various parameter data are captured, so that a data basis is provided for subsequent detection. According to the invention, through real-time data acquisition and anomaly detection, the system captures the tiny fluctuation and noise level of the sensor, and the anomaly is identified in time by using the data jitter index. The intelligent evaluation module dynamically detects deviation through machine learning and generates early warning when the deviation exceeds a threshold value. And the risk analysis module performs hierarchical response to ensure equipment stability and protect marine ecology and coastal health.

Description

Technical Field

[0001] This invention relates to the field of data fusion processing technology, and more specifically to a fusion processing system based on multi-source heterogeneous data. Background Technology

[0002] Data fusion processing based on multi-source heterogeneous data refers to the process of integrating, analyzing, and utilizing data from multiple sources, formats, structures, and semantics. This data may originate from various channels such as sensors, databases, networks, and social media, and may exist in structured, semi-structured, or unstructured forms. The goal of fusion processing is to eliminate data barriers, address the challenges posed by data diversity and inconsistency, and extract more comprehensive, accurate, and valuable information to support decision-making and applications. This requires the use of advanced data integration technologies, algorithms, and tools to solve problems related to data compatibility, redundancy, and conflicts, achieving unified data management and in-depth data mining.

[0003] The fusion processing of multi-source heterogeneous data is highly suitable for marine environmental monitoring. Its role is to integrate data from different types of sources (such as water quality, temperature, meteorology, satellite imagery, ship activity, and biosensors) to construct a comprehensive, dynamic monitoring system. This system can track and analyze multi-level information about the marine environment in real time, such as water quality changes, pollutant diffusion, and changes in biological communities, enabling precise early warning of potential pollution events and early detection of ecological anomalies. By fusing multi-source data, not only can the coverage and accuracy of monitoring be improved, but it can also provide scientific support for decision-making in ecological protection, pollution control, and resource management, thereby more effectively maintaining the stability and sustainability of marine ecosystems.

[0004] The existing technology has the following shortcomings:

[0005] In marine environmental monitoring, various sensors (such as those for water quality, temperature, and weather) operate in harsh environments for extended periods, making them prone to minor deviations. These deviations are usually not noticeable in the short term and are difficult to detect through verification by a single device. However, as multi-source data is fused, these deviations accumulate, potentially leading the system to misjudge the true state of the marine environment, such as misjudging water quality safety and missing critical pollution warnings. If this accumulation of deviations is not detected in time, it may cause actual pollution signals to be ignored as normal fluctuations, delaying the detection and control of pollution sources and allowing pollutants to continue to be released and spread. Over time, what was initially a controllable localized pollution event may evolve into a large-scale ecological disaster, affecting a wider marine ecosystem. Failure to respond promptly to oil spills, chemical pollution, or heavy metal spread could lead to mass mortality of marine life, further endangering nearby fishery resources and even threatening the health of coastal humans.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a fusion processing system based on multi-source heterogeneous data. Through real-time data acquisition and anomaly detection modules, the system can capture the frequency of minute data fluctuations from sensors in stable environments and the noise level when there is no signal input. It also identifies potential abnormal fluctuations in a timely manner through data anomaly jitter index and self-noise index. Based on this, an intelligent assessment module uses machine learning to dynamically evaluate the sensor status. Once the potential anomaly coefficient exceeds a threshold, an anomaly signal is immediately generated. Compared to single-sensor monitoring, the system fusion of multi-source data reacts faster, avoids misjudgments due to accumulated biases, and achieves more accurate pollution early warning. The risk analysis and response module classifies abnormal fluctuations into low, medium, and high risk levels, responding progressively to ensure stable equipment operation, reduce the impact of accumulated biases on judgment, protect marine ecology and the health of coastal residents, and thus solve the problems mentioned in the background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a fusion processing system based on multi-source heterogeneous data, comprising a real-time data acquisition module, an anomaly detection module, an intelligent assessment module, and a risk analysis and response module.

[0009] The real-time data acquisition module first establishes a monitoring system to collect various parameter data generated by various sensors during operation, capturing subtle changes in these parameters to provide a data foundation for subsequent detection.

[0010] The anomaly detection module performs anomaly analysis on the preprocessed sensor operating parameters to identify potential abnormal fluctuations.

[0011] The intelligent evaluation module uses a pre-learned machine learning model to intelligently evaluate the sensor's operating status based on the sensor's operating parameters after anomaly analysis.

[0012] The risk analysis and response module further analyzes sensors with potential abnormal fluctuations, determines the risk level of potential abnormal fluctuations, and formulates different response measures for different risk levels of potential abnormal fluctuations.

[0013] Preferably, the parameters generated by the sensor during operation include the minute fluctuation frequency of the sensor data under stable conditions and the noise level of the sensor when there is no signal input. The minute fluctuation frequency of the sensor data under stable conditions refers to the frequency of minute fluctuations in the output signal when the sensor keeps the measured environmental parameters constant. The noise level of the sensor when there is no signal input refers to the amount of random noise that still exists in the output signal when the sensor does not receive any external signals.

[0014] Preferably, after obtaining the data anomaly jitter index and self-noise index generated by anomaly analysis of the operating data of various sensors, the data anomaly jitter index and self-noise index are input into a pre-learned machine learning model. The machine learning model generates potential anomaly coefficients, and the operating status of various sensors is intelligently evaluated through the potential anomaly coefficients.

[0015] Preferably, the potential anomaly coefficients generated after anomaly analysis of the current sensor operating parameters under the detection window are compared and analyzed with a pre-set reference threshold for potential anomaly coefficients to identify potential abnormal fluctuations in the sensor operation. The specific steps are as follows:

[0016] If the potential anomaly coefficient is greater than or equal to the preset potential anomaly coefficient reference threshold, a potential anomaly operation signal is generated, indicating that there is a potential anomaly in the current operation of the sensor;

[0017] If the potential anomaly coefficient is less than the preset potential anomaly coefficient reference threshold, a normal operation signal is generated, indicating that the sensor is currently in a high-efficiency operating state.

