Edible oil processing monitoring system
By constructing a dual-domain structure of physical time parameter sequence and event time behavior sequence for the edible oil processing monitoring system, the problem of identifying reaction jumps during oil batch switching periods was solved, enabling effective monitoring of non-periodic jump events and improving the accuracy of risk assessment and the predictive ability of the system.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-14
- Publication Date
- 2026-03-31
AI Technical Summary
Existing edible oil processing monitoring systems cannot identify reaction jumps during oil batch switching periods, leading to broken risk causal chains and distorted assessments, and making it impossible to identify abnormal risks in a timely manner.
A dual-domain structure of physical time parameter sequence and event time behavior sequence is constructed. By extracting response direction consistency index, normalized response difference and response offset delay index, abnormal event segments are identified and their causal chain paths are traced.
It effectively identifies non-periodic surge events in edible oil processing, improves the robustness of identifying risk causal chain breaks, avoids misjudgment of short-term fluctuations and omission of slow anomalies, and improves the accuracy of risk assessment and the predictive effectiveness of the system.
Smart Images
Figure CN121765531A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of food processing monitoring and risk assessment technology, and more specifically, to an edible oil processing monitoring system. Background Technology
[0002] In current edible oil processing monitoring systems, during the batch switching period of oilseeds, reaction behavior may undergo short-term abrupt changes, such as a surge in local oxidation rate or retention of intermediate products. These processing abrupt changes often do not have stable time periodicity, making them difficult to capture by traditional time window models. Without a multi-dimensional time modeling mechanism, the system is very likely to classify such abrupt changes as statistical fluctuations, thus directly filtering them in the risk assessment logic. This results in early chemical deviations not being recorded. As this hidden risk evolves and spreads to the finished product stage, the system is unable to trace its causal path, forming a risk chain that cannot be attributed, weakening the accountability and predictive effectiveness of the risk assessment mechanism.
[0003] Therefore, the current risk assessment mechanism of the monitoring system cannot detect sudden transitions in response behavior in the non-physical time domain, which leads to the failure to identify and effectively model abnormal risks in a timely manner during the processing. As a result, the causal chain is broken and the assessment is distorted, which is the main problem to be solved. Summary of the Invention
[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an edible oil processing monitoring system. By constructing a dual-domain structure of physical time parameter sequence and event time behavior sequence, the system extracts and compares response direction consistency index, normalized response difference, and response offset delay index to identify abnormal event segments and trace their causal chain paths. This solves the problem that traditional monitoring systems cannot identify the suddenness and non-periodicity of reaction transition behavior during oil batch switching periods, which leads to the breakage of risk causal chains and the distortion of assessments.
[0005] To achieve the above objectives, the present invention provides the following technical solution: an edible oil processing monitoring system, comprising a parameter acquisition module, a behavior labeling module, a segmentation module, a difference comparison module, and a risk identification module;
[0006] The parameter acquisition module continuously collects data on temperature, acid value, pressure, and volatile component concentration during the edible oil processing by setting a fixed sampling period, acquires raw data streams, and organizes the raw data streams into a physical time parameter sequence according to the sampling order;
[0007] The behavior labeling module performs second derivative rate of change calculation on each type of parameter in the physical time parameter sequence, and identifies response jump segments based on the sudden rise segments of the second derivative rate of change. By labeling the corresponding processing behavior type in each response jump segment, an event time behavior sequence is constructed.
[0008] The segmentation module divides the physical time parameter sequence and the event time behavior sequence into equal-length sliding segments, and extracts the parameter normalization amplitude, normalization rate of change value and normalization coupling response value in each sliding segment. After performing sequence encoding, it generates physical time trajectory set and event time trajectory set respectively.
[0009] The difference comparison module compares the corresponding segments of the physical time trajectory set with the event time trajectory set, calculates the response direction consistency index, normalized response difference, and response offset delay index for each pair of trajectories under the same sliding segment, and generates a response difference vector sequence.
[0010] The risk identification module analyzes the consistency decline trend and offset delay clustering pattern in the response difference vector sequence to identify abnormal event segments with response jumps but lacking physical temporal causal support. It determines that the abnormal event segment is a risk causal chain break area and outputs risk warning information.
[0011] In a preferred embodiment, in the parameter acquisition module, by setting a fixed sampling period, the raw data streams of four process parameters—temperature, acid value, pressure, and volatile component concentration—are collected from the edible oil processing process. The acquisition time point of each process parameter is marked by a timestamp, and the corresponding measurement value and timestamp information are retained in the raw data stream record.
