Dynamic trend evaluation method for multi-source monitoring data

Abstract

The invention relates to the field of data analysis, in particular to a multi-source monitoring data-oriented dynamic trend assessment method, which comprises the following steps of: acquiring and preprocessing multi-source monitoring data to obtain a historical data sliding window and a basic smoothing coefficient for trend assessment; performing weighted correlation analysis on the disturbance variable and the measured variable change sequence to obtain a working condition response decoupling factor; obtaining a trend stability factor by evaluating the geometric morphology of the smooth trajectory of the measured variable; performing working condition response and trend stability combined correction on the basic smoothing coefficient to obtain a dynamic smoothing coefficient; a smooth trend value and a heat exchanger sub-health early warning signal are obtained by performing exponential weighted moving average and change rate judgment on a measured variable, and the problem that a fixed parameter EWMA cannot distinguish a working condition adjustment response and a tiny fault trend under a multivariable complex working condition is solved.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method for dynamic trend assessment of multi-source monitoring data. Background Technology

[0002] In continuous industrial production processes such as chemical, power, and metallurgical industries, shell-and-tube heat exchangers are key equipment for maintaining process thermal balance and energy utilization efficiency. To ensure the safe and stable operation of the production process and the achievement of product quality standards, enterprises typically install online monitoring points for temperature, flow rate, and other parameters at the inlet and outlet of the heat exchanger. A distributed control system is used to continuously collect and record key operating indicators over a long period. Among these, the heat exchanger outlet temperature is one of the core monitoring quantities reflecting heat exchange efficiency and equipment health. Operators often use the trend of outlet temperature changes to determine the presence of potential faults such as scaling, blockage, and leakage, and accordingly arrange maintenance measures such as cleaning, repair, or load adjustment. Under actual process conditions, changes in the heat exchanger outlet temperature do not occur in isolation; its value is simultaneously affected by multiple factors such as process load, medium properties, and control strategies. In particular, it has a significant physical coupling relationship with the flow rate of the refrigerant or heat medium entering the heat exchanger. Therefore, the field monitoring system often simultaneously collects tube-side outlet temperature data as the measured variable and shell-side refrigerant inlet flow rate data as the disturbance variable, using both as typical multi-source monitoring data to reflect the operating status of the heat exchange unit under complex operating conditions. For industrial process monitoring scenarios like this, existing technologies widely employ the exponentially weighted moving average method to extract trends and smooth time-series signals such as outlet temperature. The exponentially weighted moving average method is typically implemented with fixed parameters within a distributed control system or local control unit. This means that engineers pre-set sample values ​​for the smoothing coefficient, and the current trend value is updated in real-time by weighting and superimposing the observed data of the measured variable at the current moment with the trend value at the previous moment. The advantages of this method are low computational overhead, simplicity of implementation, and the ability to perform online trend assessment on a large number of monitoring points under limited processor resources and storage space. Therefore, it has been widely used in alarm management, quality monitoring, and equipment operation analysis in process industries. However, this trend assessment method based on a fixed smoothing coefficient is designed under the assumption of a univariate stationary environment. Its calculation strategy does not directly consider the physical coupling relationship between the observed data of the measured variable and the observed data of the disturbance variable, and it also lacks the ability to adaptively identify the fluctuation characteristics of the monitoring signal.

[0003] In the actual operation of heat exchangers, to cope with changes in upstream load or downstream process demands, the process control system often needs to frequently or even significantly adjust disturbance variables such as the shell-side refrigerant inlet flow rate. Due to the inherent inertia and lag in the heat exchange process and fluid heat transfer, these flow rate adjustments will cause transient changes in the outlet temperature over a period of time, manifesting as temperature fluctuations of a certain magnitude and duration. From a physical perspective, this type of temperature change is a passive response to control actions and a normal adjustment behavior generated by the system under the premise of meeting the set process parameters. However, the existing exponentially weighted moving average method only considers the numerical changes of the observed data of the measured variable when performing trend updates, without analyzing whether there is synchronous adjustment of the observed data of the disturbance variable behind it. The fixed smoothing coefficient cannot adjust its sensitivity according to the flow rate changes. Once a large flow rate adjustment occurs, the transient process of the outlet temperature will be interpreted by the algorithm as a sharp rise or fall in the trend, which can easily be misjudged by the upper-level system as a rapid deterioration of equipment performance, thereby triggering false alarms or unnecessary operation and maintenance interventions, causing interference to the on-site operation. On the other hand, even when the observed data of the disturbance variable remain relatively stable, the turbulent characteristics of the fluid inside the heat exchanger, measurement noise, and environmental interference can still cause the observed data of the measured variable, outlet temperature, to exhibit high-frequency, irregular, small-amplitude oscillations. Early equipment faults such as scaling and micro-clogging often manifest only as slow, continuous unidirectional temperature drift. Within the framework of a fixed-parameter exponentially weighted moving average, the smoothing coefficient needs to be chosen as a trade-off between "suppressing high-frequency noise" and "preserving weak trends": when the smoothing coefficient is set small, the trend curve has a strong ability to smooth and filter noise, but it also smooths out small long-term shifts, making it difficult to reflect early fault characteristics in the trend; when the smoothing coefficient is set large, the trend curve is more sensitive to small shifts, but it frequently follows random fluctuations during the noise-dominated phase, thus amplifying ineffective jitter and reducing the reliability of the trend assessment results. Existing technologies generally rely on human experience to select a fixed smoothing coefficient that is "usable as a compromise" under specific operating conditions. However, since it is impossible to adaptively adjust for different operating conditions and different fluctuation patterns, the accuracy and stability of trend assessment will decrease significantly when the process conditions change or when noise characteristics are superimposed with fault characteristics. This makes it difficult to meet the needs of complex industrial sites for sophisticated equipment health monitoring and early warning. Summary of the Invention

[0004] In view of this, the present invention aims to propose a dynamic trend evaluation method for multi-source monitoring data to solve the problem that fixed parameter EWMA cannot distinguish between the operating condition adjustment response and the trend of minor faults under complex multivariable operating conditions.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0006] A dynamic trend assessment method for multi-source monitoring data, the method comprising:

[0007] Step S1: Obtain historical data sliding window and basic smoothing coefficient for trend assessment by collecting and preprocessing multi-source monitoring data;

[0008] Step S2: Obtain the decoupling factor of the working condition response by performing weighted correlation analysis on the change sequences of the disturbance variable and the measured variable;

[0009] Step S3: Obtain the trend stability factor by evaluating the geometric shape of the smoothed trajectory of the measured variable;

[0010] Step S4: Obtain the dynamic smoothing coefficient by jointly correcting the basic smoothing coefficient based on operating condition response and trend stability;

[0011] Step S5: Obtain smoothing trend values ​​and heat exchanger sub-health warning signals by performing exponential weighted moving average and rate of change determination on the measured variables.