[0018] Preferably, when the current sensor generates a potential abnormal operation signal during operation under the detection window, several potential abnormal coefficients generated by the current sensor during operation under the detection window are acquired to establish an analysis set. The potential abnormal coefficients in the analysis set are compared with the first-level reference threshold, the second-level reference threshold, and the potential abnormal coefficient reference threshold. The second-level reference threshold is greater than the first-level reference threshold, and the first-level reference threshold is greater than the potential abnormal coefficient reference threshold. The potential abnormal coefficients are compared and analyzed with the second-level reference threshold, the first-level reference threshold, and the potential abnormal coefficient reference threshold. The number of potential abnormal coefficients that are less than the first-level reference threshold and greater than or equal to the potential abnormal coefficient reference threshold is labeled as Qa, the number of potential abnormal coefficients that are less than the second-level reference threshold and greater than or equal to the first-level reference threshold is labeled as Qb, and the number of potential abnormal coefficients that are greater than or equal to the second-level reference threshold is labeled as Qc.

[0019] A comprehensive analysis of Qa, Qb, and Qc is performed to generate the Potential Anomaly Risk Level Coefficient (PARLC), based on the following formula:

[0020]

[0021] In the formula, k1, k2, and k3 are the preset proportional coefficients of Qa, Qb, and Qc, respectively, and k1, k2, and k3 are all greater than 0.

[0022] Preferably, the anomaly risk level coefficient generated by further comprehensive analysis of sensors with potential anomalies is compared with a pre-set first anomaly risk level coefficient reference threshold and a second anomaly risk level coefficient reference threshold to determine the risk level of potential sensor malfunction. The results of the comparison analysis are as follows:

[0023] If the potential anomaly risk level coefficient is less than the reference threshold of the first potential anomaly risk level coefficient, then the risk level of the current sensor operation potential anomaly is classified as a low-risk potential anomaly.

[0024] If the potential anomaly risk level coefficient is greater than or equal to the first potential anomaly risk level coefficient reference threshold and less than the second potential anomaly risk level coefficient reference threshold, then the risk level of the current sensor operation potential anomaly is classified as medium risk potential anomaly.

[0025] If the potential anomaly risk level coefficient is greater than or equal to the reference threshold of the second potential anomaly risk level coefficient, then the risk level of the current sensor operation potential anomaly is classified as a high-risk potential anomaly.

[0026] Preferably, within the detection window, the specific steps for performing anomaly analysis on the minute fluctuation frequencies of sensor data under stable conditions to generate a data anomaly jitter index are as follows:

[0027] Under the detection window, the signal data acquired by the sensor in a stable environment is calibrated as x(t), and the high-frequency component and low-frequency component of the signal data acquired by the sensor in a stable environment are separated by wavelet transform. The separated high-frequency component and low-frequency component are calibrated as H(t) and L(t) respectively. Among them, the high-frequency component H(t) is the sudden jitter information in the signal, and the low-frequency component L(t) is the smooth part of the signal.

[0028] Using the energy of the high-frequency component H(t) as a measure of jitter intensity, the jitter energy at each moment is calculated, and a nonlinear transformation is applied to amplify subtle jitter. The calculation expression is as follows:

[0029]

[0030] In the formula, E(t) is the jitter energy, representing the jitter energy at time point t, i is the i-th high-frequency component, n is the total number of high-frequency components, and α is the nonlinear amplification factor.

[0031] Transient peak detection is performed on the jitter energy E(t) to identify jitter peaks in the signal. A nonlinear activation function is used to improve the detection sensitivity of the peaks. The calculation expression is as follows:

[0032] P(t) = f(E(t)) = ln(1 + e) β·E(t) )

[0033] In the formula, P(t) is the peak value of the jitter wave, that is, the peak value of the jitter wave at time t, and f(E(t))=ln(1+e β·E(t) ) is a nonlinear activation function that improves detection sensitivity by transforming the jitter energy E(t). e is the natural base, and β is the amplification factor of the nonlinear activation function, which controls the response strength of the nonlinear activation function to the jitter energy E(t).

[0034] The cumulative jitter intensity within the detection window is calculated, and the overall trend of sensor fluctuation frequency change is comprehensively considered. The calculation expression is as follows:

[0035]

[0036] In the formula, C(T) is the cumulative jitter intensity, T is the end time of the detection window, and 0 is the start time of the detection window;

[0037] A nonlinear transformation is performed on the cumulative jitter intensity C(T) to generate an adaptive threshold, which is used to determine abnormal jitter levels. The calculation expression is as follows:

[0038] θ(T)=γ·e -δ·T +λ·C(T-1)

[0039] In the formula, θ(T) is the adaptive threshold, γ is the initial threshold coefficient, δ is the time decay coefficient, which controls the rate at which the initial threshold decays over time, and λ is the cumulative influence coefficient, which depends on the cumulative jitter intensity C(T-1) at the previous time point and is used to capture the continuity of jitter.

[0040] The data anomaly jitter index is generated by comparing the cumulative jitter intensity C(T) and the adaptive threshold θ(T). The calculation expression is as follows:

[0041]

[0042] In the formula, DAOM is the data anomaly jitter index, and η is the sensitivity coefficient.

[0043] Preferably, under the detection window, the specific steps for performing anomaly analysis on the noise level of the sensor when there is no signal input, and generating the self-noise index, are as follows:

[0044] Within the detection window, the noise signal from the sensor when there is no signal input is decomposed into noise components of different frequencies using Fourier transform. The calculation expression is as follows:

[0045]

[0046] In the formula, F(ω) is the signal representation in the frequency domain, that is, the Fourier transform of the original signal S(t) at frequency ω, and S(t) is the original signal in the time domain, the signal value at time point t, e -jωt It is the complex exponential kernel in the Fourier transform, used to project a time-domain signal into the frequency domain. e is the natural base, and j is the imaginary unit. This represents the integration over all values ​​from negative infinity to positive infinity at time t.