[0012] The raw data stream of process parameters is sorted according to the timestamps within the sampling period. Interpolation algorithms are used to fill in the missing measurement values due to sampling gaps, and each measurement value is assigned a corresponding time label, so that the raw data stream forms a continuous time series according to the time dimension.
[0013] In time series analysis, smoothing filtering techniques are used to denoise abnormal fluctuations and remove short-term abrupt changes caused by environmental changes or equipment errors. At the same time, the reliability of the original data stream is verified by calculating the mean and standard deviation.
[0014] The denoised time series data is reorganized in physical time order to generate a physical time parameter sequence with time labels.
[0015] In a preferred embodiment, in the behavior labeling module, the second derivative rate of change of each process parameter is calculated by taking the four types of process parameters (temperature, acid value, pressure, and volatile component concentration) in the physical time parameter sequence and their corresponding time labels as input. The second derivative rate of change is calculated by performing two difference operations on adjacent sampling points in sequence, and outputting the second derivative sequence of each process parameter.
[0016] The calculated value of each process parameter in the second derivative sequence is compared with the preset surge amplitude threshold. The judgment is made in combination with the time period in which the rate of change of the second derivative continuously exceeds the preset surge amplitude threshold. The time period is the time segment that meets the surge amplitude threshold condition within three or more consecutive sampling periods, which constitutes the response surge candidate segment of the process parameter. For the time segment that does not simultaneously meet the surge amplitude and minimum duration conditions, it is marked as a non-surge segment and does not enter the subsequent processing behavior identification process.
[0017] In a preferred embodiment, in the behavior labeling module, each response jump candidate segment is cross-validated with the second derivative sequences of the other three types of process parameters to determine whether there are synchronous jump trends of two or more process parameters within the same time range. If the condition is met, the time segment is confirmed as a response jump segment; if not, the candidate segment is marked as a low-correlation response event.
[0018] The time range corresponding to each confirmed response jump segment is indexed and matched with the physical time parameter sequence. The actual parameter change pattern within the time range is extracted, and processing behavior labels are constructed based on the predefined change patterns in the rule base. The labels are clearly distinguished into temperature surge behavior, acid value shift behavior, pressure shift behavior, and volatile component concentration drastic change behavior. The processing behavior labels are combined with the corresponding time range to construct the event time behavior sequence.
[0019] In a preferred embodiment, in the segment division module, the physical time parameter sequence is divided into multiple continuous sliding segments by setting a sliding segment of fixed length. In each sliding segment, the measurement data of four types of process parameters, namely temperature, acid value, pressure and volatile component concentration, are extracted in that segment. The amplitude change value of each process parameter is calculated, and normalization is performed within its corresponding historical lower and lower limit intervals to obtain multiple normalized amplitude values.
[0020] Based on multiple normalized amplitude values, within each sliding segment, the normalized difference between the change direction of the four types of process parameters and the adjacent sampling points is calculated to form multiple normalized rate of change values. The normalized amplitude values and normalized rate of change values within each sliding segment are combined to generate vector data that represents the parameter response characteristics.
[0021] Within each sliding section, cross-parameter linkage calculation is performed based on vector data. The number of times the change direction of the four types of process parameters is consistent is counted, and the average relative difference of the normalized difference is calculated. Based on the statistical results and calculation results, the normalized coupling response value of the sliding section is generated.
[0022] The normalized amplitude value, normalized rate of change value, and normalized coupled response value corresponding to each sliding segment are used to form feature encoding data, which are then bound according to the time index of the physical time parameter sequence and the event time behavior sequence to construct the physical time trajectory set and the event time trajectory set.
[0023] In a preferred embodiment, in the difference comparison module, by establishing a time index mapping relationship, the physical time trajectory set and the event time trajectory set are paired according to the sliding segment order, and the normalized amplitude value and normalized rate of change value of the four types of process parameters in the corresponding segment are extracted to construct the trajectory pair response vector group. The trajectory pair response vector group contains the physical time response vector and the event time response vector under the same time index.
[0024] For the physical time response vector and the event time response vector in each response vector group, four types of process parameters are constructed, namely, a direction comparison array and a numerical difference array. The direction comparison array consists of the consistency of the change direction of each parameter in the two response vectors, and the numerical difference array consists of the absolute difference between the normalized amplitude value and the normalized rate of change value of each parameter in the two response vectors.
[0025] The response direction consistency index of the sliding segment is obtained by calculating the ratio of the number of parameters with consistent change direction in the direction comparison array to the total number of directions; the normalized response difference of the sliding segment is obtained by weighted averaging the absolute differences corresponding to each parameter in the numerical difference array; and the response start offset of the physical time response vector relative to the event time response vector is calculated according to the time sequence of the normalized change rate to form the response offset delay index.
