Feed proportioning abnormity identification method and system

Abstract

The invention discloses a feed proportion abnormity identification method and system, and particularly relates to the field of feed proportion abnormity identification, and the method comprises the steps: extracting the disturbance variables of vibration amplitude, feeding speed and voiceprint waveform in multi-source disturbance data, constructing the disturbance time sequence expression of a frequency, form and amplitude triple feature structure, and carrying out the recognition of the vibration amplitude, the feeding speed and the voiceprint waveform. Extracting a disturbance characteristic track with center characterization capability in a cluster through atlas clustering, and writing the disturbance characteristic track into a track benchmark expression library; constructing an aligned trajectory expression set by analyzing the frequency domain structure and the time slice expression of each channel signal in the current batch of disturbance data and combining the binding relationship between the batch number and the formula number; by constructing an offset matching and track consistency judgment mechanism based on multi-source disturbance track expression, structured recognition and path logic verification of a formula drift phenomenon are realized, and the problem that the actual feed nutrition state is misjudged due to the fact that real proportion offset caused by residual feed cannot be recognized is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of feed ratio abnormality identification, and more particularly, to a feed ratio abnormality identification method and system. Background Art

[0002] In centralized feed production processes, there is a hidden risk of starting a new formula before completely clearing the stockpile after batch switching, resulting in residual materials from the previous batch mixing with the current formula, causing "formula drift" problems;

[0003] This type of anomaly can easily escape conventional detection mechanisms in the production system because industrial data such as recipe numbers, material codes, and scheduling logs all appear normal, causing the system to misjudge the actual ratio status at the industrial data processing level, ultimately leading to animals consuming feed that does not meet nutritional standards for a long time and causing hidden health damage. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a feed ratio anomaly identification method and system. By constructing an offset matching and trajectory consistency judgment mechanism based on multi-source disturbance trajectory expression, structured identification of recipe drift phenomenon and path logic verification are achieved, thereby solving the problem that the system cannot identify the actual ratio deviation caused by residual materials when industrial data such as the recipe number shows normal, which leads to misjudgment of the actual nutritional status of the feed.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for identifying abnormal feed ratio, comprising:

[0006] By extracting disturbance variables such as vibration amplitude, feeding speed, and soundprint waveform from multi-source disturbance data, a disturbance time series expression with a triple feature structure of frequency, morphology, and amplitude is constructed. Then, through graphical clustering, the disturbance feature trajectory with central representation capability in the cluster is extracted and written into the trajectory benchmark expression library.

[0007] By analyzing the frequency domain structure and time segment expression of each channel signal in the current batch of disturbance data, combined with the binding relationship between the batch number and the recipe number, an aligned trajectory expression set is constructed. Trajectory shape and amplitude offset matching is performed based on a sliding window, generating a one-to-one mapping relationship between the current batch disturbance trajectory and the trajectory benchmark, as well as the corresponding disturbance matching offset structure.

[0008] By clustering the deviation results between the current batch disturbance trajectory and the trajectory benchmark according to the disturbance type and constructing a disturbance deviation aggregation map, combined with the trajectory trend change characteristics and scheduling path constraints, trajectory consistency judgment and path switching consistency judgment are performed to form a structural level judgment result of the recipe drift behavior;

[0009] By jointly annotating the abnormal trajectory sections with the recipe abnormal sections, locating the boundary structure of the abnormal trajectory in the disturbance feature trajectory and performing correction, an updated trajectory benchmark expression library is constructed. At the same time, trajectory samples containing ratio drift features are extracted, the recognition strategy is reconstructed, and the method's fault-tolerant response capability to abnormal offsets is enhanced.

[0010] In a preferred embodiment, historical disturbance data is obtained, including equipment structure vibration data, feed path flow rate data, and silo wall soundprint data. Time normalization and dimension alignment processing are performed on the historical disturbance data to generate a multi-source historical disturbance data sequence.

[0011] Extract disturbance variables from multi-source historical disturbance data sequences. These disturbance variables include equipment vibration amplitude, which characterizes changes in mechanical structure state; feed speed, which reflects material conveying stability; and soundprint waveform, which identifies the characteristics of silo wall impact reverberation. These three types of disturbance variables are then reconstructed in the time dimension as vibration amplitude time series, feed speed time series, and soundprint waveform time series.

[0012] Frequency decomposition operations are performed on the vibration amplitude time series, feed speed time series, and soundprint waveform time series in sequence. The frequency band boundaries are delineated according to the frequency change intervals of various types. Then, the time segments are divided into fixed step sizes within each frequency band to extract the set of disturbance trajectory segments formed under the conditions of frequency consistency and energy stability.

[0013] In a preferred embodiment, morphological feature extraction and amplitude distribution calculation are performed sequentially for each type of time segment in the disturbance trajectory segment set. The morphological features include the waveform envelope structure, local oscillation boundary points, and repetitive period segments of the time segment on the time axis. The amplitude distribution includes the local peak amplitude, the average amplitude level of the entire segment, and the amplitude change rate within the time segment. A disturbance feature spectrum set is generated by mapping the vibration amplitude time series, the feed speed time series, and the voiceprint waveform time series respectively.

[0014] For all time segments mapped in the disturbance feature map set, feature space reconstruction and multidimensional clustering analysis are performed based on the morphological contour curvature and amplitude density distribution structure of the corresponding time segments. The most representative disturbance feature trajectory is extracted from the clustering results of each time segment as a matching reference sample and written into the trajectory benchmark expression library.

