Big data real-time anomaly detection method for industrial mechanism model

Abstract

The application discloses a big data real-time anomaly detection method for an industrial mechanism model, relates to the technical field of industrial big data analysis and intelligent anomaly detection, and comprises the following steps: acquiring multi-dimensional change numerical values under continuous time scales in the running process of the industrial mechanism model, recording change amplitude difference values and change rate numerical values corresponding to each time position point by point, and calibrating decay gradient numerical values of abnormal signals in the time advancing process.The application realizes the continuous existence of abnormal signals in time series by constructing an abnormality reservation mechanism driven by the decay gradient, reversely amplifying and backtracking the abnormal signals, simultaneously combining abnormality highlighting section division, forward traction and reverse rhythm suppression processing, suppressing the coverage of normal fluctuations on the abnormality, realizing the continuous expression and traceable presentation of abnormal characteristics, and thus improving the stability and accuracy of abnormality identification.

Description

Technical Field

[0001] This invention relates to the field of industrial big data analysis and intelligent anomaly detection technology, specifically to a real-time anomaly detection method for big data based on industrial mechanism models. Background Technology

[0002] Real-time anomaly detection based on industrial mechanism models refers to combining the understanding of mechanisms reflecting equipment operation patterns and physical processes with continuously collected massive amounts of data during the operation of industrial systems. This involves online analysis and status assessment of the data stream: constraining and characterizing normal operating behavior using mechanistic relationships, while continuously comparing and dynamically evaluating real-time incoming data. Once data deviates from established mechanistic relationships or evolutionary trends, it is determined to be an abnormal state. This approach differs from detection methods that solely rely on historical statistical patterns, emphasizing the synergistic expression of physical process logic, operational constraints, and data change characteristics. This enables rapid identification and stable judgment of abnormal behavior even under complex operating conditions and highly volatile environments, making it suitable for scenarios such as industrial equipment monitoring, process control, and operational risk early warning.

[0003] The existing technology has the following shortcomings: In the real-time anomaly detection process of big data for industrial mechanism models, when anomaly signals are embedded in continuous data streams in the form of short-term weak amplitude, subsequent normal fluctuations will numerically cover and trend-guide the anomaly signals, causing the original anomaly characteristics to gradually weaken and smooth out over time. At the same time, when the overall data performance stabilizes, the mechanism matching process will determine the current state as conforming to existing operating rules, thus no longer responding to early anomalies. Ultimately, key anomaly characteristics are completely submerged in the continuous evolution, forming a traceless failure phenomenon that is difficult to trace, and misleading the subsequent judgment of operating status.

[0004] 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

[0005] The purpose of this invention is to provide a real-time anomaly detection method for big data based on industrial mechanism models, so as to solve the problems in the background art mentioned above.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a real-time anomaly detection method for big data based on industrial mechanism models, comprising the following steps: The system acquires multidimensional change values ​​at a continuous time scale during the operation of the industrial mechanism model, records the difference in change amplitude and the change rate at each time position, and calibrates the attenuation gradient value of the abnormal signal during the time progression. Based on the attenuation gradient value, the difference in the change amplitude at each time position is inversely amplified. The degree of anomaly influence is progressively backtracked and expanded within the continuous attenuation segment to form an anomaly retention scale that retains the original anomaly intensity characteristics. For the time range corresponding to the abnormal retained scale, local rhythm separation processing is performed on the change rate value. The stable fluctuation section and the abnormal fluctuation section are segmented and expanded according to the degree of change rate mutation, and the distribution results of the abnormal manifestation section are obtained. Based on the distribution results of the abnormal manifestation sections, the difference in the change amplitude at each time position is processed by forward shifting, and the abnormal impact covered by subsequent fluctuations is redistributed to the previous time position to obtain a continuous and traceable abnormal diffusion trajectory. By using the abnormal diffusion trajectory to perform reverse rhythm suppression processing on the subsequent changes in value, the release amplitude of normal fluctuations during the time process is progressively reduced, and the abnormal influence is continuously maintained, so as to achieve the explicit maintenance and stable presentation of abnormal characteristics in the dynamic evolution process.

[0007] Preferably, the change characterization and attenuation relationship are constructed for multidimensional change values ​​under continuous time scale to obtain continuous calibration results of the abnormal signal in the time progression process. The steps are as follows: Obtain multidimensional change values ​​under continuous time scale, perform time base alignment processing, extract the change amplitude difference and change rate values ​​corresponding to each time position, and record them point by point; Correlation analysis is performed on the difference in change magnitude and the change rate along the time dimension to characterize the evolution trend of abnormal signals and map the degree of abnormal influence at each time position into a time series relationship; Based on the aforementioned time series relationship, the degree of impact of anomalies at adjacent time locations is characterized by difference, and a gradient change relationship reflecting the attenuation trend of the abnormal signal is constructed to obtain the corresponding attenuation gradient value; Based on the time index relationship, the attenuation gradient values ​​are correlated with the difference in change amplitude and the change rate values, and integrated into a unified time series record result to complete the calibration of the attenuation gradient values ​​of abnormal signals.