[0026] The response direction consistency index, normalized response difference, and response offset delay index corresponding to each sliding segment are combined into vectors to construct the response difference vector corresponding to the sliding segment. The response difference vector sequence is then output according to the time index order of the sliding segment.
[0027] In a preferred embodiment, in the risk identification module, by setting a trend evaluation window of fixed length, the response difference vector sequence is divided into multiple continuous trend evaluation windows, and in each trend evaluation window, the response direction consistency index sequence, normalized response difference sequence and response offset delay index sequence of the corresponding sliding segment are extracted to construct the trend evaluation vector.
[0028] In each trend evaluation vector, statistical calculations are performed based on the response direction consistency index sequence to generate the mean, variance, and trend slope values of the sequence; for the normalized response difference sequence and the response offset delay index sequence, their upper limit and the duration of continuous high value segments are extracted respectively to construct a trend stability feature set.
[0029] In a preferred embodiment, in the risk identification module, it is determined whether the following two conditions are met simultaneously in each trend evaluation vector: the mean value of the response direction consistency index is lower than the preset lower limit threshold of direction consistency, and the upper limit value of the response offset delay index exceeds the preset upper limit threshold of response offset delay.
[0030] If both conditions are met, the time segment corresponding to the current trend assessment window will be marked as an abnormal event segment with a surge in response but lacking physical temporal causal support, and this time segment will be constructed as a risk causal chain break region.
[0031] Otherwise, if only one of the above two conditions is met, and the variance of the directional consistency index sequence is higher than the preset volatility upper limit threshold, it is marked as a response trend unstable segment and constructed as a secondary abnormal event record; if none of the above conditions are met, the time segment is constructed as a causal chain structure preservation segment.
[0032] For all abnormal event segments in the trend assessment window that are marked as areas of risk causal chain breakage, establish a mapping relationship between the time index and the original physical time parameter sequence, mark the corresponding segment positions in the original physical time parameter sequence with risk warning labels, and use the risk warning labels as the final output of the monitoring results.
[0033] The technical effects and advantages of this invention are as follows:
[0034] 1. This invention establishes a dual-sequence structure of physical time parameter sequence and event time behavior sequence, generates normalized response vectors in each sliding segment, and constructs a response difference vector sequence across time domains. In this way, it identifies whether non-periodic and discontinuous leap events have causal support in the trend evaluation window, thus solving the problem that traditional systems cannot capture the hidden risks in the processing leap process and cause the causal chain to break.
[0035] 2. This solution extracts three types of feature indicators—response direction consistency index sequence, normalized response difference, and response offset delay index—based on a trend assessment window in the risk identification module. Statistical calculations are performed within each trend assessment window to determine whether there are anomalous event segments with abrupt response increases but lacking physical causal support. Furthermore, by combining the variance of the response direction consistency index sequence with numerical limit conditions, multi-condition hierarchical judgment is performed, comprehensively outputting risk causal chain break regions, secondary anomalous event records, and causal chain structure preservation segments, thus achieving a closed-loop expression from risk identification to structural classification.
[0036] 3. This solution calculates the mean, variance, and slope of the response direction consistency index sequence in each trend assessment window, and extracts the upper limit of the normalized response shift delay index and the length of the high-value segment to construct a set of trend stability features. By comprehensively judging the degree and persistence of trend anomalies, it improves the robustness of identifying risk causal chain breaks in a sudden change environment. Attached Figure Description
[0037] Figure 1 This is a system module diagram of the present invention.
[0038] Figure 2 This is a flowchart of the parameter acquisition and behavior recognition process of the present invention.
[0039] Figure 3 This is a flowchart of the trajectory generation and response difference evaluation process of the present invention.
[0040] Figure 4 This is a flowchart of the trend assessment and risk labeling process for this invention. Detailed Implementation
[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0042] Refer to the instruction manual appendix Figure 1-4 An embodiment of the present invention provides an edible oil processing monitoring system, comprising a parameter acquisition module, a behavior labeling module, a segmentation module, a difference comparison module, and a risk identification module.
[0043] The parameter acquisition module continuously collects data on temperature, acid value, pressure, and volatile component concentration during the edible oil processing by setting a fixed sampling period, acquires raw data streams, and organizes the raw data streams into a physical time parameter sequence according to the sampling order;
[0044] The behavior labeling module performs second derivative rate of change calculation on each type of parameter in the physical time parameter sequence, and identifies response jump segments based on the sudden rise segments of the second derivative rate of change. By labeling the corresponding processing behavior type in each response jump segment, an event time behavior sequence is constructed.