[0015] In a preferred embodiment, current disturbance data is collected, and the current disturbance data includes real-time vibration response data, feeding operation sound spectrum data and silo cavity echo data;

[0016] Based on the timestamp and task registration identifier in the acquisition process, the batch number corresponding to the current disturbance data is extracted, and then the recipe number registered for the current batch is parsed based on the batch number to establish a binding relationship between the batch number and the recipe number;

[0017] Then, multi-dimensional compression and feature mapping are performed on the current disturbance data to generate the current disturbance expression vector group. The real-time vibration response data, feeding operation sound spectrum data, and silo cavity echo data in the current disturbance expression vector group are then subjected to frequency domain restoration and time window partitioning operations in sequence to reconstruct each type of data into a set of disturbance time segments arranged in chronological order.

[0018] Each time segment is bound to the batch number extracted from the current disturbance data, and finally all the time segments bound to the batch number are aggregated to construct the trajectory expression set corresponding to the current batch.

[0019] In a preferred embodiment, a trajectory shape matching operation within a sliding time window is performed on each time segment in the trajectory expression set and the disturbance feature trajectory in the trajectory reference expression library, and a one-to-one correspondence between each time segment and its corresponding disturbance feature trajectory is constructed to generate a trajectory matching relationship set;

[0020] For each time segment in the trajectory expression set and its corresponding disturbance feature trajectory, perform shape curvature difference calculation and time alignment offset solution, extract the initial matching offset value of each pair of time segments on the trajectory geometry, and construct a preliminary disturbance offset response structure;

[0021] Based on the preliminary disturbance offset response structure, trajectory-guided coincidence analysis and amplitude response offset rate evaluation are further performed on each time segment in the trajectory expression set and the corresponding disturbance feature trajectory. The difference indicators of shape change trend and energy response characteristics of each time segment are extracted to construct a complete disturbance matching offset result. All disturbance matching offset results are summarized in chronological order to generate a disturbance matching offset vector set.

[0022] In a preferred embodiment, each disturbance matching offset result in the disturbance matching offset vector set is divided into three types of disturbances: vibration, flow rate, and voiceprint, based on its original source in the trajectory expression set. Clustering and aggregation operations are performed according to the disturbance type to construct a disturbance offset aggregation map. Based on the disturbance offset aggregation map, normalized threshold analysis and sliding trend extraction are performed to generate a disturbance offset trend judgment result.

[0023] Determine whether the disturbance offset trend judgment results meet the condition that the offset degree is lower than the preset tolerance threshold under all three types of disturbance types. If all conditions are met, a trajectory consistency judgment result is generated, indicating that the current batch disturbance state is consistent with the trajectory benchmark;

[0024] If the deviation degree of any disturbance type in the disturbance deviation trend judgment result exceeds the corresponding tolerance threshold in a continuous time segment, and the deviation degree shows a monotonically increasing trend, then it is determined that the disturbance type has a persistent anomaly and a trajectory drift judgment result is generated.

[0025] In a preferred embodiment, based on the binding relationship established between the batch number and the recipe number, the switching path information related to the recipe number registered for the current batch is extracted from the task registration record;

[0026] The switching path information includes the trajectory offset mapping relationship between the recipe number registered in the current batch and the recipe number registered in the previous batch, as well as the recipe switching time interval between the recipe number registered in the current batch and the recipe number registered in the previous batch as marked in the task registration record;

[0027] The trajectory drift determination result is jointly compared with the switching path information, and the path consistency determination step is performed to evaluate whether the current trajectory state meets the recipe switching logic requirements;

[0028] In the path consistency judgment, determine whether any of the following structural anomalies exist: there is no valid switching relationship registered between the recipe number registered for the current batch and the recipe number of the previous round in the switching path parameters, the execution time range corresponding to the recipe number of the previous round overlaps with the switching time interval, or the current recipe number and switching time combination does not meet the switching sequence rules set in the trajectory reference expression library;

[0029] If any of the above anomalies exists, a final judgment result is generated that the recipe drift behavior is established, indicating that the current trajectory state has constituted a consistency conflict in the scheduling path structure.

[0030] In a preferred embodiment, the time segments in the trajectory expression set where the disturbance type involved in the matching trajectory drift judgment result forms a monotonically increasing structure in the offset trend are marked as trajectory abnormal segments. At the same time, the time segments covered by the recipe switching time interval corresponding to the recipe drift behavior judgment result in the trajectory expression set are regarded as recipe abnormal segments.

[0031] Generate abnormal state labels based on the abnormal state type and time segment information corresponding to the trajectory abnormal segment and the recipe abnormal segment, and summarize them to form an abnormal behavior feedback label set;

[0032] Extract the disturbance feature trajectory associated with the corresponding matching time segment in the trajectory expression set from the abnormal time segment indicated by each abnormal state label in the abnormal behavior feedback label set; use the disturbance feature trajectory as the reference trajectory for trajectory correction, analyze the boundary position of the abnormal time segment in the corresponding disturbance feature trajectory, perform boundary extension and expression content update processing, and generate a corrected trajectory reference template set;

[0033] The modified trajectory benchmark template set is written into the trajectory benchmark expression library as a whole, and the current version information is synchronously annotated to generate a trajectory expression version sequence arranged in chronological order.

[0034] In a preferred embodiment, the abnormal state type and time segment information corresponding to the abnormal behavior feedback tag set are combined with the recipe number information registered for the current batch to construct a behavior archive record containing the recipe abnormal segment trajectory positioning information, the recipe switching path offset structure, and the disturbance feature trajectory revision process;

[0035] Extract trajectory revision samples containing recipe abnormality segments and trajectory abnormality segment identification information from the behavior archive record, where the recipe abnormality segment is the time segment corresponding to the recipe switching time interval in the trajectory expression set, and the trajectory abnormality segment is the continuous time segment in the trajectory expression set that forms a monotonic offset trend;

[0036] Taking trajectory revision samples as training input sets, parameter structure update and evolutionary recognition strategy reconstruction based on anomaly feedback information are performed to enhance the fault tolerance of the ratio anomaly recognition method.