[0047] Since high-frequency noise is often the result of environmental interference, suppressing high-frequency noise components to obtain a stable low-frequency noise signal is achieved. The calculation expression is as follows:

[0048]

[0049] In the formula, G(ω) is the frequency domain signal after filtering, and Ω is the cutoff frequency of high-frequency noise. It is the high-frequency attenuation coefficient;

[0050] The filtered frequency domain signal G(ω) is restored to the time domain through inverse Fourier transform to obtain the processed noise signal, as shown in the following expression:

[0051]

[0052] In the formula, N(t) is the time-domain signal after filtering, that is, the performance of the noise signal at time point t, e -jωt It is the complex exponential kernel in the Fourier transform, used to convert frequency domain signals back to the time domain;

[0053] Then, the instantaneous energy of the signal within the detection window is calculated to capture the noise fluctuation amplitude when there is no signal input. The calculation expression is as follows:

[0054]

[0055] In the formula, U(t) is the noise energy at time t, T is the end time of the detection window, 0 is the start time of the detection window, and ρ is the nonlinear coefficient.

[0056] To highlight minute changes in the noise signal, the noise energy U(t) is nonlinearly amplified to generate an amplified energy index, the calculation expression of which is as follows:

[0057] M(t) = ln(1 + μU(t) 2 )

[0058] In the formula, M(t) is the amplified energy index, and μ is the amplification factor;

[0059] Finally, the amplified energy index M(t) is integrated and summed to obtain the self-noise index, which is calculated as follows:

[0060]

[0061] In the formula, INM is the self-noise index.

[0062] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0063] This invention, through a real-time data acquisition and anomaly detection module, can capture the minute fluctuation frequency of sensor data under stable conditions and the noise level when there is no signal input. It also identifies potential abnormal fluctuations in a timely manner using parameters such as the data anomaly jitter index and the self-noise index. Based on this, the intelligent evaluation module uses a machine learning model to dynamically evaluate the sensor state. Once the potential anomaly coefficient exceeds a set threshold, a potential abnormal operation signal is immediately generated. Compared to traditional single-sensor monitoring methods, the system integrating multi-source data can react quickly, avoiding misjudgments caused by the accumulation of small deviations, and achieving more accurate and timely pollution warnings. This not only improves the sensitivity of the monitoring system in detecting pollution sources but also effectively shortens the response time from detection to warning, ensuring that pollution is addressed immediately before it spreads.

[0064] This invention, through a risk analysis and response module, further integrates and categorizes potential abnormal fluctuations into low-risk, medium-risk, and high-risk levels, and formulates corresponding response measures based on the risk level. This tiered system ensures a gradual response to potential sensor anomalies, avoiding over-intervention in low-risk anomalies while enabling timely action when high-risk anomalies occur. Through multi-level risk assessment and tiered response, the system better maintains the long-term stable operation of marine monitoring equipment, reduces the impact of accumulated deviations on the overall system judgment, and effectively guarantees the accuracy and reliability of monitoring data. This refined risk management approach has profound significance for protecting marine ecosystems, controlling pollution sources, safeguarding nearby fishery resources, and protecting the health of coastal residents. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0066] Figure 1 This is a schematic diagram of the module of the multi-source heterogeneous data fusion processing system of the present invention. Detailed Implementation

[0067] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0068] This invention provides, for example Figure 1 The system shown is a multi-source heterogeneous data fusion processing system, including a real-time data acquisition module, an anomaly detection module, an intelligent assessment module, and a risk analysis and response module.

[0069] The real-time data acquisition module first establishes a monitoring system to collect various parameter data generated by various sensors during operation, capturing subtle changes in these parameters to provide a data foundation for subsequent detection.

[0070] The various parameters generated by the sensor during operation include the frequency of minute fluctuations in the sensor's data under stable conditions and the noise level of the sensor when there is no signal input. The frequency of minute fluctuations in the sensor's data under stable conditions refers to the frequency of minute fluctuations in the output signal when the sensor's measurement environment parameters remain constant. The noise level of the sensor when there is no signal input refers to the amount of random noise that still exists in the output signal when the sensor does not receive any external signal (i.e., the measured value should be zero or remain unchanged).

[0071] The anomaly detection module performs anomaly analysis on the preprocessed sensor operating parameters to identify potential abnormal fluctuations.

[0072] After obtaining the minute fluctuation frequency of the sensor data under stable conditions and the noise level of the sensor when there is no signal input, the system performs anomaly analysis on the minute fluctuation frequency of the sensor data under stable conditions to generate a data anomaly jitter index, and performs anomaly analysis on the noise level of the sensor when there is no signal input to generate a self-noise index. The system identifies potential abnormal fluctuations in various sensors through the data anomaly jitter index and the self-noise index.

[0073] Operating sensors in harsh environments over extended periods can lead to minute deviations in the acquired data. This is because harsh environments (such as high humidity, salt corrosion, and temperature fluctuations) cause gradual wear and tear and performance degradation to the sensor's internal components and external structure. For example, the high salinity of marine environments accelerates the oxidation and corrosion of sensor electrodes, reducing electrode sensitivity and affecting data accuracy. Drastic changes in humidity and temperature can also cause aging of internal circuit components or deterioration of material properties, affecting the sensor's response time and accuracy. Furthermore, prolonged exposure to such environments can lead to the gradual accumulation of minute mechanical vibrations or microbial adhesion, resulting in increased jitter in the sensor's output signal, meaning that even in stable environments, abnormal minute fluctuations may occur. These phenomena indicate that the sensor's operating state has deviated from its optimal calibration state, thus the acquired data will carry continuously accumulating minute deviations, affecting the accuracy and reliability of the monitoring results.

[0074] Within the detection window, the specific steps for performing anomaly analysis on the minute fluctuation frequencies of sensor data under stable conditions to generate a data anomaly jitter index are as follows:

[0075] Under the detection window, the signal data acquired by the sensor in a stable environment is calibrated as x(t), and the high-frequency component and low-frequency component of the signal data acquired by the sensor in a stable environment are separated by wavelet transform. The separated high-frequency component and low-frequency component are calibrated as H(t) and L(t) respectively. Among them, the high-frequency component H(t) is the sudden jitter information in the signal, and the low-frequency component L(t) is the smooth part of the signal.