[0008] Preferably, the attenuation gradient value characterizes the continuous change direction and amplitude of the abnormal signal along the time progression, and through the synergistic relationship between the difference in amplitude and the value of the rate of change, the degree of abnormal influence is described in layers, so that the abnormal influence corresponding to each time position has the characteristic of continuous temporal correlation.

[0009] Preferably, enhancement and backtracking expansion processing are performed on the difference in the magnitude of change to obtain anomaly-preserving scales, as follows: A one-to-one correspondence is established between the attenuation gradient value and the difference in the change amplitude at each time position. Time segments with consistent attenuation trends are identified and defined as continuous attenuation segments. The difference in the amplitude of change within the continuous attenuation section is adjusted in reverse according to the corresponding attenuation gradient value, and the attenuation trend is mapped in reverse to obtain the continuously changing amplified result. The continuously changing amplified results are combined with the attenuation gradient value to perform progressive backtracking and expansion processing, which gradually backtracks the degree of abnormal impact along the time direction and expands the distribution range; The progressive backtracking expansion processing results are integrated with the continuously changing amplified results, and the difference in the change amplitude corresponding to each time position is scaled and arranged to obtain an anomaly retention scale that retains the original anomaly intensity characteristics.

[0010] Preferably, in the progressive backtracking and expansion process, the degree of anomaly impact is continuously transmitted across time positions along the time direction, and the continuously changing amplified results synchronously participate in the adjustment of the distribution of anomaly impact at each time position. The difference in the magnitude of change at each time position maintains correlation and consistency, thereby enhancing the ability of the anomaly retention scale to continuously penetrate the time dimension.

[0011] Preferably, rhythm separation and segmentation processing are performed on the numerical rate of change to obtain the distribution results of abnormally manifested segments. The steps are as follows: The rate of change values ​​corresponding to the anomaly retention scale within the time range are extracted time-by-time, and synchronously associated with the anomaly retention scale through the time position index relationship. Each time position corresponds to the anomaly retention scale value and the rate of change value. The numerical rate of change is combined with the abnormal retention scale to perform local rhythm separation processing, refine the continuous change relationship, obtain several continuous rhythm segments and maintain time continuity; Continuous rhythm segments are distinguished by the degree of abrupt change in the rate of change. Stable fluctuation segments and abnormal fluctuation segments are divided by rhythmic turning points, and each segment is further subdivided.

[0012] The abnormal fluctuation segments are arranged in chronological order, and each segment is marked accordingly based on the abnormal retention scale to obtain the distribution results of the abnormal manifestation segments.

[0013] Preferably, the change rate value is introduced into the abnormal retention scale corresponding to the abnormal intensity information during the rhythm separation process, and the rhythm turning point is dynamically correlated and adjusted to strengthen the temporal continuity relationship and the consistency relationship of the segment boundary in the abnormal fluctuation segment identification process.

[0014] Preferably, the difference in the magnitude of the change is subjected to forward traction and redistribution processing to obtain a continuous and traceable anomaly diffusion trajectory. The steps are as follows: The difference in the magnitude of change within the time range of the abnormal manifestation segment distribution results is extracted time-by-time, and the abnormal manifestation segment distribution results are identified by time index relationship, with each time position corresponding to an abnormal fluctuation segment or a stable fluctuation segment. The difference in the amplitude of abnormal fluctuations within the range establishes a traction relationship from the subsequent time position to the previous time position, and performs forward traction processing to extend the impact of the abnormality forward along the time series; The forward traction processing results are combined with the difference in the change amplitude within the abnormal fluctuation section for redistribution processing. Each time position carries the abnormal influence from subsequent time positions and maintains the temporal continuity. The results of the redistribution processing are integrated with the results of the forward traction processing. The difference in the change amplitude corresponding to each time position is arranged in chronological order to obtain a continuous and traceable abnormal diffusion trajectory.

[0015] Preferably, during the forward traction process, the difference in the change amplitude corresponding to the abnormal fluctuation section is established as a continuous transmission relationship according to the time position sequence. The abnormal impact is transmitted step by step along the time direction and superimposed to the previous time position. The difference in the change amplitude corresponds to the abnormal impact in a continuous distribution state.

[0016] Preferably, the changing values ​​are subjected to reverse rhythm suppression and amplitude adjustment processing to obtain a result in which the abnormal features are continuously presented. The steps are as follows: The abnormal spread trajectory corresponds to the change value within the time range, and the change value and the distribution information of the abnormal impact are accessed time by time position. The time index relationship is used to map the abnormal spread trajectory. The changing numerical values ​​are combined with the abnormal diffusion trajectory to perform reverse rhythm suppression processing, construct rhythm adjustment relationship, adjust the changing trend, and limit the release amplitude of normal fluctuations; The rhythm adjustment relationship acts on the changing values, progressively reducing the amplitude of normal fluctuations while maintaining the continuity of the distribution of abnormal influences; The progressive reduction processing results are integrated with the reverse rhythm suppression processing results, and the change values ​​at each time position are arranged in chronological order to achieve continuous presentation of abnormal features.