[0045] The segmentation module divides the physical time parameter sequence and the event time behavior sequence into equal-length sliding segments, and extracts the parameter normalization amplitude, normalization rate of change value and normalization coupling response value in each sliding segment. After performing sequence encoding, it generates physical time trajectory set and event time trajectory set respectively.
[0046] The difference comparison module compares the corresponding segments of the physical time trajectory set with the event time trajectory set, calculates the response direction consistency index, normalized response difference, and response offset delay index for each pair of trajectories under the same sliding segment, and generates a response difference vector sequence.
[0047] The risk identification module analyzes the consistency decline trend and offset delay clustering pattern in the response difference vector sequence to identify abnormal event segments with response jumps but lacking physical temporal causal support. It determines that the abnormal event segment is a risk causal chain break area and outputs risk warning information.
[0048] In the parameter acquisition module, by setting a fixed sampling period, the raw data streams of four process parameters—temperature, acid value, pressure, and volatile component concentration—are collected from the edible oil processing process. The acquisition time point of each process parameter is marked by a timestamp, and the corresponding measurement value and timestamp information are retained in the raw data stream record.
[0049] The raw data stream of process parameters is sorted according to the timestamps within the sampling period. Interpolation algorithms are used to fill in the missing measurement values due to sampling gaps. The interpolation algorithms include linear interpolation, spline interpolation, or polynomial interpolation. The most suitable interpolation method is automatically selected according to the size of the sampling gap. Each measurement value is assigned a corresponding time label, so that the raw data stream forms a continuous time series according to the time dimension.
[0050] In time series, smoothing filtering techniques are used to denoise abnormal fluctuations and remove short-term abrupt changes caused by environmental changes or equipment errors. At the same time, the reliability of the original data stream is verified by calculating the mean and standard deviation to ensure that the original data stream meets the set error range. The error range is determined by the statistical regularity of historical data or by setting a fault tolerance threshold.
[0051] The denoised time series data is reorganized in physical time order to generate a physical time parameter sequence with time labels.
[0052] In the behavior labeling module, the four types of process parameters in the physical time parameter sequence, namely temperature, acid value, pressure and volatile component concentration, and their corresponding time labels are used as input. The second derivative rate of change of each type of process parameter is calculated. The second derivative rate of change is calculated by performing two difference operations on adjacent sampling points in sequence, and the second derivative sequence of each type of process parameter is output.
[0053] The calculated value of each process parameter in the second derivative sequence is compared with the preset surge amplitude threshold. The judgment is made in conjunction with the time period in which the rate of change of the second derivative continuously exceeds the preset surge amplitude threshold. The time period is the time segment that meets the surge amplitude threshold condition within three or more consecutive sampling periods. The sampling period is a fixed time interval set according to the process. This constitutes the response surge candidate segment of the process parameter. For the time segment that does not simultaneously meet the surge amplitude and minimum duration conditions, it is marked as a non-surge segment and does not enter the subsequent processing behavior identification process.
[0054] In the behavior labeling module, each candidate segment of response jump is cross-validated with the second derivative sequences of the other three types of process parameters to determine whether there are synchronous jump trends of two or more process parameters within the same time range. If the condition is met, the time segment is confirmed as a response jump segment. If not, the candidate segment is marked as a low-correlation response event and recorded as a secondary event in the event time behavior sequence to support subsequent model adaptive adjustment and confidence learning.
[0055] The time range corresponding to each confirmed response jump segment is indexed and matched with the physical time parameter sequence to extract the actual parameter change pattern within the time range. Processing behavior labels are constructed based on the predefined change patterns in the rule base. The labels are clearly distinguished into temperature surge behavior, acid value shift behavior, pressure upward shift behavior, and volatile component concentration drastic change behavior. The processing behavior labels are combined with the corresponding time range to construct the event time behavior sequence for subsequent event time modeling and risk trend identification.
[0056] It should be noted that the rule base is used to define the processing behavior labels corresponding to the response jump segment. It is constructed based on the evolution characteristics of four types of process parameters—temperature, acid value, pressure, and volatile component concentration—during historical processing. By extracting the normalized trend trajectory of each parameter in typical processing behaviors, a behavior feature prototype is constructed and converted into a set of rules containing constraints such as amplitude threshold, duration, and direction of change. Subsequently, after identifying the response jump segment, the physical time parameter sequences of temperature, acid value, pressure, and volatile component concentration within the segment are extracted to construct its normalized change trajectory vector group. The trajectory vector group is then used as input to perform sequence fitting with the feature prototypes of each processing behavior in the rule base. The corresponding processing behavior label is determined by calculating the fitting confidence score.