[0012] Furthermore, the step of acquiring and preprocessing multi-source monitoring data to obtain a historical data sliding window and basic smoothing coefficient for trend assessment includes:

[0013] For the target heat exchanger system, a floating head heat exchanger was selected as the monitoring object. A sampling frequency of 1 Hz was set on the distributed control system at the production site where the floating head heat exchanger was located. Continuous online monitoring and data acquisition were performed on the floating head heat exchanger according to this sampling frequency. The tube-side outlet temperature data, as the measured variable, was used as the observed measured variable, and the shell-side refrigerant inlet flow rate data, as the observed disturbance variable, was used as the observed disturbance variable. Amplitude limiting filtering was applied to both the measured variable and disturbance variable observed data. The time series of the measured variable observed data and the time series of the disturbance variable observed data obtained after amplitude limiting filtering were used as the basic monitoring data for trend assessment. Based on the rated thermal response lag time of the floating head heat exchanger, the lag observation window length was set to a target window length covering the rated thermal response lag time. Combined with the preset lag observation window length, a system was constructed in memory to store the most recent multi-times measured variable and disturbance variable observed data. A historical data sliding window is used to ensure that for any target monitoring time, the historical data sliding window contains multi-source monitoring data segments covering the length of the lag observation window. Example values ​​for the thermal decay time constant are set based on the typical physical and thermal characteristics of the floating head heat exchanger, and example values ​​for the fluctuation characteristic analysis window length are set based on the typical turbulent fluctuation period of industrial fluids at standard flow rates. These thermal decay time constants and fluctuation characteristic analysis window lengths are used as window parameters for subsequent calculations of the operating condition response decoupling factor and trend stability factor. Example values ​​for the basic smoothing coefficient are preset, and the observed data of the measured variable at the initial monitoring time are used as the initial smoothing trend value. The basic smoothing coefficient and the initial smoothing trend value are used together as the basic smoothing coefficient and initial smoothing trend value for subsequent exponentially weighted moving averages of the measured variable, thus completing the acquisition of the historical data sliding window and basic smoothing coefficient for trend assessment.

[0014] Furthermore, the step of obtaining the decoupling factor for the operating condition response by performing weighted correlation analysis on the change sequences of the disturbance variable and the measured variable includes:

[0015] By performing difference and time decay weighted processing on the observation data of the disturbance variable and the measured variable, correlation analysis data of the working condition response pattern is obtained.

[0016] By performing sensitivity amplification and normalization on the correlation analysis data of operating condition response patterns, the decoupling factor of the operating condition response is obtained.

[0017] Furthermore, the process of obtaining correlation analysis data of the working condition response morphology by performing difference and time decay weighted processing on the observed data of the disturbance variable and the measured variable includes:

[0018] For any target monitoring time, the time series of observed data for the perturbation variable and the observed data for the measured variable, covering the length of the lag observation window, are extracted from the historical data sliding window used for trend assessment. The difference between adjacent times in the time series of observed data for the perturbation variable is used as the assessment of the change in the perturbation variable at the corresponding time, and the difference between adjacent times in the time series of observed data for the measured variable is used as the assessment of the change in the measured variable at the corresponding time. A thermal decay time constant and a lag observation window length are set. Based on the set thermal decay time constant and lag observation window length, a time decay weight coefficient is calculated for each difference index within the lag observation window length. The time decay weight coefficient, which monotonically decays with the difference index in the form of a natural exponential function, is normalized and used as the time decay weight corresponding to each difference time. The assessments of the change in the perturbation variable and the assessments of the change in the measured variable at each difference time are then compared. The product of the values ​​is multiplied by the corresponding time decay weight coefficient and summed within the lag observation window to serve as the first weighted inner product assessment between the change in the perturbation variable and the change in the measured variable. The square of the perturbation variable change assessment at each difference time is multiplied by the corresponding time decay weight coefficient and summed within the lag observation window to serve as the weighted energy assessment of the perturbation variable change. The square of the measured variable change assessment at each difference time is summed within the lag observation window to serve as the energy assessment of the measured variable change. The absolute value of the first weighted inner product assessment is used as the numerator, and the product of the square roots of the weighted energy assessment of the perturbation variable change and the energy assessment of the measured variable change is added to the preset minimum positive stability protection term. The resulting fraction is used as the correlation analysis data of the working condition response mode corresponding to the target monitoring time.

[0019] Furthermore, the process of obtaining the decoupling factor of the operating condition response by performing sensitivity amplification and normalization on the correlation analysis data of the operating condition response pattern includes:

[0020] For any target monitoring time, the correlation analysis value of the operating condition response form corresponding to the target monitoring time is extracted from the correlation analysis data of the operating condition response form. A correlation sensitivity gain coefficient is set, and the product of the correlation sensitivity gain coefficient and the correlation analysis value of the operating condition response form is used as the correlation sensitivity amplification assessment corresponding to the target monitoring time. The square of the correlation sensitivity amplification assessment is used as the first normalized correlation strength assessment corresponding to the target monitoring time. When the first normalized correlation strength assessment is greater than a constant 1, the constant 1 is used as the second normalized correlation strength assessment corresponding to the target monitoring time. When the first normalized correlation strength assessment is less than or equal to a constant 1, the first normalized correlation strength assessment is used as the second normalized correlation strength assessment corresponding to the target monitoring time. The difference between the constant 1 and the second normalized correlation strength assessment is used as the operating condition response decoupling factor corresponding to the target monitoring time.

[0021] Furthermore, the step of obtaining the trend stability factor by evaluating the geometric shape of the smoothed trajectory of the measured variable includes:

[0022] By performing sliding window truncation and micro-smoothing on the original temperature time series data of the measured variable, smooth temperature trajectory data within the window of the measured variable is obtained.

[0023] By calculating the net geometric displacement and total path length of the smoothed temperature trajectory data within the measured variable window, trend geometric feature analysis data is obtained.

[0024] The trend stability factor is obtained by performing ratio normalization and squaring on the trend geometric feature analysis data.