[0017] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention introduces a decay gradient and performs inverse amplification on the difference in amplitude, progressively extending the degree of anomaly impact within a continuous decay range. This prevents the anomalous signal from gradually weakening with data updates over time, thus preserving the original anomaly intensity characteristics continuously in the continuous data stream. In this way, the existence of the anomalous signal in the time dimension changes from an instantaneous expression to a continuous expression, allowing early-appearing anomalies to be stably recorded and persist throughout subsequent data evolution. This effectively avoids the problem of anomalous features being weakened or even disappearing over time, thereby improving the continuity and stability of the anomaly identification process.

[0018] This invention constructs anomaly manifestation segment distribution results and combines forward-moving traction processing with reverse rhythm suppression processing, enabling the anomalous impact to form a continuous diffusion trajectory in the time series. Simultaneously, it rhythmically adjusts subsequent changes in numerical values, constraining the release amplitude of normal fluctuations during time progression and maintaining the continuous existence of the anomalous impact during the change process. Through this method, anomalous features remain clearly expressed throughout the dynamic evolution process, allowing anomalous signals to continuously present themselves in complex fluctuation environments and possess traceability. This improves the accuracy of identifying anomalous states and reduces the possibility of interference in subsequent operational status judgments. Attached Figure Description

[0019] 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.

[0020] Figure 1 This is a flowchart of the method for real-time anomaly detection of big data based on industrial mechanism models according to the present invention. Detailed Implementation

[0021] 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.

[0022] This invention provides, for example Figure 1 The big data real-time anomaly detection method for industrial mechanism models shown includes the following steps: The system acquires multidimensional change values ​​at a continuous time scale during the operation of the industrial mechanism model, records the difference in change amplitude and the change rate at each time position, and calibrates the attenuation gradient value of the abnormal signal during the time progression. In the operation of the industrial mechanism model, to achieve continuous characterization of anomalous signals over time, it is necessary to systematically process the multidimensional changes in values ​​over continuous time scales and simultaneously construct the attenuation gradient values ​​of the anomalous signals during the changes. This provides a foundation for subsequent anomaly identification and feature preservation. The steps are as follows: Multidimensional change values ​​under continuous time scales are acquired during the operation of the industrial mechanism model, and each time position is aligned with a unified time reference. Based on this time alignment, for the multidimensional change values ​​corresponding to each time position, the change amplitude difference and change rate values ​​reflecting the state evolution characteristics are extracted. The change amplitude difference is used to describe the degree of numerical change between adjacent time positions, and the change rate value is used to characterize the intensity of time progression during the change process. By unfolding the data point by point, the above change amplitude difference and change rate values ​​are recorded point by point, so that each time position corresponds to a complete set of change description data, thereby forming a continuous and time-related change value sequence. This change value sequence not only retains the dynamic evolution information of the original data, but also reflects the continuity of change between each time position, providing a unified data expression basis for subsequent identification of abnormal signals and extraction of attenuation features.

[0023] Based on the completed point-by-point recording of the difference in amplitude and the numerical values ​​of the rate of change, the evolution trend of the anomalous signal over time is continuously characterized. By introducing the continuity of change in the time dimension, the direction and degree of change of the difference in amplitude between multiple adjacent time positions are correlated and analyzed. Combined with the time progression rhythm reflected by the numerical values ​​of the rate of change, the intensity change trend of the anomalous signal over time is serialized and expressed. In this expression process, the degree of anomalous influence corresponding to each time position is mapped to a sequence relationship that gradually changes over time. This sequence relationship reflects the evolution trajectory of the anomalous signal from its initial appearance to its gradual decay, making the change process of the anomalous signal in the time dimension continuously describable, thus providing basic data support for the subsequent calibration of the decay gradient value.

[0024] Based on the evolutionary sequence of anomalous signals over time, the degree of anomalous influence at each time point is hierarchically characterized. By continuously expressing the differences in the degree of anomalous influence between adjacent time points, a gradient change relationship reflecting the attenuation trend of the anomalous signal is constructed. In this process, the difference in change amplitude and the value of change rate are used as common factors to refine the description of the attenuation state of the anomalous signal during the time progression. This ensures that each time point corresponds to a gradient value that reflects the attenuation trend of the anomalous signal. This gradient value not only reflects the influence state of the anomalous signal at the current time point but also reflects its direction and magnitude of change relative to the previous and subsequent time points. Thus, the attenuation behavior of the anomalous signal during the time progression is continuously and completely characterized.

[0025] After constructing the attenuation gradient value, this attenuation gradient value is correlated with the original point-by-point recorded difference in amplitude and rate of change values. This ensures that each time position simultaneously possesses three types of descriptive information: difference in amplitude, rate of change, and attenuation gradient value. Through a unified time index relationship, these three types of data are integrated into a unified time series recording result. This enables the calibration of the attenuation gradient value of the anomalous signal during the time progression, allowing the anomalous signal to be recorded not only at the numerical level but also continuously expressed at the temporal evolution level. This calibration result can be directly used in subsequent processing to describe the attenuation characteristics and temporal distribution of the anomalous signal, thus providing complete data support for the continuous identification and dynamic presentation of anomalous signals.