[0037] In a preferred embodiment, a feed ratio abnormality identification system includes a track raising and database building module, a deconstruction and matching module, a bias and drift judgment module, and a standard correction and evolution module;

[0038] The track extraction and database construction module extracts disturbance variables such as vibration amplitude, feed speed, and soundprint waveform from multi-source disturbance data, constructs a disturbance time series expression with a triple feature structure of frequency, morphology, and amplitude, and extracts disturbance feature trajectories with central representation capabilities in the cluster through graphical clustering. These are written into the trajectory reference expression library and serve as a static comparison reference for subsequent trajectory matching and disturbance offset evaluation.

[0039] The deconstruction matching module analyzes the frequency domain structure and time segment expression of each channel signal in the current batch of disturbance data, combines the binding relationship between the batch number and the recipe number, constructs an aligned trajectory expression set, and performs trajectory shape and amplitude offset matching based on a sliding window. It generates a one-to-one mapping relationship between the current batch disturbance trajectory and the trajectory benchmark, as well as the corresponding disturbance matching offset structure.

[0040] The drift detection module clusters the deviation results between the current batch disturbance trajectory and the trajectory benchmark by disturbance type and constructs a disturbance deviation aggregation map. It then combines trajectory trend change characteristics with scheduling path constraints to perform trajectory consistency judgment and path switching consistency judgment, forming a structural-level judgment result on the recipe drift behavior, which serves as the core basis for determining whether the ratio is abnormal.

[0041] The calibration and correction evolution module jointly labels the abnormal trajectory sections with the recipe abnormal sections, locates the boundary structure of the abnormal trajectory in the disturbance feature trajectory and performs correction, constructs an updated trajectory benchmark expression library, and extracts trajectory samples containing ratio drift features to reconstruct the recognition strategy and enhance the method's fault-tolerant response capability to abnormal offsets.

[0042] Technical effects and advantages of the present invention:

[0043] 1. By constructing the frequency-morphology-amplitude triple trajectory features of multi-source disturbance variables such as vibration amplitude, feeding speed, and soundprint waveform, and writing them into the trajectory reference expression library, it is possible to model the structured abnormal state of "normal recipe number but actual material has drifted", effectively filling the detection blind spots of traditional industrial data processing;

[0044] 2. By constructing a trajectory representation set in the current batch data and introducing a sliding window trajectory alignment mechanism, a structural mapping relationship between the perturbation trajectory and the reference trajectory is generated. This captures abnormal offsets from two dimensions: time series segments and trajectory shapes, improving the recognition process's adaptability to the ambiguity of recipe switching timing.

[0045] 3. By clustering disturbance-matching offset vectors by type and constructing an offset aggregation graph, trend differences are extracted for multiple disturbance types, such as vibration, soundprint, and flow rate. This is combined with dispatch path information to perform path consistency judgment, resulting in anomaly determination results with structural decision-making capabilities, achieving closed-loop verification from trajectory offset to recipe behavior logic.

[0046] 4. By jointly annotating the abnormal trajectory sections and the abnormal recipe sections and feeding them back to form trajectory revision samples, an evolutionary strategy recognition model is constructed. After the abnormal state is identified and located, the trajectory benchmark expression library is continuously updated to enhance the system's adaptive evolution capabilities in fault-tolerant training and model robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The figure is a flow chart of the method steps of the present invention.

[0048] Figure 2 Schematic diagram of the system module of the present invention.

[0049] Figure 3 This is a flow chart of trajectory reference extraction of the present invention.

[0050] Figure 4 Build and calculate the offset graph for the current trajectory of the present invention.

[0051] Figure 5 This is a drift judgment and path consistency verification diagram of the present invention.

[0052] Figure 6 This is the trajectory revision and identification strategy evolution diagram of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] Refer to the instruction manual Figure 1-6 According to one embodiment of the present invention, a method and system for identifying abnormal feed ratios include:

[0055] By extracting disturbance variables such as vibration amplitude, feeding speed, and soundprint waveform from multi-source disturbance data, a disturbance time series expression with triple feature structures of frequency, morphology, and amplitude is constructed. The disturbance feature trajectory with central representation capability in the cluster is extracted through graphical clustering and written into the trajectory reference expression library, which serves as a static comparison reference for subsequent trajectory matching and disturbance offset evaluation.

[0056] By analyzing the frequency domain structure and time segment expression of each channel signal in the current batch of disturbance data, combined with the binding relationship between the batch number and the recipe number, an aligned trajectory expression set is constructed. Trajectory shape and amplitude offset matching is performed based on a sliding window, generating a one-to-one mapping relationship between the current batch disturbance trajectory and the trajectory benchmark, as well as the corresponding disturbance matching offset structure.

[0057] By clustering the deviation results between the current batch disturbance trajectory and the trajectory benchmark by disturbance type and constructing a disturbance deviation aggregation map, combined with the trajectory trend change characteristics and scheduling path constraints, trajectory consistency judgment and path switching consistency judgment are performed to form a structural level judgment result of the recipe drift behavior, which serves as the core basis for whether the ratio anomaly is established.

[0058] By jointly annotating the abnormal trajectory sections with the recipe abnormal sections, locating the boundary structure of the abnormal trajectory in the disturbance feature trajectory and performing correction, an updated trajectory benchmark expression library is constructed. At the same time, trajectory samples containing ratio drift features are extracted, the recognition strategy is reconstructed, and the method's fault-tolerant response capability to abnormal offsets is enhanced.