[0057] In the segment division module, by setting a sliding segment of fixed length, the physical time parameter sequence is divided into multiple continuous sliding segments. Within each sliding segment, the measurement data of four types of process parameters, namely temperature, acid value, pressure and volatile component concentration, are extracted. The amplitude change value of each process parameter is calculated, and normalization is performed within its corresponding historical lower and lower limit intervals to obtain multiple normalized amplitude values.
[0058] Based on multiple normalized amplitude values, within each sliding segment, the normalized difference between the change direction of the four types of process parameters and the adjacent sampling points is calculated to form multiple normalized rate of change values. The normalized amplitude values and normalized rate of change values within each sliding segment are combined to generate vector data that represents the parameter response characteristics.
[0059] Within each sliding section, cross-parameter linkage calculation is performed based on vector data. The number of times the four types of process parameters change in the same direction is counted, and the average relative difference of the normalized difference is calculated. Based on the statistical results and calculation results, the normalized coupling response value of the sliding section is generated to characterize the linkage response characteristics of the four types of process parameters in the current section.
[0060] The normalized amplitude value, normalized rate of change value, and normalized coupled response value corresponding to each sliding segment are used to form feature encoding data. These data are then bound according to the time index of the physical time parameter sequence and the event time behavior sequence to construct a physical time trajectory set and an event time trajectory set, which are used to support subsequent difference analysis and risk identification calculations.
[0061] In the difference comparison module, by establishing a time index mapping relationship, the physical time trajectory set and the event time trajectory set are paired according to the sliding segment order, and the normalized amplitude value and normalized rate of change value of the four types of process parameters in the corresponding segment are extracted to construct the trajectory pair response vector group. The trajectory pair response vector group contains the physical time response vector and the event time response vector under the same time index.
[0062] For the physical time response vector and the event time response vector in each response vector group, four types of process parameters are constructed, namely, a direction comparison array and a numerical difference array. The direction comparison array consists of the consistency of the change direction of each parameter in the two response vectors, and the numerical difference array consists of the absolute difference between the normalized amplitude value and the normalized rate of change value of each parameter in the two response vectors.
[0063] The response direction consistency index of the sliding segment is obtained by calculating the ratio of the number of parameters with consistent change direction in the direction comparison array to the total number of directions; the normalized response difference of the sliding segment is obtained by weighted averaging the absolute differences corresponding to each parameter in the numerical difference array; and the response start offset of the physical time response vector relative to the event time response vector is calculated according to the time sequence of the normalized change rate to form the response offset delay index.
[0064] The response direction consistency index, normalized response difference, and response offset delay index corresponding to each sliding segment are vectorized to construct the response difference vector corresponding to the sliding segment. The response difference vector sequence is output according to the time index order of the sliding segment to support subsequent risk causal chain determination and trend evolution modeling.
[0065] In the risk identification module, by setting a trend evaluation window of fixed length, the response difference vector sequence is divided into multiple continuous trend evaluation windows. In each trend evaluation window, the response direction consistency index sequence, normalized response difference sequence and response offset delay index sequence of the corresponding sliding segment are extracted to construct the trend evaluation vector.
[0066] In each trend evaluation vector, statistical calculations are performed based on the response direction consistency index sequence to generate the mean, variance, and trend slope values of the sequence; for the normalized response difference sequence and the response offset delay index sequence, their upper limit and the duration of continuous high value segments are extracted respectively to construct a trend stability feature set.
[0067] In the risk identification module, it is determined whether the following two conditions are met simultaneously in each trend evaluation vector: the mean value of the response direction consistency index is lower than the preset lower limit threshold of direction consistency, and the upper limit value of the response offset delay index exceeds the preset upper limit threshold of response offset delay.
[0068] If both conditions are met, the time segment corresponding to the current trend assessment window will be marked as an abnormal event segment with a surge in response but lacking physical temporal causal support, and this time segment will be constructed as a risk causal chain break region.
[0069] Otherwise, if only one of the above two conditions is met, and the variance of the directional consistency index sequence is higher than the preset volatility upper limit threshold, it is marked as a response trend unstable segment and constructed as a secondary abnormal event record; if none of the above conditions are met, the time segment is constructed as a causal chain structure preservation segment and excluded from the subsequent risk labeling process.
[0070] For all abnormal event segments in the trend assessment window that are marked as risk causal chain break areas, establish a mapping relationship between the time index and the original physical time parameter sequence, mark the corresponding segment positions in the original physical time parameter sequence with risk warning labels, and use the risk warning labels as the final output of the monitoring results.