[0025] Furthermore, the step of obtaining smoothed temperature trajectory data within the window of the measured variable by performing sliding window truncation and micro-smoothing on the original temperature time series data of the measured variable includes:

[0026] For any target monitoring time, the time series of observed data of the measured variable covering the length of the fluctuation feature analysis window is extracted from the historical data sliding window used for trend assessment. This time series of observed data of the measured variable is used as the original temperature time series data of the measured variable corresponding to the target monitoring time. A micro-smoothing neighborhood length is set. Based on the micro-smoothing neighborhood length, the arithmetic mean of the observed data of the measured variable at each sample point in the original temperature time series data of the measured variable and the observed data of the measured variable at multiple adjacent times in its time neighborhood is calculated. The arithmetic mean corresponding to each sample point is used as the smoothed temperature data corresponding to the target monitoring time. When the sample points in the original temperature time series data of the measured variable are located at both ends of the time series and cannot meet the requirement of a complete micro-smoothing neighborhood length, the arithmetic mean of the observed data of the measured variable at the actual available sample points in the neighborhood is used for calculation, or the smoothed temperature data of the same as the nearest inner sample point is used to replace it to supplement the smoothed temperature data of the boundary sample points. The smoothed temperature time series formed by arranging the smoothed temperature data of each sample point covering the length of the fluctuation feature analysis window in chronological order is used as the smoothed temperature trajectory data within the window of the measured variable corresponding to the target monitoring time.

[0027] Furthermore, the step of obtaining trend geometric feature analysis data by calculating the net geometric displacement and total path length of the smoothed temperature trajectory data within the measured variable window includes:

[0028] For any target monitoring time, a smoothed temperature time series covering the length of the fluctuation feature analysis window is extracted from the smoothed temperature trajectory data within the measured variable window. The difference between the smoothed temperature data of adjacent times in the smoothed temperature time series is used as the smoothed temperature increment assessment for the corresponding time. The smoothed temperature increment assessments for each time within the length of the fluctuation feature analysis window are summed in the time series direction, and the absolute value of the summation is used as the net geometric displacement assessment corresponding to the target monitoring time. The absolute values ​​of the smoothed temperature increment assessments for each time within the length of the fluctuation feature analysis window are summed as the total path length assessment corresponding to the target monitoring time. The combined data of the net geometric displacement assessment and the total path length assessment is used as the trend geometric feature analysis data corresponding to the target monitoring time.

[0029] Furthermore, the process of obtaining the trend stability factor by performing ratio normalization and squaring on the trend geometric feature analysis data includes:

[0030] For any target monitoring time, the net geometric displacement assessment and total path length assessment corresponding to the target monitoring time are extracted from the trend geometric feature analysis data. The absolute value of the net geometric displacement assessment is used as the numerator, and the result of adding the total path length assessment to the preset minimum positive stability protection term is used as the denominator. The ratio of the numerator to the denominator is used as the trend geometric ratio normalization assessment corresponding to the target monitoring time. The square of the trend geometric ratio normalization assessment is used as the trend stability factor corresponding to the target monitoring time.

[0031] Furthermore, the method of obtaining the dynamic smoothing coefficient by jointly correcting the basic smoothing coefficient based on operating condition response and trend stability includes:

[0032] For any target monitoring time, a preset example value of the basic smoothing coefficient is extracted from the basic smoothing coefficient used for trend assessment. The decoupling factor of the working condition response corresponding to the target monitoring time is multiplied by the trend stability factor to obtain the joint correction weight assessment of the working condition and trend corresponding to the target monitoring time. The product of the basic smoothing coefficient and the joint correction weight assessment of the working condition and trend is obtained as the dynamic smoothing coefficient corresponding to the target monitoring time. The dynamic smoothing coefficient corresponding to each target monitoring time is used as the dynamic smoothing coefficient when performing an exponentially weighted moving average on the observed data of the measured variable.

[0033] Compared with the prior art, the present invention has the following advantages:

[0034] This invention presents a dynamic trend assessment method for multi-source monitoring data. By simultaneously introducing two types of monitoring data—outlet temperature and flow rate—it uses a condition response decoupling factor to quantitatively characterize the physical coupling relationship between the disturbance variable and the measured variable. This method can automatically identify "passive temperature changes caused by flow rate regulation" and adaptively suppress trend update weights when temperature fluctuations can be explained by flow rate changes, thereby effectively shielding transitional temperature fluctuations during process load adjustments. Compared to traditional trend algorithms that rely solely on single-variable temperature changes, this invention significantly reduces false fault trend identification caused by normal operating condition adjustments, decreases false alarms, avoids unnecessary shutdowns for inspection and cleaning, and better reflects the real-world operating environment of continuous industrial production with frequent condition adjustments and complex control linkages. Meanwhile, this invention constructs a trend stability factor to perform geometric morphology analysis on the smoothed trajectory of the outlet temperature time series within a sliding window. The ratio of net geometric displacement to total path length is used as a key indicator to distinguish between "high-frequency noise reciprocating oscillations" and "small unidirectional drift trends." This indicator is embedded nonlinearly into the smoothing coefficient of an exponentially weighted moving average, thus maintaining strong noise suppression capabilities while remaining sufficiently sensitive to slow temperature shifts caused by early faults such as scaling and blockage. Based on this dynamic smoothing mechanism, the trend curve obtained by this invention more closely approximates the actual physical health evolution process of the heat exchanger. It can issue early warnings of sub-health conditions before obvious process anomalies and safety risks occur, providing a basis for on-site maintenance personnel to arrange planned shutdowns and preventative cleaning, achieving high reliability and practicality in heat exchanger condition monitoring under complex operating conditions. Attached Figure Description

[0035] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0036] Figure 1 This is a flowchart of a dynamic trend assessment method for multi-source monitoring data according to an embodiment of the present invention. Detailed Implementation

[0037] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0038] See Figure 1 This is a flowchart of a dynamic trend assessment method for multi-source monitoring data provided in Embodiment 1 of the present invention. Figure 1 As shown, a dynamic trend assessment method for multi-source monitoring data may include:

[0039] Step S1 involves collecting and preprocessing multi-source monitoring data to obtain a historical data sliding window and basic smoothing coefficient for trend assessment.