[0059] Acquire historical disturbance data, including equipment structure vibration data, feed path flow rate data, and silo wall soundprint data. Perform time normalization and dimension alignment on the historical disturbance data to generate a multi-source historical disturbance data sequence.

[0060] Extract disturbance variables from multi-source historical disturbance data sequences. These disturbance variables include equipment vibration amplitude, which characterizes changes in mechanical structure state; feed speed, which reflects material conveying stability; and soundprint waveform, which identifies the characteristics of silo wall impact reverberation. These three types of disturbance variables are then reconstructed in the time dimension as vibration amplitude time series, feed speed time series, and soundprint waveform time series.

[0061] Frequency decomposition operations are performed on the vibration amplitude time series, feed speed time series, and soundprint waveform time series in sequence. The frequency band boundaries are delineated according to the frequency change intervals of various types. Then, the time segments are divided into fixed step sizes within each frequency band to extract the set of disturbance trajectory segments formed under the conditions of frequency consistency and energy stability.

[0062] For each type of time segment in the disturbance trajectory segment set, morphological feature extraction and amplitude distribution calculation are performed in sequence. The morphological features include the waveform envelope structure, local oscillation boundary points, and repetitive period segments of the time segment on the time axis. The amplitude distribution includes the local peak amplitude, the average amplitude level of the entire segment, and the amplitude change rate within the time segment. A disturbance feature spectrum set is generated by mapping the vibration amplitude time series, the feed speed time series, and the voiceprint waveform time series respectively.

[0063] For all time segments mapped in the disturbance feature map set, feature space reconstruction and multidimensional clustering analysis are performed based on the morphological contour curvature and amplitude density distribution structure of the corresponding time segments. The most representative disturbance feature trajectory is extracted from the clustering results of each time segment as a matching reference sample and written into the trajectory benchmark expression library. This is used as the static comparison basis for the current batch trajectory expression in trajectory alignment and offset judgment, and is used for subsequent trajectory matching and offset trend assessment.

[0064] Collect current disturbance data, including real-time vibration response data, feeding operation sound spectrum data and silo cavity echo data;

[0065] Based on the timestamp and task registration identifier in the acquisition process, the batch number corresponding to the current disturbance data is extracted, and then the recipe number registered for the current batch is parsed based on the batch number to establish a binding relationship between the batch number and the recipe number;

[0066] Then, multi-dimensional compression and feature mapping are performed on the current disturbance data to generate the current disturbance expression vector group. The real-time vibration response data, feeding operation sound spectrum data, and silo cavity echo data in the current disturbance expression vector group are then subjected to frequency domain restoration and time window partitioning operations in sequence to reconstruct each type of data into a set of disturbance time segments arranged in chronological order.

[0067] Each time segment is bound to the batch number extracted from the current disturbance data to identify its batch affiliation in the trajectory expression set. Finally, all time segments bound to the batch number are aggregated to construct the trajectory expression set corresponding to the current batch.

[0068] Perform trajectory shape matching operations within a sliding time window on each time segment in the trajectory expression set and the disturbance feature trajectory in the trajectory reference expression library, build a one-to-one correspondence between each time segment and its corresponding disturbance feature trajectory, and generate a trajectory matching relationship set;

[0069] For each time segment in the trajectory expression set and its corresponding perturbation feature trajectory, shape curvature difference calculation and time alignment offset solution are performed to extract the initial matching offset value of each pair of time segments on the trajectory geometry structure, and construct a preliminary perturbation offset response structure for subsequent enhanced extraction of feature indicators;

[0070] Based on the preliminary perturbation offset response structure, we further perform trajectory-guided coincidence analysis and amplitude response offset rate evaluation on each time segment in the trajectory expression set and the corresponding perturbation feature trajectory. We extract the difference indicators of each time segment in terms of shape change trend and energy response characteristics, construct a complete perturbation matching offset result, and summarize all perturbation matching offset results in chronological order to generate a perturbation matching offset vector set.

[0071] The formula is:

[0072]

[0073] in,

[0074] α i =max t S i (t);

[0075] β i =max t T i (t-δ i );

[0076] S i (t) is the response function in the current disturbance time segment; T i (t) is the corresponding disturbance characteristic trajectory; is the first-order velocity and second-order curvature response of the disturbance; δ i is the matching offset; θ i (t) is the local waveform guide angle; τ i is the starting point of the matching window; Δτ is the length of the matching window; α i is the maximum amplitude of the time segment; β i is the maximum response of the reference trajectory; D i is the matching offset value.

[0077] Further, The curvature difference term represents the square of the difference between the second-order derivative of the current time segment and the perturbation characteristic trajectory at time point t. It is used to capture the shape deviation degree of the trajectory at the microstructure level, reflecting the local difference in the curvature trend of the trajectory contour, and is the main criterion for trajectory morphology mismatch. is the directional structure difference term, which represents the amplification effect of the difference in structural direction between the first-order derivative angle θi(t) of the current trajectory segment and the reference trajectory segment at time point t. The product term reflects the trend of the two guiding velocities in the same or different directions. The sine square term amplifies the difference amplitude to ensure that the structural trends are amplified with high weight when they are inconsistent.

[0078] (α i -β i ) 2 The peak amplitude difference term represents the square difference between the maximum response amplitude of the current time segment αi=maxtS(t) and the maximum amplitude of the corresponding disturbance feature trajectory βi=maxtT(tδi). It is used to evaluate the degree of strong pairing of the overall response capability by measuring whether the two trajectories are consistent at the level of signal energy extremes; It is an overall integral structure, which represents the cumulative contribution of the curvature difference term and the directional structure difference term within the matching time window [τi, τi+Δτ], ensuring a unified evaluation of the trajectory differences at all moments in the matching time period and reflecting the persistent error within the complete local window.