[0026] Based on the attenuation gradient value, the difference in the change amplitude at each time position is inversely amplified. The degree of anomaly influence is progressively backtracked and expanded within the continuous attenuation segment to form an anomaly retention scale that retains the original anomaly intensity characteristics. In the multidimensional variation numerical processing under continuous time scale, to prevent the anomalous signal from being continuously attenuated during time progression, it is necessary to perform targeted enhancement processing on the variation amplitude difference, and combine it with the attenuation gradient value to continuously preserve the degree of anomalous influence, thereby constructing an anomalous preservation scale that can reflect the original anomalous intensity characteristics. The specific steps are as follows: The attenuation gradient values ​​are used to correlate and map the difference in the amplitude of change at each time position. A one-to-one correspondence is established between the difference in the amplitude of change at each time position and its corresponding attenuation gradient value. In this correspondence, the abnormal attenuation direction and degree represented by the attenuation gradient value are expressed in detail, so that the attenuation effect of the difference in amplitude of change on the time progress can be quantified. Time segments with consistent attenuation trends are identified within the continuous time scale. These time segments are defined as continuous attenuation segments. Within the continuous attenuation segments, the difference in the amplitude of change between each time position shows a continuous change relationship under the action of the attenuation gradient value, thus providing a clear range and basis for subsequent reverse amplification processing.

[0027] A reverse amplification mechanism is introduced around the difference in amplitude within a continuous attenuation range. By reverse mapping the attenuation trend represented by the attenuation gradient value, the attenuation state reflected by the difference in amplitude over time is compensated. In this reverse amplification process, the difference in amplitude at each time position is adjusted in reverse according to its corresponding attenuation gradient value. This restores the degree of abnormal influence that was originally weakened by the passage of time in the numerical expression, and forms a continuously changing amplified result within the continuous attenuation range. This makes the difference in amplitude present an intensity distribution relationship corresponding to the original abnormal state in the time dimension, thereby realizing the numerical reconstruction expression of the initial characteristics of the abnormal signal.

[0028] For the difference in change amplitude after inverse amplification, a progressive backtracking expansion process is performed on the degree of abnormal influence within the continuous attenuation segment. By taking the time progression direction as a reference, the change relationship between each time position is gradually backtracked, so that the abnormal influence reflected in the subsequent time position can be propagated forward to the previous time position. In this progressive backtracking expansion process, based on the correlation between the attenuation gradient value and the inverse amplification result, the distribution range of the degree of abnormal influence in the time dimension is gradually expanded, so that the abnormal influence originally limited to the local time position can form a penetrating time distribution state within the continuous attenuation segment. Thus, the existence trajectory of the abnormal signal in the time progression process is continuously extended and expressed, and the existence continuity of the abnormal signal in the overall time series is further strengthened.

[0029] After completing the progressive backtracking expansion process, the variation amplitude difference corresponding to each time position within the continuous attenuation segment is uniformly scaled. The reverse amplification result is integrated with the progressive backtracking expansion result, so that each time position corresponds to a scale value that can reflect the original anomaly intensity characteristics. The above scale values ​​are continuously arranged through the time position index relationship to form an anomaly retention scale. In this anomaly retention scale, not only is the numerical expression of the variation amplitude difference after reverse amplification processing included, but the time distribution extension effect brought about by the progressive backtracking expansion is also reflected. This allows the original intensity characteristics of the anomaly signal to be continuously maintained and stably presented during the time progression, thereby providing a continuous and complete anomaly expression basis for subsequent anomaly manifestation processing.

[0030] For the time range corresponding to the abnormal retained scale, local rhythm separation processing is performed on the change rate value. The stable fluctuation section and the abnormal fluctuation section are segmented and expanded according to the degree of change rate mutation, and the distribution results of the abnormal manifestation section are obtained. Based on the anomaly retention scale, to reveal the differences in the performance of anomalous signals over time, it is necessary to perform local rhythm separation processing on the change rate values ​​within the time range corresponding to the anomaly retention scale. This allows for the differentiated representation of different fluctuation patterns and outputs the distribution results of the anomaly manifestation segments. The specific process is as follows:

[0031] For the time range corresponding to the anomaly retention scale, the rate of change values ​​within this time range are extracted time-by-time. The rate of change values ​​are synchronously associated with the anomaly retention scale through a time position index relationship, so that each time position has a correspondence between the anomaly retention scale value and the rate of change value. In this correspondence, the change state of the rate of change value between adjacent time positions is continuously expressed, so that the rhythmic change of the rate of change value in the process of time progress is fully presented. At the same time, based on the anomaly intensity distribution reflected by the anomaly retention scale, the time distribution range of the rate of change value is limited, so that the subsequent local rhythm separation processing can focus on the time interval related to the anomaly, ensuring that the rhythm separation processing process has a clear time range and a consistent time reference relationship.

[0032] Based on the established correspondence between the anomaly retention scale and the rate of change values, local rhythm separation processing is performed on the rate of change values. By refining the continuous change relationship between the rate of change values ​​at each time position, the trajectory of the rate of change values ​​during the time progression is divided into several continuous rhythm segments. Within each rhythm segment, the rate of change values ​​are expressed according to their continuity of change. At the same time, by distinguishing the connection state between adjacent rhythm segments, the rhythm changes reflected by the rate of change values ​​at different time positions can form a layered expression. In this process, the anomaly retention scale serves as the reference basis for rhythm separation processing, constraining the distribution structure of the rate of change values ​​in the time dimension, so that the local rhythm separation processing can highlight the rhythm change characteristics of the time region where the anomaly signal is located, thereby forming a rhythm division result with temporal continuity.