[0071] It should be noted that in the formula structure involved in this scheme, dimensionless terms can be used as proportional or structural adjustment factors. When combined with quantities with units, they only play a role in numerical scaling and do not introduce new physical dimensions. Therefore, they will not change or confuse the overall unit system. This combination of "dimensionless terms and terms with units" can be understood as a composite structural expression commonly used in mathematical physics modeling. It conforms to the principle of dimensional consistency and has a clear physical interpretation basis.
[0072] Secondly, in the formula structure of this scheme, if multiple variables with different physical units are involved, including but not limited to time, mass or energy variables, their joint appearance is to express the collaborative modeling relationship of multiple physical mechanisms. Each variable can form a unified structure through function mapping, ratio combination or normalization adjustment, with clear units and clear meaning. The overall expression conforms to the principle of dimensional consistency and the conventional formula of engineering modeling.
[0073] In this solution, constants, weights, adjustment factors, threshold parameters, proportional coefficients, etc., are all adjustable control parameters for different application environments. Their values depend on the target equipment configuration, data input characteristics, and performance optimization goals. During the implementation phase, they are set to converge within a reasonable range through model verification, performance constraints, or engineering calibration. Although these parameters do not have a unique preset value, they have clear adjustment logic and calculation paths. They belong to the deterministic setting process in engineering implementation. The purpose of this setting is to ensure that the solution is both universally adaptable and reproducible and operable, without affecting its technical clarity and feasibility.
[0074] definition This is the trend evaluation vector, and the trend evaluation vector is in the sliding segment. It consists of three sub-indicators constitute,
[0075]
[0076]
[0077]
[0078]
[0079]
[0080] in, For trend assessment vector, This represents the comprehensive response characteristics within each sliding segment. The comprehensive response characteristics include response direction consistency, normalized response difference, and response offset delay. To respond to the consistency index of direction, Indicates the sliding section Within, the proportion of all process parameters changing in the same direction; For sliding section The number of sampling points with the same direction of change within the sample; For sliding section Total number of sampling points within; The normalized response difference index Indicates the sliding section Within, the difference in response magnitude between physical time and event time;
[0081] For the physical time trajectory at a point in time The normalized value; For the event timeline at time points The normalized value; For sliding section The total number of sampling points within; The response offset delay metric measures the offset between physical time and event time. The time offset from the start of the response event; The time offset at which the response event ends; This is an indicator of the region where the causal chain of risk is broken. As a weighting factor, Used to quantify the weight of each indicator in risk assessment;
[0082] Furthermore, in In the formula, the consistency of the changing direction of various process parameters within the sliding zone is calculated. Its value ranges between 0 and 1; a higher value indicates a more consistent changing direction of the process parameters within that zone, and a more stable response process. The formula calculates the difference in response magnitude between physical time and event time. By measuring the difference between the two, it reflects the synchronicity of the response process. The lower the value, the more consistent the response between physical time and event time.
[0083] exist In the formula, the response offset between physical time and event time is calculated. The smaller the offset, the stronger the causal relationship between physical time and event time. In the formula, the comprehensive consideration is then used to determine the final result. , and Three indicators are used to determine whether a time period is a region where the causal chain is broken. When the value is 1, it indicates that the segment is identified as a region where the risk causal chain is broken; when it is 0, it indicates that the segment is risk-free.
[0084] In practical implementation, this scheme continuously collects four process parameters during the edible oil processing process—temperature, acid value, pressure, and volatile component concentration—by setting a fixed sampling period. The raw data stream is then organized into a physical time parameter sequence according to the sampling order. Next, the behavior labeling module performs second derivative rate of change calculation on each type of process parameter to identify response jump segments and label the corresponding processing behavior types. Finally, an event time behavior sequence corresponding to the physical time parameter sequence is constructed. Then, in the segment division module, the physical time parameter sequence and the event time behavior sequence are divided into equal-length sliding segments, and the normalized amplitude, normalized rate of change, and normalized coupled response value of each sliding segment are extracted. The physical time trajectory set and the event time trajectory set are generated through sequence encoding.
[0085] In the difference comparison module, corresponding segments of the physical time trajectory set and the event time trajectory set are compared. The response direction consistency index, normalized response difference and response offset delay index of each pair of trajectories are calculated to generate a response difference vector sequence. Through the risk identification module, the consistency decline trend and offset delay clustering pattern in the response difference vector sequence are analyzed to identify abnormal event segments with response jumps but lack physical time causal support, and these segments are identified as risk causal chain break areas, and risk warning information is output.