[0079] Based on the original source of each disturbance matching offset result in the trajectory expression set, the disturbances in the disturbance matching offset vector set are divided into three types: vibration, flow rate, and voiceprint. Clustering and aggregation operations are performed according to the disturbance type to construct a disturbance offset aggregation map. Based on the disturbance offset aggregation map, normalized threshold analysis and sliding trend extraction are performed to generate disturbance offset trend judgment results.

[0080] Determine whether the disturbance offset trend judgment results meet the condition that the offset degree is lower than the preset tolerance threshold under all three types of disturbance types. If all conditions are met, a trajectory consistency judgment result is generated, indicating that the current batch disturbance state is consistent with the trajectory benchmark;

[0081] If the deviation degree of any disturbance type in the disturbance deviation trend judgment result exceeds the corresponding tolerance threshold in a continuous time segment, and the deviation degree shows a monotonically increasing trend, then the disturbance type is judged to have a persistent anomaly and a trajectory drift judgment result is generated, indicating that the current batch disturbance state has significantly deviated from the trajectory reference;

[0082] The formula is:

[0083]

[0084] in,

[0085] Λ k (t)>Θ k ;

[0086] V k (t) is the offset response curve of disturbance type k; is the disturbance offset rate; is the disturbance trend acceleration; η k (t) is the disturbance jump intensity index; Δt is the trend window length; Λ k (t) is the trend response function; Θ k is the trigger threshold.

[0087] Further, The oscillation rate enhancement term represents the oscillation change rate at time point t' in the oscillation aggregation graph of oscillation type k. It is used to identify whether the oscillation offset is increasing rapidly locally and is a basic indicator of trend direction changes. is the trend mutation adjustment item, where is the oscillation acceleration, which indicates the mutation rate of the oscillation trajectory trend near t'. Adding 1 after the absolute value avoids taking negative logarithmic values, which serves as a logarithmic modulation of the severity of the change and enhances the identification of "trend turning points" or "continuous deviation acceleration";

[0088] tanh(η k (t′)) is the jump response amplification term, where η k (t′) is the structural jump response index of oscillation type k at time t′, which is an index of morphological discontinuous change extracted from the oscillation aggregation spectrum. Its constant value range is (0, 1). The hyperbolic tangent function is used to compress excessive responses and suppress occasional spike misjudgments. The local trend response accumulation term represents the accumulation of the local trend change intensity of the oscillation response within the trend time window [tΔt, t], which is used to extract the continuity and structure of the trend from the short-term oscillation, rather than the deviation behavior of isolated points;

[0089] The outer derivative corresponding to the global trend evolution term is used to differentiate the entire trend accumulation structure at time t to detect whether the trend change is strengthening, that is, whether the trend is "growing or declining faster", making Λ_k(t) more sensitive to the rate of change of the trend; at the same time, Λ k (t)>Θ k Indicates that when the trend response function value exceeds the trigger threshold Θ set by the disturbance type k k When the system deviates from the trajectory and enters the "structural drift state", it will be considered that the trajectory has deviated into the "structural drift state".

[0090] Based on the binding relationship established between the batch number and the recipe number, the switching path information related to the recipe number registered for the current batch is extracted from the task registration record;

[0091] The switching path information includes the trajectory offset mapping relationship between the recipe number registered in the current batch and the recipe number registered in the previous batch, as well as the recipe switching time interval between the recipe number registered in the current batch and the recipe number registered in the previous batch as marked in the task registration record;

[0092] The trajectory drift determination result is jointly compared with the switching path information, and the path consistency determination step is performed to evaluate whether the current trajectory state meets the recipe switching logic requirements;

[0093] In the path consistency judgment, determine whether any of the following structural anomalies exist: there is no valid switching relationship registered between the recipe number registered for the current batch and the recipe number of the previous round in the switching path parameters, the execution time range corresponding to the recipe number of the previous round overlaps with the switching time interval, or the current recipe number and switching time combination does not meet the switching sequence rules set in the trajectory reference expression library;

[0094] If any of the above anomalies exists, a final judgment result is generated that the recipe drift behavior is established, indicating that the current trajectory state has constituted a consistency conflict in the scheduling path structure.

[0095] The time segments in the trajectory expression set where the disturbance type involved in the trajectory drift judgment result forms a monotonically increasing structure in the offset trend are marked as trajectory abnormal segments. At the same time, the time segments covered by the recipe switching time interval corresponding to the recipe drift behavior judgment result in the trajectory expression set are regarded as recipe abnormal segments.

[0096] Generate abnormal state labels based on the abnormal state type and time segment information corresponding to the trajectory abnormal segment and the recipe abnormal segment, and summarize them to form an abnormal behavior feedback label set;

[0097] Extract the disturbance feature trajectory associated with the corresponding matching time segment in the trajectory expression set from the abnormal time segment indicated by each abnormal state label in the abnormal behavior feedback label set; use the disturbance feature trajectory as the reference trajectory for trajectory correction, analyze the boundary position of the abnormal time segment in the corresponding disturbance feature trajectory, perform boundary extension and expression content update processing, and generate a corrected trajectory reference template set;

[0098] The modified trajectory benchmark template set is written into the trajectory benchmark expression library as a whole, and the current version information is synchronously annotated to generate a trajectory expression version sequence arranged in chronological order, which is used to record the structural update of the trajectory benchmark during the anomaly evolution process.