[0076] Wavelet transform is a signal processing technique used to decompose a raw signal into components of different frequencies while preserving temporal information, enabling localized analysis in the time-frequency domain. Unlike Fourier transform, wavelet transform can capture high-frequency details in a short time while capturing low-frequency changes over a long time. Therefore, it is particularly effective in analyzing complex, non-stationary signals. Here, the role of wavelet transform is to decompose sensor data into high-frequency and low-frequency components. The high-frequency components represent transient fluctuations or jitter in the data (which may reflect abnormal signals), while the low-frequency components represent stationary trends (used to filter noise). Through this decomposition, subtle fluctuations in sensor data can be identified more accurately, thereby detecting potential abnormal jitter.

[0077] Using the energy of the high-frequency component H(t) as a measure of jitter intensity, the jitter energy at each moment is calculated, and a nonlinear transformation is applied to amplify subtle jitter. The calculation expression is as follows:

[0078]

[0079] In the formula, E(t) is the jitter energy, which represents the jitter energy at time t and is used to quantify the fluctuation intensity of high-frequency components in the sensor signal. i is the i-th high-frequency component, n is the total number of high-frequency components, and α is the nonlinear amplification factor.

[0080] The nonlinear amplification factor, denoted as α, is a parameter used to nonlinearly amplify a signal. Its purpose is to amplify subtle changes or fluctuations in the original signal, giving these minute fluctuations a greater weight in the calculation results. This amplification is "nonlinear," meaning the amplification ratio does not increase linearly but rather at a higher rate as the value increases, allowing even originally weak fluctuations in the signal to significantly affect the results.

[0081] Here, the nonlinear amplification factor α amplifies subtle high-frequency fluctuations by increasing the absolute value of the high-frequency component α to the power of α. This makes minute deviations in the sensor data more prominent in the calculation of the jitter energy E(t). Thus, even slight jitters or deviations accumulated by the sensor during operation in harsh environments can be effectively identified after nonlinear amplification, leading to more sensitive detection of potential anomalies.

[0082] In simple terms, the nonlinear amplification factor here amplifies minute vibrations, improves sensitivity, and makes the vibration energy more reflective of subtle deviations in the sensor data, thereby helping to monitor the sensor's health status more accurately.

[0083] Transient peak detection is performed on the jitter energy E(t) to identify jitter peaks in the signal. A nonlinear activation function is used to improve the detection sensitivity of the peaks. The calculation expression is as follows:

[0084] P(t) = f(E(t)) = ln(1 + e) β·E(t) )

[0085] In the formula, P(t) is the peak value of the jitter wave, that is, the peak value of the jitter wave at time t, and f(E(t))=ln(1+e β·E(t) ) is a nonlinear activation function that improves detection sensitivity by transforming the jitter energy E(t). e is the natural base, and β is the amplification factor of the nonlinear activation function, which controls the response strength of the nonlinear activation function to the jitter energy E(t).

[0086] The amplification factor of a nonlinear activation function is a parameter used to control the degree of response of the activation function to the input value. It is represented by β in the formula, and its function is to adjust the amplification of the activation function; that is, when the input value increases, the amplification factor makes the increase in the output value more significant. For example, under high jitter energy conditions, a larger amplification factor β will amplify the activation function's response to the input signal, making abnormal signals more obvious. This can enhance the system's sensitivity to abnormal fluctuations, making it easier to detect potential anomalies in sensor data, thereby improving the accuracy of monitoring and early warning.

[0087] The cumulative jitter intensity within the detection window is calculated, and the overall trend of sensor fluctuation frequency change is comprehensively considered. The calculation expression is as follows:

[0088]

[0089] In the formula, C(T) is the cumulative jitter intensity, T is the end time of the detection window, and 0 is the start time of the detection window;

[0090] A nonlinear transformation is performed on the cumulative jitter intensity C(T) to generate an adaptive threshold, which is used to determine abnormal jitter levels. The calculation expression is as follows:

[0091] θ(T)=γ·e -δ·T +λ·C(T-1)

[0092] In the formula, θ(T) is the adaptive threshold, which is used to dynamically evaluate whether the current jitter intensity is abnormal, γ is the initial threshold coefficient, δ is the time decay coefficient, which controls the rate at which the initial threshold decays over time, and λ is the cumulative influence coefficient, which depends on the cumulative jitter intensity C(T-1) at the previous time point and is used to capture the continuity of jitter.

[0093] The data anomaly jitter index is generated by comparing the cumulative jitter intensity C(T) and the adaptive threshold θ(T). The calculation expression is as follows:

[0094]

[0095] In the formula, DAOM is the data anomaly jitter index, and η is the sensitivity coefficient, which is used to amplify the jitter index and make it more sensitive to jitter exceeding the threshold.

[0096] Within the detection window, anomaly analysis of the minute fluctuations in sensor data under stable conditions generates a data anomaly jitter index. A higher index value indicates increased data fluctuation frequency under stable conditions, potentially reflecting long-term operation in harsh environments, leading to a gradual deterioration of the sensor's operating status and consequently affecting data accuracy. Prolonged exposure to harsh environments (such as high humidity, salt corrosion, and temperature fluctuations) causes wear and tear on internal sensor components, resulting in the accumulation of minute deviations in the data. Therefore, a high data anomaly jitter index indicates that the sensor's data reliability has been affected by the environment, resulting in deviations; conversely, a low and stable jitter index usually indicates normal sensor operation and that the acquired data has not been significantly affected by deviations.

[0097] An abnormal increase in sensor noise levels when there is no signal input usually indicates that the sensor has been operating in a harsh environment for a long time, which can easily lead to slight deviations in the acquired data. Harsh environments (such as the high salinity, high humidity, and drastic temperature changes of seawater) gradually corrode the internal components of the sensor, causing physical or chemical damage to its sensitive parts, thus affecting the stability of signal processing. Over time, the deterioration of the sensor's internal circuitry and changes in its microstructure lead to an increase in background noise levels, which manifests as abnormal noise fluctuations when there is no signal input. Increased noise levels may mask or interfere with weak signals, introducing imperceptible errors into normal data and causing the accumulation of small deviations. In addition, electrode oxidation, aging of internal circuitry, or microbial adhesion can also weaken the sensor's response sensitivity, further amplifying noise and distortion. Such deviations may be misinterpreted as normal fluctuations and go undetected, affecting the accuracy of monitoring data, leading to misjudgments of the environment, or even delays in pollution warnings.