[0033] Based on the rhythm segmentation results formed by local rhythm separation processing, each rhythm segment is further distinguished according to the degree of abrupt change in the rate of change. By continuously comparing the change state of the rate of change value at the boundary of the rhythm segment, the time position where the rate of change value changes abruptly during the time progression is identified as the rhythm turning point. The rhythm segment is then re-segmented and expanded using the rhythm turning point as the boundary. During the segmentation and expansion process, time segments where the rate of change value changes continuously and fluctuates in a consistent manner are divided into stable fluctuation segments, and time segments where the rate of change value shows abrupt change characteristics during the time progression are divided into abnormal fluctuation segments. This creates an alternating distribution structure of stable fluctuation segments and abnormal fluctuation segments in the time dimension. At the same time, during this segmentation and expansion process, the anomaly retention scale is continuously used to identify the degree of abnormal influence in each segment, so that each segmentation result can reflect the performance characteristics of the abnormal signal in the time dimension.

[0034] Based on the segmented expansion results of stable and abnormal fluctuation segments, the segment attributes corresponding to each time position are uniformly integrated and expressed. All abnormal fluctuation segments are arranged in chronological order, and combined with the abnormal intensity information reflected by the abnormal retention scale, the distribution state of abnormal fluctuation segments in the time dimension is continuously described, thus forming the abnormal manifestation segment distribution result. This distribution result not only includes the positional relationship of abnormal fluctuation segments in the time series, but also reflects the rhythmic change characteristics of the change rate value in each segment, so that the abnormal signal can be clearly presented in the time dimension in the form of segments. At the same time, by maintaining the time index relationship between the abnormal manifestation segment distribution result and the original abnormal retention scale, the subsequent processing can directly use the distribution result to further expand and strengthen the abnormal signal, thereby realizing the continuous expression and stable presentation of abnormal features in the time series.

[0035] Based on the distribution results of the abnormal manifestation sections, the difference in the change amplitude at each time position is processed by forward shifting, and the abnormal impact covered by subsequent fluctuations is redistributed to the previous time position to obtain a continuous and traceable abnormal diffusion trajectory. After obtaining the distribution results of the anomaly manifestation segments, in order to avoid the anomaly impact being covered by subsequent fluctuations as time progresses, it is necessary to reallocate the difference in change magnitude over time so that the anomaly impact can form a continuous expression in the time series, thereby constructing a continuous and traceable anomaly diffusion trajectory. The specific steps are as follows: For the time range covered by the distribution results of the abnormal manifestation segment, the difference in the amplitude of change corresponding to each time position is extracted time-by-time. The difference in amplitude of change is then mapped to the distribution results of the abnormal manifestation segment through a time index relationship, so that each time position can be clearly identified as belonging to the abnormal fluctuation segment or the stable fluctuation segment. In this correspondence, the difference in amplitude of change within the abnormal fluctuation segment is centrally marked, so that the distribution of the abnormal impact in the time dimension is uniformly expressed. At the same time, the continuous change relationship of the difference in amplitude of change within the stable fluctuation segment is preserved, so that the time distribution characteristics of the abnormal impact and normal fluctuation can be distinguished in the subsequent processing, providing a clear target and time reference basis for forward-looking traction processing.

[0036] Based on the corresponding identification of the variation amplitude difference and the distribution results of the abnormal manifestation segment, the variation amplitude difference within the abnormal fluctuation segment is subjected to forward traction processing. By establishing a traction relationship from the subsequent time position to the previous time position in the time progression direction, the abnormal influence covered by the subsequent fluctuation is directionally guided, so that the abnormal influence originally distributed in the subsequent time position can extend forward along the time series. In this forward traction process, the time boundary of each abnormal fluctuation segment in the abnormal manifestation segment distribution results is used as the traction starting point, and the previous time position in the time series is used as the traction target. By continuously expressing the correlation between the variation amplitude difference at different time positions, the distribution of the abnormal influence in the time dimension is no longer limited to its original occurrence position, but forms a continuous expansion state across time positions under the forward traction action, thereby realizing the directional extension of the abnormal influence in the time series.

[0037] In the process of performing forward shifting processing on the difference in variation amplitude at each time position, in order to achieve the orderly allocation of the abnormal impact under subsequent fluctuation coverage to the forward time position, the correlation between the difference in variation amplitude at different time positions is quantified. Specifically, the following expression is used to describe the forward shifting result: ; in, Indicates the first The difference in the magnitude of change at each time position after the forward traction treatment. This represents the difference in the original change magnitude at the i-th time position. This represents the difference in the magnitude of change of consecutive time positions after the i-th time position. This represents the anomalous impact propagation factor corresponding to the k-th time position. The anomalous impact propagation factor is used to describe the degree to which the anomalous impact of subsequent time positions propagates to the previous time position. This indicates the length of the continuous time range involved in the forward traction processing. Through the above expression method, the abnormal effects contained in the subsequent time position can be transmitted forward in the time dimension and superimposed to the current time position, so that the difference in the magnitude of change forms a continuous expansion relationship in the time series, and the abnormal effects that were originally covered by subsequent fluctuations are reconstructed and expressed at the previous time position.