[0040] For the target heat exchanger system, a floating head heat exchanger is selected as the monitoring object. On the distributed control system at the production site where the floating head heat exchanger is located, a sampling frequency of 1 Hz is set. Continuous online monitoring and data acquisition are performed on the floating head heat exchanger according to the sampling frequency. The tube-side outlet temperature data, as the measured variable, is used as the observed measured variable, and the shell-side refrigerant inlet flow rate data, as the observed disturbance variable, is used as the observed disturbance variable. Amplitude limiting filtering is applied to both the measured variable and disturbance variable observed data. The time series of the measured variable and disturbance variable observed data obtained after amplitude limiting filtering are used as the basic monitoring data for trend assessment. Based on the rated thermal response lag time of the floating head heat exchanger, the lag observation window length is set to the target window length covering the rated thermal response lag time. In this embodiment, the lag observation window length is set to 10. Combined with the preset lag observation window length, a historical data sliding window is constructed in memory to store the most recent multi-times measured variable and disturbance variable observed data, so that for any given... At a target monitoring time, the historical data sliding window contains multi-source monitoring data segments covering the length of the lag observation window. An example value for the thermal decay time constant is set based on the typical physical and thermal characteristics of the floating head heat exchanger, and an example value for the fluctuation characteristic analysis window length is set based on the typical turbulent fluctuation period of industrial fluids at standard flow rates. In this embodiment, the fluctuation characteristic analysis window length is set to 30, and the thermal decay time constant is set to 3.3. The thermal decay time constant and the fluctuation characteristic analysis window length are used as window parameters for subsequent calculations of the operating condition response decoupling factor and trend stability factor. An example value for the preset basic smoothing coefficient is set to 0.15 in this embodiment. The observed data of the measured variable at the initial monitoring time is used as the initial smoothing trend value. The basic smoothing coefficient and the initial smoothing trend value are used together as the basic smoothing coefficient and the initial smoothing trend value for subsequent exponentially weighted moving averages of the measured variable, thus completing the acquisition of the historical data sliding window and the basic smoothing coefficient for trend assessment.

[0041] This completes the process of acquiring historical data sliding windows and basic smoothing coefficients for trend assessment by collecting and preprocessing multi-source monitoring data.

[0042] Step S2: Obtain the decoupling factor of the working condition response by performing weighted correlation analysis on the change sequences of the disturbance variable and the measured variable.

[0043] To address the problem of existing technologies in multi-source monitoring data processing—namely, the inability to distinguish between passive responses caused by external control adjustments and spontaneous drift caused by internal equipment faults—this invention addresses the issue of the inability to distinguish between passive responses caused by external control adjustments and spontaneous drift caused by internal equipment faults in heat exchanger operation. In actual operation of heat exchangers, there is a significant physical coupling between the disturbance variable (flow rate) and the measured variable (temperature). When the control system significantly adjusts the flow rate to maintain production targets, the outlet temperature will inevitably change according to thermodynamic principles. Although this change manifests as significant fluctuations in numerical value, it is essentially a normal physical response of the system to external control actions, rather than a sign of deterioration in equipment health. However, traditional trend assessment algorithms only focus on the numerical changes in temperature data, ignoring the driving factors behind these changes, thus easily leading to false alarms. To solve this problem, this invention utilizes historical data within a sliding time window to analyze the correlation between flow rate and temperature change trends in real time (the system uses a sliding time window mechanism to extract continuously collected single-point scalar data into short time-series segments containing historical information, thereby transforming one-dimensional scalar values ​​into vector forms in a high-dimensional feature space). If the changes in both flow rate and temperature show a high degree of consistency over time (e.g., the temperature decreases as expected according to physical laws while the flow rate decreases), it indicates that the current temperature fluctuation is passively controlled. In this case, the weight of trend updates should be suppressed to mask operational interference. Conversely, if the flow rate remains stable, or the flow rate change pattern is completely unrelated to the temperature change, it indicates that the current temperature fluctuation originates from spontaneous evolution within the system (such as scaling). In this case, the weight of trend updates should be retained or even increased to capture real fault signals.

[0044] In summary, this invention first obtains correlation analysis data on the working condition response pattern by performing difference and time decay weighted processing on the observation data of the disturbance variable and the measured variable. Specifically, for any target monitoring time, the time series of the disturbance variable observation data and the time series of the measured variable observation data covering the lag observation window length are extracted from the historical data sliding window used for trend assessment. The difference between adjacent times in the disturbance variable observation data time series is used as the evaluation of the change in the disturbance variable at the corresponding time, and the difference between adjacent times in the measured variable observation data time series is used as the evaluation of the change in the measured variable at the corresponding time. A thermal decay time constant and a lag observation window length are set. Based on the set thermal decay time constant and lag observation window length, a time decay weight coefficient is calculated for each difference index within the lag observation window length. The time decay weight coefficient, which monotonically decays with the difference index in the form of a natural exponential function, is normalized and used as the time decay corresponding to each difference time. The weighting is reduced; the product of the estimated change in the disturbance variable and the estimated change in the measured variable at each difference time is multiplied by the corresponding time decay weight coefficient and summed within the lag observation window length to obtain the first weighted inner product assessment between the changes in the disturbance variable and the measured variable; the square of the estimated change in the disturbance variable at each difference time is multiplied by the corresponding time decay weight coefficient and summed within the lag observation window length to obtain the weighted energy assessment of the change in the disturbance variable; the square of the estimated change in the measured variable at each difference time is summed within the lag observation window length to obtain the energy assessment of the change in the measured variable; the absolute value of the first weighted inner product assessment is used as the numerator, and the product of the square roots of the weighted energy assessment of the change in the disturbance variable and the energy assessment of the change in the measured variable is added to the preset minimum positive stability protection term as the denominator. The resulting fraction is used as the correlation analysis data of the working condition response mode corresponding to the target monitoring time.

[0045] After obtaining the correlation analysis data of the operating condition response pattern, the data is further processed by sensitivity amplification and normalization to obtain the operating condition response decoupling factor. Specifically, for any target monitoring time, the operating condition response pattern correlation analysis value corresponding to the target monitoring time is extracted from the operating condition response pattern correlation analysis data. A correlation sensitivity gain coefficient is set, and the product of the correlation sensitivity gain coefficient and the operating condition response pattern correlation analysis value is used as the correlation sensitivity amplification assessment corresponding to the target monitoring time. The square of the correlation sensitivity amplification assessment is used as the first normalized correlation strength assessment corresponding to the target monitoring time. When the first normalized correlation strength assessment is greater than a constant 1, the constant 1 is used as the second normalized correlation strength assessment corresponding to the target monitoring time. When the first normalized correlation strength assessment is less than or equal to a constant 1, the first normalized correlation strength assessment is used as the second normalized correlation strength assessment corresponding to the target monitoring time. The difference between the constant 1 and the second normalized correlation strength assessment is used as the operating condition response decoupling factor corresponding to the target monitoring time.