[0086] This solution considers the changing trends over multiple time periods, the interrelationships between different parameters, and whether there is a causal relationship between processing behavior and parameter changes. This avoids misjudging short-term fluctuations, missing slow anomalies, and misidentifying invalid events during risk assessment. When the system identifies inconsistent parameter changes, significant delays, and non-compliant causal relationships within a certain time period, it marks this as a risk segment and writes this mark back to the original parameter record as part of the risk assessment results. This allows operators to directly see which time period is at risk, facilitating timely equipment adjustments, operational corrections, or inspections. This enables more accurate problem detection and improves the monitoring effectiveness and operational safety of the entire edible oil processing process.
[0087] In conclusion, 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. An edible oil processing monitoring system, comprising a parameter acquisition module, a behavior labeling module, a segmentation module, a difference comparison module, and a risk identification module, characterized in that: The parameter acquisition module continuously collects data on temperature, acid value, pressure, and volatile component concentration during the edible oil processing by setting a fixed sampling period, acquires raw data streams, and organizes the raw data streams into a physical time parameter sequence according to the sampling order; The behavior labeling module performs second derivative rate of change calculation on each type of parameter in the physical time parameter sequence, and identifies response jump segments based on the sudden rise segments of the second derivative rate of change. By labeling the corresponding processing behavior type in each response jump segment, an event time behavior sequence is constructed. The segmentation module divides the physical time parameter sequence and the event time behavior sequence into equal-length sliding segments, and extracts the parameter normalization amplitude, normalization rate of change value and normalization coupling response value in each sliding segment. After performing sequence encoding, it generates physical time trajectory set and event time trajectory set respectively. The difference comparison module compares the corresponding segments of the physical time trajectory set with the event time trajectory set, calculates the response direction consistency index, normalized response difference, and response offset delay index for each pair of trajectories under the same sliding segment, and generates a response difference vector sequence. The risk identification module analyzes the consistency decline trend and offset delay clustering pattern in the response difference vector sequence to identify abnormal event segments with response jumps but lacking physical temporal causal support. It determines that the abnormal event segment is a risk causal chain break area and outputs risk warning information.
2. The edible oil processing monitoring system according to claim 1, characterized in that: In the parameter acquisition module, by setting a fixed sampling period, the raw data streams of four process parameters—temperature, acid value, pressure, and volatile component concentration—are collected from the edible oil processing process. The acquisition time point of each process parameter is marked by a timestamp, and the corresponding measurement value and timestamp information are retained in the raw data stream record. The raw data stream of process parameters is sorted according to the timestamps within the sampling period. Interpolation algorithms are used to fill in the missing measurement values due to sampling gaps, and each measurement value is assigned a corresponding time label, so that the raw data stream forms a continuous time series according to the time dimension. In time series analysis, smoothing filtering techniques are used to denoise abnormal fluctuations and remove short-term abrupt changes caused by environmental changes or equipment errors. At the same time, the reliability of the original data stream is verified by calculating the mean and standard deviation. The denoised time series data is reorganized in physical time order to generate a physical time parameter sequence with time labels.
3. The edible oil processing monitoring system according to claim 2, characterized in that: In the behavior labeling module, the four types of process parameters in the physical time parameter sequence, namely temperature, acid value, pressure and volatile component concentration, and their corresponding time labels are used as input. The second derivative rate of change of each type of process parameter is calculated. The second derivative rate of change is calculated by performing two difference operations on adjacent sampling points in sequence, and the second derivative sequence of each type of process parameter is output. The calculated value of each process parameter in the second derivative sequence is compared with the preset surge amplitude threshold. The judgment is made in combination with the time period in which the rate of change of the second derivative continuously exceeds the preset surge amplitude threshold. The time period is the time segment that meets the surge amplitude threshold condition within three or more consecutive sampling periods, which constitutes the response surge candidate segment of the process parameter. For the time segment that does not simultaneously meet the surge amplitude and minimum duration conditions, it is marked as a non-surge segment and does not enter the subsequent processing behavior identification process.
4. The edible oil processing monitoring system according to claim 3, characterized in that: In the behavior labeling module, each response jump candidate segment is cross-validated with the second derivative sequences of the other three types of process parameters to determine whether there are synchronous jump trends of two or more process parameters within the same time range. If the condition is met, the time segment is confirmed as a response jump segment; if not, the candidate segment is marked as a low-correlation response event. The time range corresponding to each confirmed response jump segment is indexed and matched with the physical time parameter sequence. The actual parameter change pattern within the time range is extracted, and processing behavior labels are constructed based on the predefined change patterns in the rule base. The labels are clearly distinguished into temperature surge behavior, acid value shift behavior, pressure shift behavior, and volatile component concentration drastic change behavior. The processing behavior labels are combined with the corresponding time range to construct the event time behavior sequence.