[0098] Within the detection window, the specific steps for performing anomaly analysis on the sensor's noise level when there is no signal input, and generating the self-noise index, are as follows:

[0099] Within the detection window, the noise signal from the sensor when there is no signal input is decomposed into noise components of different frequencies using Fourier transform. The calculation expression is as follows:

[0100]

[0101] In the formula, F(ω) is the signal representation in the frequency domain, that is, the Fourier transform of the original signal S(t) at frequency ω, and S(t) is the original signal in the time domain, the signal value at time point t, e -jωt It is the complex exponential kernel in the Fourier transform, used to project a time-domain signal into the frequency domain. e is the natural base, and j is the imaginary unit. This means that by integrating all values ​​from negative infinity to positive infinity over time point t, the contribution of the original signal S(t) at all time points is summarized.

[0102] The Fourier transform decomposes a time-domain signal into a series of frequency components. The core idea of ​​the Fourier transform is to view the original signal as a superposition of multiple sine and cosine waves, each with a specific frequency, amplitude, and phase. In the Fourier transform formula, the contribution of the signal at each frequency is calculated through integration, thus obtaining the amplitude and phase information in the frequency domain. This process can be understood as "projection," projecting the time-domain signal onto different frequencies one by one to obtain the amplitude F(ω) at the corresponding frequency. After the Fourier transform, the signal appears as a series of frequency components in the frequency domain, each corresponding to the amplitude of a sine or cosine wave. The combination of these frequency components completely describes the original signal. In this way, we can analyze the signal from a frequency perspective, identifying the intensity of different frequency components, such as separating low-frequency components (usually gently changing) and high-frequency components (usually rapidly changing or noise).

[0103] Since high-frequency noise is often the result of environmental interference, suppressing high-frequency noise components to obtain a stable low-frequency noise signal is achieved. The calculation expression is as follows:

[0104]

[0105] In the formula, G(ω) is the frequency domain signal after filtering, and Ω is the cutoff frequency for high-frequency noise. Signal components with frequencies greater than Ω are suppressed to remove high-frequency noise. It is a high-frequency attenuation coefficient, used to reduce the amplitude of high-frequency components, thereby reducing the impact of high-frequency noise on the signal;

[0106] The filtered frequency domain signal G(ω) is restored to the time domain through inverse Fourier transform to obtain the processed noise signal, as shown in the following expression:

[0107]

[0108] In the formula, N(t) is the time-domain signal after filtering, that is, the performance of the noise signal at time point t, e -jωt It is the complex exponential kernel in the Fourier transform, used to convert frequency domain signals back to the time domain;

[0109] Then, the instantaneous energy of the signal within the detection window is calculated to capture the noise fluctuation amplitude when there is no signal input. The calculation expression is as follows:

[0110]

[0111] In the formula, U(t) is the noise energy at time t, T is the end time of the detection window, 0 is the start time of the detection window, and ρ is the nonlinear coefficient.

[0112] The nonlinear coefficient ρ is an exponential parameter used to amplify small fluctuations in noise energy calculations. In the noise energy formula, ρ is typically greater than 1 to improve sensitivity to small noise variations. By nonlinearly boosting the absolute value of the noise signal N(t), ρ amplifies the contribution of smaller noise fluctuations to the energy, making subtle noise changes easier to detect. This is particularly important in harsh environments, where long-running sensors may exhibit subtle but persistent deviations under adverse conditions. The introduction of the nonlinear coefficient ρ significantly improves the precision of noise analysis, thereby enabling more accurate identification of potential anomalies.

[0113] To highlight minute changes in the noise signal, the noise energy U(t) is nonlinearly amplified to generate an amplified energy index, the calculation expression of which is as follows:

[0114] M(t) = ln(1 + μU(t) 2 )

[0115] In the formula, M(t) is the amplified energy index, which is used to highlight small changes in noise energy after nonlinear processing, and μ is the amplification coefficient, used to adjust the intensity of the amplification effect.

[0116] Nonlinear amplification of noise energy enhances sensitivity to minute noise fluctuations. Sensors operating in harsh environments over extended periods often exhibit slight deviations or noise fluctuations; these minute changes can be early signs of sensor performance degradation. Nonlinear amplification makes these initially insignificant fluctuations more apparent in the energy metrics, making them easier for monitoring systems to detect. This amplification process assigns higher weight to small-amplitude noise, facilitating timely identification and response to potential sensor anomalies and preventing the accumulation of small deviations over time from developing into widespread data distortion.

[0117] Generating amplified energy metrics provides a more accurate reference for assessing sensor health. The amplified energy metric M(t) is highly sensitive to noise fluctuations, effectively reflecting subtle increases and trends in noise levels. Integrating these amplified energy metrics yields a comprehensive self-noise index, used to quantify the sensor's noise level in the absence of signal input. A higher index value indicates greater noise fluctuations, potentially pointing to sensor aging or environmental influences, facilitating timely warnings and maintenance interventions from the monitoring system.

[0118] The physical meaning of the amplification factor μ lies in adjusting the amplification intensity during the nonlinear amplification process. Specifically, it amplifies subtle changes in noise energy, making previously insignificant noise fluctuations more prominent in the calculation process, thereby improving the sensitivity to minute noise anomalies in the sensor. In the noise energy formula, a larger μ value significantly amplifies changes in noise energy, suitable for noise monitoring scenarios requiring high sensitivity, but may also introduce unnecessary noise interference; a smaller μ value weakens this amplification effect, making it more suitable for environments with smaller noise fluctuations. Recommended μ values ​​are typically between 0.1 and 1, and can be adjusted according to the actual noise environment and monitoring requirements to achieve a balance between sensitivity and stability.