[0046] In one embodiment, it is assumed that the correlation sensitivity gain coefficient is The length of the lag observation window is ;No. The observed values ​​of the disturbance variable at each time point are ;No. The observed values ​​of the disturbance variable at each time point are ;No. The observed value of the measured variable at each time point is ;No. The observed value of the measured variable at each time point is ;No. The time decay weight at each moment is: Then the first The expression for calculating the decoupling factor of the operating condition response at time t is:

[0047]

[0048] in, Indicates the first Decoupling factor of operating condition response at each moment; The correlation sensitivity gain coefficient is set to 10 in this embodiment of the invention. Since the thermal hysteresis in the actual physical process will cause the temperature response waveform to be stretched and deformed relative to the flow regulation waveform, the directly calculated mathematical correlation coefficient is too low. This coefficient is used to amplify this weak correlation feature and ensure that once a morphological causal relationship is detected, the operating condition response decoupling factor can quickly approach 0. Indicates the length of the lag observation window; Indicates the first The observed values ​​of the disturbance variable at each time point; Indicates the first The observed values ​​of the disturbance variable at each time point; Indicates the first The observed values ​​of the measured variable at each time point; Indicates the first The observed values ​​of the measured variable at each time point; Indicates the first The time decay weight at each moment.

[0049] It should be noted that, firstly, the formula introduces a time decay weight. Because the temperature response of a heat exchanger has a hysteresis, recent flow rate changes have the greatest impact on the current temperature, while the impact gradually weakens in the long term. By leveraging the advantages of this invention, the method can more accurately align the relationship between flow rate and temperature over time. Based on this, the formula calculates the weighted morphological correlation between the flow rate change sequence and the temperature change sequence. Considering that physical lag can cause waveform slowing, leading to a smaller calculated inner product value, this invention introduces a sensitivity gain coefficient. The correlation is amplified. When a significant adjustment in flow rate directly leads to a corresponding change in temperature, the two exhibit synchronicity in morphology. The magnified values ​​of the relevant terms will quickly approach or reach [the desired value]. Because the formula uses The structure at this time It will drop rapidly and approach [the point]. This corresponds to the suppression requirement: when the algorithm detects that a temperature change is explained by a flow rate change, it uses minimal... By forcibly lowering the smoothing coefficient, the monitoring system temporarily ignores this passive fluctuation, successfully avoiding false alarms caused by operating condition adjustments. Conversely, when the system is in a steady-state spontaneous drift state, flow data typically remains relatively stable or exhibits only minor random fluctuations, while temperature data may show slow unidirectional shifts due to equipment scaling. In this situation, changes in flow and temperature lose synchronization, exhibiting extremely low correlation in statistical analysis, even after... Even when magnified, its value remains extremely small. Furthermore, the overall score approaches... , making Maintain at Nearby. This means the algorithm determines that the current temperature change cannot be explained by the flow rate change, and is an abnormal trend that should be monitored. At this point, Allowing the base smoothing coefficient to function at its full potential ensures the monitoring system's ability to keenly detect minor fault trends.

[0050] Thus, the decoupling factor for the operating condition response was obtained by performing weighted correlation analysis on the change sequences of the disturbance variable and the measured variable.

[0051] Step S3: Obtain the trend stability factor by evaluating the geometric shape of the smoothed trajectory of the measured variable.

[0052] After successfully eliminating passive response interference caused by external control regulation through the operating condition response decoupling factor in step S2, the data environment faced by the monitoring system has been initially purified into steady-state operating data. However, a problem remains: after removing the influence of large external fluctuations, the remaining small fluctuation components are still mixed with high-frequency random noise caused by fluid turbulence. This is extremely difficult to distinguish from the small unidirectional trends caused by early equipment failures (such as slight increases in thermal resistance due to scaling). In industrial settings, even after eliminating the interference of flow regulation, the turbulent characteristics of the fluid inside the heat exchanger will still cause unavoidable random oscillations in the outlet temperature reading. Although these oscillations have limited amplitude, their frequency is extremely high, easily masking the slow temperature drift signals caused by early scaling or blockage. At this point, if the algorithm lacks further identification capabilities, the non-operating condition fluctuations retained in step S2 are easily submerged by noise. Therefore, it is necessary to further introduce an identification strategy based on trajectory geometry on top of step S2. The system extracts the temperature data sequence through a sliding time window mechanism, first performing microscopic smoothing preprocessing on the data within the window to extract the skeleton shape of the signal and filter out high-frequency noise interference. Subsequently, geometric feature analysis is performed on the preprocessed skeleton trajectory: the net geometric displacement of the trajectory within the window and the actual total path length are calculated. A typical characteristic of turbulent noise is oscillation, with data points repeatedly jumping up and down around the mean, resulting in a long total path but a very small net displacement. Conversely, a typical characteristic of fault trends is unidirectional drift, with data points exhibiting a continuous rise or fall, and their net displacement and total path length being highly similar in value. Based on this physical difference, this invention constructs a geometric ratio factor; when the signal shape approaches a straight line (trend), the factor tends to... This ensures that subtle trends are not drowned out by noise; when the signal shape approaches a clumpy (noise) pattern, the factor tends to... It serves as an adaptive filter.

[0053] In summary, this invention first obtains smoothed temperature trajectory data within the measured variable window by performing sliding window truncation and micro-smoothing on the original temperature time series data of the measured variable. Specifically, for any target monitoring time, the measured variable observation data time series covering the fluctuation feature analysis window length is extracted from the historical data sliding window used for trend assessment, and this measured variable observation data time series is used as the original temperature time series data of the measured variable corresponding to the target monitoring time. A micro-smoothing neighborhood length is set, and based on this micro-smoothing neighborhood length, the measured variable observation data of each sample point in the original temperature time series data of the measured variable is compared with the measured variable observation data of multiple adjacent times within its time neighborhood. The arithmetic mean of the measured observation data is calculated, and the arithmetic mean of each sample point is used as the smoothed temperature data corresponding to the target monitoring time. When the sample points in the original temperature time series data of the measured variable are located at both ends of the time series and cannot meet the requirement of a complete micro-smoothing neighborhood length, the arithmetic mean of the measured variable observation data of the actual available sample points in the neighborhood is used, or the smoothed temperature data of the same as the nearest inner sample point is used to replace it to supplement the smoothed temperature data of the boundary sample points. The smoothed temperature time series formed by arranging the smoothed temperature data of each sample point covering the length of the fluctuation feature analysis window in chronological order is used as the smoothed temperature trajectory data within the window of the measured variable corresponding to the target monitoring time.