[0038] By expressing the above relationships, the difference in the magnitude of change between different time positions is correlated across time positions, so that the impact of anomalies in the time series is no longer limited to a single time position, but forms a continuous expansion state along the time direction under the forward pulling effect, thus providing basic data support for the construction of subsequent anomaly diffusion trajectories.

[0039] Based on the temporal expansion relationship formed by the forward-moving traction processing, the abnormal impacts between each time position are redistributed. The abnormal impacts covered by subsequent fluctuations are gradually mapped to the previous time positions in chronological order. This ensures that each time position, while maintaining the original difference in the magnitude of change, also carries the abnormal impacts from subsequent time positions. In this redistribution process, by integrating the continuous expression of the difference in the magnitude of change within the abnormal fluctuation segment, the transmission process of abnormal impacts in the time dimension forms a continuous chain structure. This results in the abnormal impacts forming a distribution pattern that gradually expands from back to front in the time series, while ensuring that the transmission of abnormal impacts between each time position maintains temporal continuity, and that the correlation between abnormal impacts at different time positions is fully preserved.

[0040] After the abnormal impact is redistributed to the previous time position, the difference in the change amplitude corresponding to each time position is integrated and expressed as a whole. The time expansion results formed by the forward traction processing and the redistribution processing are arranged in a unified manner, so that the abnormal impact between each time position forms a continuous connection relationship in chronological order, thereby constructing a continuous and traceable abnormal diffusion trajectory. In this abnormal diffusion trajectory, not only is the propagation path of the abnormal impact in the time series reflected, but also the distribution relationship of the abnormal impact between different time positions is reflected. This allows the evolution state of the abnormal signal in the time process to be continuously expressed in the form of a trajectory. At the same time, by maintaining a consistent time index relationship between the distribution results of the abnormal manifestation segment and the abnormal diffusion trajectory, the subsequent processing can further regulate and maintain the abnormal signal based on the abnormal diffusion trajectory, thereby realizing the continuous presentation and stable expression of abnormal features in the time dimension.

[0041] Based on the forward-looking processing results, in order to achieve continuous expression of the abnormal impact in the time series, the abnormal impact at each time position is progressively accumulated to construct a continuous and traceable abnormal diffusion trajectory. Specifically, the abnormal diffusion trajectory is described by the following expression: ; in, Indicates the first The abnormal diffusion trajectory value corresponding to each time location. This represents the difference in the magnitude of change after the forward traction treatment. This represents the trajectory extension factor corresponding to the i-th time position, which describes the degree of persistence of the abnormal influence at the current time position in the time series. This represents the anomalous diffusion trajectory value at the previous time position. By correlating the difference in the change magnitude at the current time position with the anomalous diffusion trajectory at the previous time position, it establishes a continuous transmission relationship of the anomalous impact over time, thereby constructing a continuous anomalous diffusion trajectory in the time series. Through the above expression method, the abnormal influence forms a trajectory structure that accumulates gradually between different time positions. This allows the abnormal state at each time position to not only reflect the current change characteristics but also include information on the abnormal influence of previous time positions. This enables the continuous expression and traceable presentation of abnormal signals during the time progression process and further strengthens the correlation of the abnormal diffusion trajectory in the overall time series.

[0042] By using the abnormal diffusion trajectory to perform reverse rhythm suppression processing on the subsequent incoming change values, the release amplitude of normal fluctuations during the time process is progressively reduced, and the abnormal influence is continuously maintained, so as to achieve the explicit maintenance and stable presentation of abnormal characteristics in the dynamic evolution process. Based on the established continuous and traceable anomaly propagation trajectory, to prevent subsequent changes from further overshadowing the anomaly's impact, it is necessary to intervene in the change process through temporal rhythm regulation, ensuring that the anomaly characteristics remain continuously expressed during dynamic evolution. The specific steps are as follows: For the time range corresponding to the abnormal diffusion trajectory, subsequent incoming change values ​​are accessed time-by-time, and the change values ​​are mapped to the abnormal diffusion trajectory through a time index relationship. This ensures that each time position simultaneously contains the abnormal impact distribution information described by the change value and the abnormal diffusion trajectory. In this correspondence, by expressing the continuous change state of the change value in the process of time progression, the evolution path of the change value in the time dimension is made consistent with the abnormal diffusion trajectory. This allows subsequent incoming change values ​​to be expressed in an orderly manner within the time range defined by the abnormal diffusion trajectory, providing a unified data foundation and time reference relationship for reverse rhythm suppression processing.

[0043] Based on the distribution of abnormal influences described by the abnormal diffusion trajectory, a reverse rhythm suppression process is applied to the changing values. By constructing a rhythm adjustment relationship corresponding to the abnormal diffusion trajectory in the direction of time progression, the rhythm of change of the changing values ​​in the time dimension is constrained. In this reverse rhythm suppression process, the intensity of abnormal influences at each time position in the abnormal diffusion trajectory is used as a reference to adjust the trend of change of the changing values ​​at the corresponding time positions, so that the release amplitude of normal fluctuations during the time progression is suppressed. Furthermore, by reconstructing the continuous change relationship of the changing values, the rhythm of change in the time dimension gradually approaches the distribution of abnormal influences reflected by the abnormal diffusion trajectory, thereby realizing the intervention expression of the normal fluctuation release process.