[0119] Finally, the amplified energy index M(t) is integrated and summed to obtain the self-noise index, which is calculated as follows:

[0120]

[0121] In the formula, INM is the self-noise index;

[0122] Within the detection window, a higher self-noise index (SNOI) value generated after anomaly analysis of the sensor's noise level when there is no signal input indicates a higher noise level for the sensor. This typically signifies that the sensor has been operating in a harsh environment for a long time, causing damage or aging of internal components, leading to decreased data acquisition accuracy and resulting in minor deviations. In harsh environments (such as high humidity, salt spray, and temperature fluctuations), the sensor's circuitry and sensitive elements may gradually corrode or deteriorate, causing an increase in self-noise levels. Therefore, a higher SNOI indicates that the sensor's operating state has deviated from normal, and the acquired data may contain biases. Conversely, a low and stable SNOI value indicates that the sensor's internal condition is good, and the data does not show significant deviations.

[0123] The intelligent evaluation module uses a pre-learned machine learning model to intelligently evaluate the sensor's operating status based on the sensor's operating parameters after anomaly analysis.

[0124] After obtaining the Data Anomaly Jitter Index (DAOM) and Self-Noise Index (INM) generated by anomaly analysis of various sensor operating data, the Data Anomaly Jitter Index (DAOM) and Self-Noise Index (INM) are input into a pre-learned machine learning model. The machine learning model generates the Latent Anomaly Factor (LAF), and the Latent Anomaly Factor (LAF) is used to intelligently evaluate the operating status of various sensors.

[0125] The machine learning model is not limited here; any machine learning model capable of generating the latent anomaly coefficient (LAF) after comprehensively analyzing the data anomaly jitter index (DAOM) and the self-noise index (INM) is acceptable. To achieve the technical solution of this invention, this invention provides a specific implementation method:

[0126] The formula for generating the latent anomaly coefficient (LAF) is as follows:

[0127]

[0128] In the formula, f1 and f2 are the preset proportional coefficients of the data anomaly jitter index DAOM and the self-noise index INM, respectively, and both f1 and f2 are greater than 0.

[0129] As can be seen from the expression for calculating the potential anomaly coefficient, under the detection window, the larger the value of the data anomaly jitter index generated after anomaly analysis of the small fluctuation frequency of the sensor data in a stable environment, and the larger the value of the self-noise index generated after anomaly analysis of the noise level of the sensor when there is no signal input, the larger the value of the potential anomaly coefficient generated after anomaly analysis of the current sensor operating parameters under the detection window, the greater the probability that the current sensor is operating under harsh conditions, leading to anomalies in its operating state. Conversely, the smaller the value, the smaller the probability that its operating state is abnormal.

[0130] The potential anomaly coefficients generated after anomaly analysis of the current sensor operating parameters under the detection window are compared with a pre-set reference threshold for potential anomaly coefficients to identify potential abnormal fluctuations during sensor operation. The specific steps are as follows:

[0131] If the potential anomaly coefficient is greater than or equal to the preset potential anomaly coefficient reference threshold, a potential anomaly operation signal is generated, indicating that there is a potential anomaly in the current operation of the sensor;

[0132] If the potential anomaly coefficient is less than the preset potential anomaly coefficient reference threshold, a normal operation signal is generated, indicating that the sensor is currently in a high-efficiency operating state.

[0133] The risk analysis and response module further analyzes sensors with potential abnormal fluctuations, determines the risk level of potential abnormal fluctuations, and formulates different response measures for different risk levels of potential abnormal fluctuations.

[0134] When the sensor generates a potential abnormal operation signal during operation within the detection window, several potential abnormal coefficients generated by the sensor during operation within the detection window are acquired to establish an analysis set. The potential abnormal coefficients in the analysis set are compared with the first-level reference threshold, the second-level reference threshold, and the potential abnormal coefficient reference threshold. The second-level reference threshold is greater than the first-level reference threshold, and the first-level reference threshold is greater than the potential abnormal coefficient reference threshold. The potential abnormal coefficients are compared and analyzed with the second-level reference threshold, the first-level reference threshold, and the potential abnormal coefficient reference threshold. The number of potential abnormal coefficients that are less than the first-level reference threshold and greater than or equal to the potential abnormal coefficient reference threshold is labeled as Qa, the number of potential abnormal coefficients that are less than the second-level reference threshold and greater than or equal to the first-level reference threshold is labeled as Qb, and the number of potential abnormal coefficients that are greater than or equal to the second-level reference threshold is labeled as Qc.

[0135] A comprehensive analysis of Qa, Qb, and Qc is performed to generate the Potential Anomaly Risk Level Coefficient (PARLC), based on the following formula:

[0136]

[0137] In the formula, k1, k2, and k3 are the preset proportional coefficients of Qa, Qb, and Qc, respectively, and k1, k2, and k3 are all greater than 0.

[0138] As can be seen from the expression for calculating the anomaly risk level coefficient, the larger the value of the anomaly risk level coefficient, the more serious the potential anomaly is during the operation of the sensor, and vice versa.

[0139] The anomaly risk level coefficient generated by further comprehensive analysis of sensors with potential anomalies will be compared with a pre-set first anomaly risk level coefficient reference threshold and a second anomaly risk level coefficient reference threshold to determine the risk level of potential sensor malfunction. The results of the comparison analysis are as follows:

[0140] If the potential anomaly risk level coefficient is less than the reference threshold of the first potential anomaly risk level coefficient, then the risk level of the current sensor operation potential anomaly is classified as a low-risk potential anomaly.

[0141] If the potential anomaly risk level coefficient is greater than or equal to the first potential anomaly risk level coefficient reference threshold and less than the second potential anomaly risk level coefficient reference threshold, then the risk level of the current sensor operation potential anomaly is classified as medium risk potential anomaly.

[0142] If the potential anomaly risk level coefficient is greater than or equal to the reference threshold of the second potential anomaly risk level coefficient, then the risk level of the current sensor operation potential anomaly is classified as a high-risk potential anomaly.

[0143] For sensors classified into different risk levels, the following tiered response measures can be taken:

[0144] For sensors classified as having low-risk potential anomalies, these anomalies are addressed through regular monitoring and recording. Periodic reviews are conducted to ensure data stability; for example, sensor data calibration checks are performed every few weeks, and observed changes are recorded. Simultaneously, a mild automatic correction function is implemented to ensure the system can fine-tune itself under low-risk conditions, thereby extending the equipment's uptime and monitoring accuracy.