[0099] The abnormal state type and time segment information corresponding to the abnormal behavior feedback tag set are combined with the recipe number information registered for the current batch to construct a behavioral archive record containing the recipe abnormal segment trajectory location information, the recipe switching path offset structure, and the disturbance feature trajectory revision process. This is used to support the traceability of ratio anomalies and scheduling path backtracking analysis in the industrial data processing process.

[0100] Extract trajectory revision samples containing recipe abnormality segments and trajectory abnormality segment identification information from the behavior archive record, where the recipe abnormality segment is the time segment corresponding to the recipe switching time interval in the trajectory expression set, and the trajectory abnormality segment is the continuous time segment in the trajectory expression set that forms a monotonic offset trend;

[0101] Using trajectory revision samples as training input, we perform parameter structure updates and evolutionary recognition strategy reconstruction based on anomaly feedback information to enhance the fault tolerance of the matching anomaly recognition method.

[0102] The formula is:

[0103]

[0104] in,

[0105]

[0106] Revise the sample for the jth trajectory; ΔD j is the disturbance offset amplitude; L j is the abnormal duration; K j is the trajectory jump amplitude; γ j is the misidentification weight factor;κ(·) is the structural response function; is the set of model parameters; is the recognition performance index under the t-th training;

[0107] in, Represents the three-dimensional features of the trajectory revision sample, the three-dimensional features are: ΔD j L is the trajectory matching offset amplitude; j is the duration of the abnormal trajectory section; K j is the morphological jump intensity of the corresponding trajectory boundary;

[0108] Furthermore, the structural response function term For the evolutionary recognition method, the current parameters Next, for the input sample The response output is a nonlinear recognition scoring function based on feature inputs such as trajectory offset, span and boundary mutation; the historical misjudgment weight adjustment term γ j Representation sample The weight of the number of misidentifications in historical matching recognition is greater. The greater the weight, the more likely this type of trajectory is to be misidentified in history, and the system should give it a higher level of training reinforcement.

[0109] Feature Driver It is used to combine the recognition response value with the historical misjudgment frequency, strengthen the learning effect of high-error samples in a "band exponential suppression" manner, and enhance the robustness of the model to difficult-to-recognize samples; abnormal morphology penalty term The square of the ratio of the trajectory boundary jump intensity to the abnormal segment duration is used to penalize samples with "short but high-intensity jumps" to prevent the model from overfitting to extreme anomalies with high mutation and low persistence.

[0110] Training objective function Indicates that in the training cycle t, the structural parameters The fault-tolerant performance evolution function driven by it forms an optimal suppression structure for the risk of misidentification of trajectory disturbances by enhancing the consistency of sample responses and the elasticity of recognition boundaries, which is used to reflect the stable convergence trend of strategy recognition ability with the training process.

[0111] A feed ratio abnormality identification system includes a track building module, a deconstruction matching module, a bias detection module, and a standard correction and evolution module;

[0112] The track extraction and database construction module extracts disturbance variables such as vibration amplitude, feed speed, and soundprint waveform from multi-source disturbance data, constructs a disturbance time series expression with a triple feature structure of frequency, morphology, and amplitude, and extracts disturbance feature trajectories with central representation capabilities in the cluster through graphical clustering. These are written into the trajectory reference expression library and serve as a static comparison reference for subsequent trajectory matching and disturbance offset evaluation.

[0113] The deconstruction matching module analyzes the frequency domain structure and time segment expression of each channel signal in the current batch of disturbance data, combines the binding relationship between the batch number and the recipe number, constructs an aligned trajectory expression set, and performs trajectory shape and amplitude offset matching based on a sliding window. It generates a one-to-one mapping relationship between the current batch disturbance trajectory and the trajectory benchmark, as well as the corresponding disturbance matching offset structure.

[0114] The drift detection module clusters the deviation results between the current batch disturbance trajectory and the trajectory benchmark by disturbance type and constructs a disturbance deviation aggregation map. It then combines trajectory trend change characteristics with scheduling path constraints to perform trajectory consistency judgment and path switching consistency judgment, forming a structural-level judgment result on the recipe drift behavior, which serves as the core basis for determining whether the ratio is abnormal.

[0115] The calibration and correction evolution module jointly labels the abnormal trajectory sections with the recipe abnormal sections, locates the boundary structure of the abnormal trajectory in the disturbance feature trajectory and performs correction, constructs an updated trajectory benchmark expression library, and extracts trajectory samples containing ratio drift features to reconstruct the recognition strategy and enhance the method's fault-tolerant response capability to abnormal offsets.

[0116] It should be noted that in the formula structure involved in this solution, dimensionless terms can serve as proportionality or structural adjustment factors. When combined with quantities with units, they only play a numerical scaling role and do not introduce new physical dimensions. Therefore, they will not change or confuse the overall unit system of expression. This combination of "dimensionless terms and units" can be understood as a composite structural expression commonly used in mathematical and physical modeling, conforming to the principle of dimensional consistency and having a clear physical interpretation basis.

[0117] 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 be formed into a unified structure through function mapping, ratio combination or normalization adjustment. The units and meanings are clear, and the overall expression conforms to the principle of dimensional consistency and the common formula of engineering modeling.

[0118] In this solution, any design constants, weights, adjustment factors, threshold parameters, and proportional coefficients are adjustable control parameters for different application environments. Their values ​​depend on the target device configuration, data input characteristics, and performance optimization goals. During the implementation phase, they are set within a reasonable range through model verification, performance constraints, or engineering calibration. Although these parameters do not have unique preset values, they have clear adjustment logic and calculation paths, and are part of the deterministic setting process in engineering implementation. The purpose of such setting is to ensure that the solution is both universally adaptable, reproducible, and operable, without affecting its technical clarity and feasibility.

[0119] 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 in the scope of protection of the present invention.