5. The edible oil processing monitoring system according to claim 4, characterized in that: In the segment division module, by setting a sliding segment of fixed length, the physical time parameter sequence is divided into multiple continuous sliding segments. Within each sliding segment, the measurement data of four types of process parameters, namely temperature, acid value, pressure and volatile component concentration, are extracted. The amplitude change value of each process parameter is calculated, and normalization is performed within its corresponding historical lower and lower limit intervals to obtain multiple normalized amplitude values. Based on multiple normalized amplitude values, within each sliding segment, the normalized difference between the change direction of the four types of process parameters and the adjacent sampling points is calculated to form multiple normalized rate of change values. The normalized amplitude values and normalized rate of change values within each sliding segment are combined to generate vector data that represents the parameter response characteristics. Within each sliding section, cross-parameter linkage calculation is performed based on vector data. The number of times the change direction of the four types of process parameters is consistent is counted, and the average relative difference of the normalized difference is calculated. Based on the statistical results and calculation results, the normalized coupling response value of the sliding section is generated. The normalized amplitude value, normalized rate of change value, and normalized coupled response value corresponding to each sliding segment are used to form feature encoding data, which are then bound according to the time index of the physical time parameter sequence and the event time behavior sequence to construct the physical time trajectory set and the event time trajectory set.
6. The edible oil processing monitoring system according to claim 5, characterized in that: In the difference comparison module, by establishing a time index mapping relationship, the physical time trajectory set and the event time trajectory set are paired according to the sliding segment order, and the normalized amplitude value and normalized rate of change value of the four types of process parameters in the corresponding segment are extracted to construct the trajectory pair response vector group. The trajectory pair response vector group contains the physical time response vector and the event time response vector under the same time index. For the physical time response vector and the event time response vector in each response vector group, four types of process parameters are constructed, namely, a direction comparison array and a numerical difference array. The direction comparison array consists of the consistency of the change direction of each parameter in the two response vectors, and the numerical difference array consists of the absolute difference between the normalized amplitude value and the normalized rate of change value of each parameter in the two response vectors. The response direction consistency index of the sliding segment is obtained by calculating the ratio of the number of parameters with consistent change direction in the direction comparison array to the total number of directions; the normalized response difference of the sliding segment is obtained by weighted averaging the absolute differences corresponding to each parameter in the numerical difference array; and the response start offset of the physical time response vector relative to the event time response vector is calculated according to the time sequence of the normalized change rate to form the response offset delay index. The response direction consistency index, normalized response difference, and response offset delay index corresponding to each sliding segment are combined into vectors to construct the response difference vector corresponding to the sliding segment. The response difference vector sequence is then output according to the time index order of the sliding segment.
7. The edible oil processing monitoring system according to claim 6, characterized in that: In the risk identification module, by setting a trend evaluation window of fixed length, the response difference vector sequence is divided into multiple continuous trend evaluation windows. In each trend evaluation window, the response direction consistency index sequence, normalized response difference sequence and response offset delay index sequence of the corresponding sliding segment are extracted to construct the trend evaluation vector. In each trend assessment vector, statistical calculations are performed based on the response direction consistency index sequence to generate the mean, variance, and trend slope values of the sequence. For the normalized response difference sequence and the response offset delay index sequence, extract their upper limit and the duration of continuous high value segments respectively, and construct a set of trend stability features.
8. The edible oil processing monitoring system according to claim 7, characterized in that: In the risk identification module, it is determined whether the following two conditions are met simultaneously in each trend evaluation vector: the mean value of the response direction consistency index is lower than the preset lower limit threshold of direction consistency, and the upper limit value of the response offset delay index exceeds the preset upper limit threshold of response offset delay. If both conditions are met, the time segment corresponding to the current trend assessment window will be marked as an abnormal event segment with a surge in response but lacking physical temporal causal support, and this time segment will be constructed as a risk causal chain break region. Otherwise, if only one of the above two conditions is met, and the variance of the directional consistency index sequence is higher than the preset volatility upper limit threshold, it is marked as a response trend unstable segment and constructed as a secondary abnormal event record; if none of the above conditions are met, the time segment is constructed as a causal chain structure preservation segment. For all abnormal event segments in the trend assessment window that are marked as areas of risk causal chain breakage, establish a mapping relationship between the time index and the original physical time parameter sequence, mark the corresponding segment positions in the original physical time parameter sequence with risk warning labels, and use the risk warning labels as the final output of the monitoring results.