[0044] Based on the rhythm regulation relationship formed by the reverse rhythm suppression process, the release amplitude of normal fluctuations during the time progression is progressively reduced. By continuously adjusting the amplitude of changes in values ​​between different time positions, the release amplitude of normal fluctuations at different time positions gradually weakens in chronological order. At the same time, in this progressive reduction process, relying on the abnormal influence distribution information provided by the abnormal diffusion trajectory, the abnormal influence contained in the changing values ​​is continuously maintained, so that the expression of abnormal influence in the time dimension does not weaken with the change of normal fluctuations. Thus, a distribution state in the same time series is formed in which normal fluctuations gradually weaken while abnormal influences continue to exist, so that the abnormal characteristics are strengthened in the overall change process.

[0045] After completing the progressive reduction of normal fluctuation release amplitude and the maintenance of abnormal influence, the change values ​​corresponding to each time position are uniformly integrated and expressed. The results of the reverse rhythm suppression processing are arranged continuously in chronological order, so that the evolution of change values ​​in the time dimension can simultaneously reflect the distribution of abnormal influence described by the abnormal diffusion trajectory and the adjustment results of normal fluctuation release amplitude. This achieves the explicit maintenance and stable presentation of abnormal features in the dynamic evolution process. In this expression result, the abnormal influence persists throughout the entire time series, while normal fluctuations are gradually constrained as time progresses, so that the overall change process presents a time distribution structure dominated by abnormal features, providing a clear and continuous data expression basis for subsequent abnormal state judgment.

[0046] This invention introduces a decay gradient and performs inverse amplification on the difference in amplitude, progressively extending the degree of anomaly impact within a continuous decay range. This prevents the anomalous signal from gradually weakening with data updates over time, thus preserving the original anomaly intensity characteristics continuously in the continuous data stream. In this way, the existence of the anomalous signal in the time dimension changes from an instantaneous expression to a continuous expression, allowing early-appearing anomalies to be stably recorded and persist throughout subsequent data evolution. This effectively avoids the problem of anomalous features being weakened or even disappearing over time, thereby improving the continuity and stability of the anomaly identification process.

[0047] This invention constructs anomaly manifestation segment distribution results and combines forward-moving traction processing with reverse rhythm suppression processing, enabling the anomalous impact to form a continuous diffusion trajectory in the time series. Simultaneously, it rhythmically adjusts subsequent changes in numerical values, constraining the release amplitude of normal fluctuations during time progression and maintaining the continuous existence of the anomalous impact during the change process. Through this method, anomalous features remain clearly expressed throughout the dynamic evolution process, allowing anomalous signals to continuously present themselves in complex fluctuation environments and possess traceability. This improves the accuracy of identifying anomalous states and reduces the possibility of interference in subsequent operational status judgments.

[0048] 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 method for real-time anomaly detection of big data oriented industrial mechanistic models, characterized in that, Includes the following steps: The system acquires multidimensional change values ​​at a continuous time scale during the operation of the industrial mechanism model, records the difference in change amplitude and the change rate at each time position, and calibrates the attenuation gradient value of the abnormal signal during the time progression. Based on the attenuation gradient value, the difference in the change amplitude at each time position is inversely amplified. The degree of anomaly influence is progressively backtracked and expanded within the continuous attenuation segment to form an anomaly retention scale that retains the original anomaly intensity characteristics. For the time range corresponding to the abnormal retained scale, local rhythm separation processing is performed on the change rate value. The stable fluctuation section and the abnormal fluctuation section are segmented and expanded according to the degree of change rate mutation, and the distribution results of the abnormal manifestation section are obtained. Based on the distribution results of the abnormal manifestation sections, the difference in the change amplitude at each time position is processed by forward shifting, and the abnormal impact covered by subsequent fluctuations is redistributed to the previous time position to obtain a continuous and traceable abnormal diffusion trajectory. The abnormal diffusion trajectory is used to perform reverse rhythm suppression on the subsequent changes in value. This is achieved by progressively reducing the release amplitude of normal fluctuations over time and continuously maintaining the abnormal impact.

2. The method of claim 1, wherein, The following steps are taken to construct a change characterization and attenuation relationship for multidimensional changes in values ​​at a continuous time scale, thereby obtaining continuous calibration results of the anomalous signal during the time progression process: Obtain multidimensional change values ​​under continuous time scale, perform time base alignment processing, extract the change amplitude difference and change rate values ​​corresponding to each time position, and record them point by point; Correlation analysis is performed on the difference in change magnitude and the change rate along the time dimension to characterize the evolution trend of abnormal signals and map the degree of abnormal influence at each time position into a time series relationship; Based on the aforementioned time series relationship, the degree of impact of anomalies at adjacent time locations is characterized by difference, and a gradient change relationship reflecting the attenuation trend of the abnormal signal is constructed to obtain the corresponding attenuation gradient value; Based on the time index relationship, the attenuation gradient values ​​are correlated with the difference in change amplitude and the change rate values, and integrated into a unified time series record result to complete the calibration of the attenuation gradient values ​​of abnormal signals.