[0054] After obtaining the smoothed temperature trajectory data within the measured variable window, the net geometric displacement and total path length are calculated on the smoothed temperature trajectory data within the measured variable window to obtain trend geometric feature analysis data. Specifically, for any target monitoring time, a smoothed temperature time series covering the length of the fluctuation feature analysis window is extracted from the smoothed temperature trajectory data within the measured variable window. The difference between the smoothed temperature data of adjacent times in the smoothed temperature time series is used as the smoothed temperature increment assessment for the corresponding time. The smoothed temperature increment assessments for each time within the length of the fluctuation feature analysis window are summed in the time series direction, and the absolute value of the summation is used as the net geometric displacement assessment corresponding to the target monitoring time. The absolute values ​​of the smoothed temperature increment assessments for each time within the length of the fluctuation feature analysis window are summed as the total path length assessment corresponding to the target monitoring time. The combined data of the net geometric displacement assessment and the total path length assessment is used as the trend geometric feature analysis data corresponding to the target monitoring time.

[0055] After obtaining the trend geometric feature analysis data, the trend stability factor is obtained by performing ratio normalization and squaring on the trend geometric feature analysis data. Specifically, for any target monitoring time, the net geometric displacement assessment and total path length assessment corresponding to the target monitoring time are extracted from the trend geometric feature analysis data. The absolute value of the net geometric displacement assessment is used as the numerator, and the result of adding the total path length assessment to the preset minimum positive stability protection term is used as the denominator. The ratio of the numerator to the denominator is used as the trend geometric ratio normalization assessment corresponding to the target monitoring time. The square of the trend geometric ratio normalization assessment is used as the trend stability factor corresponding to the target monitoring time.

[0056] In one implementation, assume the first The measured variable value after smoothing preprocessing at each time point is , No. The measured variable value after smoothing at each time point is The fluctuation characteristic analysis window length is Then the first The expression for calculating the trend stability factor at time point is:

[0057]

[0058] in, Indicates the first Trend stability factor at each moment; Indicates the first The measured variable values ​​after smoothing and preprocessing at each time point; Indicates the first The smoothed value of the measured variable at each time point; This indicates the length of the fluctuation characteristic analysis window.

[0059] It should be noted that this formula constructs a geometric ratio function reflecting the monotonicity of the signal trajectory. By comparing the proportion of net displacement in the total path, it achieves accurate identification of the signal pattern. When the monitoring data is dominated by turbulent noise inside the heat exchanger, the temperature value fluctuates randomly around the mean. Even after microscopic smoothing, the trajectory still exhibits a curved and zigzag pattern. This zigzag motion causes the cumulative total path length (denominator) to be much larger than the final net displacement (numerator), resulting in an extremely small ratio within the brackets. After squaring, the trend stability factor statistically remains at an extremely low level. This corresponds to the filtering requirement: the algorithm automatically identifies the current fluctuation as invalid non-monotonic noise and suppresses the smoothing coefficient through the extremely small trend stability factor value, which is equivalent to constructing an adaptive, powerful low-pass filter, effectively filtering out high-frequency interference and ensuring the stability of the trend line. Conversely, when early fouling of the equipment leads to an increase in thermal resistance, the temperature data exhibits a slow, continuous unidirectional drift. Under this condition, the trajectory macroscopically appears as a smooth straight line or curve. Although the addition of random noise causes local reversals, preventing the geometric ratio from reaching 1, the net displacement of the signal increases significantly on the statistical scale of the sliding window. Therefore, the numerical distribution of the trend stability factor undergoes a significant hierarchical shift relative to the noise region, exhibiting fluctuations around the medium-to-high mean. The algorithm determines that although the current fluctuation is weak and noisy, it possesses clear physical trend characteristics, thus restoring the smoothing coefficient to the effective range. This half-open weighting mechanism ensures that the monitoring system can sensitively accumulate this small change for accurate early warning, while retaining a certain filtering capability to prevent excessive jitter in the trend line, thus solving the dilemma of traditional algorithms in extracting weak fault signals.

[0060] This completes the process of obtaining the trend stability factor by evaluating the geometric shape of the smoothed trajectory of the measured variable.

[0061] Step S4: Obtain the dynamic smoothing coefficient by jointly correcting the basic smoothing coefficient based on operating condition response and trend stability.

[0062] After obtaining the operating condition response decoupling factor and the trend stability factor, for any target monitoring time, a preset example value of the basic smoothing coefficient is extracted from the basic smoothing coefficient used for trend assessment. The operating condition response decoupling factor and the trend stability factor corresponding to the target monitoring time are multiplied together to obtain the joint correction weight assessment of the operating condition and trend corresponding to the target monitoring time. The product of the basic smoothing coefficient and the joint correction weight assessment of the operating condition and trend is used as the dynamic smoothing coefficient corresponding to the target monitoring time. The dynamic smoothing coefficient corresponding to each target monitoring time is used as the dynamic smoothing coefficient when performing an exponentially weighted moving average on the observed data of the measured variable.

[0063] Thus, the process of obtaining dynamic smoothing coefficients by jointly correcting the basic smoothing coefficients based on operating condition response and trend stability has been completed.

[0064] Step S5: Obtain smoothing trend values ​​and heat exchanger sub-health warning signals by performing exponential weighted moving average and rate of change determination on the measured variables.