[0145] For sensors classified as having medium-risk potential anomalies, more proactive intervention measures are needed when these anomalies are detected. These measures include increasing monitoring frequency and enabling real-time alarms to ensure timely understanding of the sensor's status. Automated alarm functions should be set up to promptly alert the maintenance team to further inspect the sensors and shorten calibration and inspection cycles, allowing for rapid response to further accumulation or changes in anomalies. Simultaneously, on-site inspections should be arranged to assess whether sensor replacement or maintenance is necessary to maintain data accuracy and system reliability.

[0146] For sensors classified as high-risk potential anomalies, emergency response measures are implemented to ensure the safety of the monitoring system and the accuracy of the data. Data acquisition from the affected sensor is suspended to prevent the abnormal data from impacting the overall system data quality, and a maintenance team is assigned to conduct a comprehensive diagnosis and repair; if the repair is ineffective, the device is replaced. To ensure the continuity of the monitoring system, backup sensors are activated. Simultaneously, the status of all sensors is thoroughly reviewed to prevent similar anomalies from occurring in other devices.

[0147] This invention, through a real-time data acquisition and anomaly detection module, can capture the minute fluctuation frequency of sensor data under stable conditions and the noise level when there is no signal input. It also identifies potential abnormal fluctuations in a timely manner using parameters such as the data anomaly jitter index and the self-noise index. Based on this, the intelligent evaluation module uses a machine learning model to dynamically evaluate the sensor state. Once the potential anomaly coefficient exceeds a set threshold, a potential abnormal operation signal is immediately generated. Compared to traditional single-sensor monitoring methods, the system integrating multi-source data can react quickly, avoiding misjudgments caused by the accumulation of small deviations, and achieving more accurate and timely pollution warnings. This not only improves the sensitivity of the monitoring system in detecting pollution sources but also effectively shortens the response time from detection to warning, ensuring that pollution is addressed immediately before it spreads.

[0148] This invention, through a risk analysis and response module, further integrates and categorizes potential abnormal fluctuations into low-risk, medium-risk, and high-risk levels, and formulates corresponding response measures based on the risk level. This tiered system ensures a gradual response to potential sensor anomalies, avoiding over-intervention in low-risk anomalies while enabling timely action when high-risk anomalies occur. Through multi-level risk assessment and tiered response, the system better maintains the long-term stable operation of marine monitoring equipment, reduces the impact of accumulated deviations on the overall system judgment, and effectively guarantees the accuracy and reliability of monitoring data. This refined risk management approach has profound significance for protecting marine ecosystems, controlling pollution sources, safeguarding nearby fishery resources, and protecting the health of coastal residents.

[0149] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A fusion processing system based on multi-source heterogeneous data, characterized in that, It includes a real-time data acquisition module, an anomaly detection module, an intelligent assessment module, and a risk analysis and response module. The real-time data acquisition module first establishes a monitoring system to collect various parameter data generated by various sensors during operation, capturing subtle changes in these parameters to provide a data foundation for subsequent detection. The anomaly detection module performs anomaly analysis on the preprocessed sensor operating parameters to identify potential abnormal fluctuations. The intelligent evaluation module uses a pre-learned machine learning model to intelligently evaluate the sensor's operating status based on the sensor's operating parameters after anomaly analysis. The risk analysis and response module further analyzes sensors with potential abnormal fluctuations, determines the risk level of potential abnormal fluctuations, and formulates different response measures for different risk levels of potential abnormal fluctuations.

2. The fusion processing system based on multi-source heterogeneous data according to claim 1, characterized in that: The various parameters generated by the sensor during operation include the frequency of minute fluctuations in the sensor's data under stable conditions and the noise level of the sensor when there is no signal input. The frequency of minute fluctuations in the sensor's data under stable conditions refers to the frequency of minute fluctuations in the output signal when the measured environmental parameters remain constant. The noise level of the sensor when there is no signal input refers to the amount of random noise that still exists in the output signal when the sensor does not receive any external signals.

3. The fusion processing system based on multi-source heterogeneous data according to claim 2, characterized in that: After obtaining the data anomaly jitter index and self-noise index generated by anomaly analysis of various sensor operating data, the data anomaly jitter index and self-noise index are input into a pre-learned machine learning model. The machine learning model generates potential anomaly coefficients, and the operating status of various sensors is intelligently evaluated through the potential anomaly coefficients.

4. The fusion processing system based on multi-source heterogeneous data according to claim 3, characterized in that: The potential anomaly coefficients generated after anomaly analysis of the current sensor operating parameters under the detection window are compared with a pre-set reference threshold for potential anomaly coefficients to identify potential abnormal fluctuations during sensor operation. The specific steps are as follows: If the potential anomaly coefficient is greater than or equal to the preset potential anomaly coefficient reference threshold, a potential anomaly operation signal is generated, indicating that there is a potential anomaly in the current operation of the sensor; If the potential anomaly coefficient is less than the preset potential anomaly coefficient reference threshold, a normal operation signal is generated, indicating that the sensor is currently in a high-efficiency operating state.

5. The fusion processing system based on multi-source heterogeneous data according to claim 4, characterized in that: When the sensor generates a potential abnormal operation signal during operation within the detection window, several potential abnormal coefficients generated by the sensor during operation within the detection window are acquired to establish an analysis set. The potential abnormal coefficients in the analysis set are compared with the first-level reference threshold, the second-level reference threshold, and the potential abnormal coefficient reference threshold. The second-level reference threshold is greater than the first-level reference threshold, and the first-level reference threshold is greater than the potential abnormal coefficient reference threshold. The potential abnormal coefficients are compared and analyzed with the second-level reference threshold, the first-level reference threshold, and the potential abnormal coefficient reference threshold. The number of potential abnormal coefficients that are less than the first-level reference threshold and greater than or equal to the potential abnormal coefficient reference threshold is labeled as Qa, the number of potential abnormal coefficients that are less than the second-level reference threshold and greater than or equal to the first-level reference threshold is labeled as Qb, and the number of potential abnormal coefficients that are greater than or equal to the second-level reference threshold is labeled as Qc. By comprehensively analyzing Qa, Qb, and Qc, the potential anomaly risk level coefficient RARLC is generated, based on the following formula: ， In the formula, k1, k2, and k3 are the preset proportional coefficients of Qa, Qb, and Qc, respectively, and k1, k2, and k3 are all greater than 0.