Claims

1. A method for identifying abnormal feed ratio, characterized in that: include: By extracting disturbance variables such as vibration amplitude, feeding speed, and soundprint waveform from multi-source disturbance data, a disturbance time series expression with a triple feature structure of frequency, morphology, and amplitude is constructed. Then, through graphical clustering, the disturbance feature trajectory with central representation capability in the cluster is extracted and written into the trajectory benchmark expression library. By analyzing the frequency domain structure and time segment expression of each channel signal in the current batch of disturbance data, combined with the binding relationship between the batch number and the recipe number, an aligned trajectory expression set is constructed. Trajectory shape and amplitude offset matching is performed based on a sliding window, generating a one-to-one mapping relationship between the current batch disturbance trajectory and the trajectory benchmark, as well as the corresponding disturbance matching offset structure. By clustering the deviation results between the current batch disturbance trajectory and the trajectory benchmark according to the disturbance type and constructing a disturbance deviation aggregation map, combined with the trajectory trend change characteristics and scheduling path constraints, trajectory consistency judgment and path switching consistency judgment are performed to form a structural level judgment result of the recipe drift behavior; By jointly annotating the abnormal trajectory sections with the recipe abnormal sections, locating the boundary structure of the abnormal trajectory in the disturbance feature trajectory and performing correction, an updated trajectory benchmark expression library is constructed. At the same time, trajectory samples containing ratio drift features are extracted, the recognition strategy is reconstructed, and the method's fault-tolerant response capability to abnormal offsets is enhanced.

2. A method for identifying abnormal feed ratio according to claim 1, characterized in that: Acquire historical disturbance data, including equipment structure vibration data, feed path flow rate data, and silo wall soundprint data. Perform time normalization and dimension alignment on the historical disturbance data to generate a multi-source historical disturbance data sequence. Extract disturbance variables from multi-source historical disturbance data sequences. These disturbance variables include equipment vibration amplitude, which characterizes changes in mechanical structure state; feed speed, which reflects material conveying stability; and soundprint waveform, which identifies the characteristics of silo wall impact reverberation. These three types of disturbance variables are then reconstructed in the time dimension as vibration amplitude time series, feed speed time series, and soundprint waveform time series. Frequency decomposition operations are performed on the vibration amplitude time series, feed speed time series, and soundprint waveform time series in sequence. The frequency band boundaries are delineated according to the frequency change intervals of various types. Then, the time segments are divided into fixed step sizes within each frequency band to extract the set of disturbance trajectory segments formed under the conditions of frequency consistency and energy stability.

3. A method for identifying abnormal feed ratio according to claim 2, characterized in that: For each type of time segment in the disturbance trajectory segment set, morphological feature extraction and amplitude distribution calculation are performed in sequence. The morphological features include the waveform envelope structure, local oscillation boundary points, and repetitive period segments of the time segment on the time axis. The amplitude distribution includes the local peak amplitude, the average amplitude level of the entire segment, and the amplitude change rate within the time segment. A disturbance feature spectrum set is generated by mapping the vibration amplitude time series, the feed speed time series, and the voiceprint waveform time series respectively. For all time segments mapped in the disturbance feature map set, feature space reconstruction and multidimensional clustering analysis are performed based on the morphological contour curvature and amplitude density distribution structure of the corresponding time segments. The most representative disturbance feature trajectory is extracted from the clustering results of each time segment as a matching reference sample and written into the trajectory benchmark expression library.

4. A method for identifying abnormal feed ratio according to claim 3, characterized in that: Collect current disturbance data, including real-time vibration response data, feeding operation sound spectrum data and silo cavity echo data; Based on the timestamp and task registration identifier in the acquisition process, the batch number corresponding to the current disturbance data is extracted, and then the recipe number registered for the current batch is parsed based on the batch number to establish a binding relationship between the batch number and the recipe number; Then, multi-dimensional compression and feature mapping are performed on the current disturbance data to generate the current disturbance expression vector group. The real-time vibration response data, feeding operation sound spectrum data, and silo cavity echo data in the current disturbance expression vector group are then subjected to frequency domain restoration and time window partitioning operations in sequence to reconstruct each type of data into a set of disturbance time segments arranged in chronological order. Each time segment is bound to the batch number extracted from the current disturbance data, and finally all the time segments bound to the batch number are aggregated to construct the trajectory expression set corresponding to the current batch.

5. A method for identifying abnormal feed ratio according to claim 4, characterized in that: Perform trajectory shape matching operations within a sliding time window on each time segment in the trajectory expression set and the disturbance feature trajectory in the trajectory reference expression library, build a one-to-one correspondence between each time segment and its corresponding disturbance feature trajectory, and generate a trajectory matching relationship set; For each time segment in the trajectory expression set and its corresponding disturbance feature trajectory, perform shape curvature difference calculation and time alignment offset solution, extract the initial matching offset value of each pair of time segments on the trajectory geometry, and construct a preliminary disturbance offset response structure; Based on the preliminary disturbance offset response structure, trajectory-guided coincidence analysis and amplitude response offset rate evaluation are further performed on each time segment in the trajectory expression set and the corresponding disturbance feature trajectory. The difference indicators of shape change trend and energy response characteristics of each time segment are extracted to construct a complete disturbance matching offset result. All disturbance matching offset results are summarized in chronological order to generate a disturbance matching offset vector set.