[0065] After calculating the decoupling factor and trend stability factor for the operating condition response and obtaining the dynamic smoothing coefficients corresponding to each target monitoring time, this step performs an exponentially weighted moving average operation based on the observed data of the measured variable and the dynamic smoothing coefficients to construct a smoothed trend value sequence reflecting the evolution of the actual physical state of the heat exchanger. The rate of change is then determined on the smoothed trend value sequence to output a sub-health warning signal for the heat exchanger. Specifically, for any target monitoring time, the system extracts the observed data of the measured variable corresponding to the target monitoring time from the historical data sliding window used for trend assessment. The smoothed trend value calculated at the previous monitoring time is used as the smoothed trend value at the previous time. The dynamic smoothing coefficient corresponding to the target monitoring time is used as the current smoothing weight. Following the iterative form of the exponentially weighted moving average, the product of the dynamic smoothing coefficient and the observed data of the measured variable, and the product of the result after subtracting the dynamic smoothing coefficient and the smoothed trend value at the previous time are weighted and superimposed to obtain the smoothed trend value corresponding to the target monitoring time. The smoothed trend values ​​of each target monitoring time are then arranged in chronological order to form a smoothed trend value time series. Subsequently, the system sets the change rate judgment window length for trend evolution analysis on the smoothed trend value time series. Within the change rate judgment window length, the monotonicity and change rate of the smoothed trend values ​​are evaluated: for any target judgment time, a smoothed trend value segment covering the change rate judgment window length is extracted from the smoothed trend value time series. By differencing the smoothed trend values ​​at the start and end times of the change rate judgment window and dividing by the corresponding time interval, the trend change rate assessment corresponding to the target judgment time is obtained. A benchmark change rate threshold is set based on the aging rate of the heat exchanger under normal operating conditions and the allowable slow temperature drift range. In this embodiment of the invention, the benchmark change rate threshold is set as follows: The system compares the assessed trend change rate with a benchmark change rate threshold. When the assessed trend change rate is consistently greater than the benchmark change rate threshold in the positive direction and meets the condition of unidirectional increase within multiple consecutive change rate judgment windows, it determines that there is a continuously increasing irreversible temperature drift trend in the current smoothed trend value time series, indicating that there are potential sub-health risks such as scaling or blockage inside the floating head heat exchanger. When the system determines that there is a sub-health risk, it generates a heat exchanger sub-health warning signal and outputs it along with the current smoothed trend value to the upper-level monitoring system or operation and maintenance management system. This prompts on-site operators to arrange preventative cleaning and maintenance work in a timely manner, achieving early warning of heat exchanger sub-health based on the joint judgment of smoothed trend value and change rate.

[0066] Thus, the process of obtaining smoothed trend values ​​and heat exchanger sub-health warning signals by performing exponentially weighted moving averages and rate of change determination on the measured variables has been completed.

[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic trend assessment method for multi-source monitoring data, characterized in that, The method includes: Step S1: Obtain historical data sliding window and basic smoothing coefficient for trend assessment by collecting and preprocessing multi-source monitoring data; Step S2: Obtain the decoupling factor of the working condition response by performing weighted correlation analysis on the change sequences of the disturbance variable and the measured variable; Step S3: Obtain the trend stability factor by evaluating the geometric shape of the smoothed trajectory of the measured variable; Step S4: Obtain the dynamic smoothing coefficient by jointly correcting the basic smoothing coefficient based on operating condition response and trend stability; Step S5: Obtain smoothed trend values ​​and heat exchanger sub-health warning signals by performing exponential weighted moving average and rate of change determination on the measured variables.

2. The dynamic trend assessment method for multi-source monitoring data according to claim 1, characterized in that, The process of acquiring and preprocessing multi-source monitoring data to obtain historical data sliding windows and basic smoothing coefficients for trend assessment includes: For the target heat exchanger system, a floating head heat exchanger was selected as the monitoring object. A sampling frequency of 1 Hz was set on the distributed control system at the production site where the floating head heat exchanger was located. Continuous online monitoring and data acquisition were performed on the floating head heat exchanger according to this sampling frequency. The tube-side outlet temperature data, as the measured variable, was used as the observed measured variable, and the shell-side refrigerant inlet flow rate data, as the observed disturbance variable, was used as the observed disturbance variable. Amplitude limiting filtering was applied to both the measured variable and disturbance variable observed data. The time series of the measured variable observed data and the time series of the disturbance variable observed data obtained after amplitude limiting filtering were used as the basic monitoring data for trend assessment. Based on the rated thermal response lag time of the floating head heat exchanger, the lag observation window length was set to a target window length covering the rated thermal response lag time. Combined with the preset lag observation window length, a system was constructed in memory to store the most recent multi-times measured variable and disturbance variable observed data. A historical data sliding window is used to ensure that for any target monitoring time, the historical data sliding window contains multi-source monitoring data segments covering the length of the lag observation window. Example values ​​for the thermal decay time constant are set based on the typical physical and thermal characteristics of the floating head heat exchanger, and example values ​​for the fluctuation characteristic analysis window length are set based on the typical turbulent fluctuation period of industrial fluids at standard flow rates. These thermal decay time constants and fluctuation characteristic analysis window lengths are used as window parameters for subsequent calculations of the operating condition response decoupling factor and trend stability factor. Example values ​​for the basic smoothing coefficient are preset, and the observed data of the measured variable at the initial monitoring time are used as the initial smoothing trend value. The basic smoothing coefficient and the initial smoothing trend value are used together as the basic smoothing coefficient and initial smoothing trend value for subsequent exponentially weighted moving averages of the measured variable, thus completing the acquisition of the historical data sliding window and basic smoothing coefficient for trend assessment.

3. The dynamic trend assessment method for multi-source monitoring data according to claim 1, characterized in that, The method of obtaining the decoupling factor for the operating condition response by performing weighted correlation analysis on the change sequences of the disturbance variable and the measured variable includes: By performing difference and time decay weighted processing on the observation data of the disturbance variable and the measured variable, correlation analysis data of the working condition response pattern is obtained. By performing sensitivity amplification and normalization on the correlation analysis data of operating condition response patterns, the decoupling factor of the operating condition response is obtained.

4. The dynamic trend assessment method for multi-source monitoring data according to claim 3, characterized in that, The process involves performing difference and time-decrease weighted processing on the observed data of the disturbance variable and the measured variable to obtain correlation analysis data of the working condition response pattern, including: For any target monitoring time, the time series of observed data for the perturbation variable and the observed data for the measured variable, covering the length of the lag observation window, are extracted from the historical data sliding window used for trend assessment. The difference between adjacent times in the time series of observed data for the perturbation variable is used as the assessment of the change in the perturbation variable at the corresponding time, and the difference between adjacent times in the time series of observed data for the measured variable is used as the assessment of the change in the measured variable at the corresponding time. A thermal decay time constant and a lag observation window length are set. Based on the set thermal decay time constant and lag observation window length, a time decay weight coefficient is calculated for each difference index within the lag observation window length. The time decay weight coefficient, which monotonically decays with the difference index in the form of a natural exponential function, is normalized and used as the time decay weight corresponding to each difference time. The assessments of the change in the perturbation variable and the assessments of the change in the measured variable at each difference time are then compared. The product of the values ​​is multiplied by the corresponding time decay weight coefficient and summed within the lag observation window to serve as the first weighted inner product assessment between the change in the perturbation variable and the change in the measured variable. The square of the perturbation variable change assessment at each difference time is multiplied by the corresponding time decay weight coefficient and summed within the lag observation window to serve as the weighted energy assessment of the perturbation variable change. The square of the measured variable change assessment at each difference time is summed within the lag observation window to serve as the energy assessment of the measured variable change. The absolute value of the first weighted inner product assessment is used as the numerator, and the product of the square roots of the weighted energy assessment of the perturbation variable change and the energy assessment of the measured variable change is added to the preset minimum positive stability protection term. The resulting fraction is used as the correlation analysis data of the working condition response mode corresponding to the target monitoring time.