3. The real-time anomaly detection method for big data based on industrial mechanism models according to claim 2, characterized in that, The attenuation gradient numerical value characterizes the continuous change direction and amplitude of the abnormal signal over time, and through the synergistic relationship between the difference in amplitude and the numerical value of the rate of change, the degree of influence of the anomaly is described in layers.

4. The method of claim 2, wherein, Enhancement and backtracking expansion processing are performed on the difference in the magnitude of change to obtain anomaly-preserving scales. The steps are as follows: A one-to-one correspondence is established between the attenuation gradient value and the difference in the change amplitude at each time position. Time segments with consistent attenuation trends are identified and defined as continuous attenuation segments. The difference in the amplitude of change within the continuous attenuation section is adjusted in reverse according to the corresponding attenuation gradient value, and the attenuation trend is mapped in reverse to obtain the continuously changing amplified result. The continuously changing amplified results are combined with the attenuation gradient value to perform progressive backtracking and expansion processing, which gradually backtracks the degree of abnormal impact along the time direction and expands the distribution range; The progressive backtracking expansion processing results are integrated with the continuously changing amplified results, and the difference in the change amplitude corresponding to each time position is scaled and arranged to obtain an anomaly retention scale that retains the original anomaly intensity characteristics.

5. The method of claim 4, wherein, During the progressive backtracking and expansion process, the degree of anomaly impact is continuously transmitted across time locations along the time direction. The continuously changing amplified results synchronously participate in the adjustment of the distribution of anomaly impact at each time location, and the difference in the magnitude of change at each time location remains consistent.

6. The method of claim 4, wherein, The change rate values ​​are processed by rhythm separation and segmentation to obtain the distribution results of abnormal explicit segments. The steps are as follows: The rate of change values ​​corresponding to the anomaly retention scale within the time range are extracted time-by-time, and synchronously associated with the anomaly retention scale through the time position index relationship. Each time position corresponds to the anomaly retention scale value and the rate of change value. The numerical rate of change is combined with the abnormal retention scale to perform local rhythm separation processing, refine the continuous change relationship, obtain several continuous rhythm segments and maintain time continuity; Continuous rhythm segments are distinguished by the degree of abrupt change in the rate of change. Stable fluctuation segments and abnormal fluctuation segments are divided by rhythmic turning points, and each segment is further subdivided. The abnormal fluctuation segments are arranged in chronological order, and each segment is marked accordingly based on the abnormal retention scale to obtain the distribution results of the abnormal manifestation segments.

7. The method of claim 6, wherein, The change rate value is introduced into the rhythm separation process to retain the abnormal intensity information corresponding to the abnormal scale, and the rhythm turning point is dynamically correlated and adjusted to strengthen the temporal continuity relationship and the consistency relationship of the segment boundary in the process of abnormal fluctuation segment identification.

8. The industrial mechanism model oriented big data real-time anomaly detection method according to claim 6, characterized in that, The difference in the magnitude of the change is processed by forward traction and redistribution to obtain a continuous and traceable anomaly diffusion trajectory. The steps are as follows: The difference in the magnitude of change within the time range of the abnormal manifestation segment distribution results is extracted time-by-time, and the abnormal manifestation segment distribution results are identified by time index relationship, with each time position corresponding to an abnormal fluctuation segment or a stable fluctuation segment. The difference in the amplitude of abnormal fluctuations within the range establishes a traction relationship from the subsequent time position to the previous time position, and performs forward traction processing to extend the impact of the abnormality forward along the time series; The forward traction processing results are combined with the difference in the change amplitude within the abnormal fluctuation section for redistribution processing. Each time position carries the abnormal influence from subsequent time positions and maintains the temporal continuity. The results of the redistribution processing are integrated with the results of the forward traction processing. The difference in the change amplitude corresponding to each time position is arranged in chronological order to obtain a continuous and traceable abnormal diffusion trajectory.

9. The industrial mechanism model oriented big data real-time anomaly detection method according to claim 8, characterized in that, During the forward traction process, the difference in the change amplitude corresponding to the abnormal fluctuation section is established as a continuous transmission relationship according to the time position. The abnormal impact is transmitted step by step along the time direction and superimposed to the previous time position. The difference in the change amplitude corresponds to the abnormal impact in a continuous distribution state.

10. The industrial mechanism model oriented big data real-time anomaly detection method according to claim 8, characterized in that, The changing values ​​are subjected to reverse rhythm suppression and amplitude adjustment to obtain a result in which the abnormal features are continuously presented. The steps are as follows: The abnormal spread trajectory corresponds to the change value within the time range, and the change value and the distribution information of the abnormal impact are accessed time by time position. The time index relationship is used to map the abnormal spread trajectory. The changing numerical values ​​are combined with the abnormal diffusion trajectory to perform reverse rhythm suppression processing, construct rhythm adjustment relationship, adjust the changing trend, and limit the release amplitude of normal fluctuations; The rhythm adjustment relationship acts on the changing values, progressively reducing the amplitude of normal fluctuations while maintaining the continuity of the distribution of abnormal influences; The results of progressive reduction processing and reverse rhythm suppression processing are integrated, and the change values ​​at each time position are arranged in chronological order.

Images