6. The fusion processing system based on multi-source heterogeneous data according to claim 5, characterized in that: The anomaly risk level coefficient generated by further comprehensive analysis of sensors with potential anomalies will be compared with a pre-set first anomaly risk level coefficient reference threshold and a second anomaly risk level coefficient reference threshold to determine the risk level of potential sensor malfunction. The results of the comparison analysis are as follows: If the potential anomaly risk level coefficient is less than the reference threshold of the first potential anomaly risk level coefficient, then the risk level of the current sensor operation potential anomaly is classified as a low-risk potential anomaly. If the potential anomaly risk level coefficient is greater than or equal to the first potential anomaly risk level coefficient reference threshold and less than the second potential anomaly risk level coefficient reference threshold, then the risk level of the current sensor operation potential anomaly is classified as medium risk potential anomaly. If the potential anomaly risk level coefficient is greater than or equal to the reference threshold of the second potential anomaly risk level coefficient, then the risk level of the current sensor operation potential anomaly is classified as a high-risk potential anomaly.

7. The fusion processing system based on multi-source heterogeneous data according to claim 3, characterized in that, Within the detection window, the specific steps for performing anomaly analysis on the minute fluctuation frequencies of sensor data under stable conditions to generate a data anomaly jitter index are as follows: Under the detection window, the signal data acquired by the sensor in a stable environment is calibrated as x(t), and the high-frequency component and low-frequency component of the signal data acquired by the sensor in a stable environment are separated by wavelet transform. The separated high-frequency component and low-frequency component are calibrated as H(t) and L(t) respectively. Among them, the high-frequency component H(t) is the sudden jitter information in the signal, and the low-frequency component L(t) is the smooth part of the signal. Using the energy of the high-frequency component H(t) as a measure of jitter intensity, the jitter energy at each moment is calculated, and a nonlinear transformation is applied to amplify subtle jitter. The calculation expression is as follows: ， In the formula, E(t) is the jitter energy, representing the jitter energy at time point t, i is the i-th high-frequency component, n is the total number of high-frequency components, and α is the nonlinear amplification factor. Transient peak detection is performed on the jitter energy E(t) to identify jitter peaks in the signal. A nonlinear activation function is used to improve the detection sensitivity of the peaks. The calculation expression is as follows: P(t)=f(E(t))=ln(1+e β·E(t) ), In the formula, P(t) is the peak value of the jitter wave, that is, the peak value of the jitter wave at time t, and f(E(t))=ln(1+e β·E(t) ) is a nonlinear activation function that improves detection sensitivity by transforming the jitter energy E(t). e is the natural base, and β is the amplification factor of the nonlinear activation function, which controls the response strength of the nonlinear activation function to the jitter energy E(t). The cumulative jitter intensity within the detection window is calculated, and the overall trend of sensor fluctuation frequency change is comprehensively considered. The calculation expression is as follows: ， In the formula, C(T) is the cumulative jitter intensity, T is the end time of the detection window, and 0 is the start time of the detection window; A nonlinear transformation is performed on the cumulative jitter intensity C(T) to generate an adaptive threshold, which is used to determine abnormal jitter levels. The calculation expression is as follows: θ(T)=γ·e -δ·T +λ·C(T-1), In the formula, θ(T) is the adaptive threshold, γ is the initial threshold coefficient, δ is the time decay coefficient, which controls the rate at which the initial threshold decays over time, and λ is the cumulative influence coefficient, which depends on the cumulative jitter intensity C(T-1) at the previous time point and is used to capture the continuity of jitter. The data anomaly jitter index is generated by comparing the cumulative jitter intensity C(T) and the adaptive threshold θ(T). The calculation expression is as follows: ， In the formula, DAOM is the data anomaly jitter index, and η is the sensitivity coefficient.

8. The fusion processing system based on multi-source heterogeneous data according to claim 3, characterized in that, Within the detection window, the specific steps for performing anomaly analysis on the sensor's noise level when there is no signal input, and generating the self-noise index, are as follows: Within the detection window, the noise signal from the sensor when there is no signal input is decomposed into noise components of different frequencies using Fourier transform. The calculation expression is as follows: ， In the formula, F(ω) is the signal representation in the frequency domain, that is, the Fourier transform of the original signal S(t) at frequency ω, and S(t) is the original signal in the time domain, the signal value at time point t, e -jωt It is the complex exponential kernel in the Fourier transform, used to project a time-domain signal into the frequency domain. e is the natural base, and j is the imaginary unit. This represents the integration over all values ​​from negative infinity to positive infinity at time t. Since high-frequency noise is often the result of environmental interference, suppressing high-frequency noise components to obtain a stable low-frequency noise signal is achieved. The calculation expression is as follows: ， In the formula, G(ω) is the frequency domain signal after filtering, and Ω is the cutoff frequency of high-frequency noise. It is the high-frequency attenuation coefficient; The filtered frequency domain signal G(ω) is restored to the time domain through inverse Fourier transform to obtain the processed noise signal, as shown in the following expression: ， In the formula, N(t) is the time-domain signal after filtering, that is, the performance of the noise signal at time t, and e -jωt It is the complex exponential kernel in the Fourier transform, used to convert frequency domain signals back to the time domain; Then, the instantaneous energy of the signal within the detection window is calculated to capture the noise fluctuation amplitude when there is no signal input. The calculation expression is as follows: ， In the formula, U(t) is the noise energy at time t, T is the end time of the detection window, 0 is the start time of the detection window, and ρ is the nonlinear coefficient. To highlight minute changes in the noise signal, the noise energy U(t) is nonlinearly amplified to generate an amplified energy index, the calculation expression of which is as follows: M(t)=ln(1+μU(t) 2 ), In the formula, M(t) is the amplified energy index, and μ is the amplification factor; Finally, the amplified energy index M(t) is integrated and summed to obtain the self-noise index, which is calculated as follows: ， In the formula, INM is the self-noise index.

Images