5. The dynamic trend assessment method for multi-source monitoring data according to claim 3, characterized in that, The process of obtaining the decoupling factor of the operating condition response by performing sensitivity amplification and normalization on the correlation analysis data of the operating condition response pattern includes: For any target monitoring time, the correlation analysis value of the operating condition response form corresponding to the target monitoring time is extracted from the correlation analysis data of the operating condition response form. A correlation sensitivity gain coefficient is set, and the product of the correlation sensitivity gain coefficient and the correlation analysis value of the operating condition response form is used as the correlation sensitivity amplification assessment corresponding to the target monitoring time. The square of the correlation sensitivity amplification assessment is used as the first normalized correlation strength assessment corresponding to the target monitoring time. When the first normalized correlation strength assessment is greater than a constant 1, the constant 1 is used as the second normalized correlation strength assessment corresponding to the target monitoring time. When the first normalized correlation strength assessment is less than or equal to a constant 1, the first normalized correlation strength assessment is used as the second normalized correlation strength assessment corresponding to the target monitoring time. The difference between the constant 1 and the second normalized correlation strength assessment is used as the operating condition response decoupling factor corresponding to the target monitoring time.

6. The dynamic trend assessment method for multi-source monitoring data according to claim 1, characterized in that, The process of obtaining the trend stability factor by evaluating the geometric shape of the smoothed trajectory of the measured variable includes: By performing sliding window truncation and micro-smoothing on the original temperature time series data of the measured variable, smooth temperature trajectory data within the window of the measured variable is obtained. By calculating the net geometric displacement and total path length of the smoothed temperature trajectory data within the measured variable window, trend geometric feature analysis data is obtained. The trend stability factor is obtained by performing ratio normalization and squaring on the trend geometric feature analysis data.

7. The dynamic trend assessment method for multi-source monitoring data according to claim 6, characterized in that, The process involves performing sliding window truncation and micro-smoothing on the original temperature time series data of the measured variable to obtain smoothed temperature trajectory data within the measured variable window, including: For any target monitoring time, the time series of observed data of the measured variable covering the length of the fluctuation feature analysis window is extracted from the historical data sliding window used for trend assessment. This time series of observed data of the measured variable is used as the original temperature time series data of the measured variable corresponding to the target monitoring time. A micro-smoothing neighborhood length is set. Based on the micro-smoothing neighborhood length, the arithmetic mean of the observed data of the measured variable at each sample point in the original temperature time series data of the measured variable and the observed data of the measured variable at multiple adjacent times in its time neighborhood is calculated. The arithmetic mean corresponding to each sample point is used as the smoothed temperature data corresponding to the target monitoring time. When the sample points in the original temperature time series data of the measured variable are located at both ends of the time series and cannot meet the requirement of a complete micro-smoothing neighborhood length, the arithmetic mean of the observed data of the measured variable at the actual available sample points in the neighborhood is used for calculation, or the smoothed temperature data of the same as the nearest inner sample point is used to replace it to supplement the smoothed temperature data of the boundary sample points. The smoothed temperature time series formed by arranging the smoothed temperature data of each sample point covering the length of the fluctuation feature analysis window in chronological order is used as the smoothed temperature trajectory data within the window of the measured variable corresponding to the target monitoring time.

8. The dynamic trend assessment method for multi-source monitoring data according to claim 6, characterized in that, The process involves calculating the net geometric displacement and total path length of the smoothed temperature trajectory data within the measured variable window to obtain trend geometric feature analysis data, including: For any target monitoring time, a smoothed temperature time series covering the length of the fluctuation feature analysis window is extracted from the smoothed temperature trajectory data within the measured variable window. The difference between the smoothed temperature data of adjacent times in the smoothed temperature time series is used as the smoothed temperature increment assessment for the corresponding time. The smoothed temperature increment assessments for each time within the length of the fluctuation feature analysis window are summed in the time series direction, and the absolute value of the summation is used as the net geometric displacement assessment corresponding to the target monitoring time. The absolute values ​​of the smoothed temperature increment assessments for each time within the length of the fluctuation feature analysis window are summed as the total path length assessment corresponding to the target monitoring time. The combined data of the net geometric displacement assessment and the total path length assessment is used as the trend geometric feature analysis data corresponding to the target monitoring time.

9. The dynamic trend assessment method for multi-source monitoring data according to claim 6, characterized in that, The process of obtaining the trend stability factor by performing ratio normalization and squaring on the trend geometric feature analysis data includes: For any target monitoring time, the net geometric displacement assessment and total path length assessment corresponding to the target monitoring time are extracted from the trend geometric feature analysis data. The absolute value of the net geometric displacement assessment is used as the numerator, and the result of adding the total path length assessment to the preset minimum positive stability protection term is used as the denominator. The ratio of the numerator to the denominator is used as the trend geometric ratio normalization assessment corresponding to the target monitoring time. The square of the trend geometric ratio normalization assessment is used as the trend stability factor corresponding to the target monitoring time.

10. The dynamic trend assessment method for multi-source monitoring data according to claim 1, characterized in that, The process of obtaining a dynamic smoothing coefficient by jointly correcting the basic smoothing coefficient based on operating condition response and trend stability includes: For any target monitoring time, a preset example value of the basic smoothing coefficient is extracted from the basic smoothing coefficient used for trend assessment. The decoupling factor of the working condition response corresponding to the target monitoring time is multiplied by the trend stability factor to obtain the joint correction weight assessment of the working condition and trend corresponding to the target monitoring time. The product of the basic smoothing coefficient and the joint correction weight assessment of the working condition and trend is obtained as the dynamic smoothing coefficient corresponding to the target monitoring time. The dynamic smoothing coefficient corresponding to each target monitoring time is used as the dynamic smoothing coefficient when performing an exponentially weighted moving average on the observed data of the measured variable.

Images