6. A method for identifying abnormal feed ratio according to claim 5, characterized in that: Based on the original source of each disturbance matching offset result in the trajectory expression set, the disturbances in the disturbance matching offset vector set are divided into three types: vibration, flow rate, and voiceprint. Clustering and aggregation operations are performed according to the disturbance type to construct a disturbance offset aggregation map. Based on the disturbance offset aggregation map, normalized threshold analysis and sliding trend extraction are performed to generate disturbance offset trend judgment results. Determine whether the disturbance offset trend judgment results meet the condition that the offset degree is lower than the preset tolerance threshold under all three types of disturbance types. If all conditions are met, a trajectory consistency judgment result is generated, indicating that the current batch disturbance state is consistent with the trajectory benchmark; If the deviation degree of any disturbance type in the disturbance deviation trend judgment result exceeds the corresponding tolerance threshold in a continuous time segment, and the deviation degree shows a monotonically increasing trend, then it is determined that the disturbance type has a persistent anomaly and a trajectory drift judgment result is generated.

7. A method for identifying abnormal feed ratio according to claim 6, characterized in that: Based on the binding relationship established between the batch number and the recipe number, the switching path information related to the recipe number registered for the current batch is extracted from the task registration record; The switching path information includes the trajectory offset mapping relationship between the recipe number registered in the current batch and the recipe number registered in the previous batch, as well as the recipe switching time interval between the recipe number registered in the current batch and the recipe number registered in the previous batch as marked in the task registration record; The trajectory drift determination result is jointly compared with the switching path information, and the path consistency determination step is performed to evaluate whether the current trajectory state meets the recipe switching logic requirements; In the path consistency judgment, determine whether any of the following structural anomalies exist: there is no valid switching relationship registered between the recipe number registered for the current batch and the recipe number of the previous round in the switching path parameters, the execution time range corresponding to the recipe number of the previous round overlaps with the switching time interval, or the current recipe number and switching time combination does not meet the switching sequence rules set in the trajectory reference expression library; If any of the above anomalies exists, a final judgment result is generated that the recipe drift behavior is established, indicating that the current trajectory state has constituted a consistency conflict in the scheduling path structure.

8. A method for identifying abnormal feed ratio according to claim 7, characterized in that: The time segments in the trajectory expression set where the disturbance type involved in the trajectory drift judgment result forms a monotonically increasing structure in the offset trend are marked as trajectory abnormal segments. At the same time, the time segments covered by the recipe switching time interval corresponding to the recipe drift behavior judgment result in the trajectory expression set are regarded as recipe abnormal segments. Generate abnormal state labels based on the abnormal state type and time segment information corresponding to the trajectory abnormal segment and the recipe abnormal segment, and summarize them to form an abnormal behavior feedback label set; Extract the disturbance feature trajectory associated with the corresponding matching time segment in the trajectory expression set from the abnormal time segment indicated by each abnormal state label in the abnormal behavior feedback label set; use the disturbance feature trajectory as the reference trajectory for trajectory correction, analyze the boundary position of the abnormal time segment in the corresponding disturbance feature trajectory, perform boundary extension and expression content update processing, and generate a corrected trajectory reference template set; The modified trajectory benchmark template set is written into the trajectory benchmark expression library as a whole, and the current version information is synchronously annotated to generate a trajectory expression version sequence arranged in chronological order.

9. A method for identifying abnormal feed ratio according to claim 8, characterized in that: The abnormal state type and time segment information corresponding to the abnormal behavior feedback tag set are combined with the recipe number information registered for the current batch to construct a behavior archive record containing the recipe abnormal segment trajectory positioning information, the recipe switching path offset structure, and the disturbance feature trajectory revision process; Extract trajectory revision samples containing recipe abnormality segments and trajectory abnormality segment identification information from the behavior archive record, where the recipe abnormality segment is the time segment corresponding to the recipe switching time interval in the trajectory expression set, and the trajectory abnormality segment is the continuous time segment in the trajectory expression set that forms a monotonic offset trend; Taking trajectory revision samples as training input sets, parameter structure update and evolutionary recognition strategy reconstruction based on anomaly feedback information are performed to enhance the fault tolerance of the ratio anomaly recognition method.

10. A feed ratio anomaly identification system, comprising the feed ratio anomaly identification method according to claim 9, the system comprising a track building module, a deconstruction matching module, a bias aggregation and drift judgment module, and a standard correction and evolution module, characterized in that: The track extraction and database construction module extracts disturbance variables such as vibration amplitude, feed speed, and soundprint waveform from multi-source disturbance data, constructs a disturbance time series expression with a triple feature structure of frequency, morphology, and amplitude, and extracts disturbance feature trajectories with central representation capabilities in the cluster through graphical clustering. These are written into the trajectory reference expression library and serve as a static comparison reference for subsequent trajectory matching and disturbance offset evaluation. The deconstruction matching module analyzes the frequency domain structure and time segment expression of each channel signal in the current batch of disturbance data, combines the binding relationship between the batch number and the recipe number, constructs an aligned trajectory expression set, and performs trajectory shape and amplitude offset matching based on a sliding window. It generates a one-to-one mapping relationship between the current batch disturbance trajectory and the trajectory benchmark, as well as the corresponding disturbance matching offset structure. The drift detection module clusters the deviation results between the current batch disturbance trajectory and the trajectory benchmark by disturbance type and constructs a disturbance deviation aggregation map. It then combines trajectory trend change characteristics with scheduling path constraints to perform trajectory consistency judgment and path switching consistency judgment, forming a structural-level judgment result on the recipe drift behavior, which serves as the core basis for determining whether the ratio is abnormal. The calibration and correction evolution module jointly labels the abnormal trajectory sections with the recipe abnormal sections, locates the boundary structure of the abnormal trajectory in the disturbance feature trajectory and performs correction, constructs an updated trajectory benchmark expression library, and extracts trajectory samples containing ratio drift features to reconstruct the recognition strategy and enhance the method's fault-tolerant response capability to abnormal offsets.

Images