CN-121580256-B - Storage cabinet abnormal trend prediction system based on time sequence data analysis

CN121580256BCN 121580256 BCN121580256 BCN 121580256BCN-121580256-B

Abstract

The invention relates to the technical field of anomaly prediction, in particular to a storage cabinet anomaly trend prediction system based on time sequence data analysis. According to the invention, a state vector comprising temperature, voltage, current and door lock states is constructed, a time sequence data sequence is formed by combining timestamp information, a dynamic expression mode of state change is established, jump characteristics are analyzed by utilizing the ratio of time intervals to state change amplitude, a jump rate statistical index is combined to distinguish short-time disturbance and trend evolution paths, an evolution activating signal is identified based on trend retention and non-fallback characteristics, on the basis, an aperiodic thermal anomaly trend in a neural network structure capturing state sequence with long-term dependence learning ability is introduced, and the accuracy and timeliness of anomaly identification are improved through multidimensional parameter cooperative processing and path construction logic.

Inventors

LIU YANG
GAO ZIQI
CHEN YUBO
LIN ZHIYANG

Assignees

福建安吉达智能科技有限公司
江西泽山智能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (9)

1. A storage cabinet anomaly trend prediction system based on time series data analysis, the system comprising: The state monitoring module is used for collecting real-time monitoring data in the operation process of the storage cabinet, wherein the monitoring data comprise temperature, voltage, current and door lock states, corresponding time stamps are associated to form a state vector in a combined mode, and the state vectors at a plurality of continuous moments are arranged according to time sequence to generate the operation data of the storage cabinet; the interval sensing module calculates an actual time interval according to the time stamp of the adjacent data point in the operation data of the storage cabinet and maps the actual time interval into a dynamic offset, calculates the change amplitude of the temperature gradient in the corresponding interval, executes the bidirectional difference ratio operation of the dynamic offset and the change amplitude, and screens out the interval of which the operation result exceeds a preset reference to mark the abnormal jump segment; The path reconstruction module calculates the difference value of state vectors at adjacent moments according to the operation data of the storage cabinet, constructs a jump rate sequence in combination with a time interval, acquires the jump rate sequence, calculates an arithmetic mean value and a statistical variance, extracts a voltage fluctuation amplitude value and a current fluctuation amplitude value corresponding to the operation data of the storage cabinet, calculates and acquires a state stability index, eliminates a high-frequency disturbance interval from the abnormal jump segment according to the state stability index, and connects nodes with continuous abrupt change characteristics to construct a fault evolution path; the symptom activating module is used for carrying out sectional statistics on current instantaneous values in the operation data of the storage cabinet, calculating a mean value and a standard deviation, calculating a ratio of the current instantaneous values to the standard deviation to generate a current fluctuation variation coefficient, identifying intervals in which the current fluctuation variation coefficient keeps the same direction deviation and the deviation amplitude is in a non-return situation in a plurality of continuous time periods, and generating a trend activating signal; And the evolution prediction module responds to the trend activation signal, inputs a fault evolution path to a preset long-short-term memory network, invokes a time back propagation algorithm to calculate the evolution direction of the state vector at the future moment, and outputs an abnormal trend prediction result.
2. The system for predicting a cabinet anomaly trend based on time series data analysis of claim 1, wherein the status monitoring module comprises: The multi-dimensional parameter acquisition sub-module acquires real-time monitoring data in the running process of the storage cabinet, wherein the monitoring data comprise temperature, voltage, current and door lock states, a multi-channel signal acquisition card is called to synchronously sample and digitally convert each path of analog quantity, level state reading is carried out on discrete switching value signals, all sensing values acquired at the same sampling triggering moment are integrated in parallel and mapped with channel identifiers, and a multi-dimensional sensing parameter set is generated; The time vector construction submodule is used for calling a synchronous clock source to acquire corresponding time information based on sampling trigger time of the multidimensional sensing parameter set and converting the time information into a time stamp in a unified format, constructing a feature space comprising a temperature dimension, a voltage dimension, a current dimension and a door lock state dimension, mapping each numerical value in the multidimensional sensing parameter set to a corresponding coordinate axis position of the feature space, and embedding the time stamp as a time sequence index dimension to construct a state vector with a time mark; And the time sequence data generation sub-module circularly acquires the state vector with the time mark in a continuous monitoring period, stores the state vector into a first-in first-out queue buffer zone, reads the time stamp indexes of all the state vectors with the time mark in the buffer zone, reorders and aligns all the vectors according to the monotonically increasing characteristic of the time stamp values, eliminates conflict items repeated by the time stamp, and splices and matrix packages the ordered vector sequence after the verification according to the time dimension to generate the operation data of the storage cabinet.
3. The abnormal trend prediction system of the storage cabinet based on time series data analysis according to claim 2, wherein the construction of the feature space comprising a temperature dimension, a voltage dimension, a current dimension and a door lock state dimension comprises the steps of establishing a multi-dimensional coordinate system with a temperature parameter as a first dimension, a voltage parameter as a second dimension, a current parameter as a third dimension and a door lock state parameter as a fourth dimension, performing linear normalization operation on each item of value acquired at the current moment by using the history extremum range, and projecting the normalized value into the corresponding coordinate axis interval of the multi-dimensional coordinate system; The method comprises the steps of carrying out reordering and alignment verification on all vectors according to monotonically increasing characteristics of timestamp values, removing conflict items repeated by the timestamps, specifically, carrying out bubbling sequencing based on the timestamp values on the time-stamp-carrying state vectors in a first-in first-out queue buffer area, establishing monotonically non-decreasing sequence of the time-stamp-carrying state vectors in time sequence, traversing the sequenced vector sequence, identifying a plurality of repeated vector groups with identical timestamp values, respectively calculating Euclidean distance between each vector in the group and a previous time vector in a feature space according to each repeated vector group, selecting one time-carrying state vector with the smallest Euclidean distance as effective data to be reserved, and deleting the rest time-carrying state vectors in the group from the vector sequence.
4. The storage cabinet anomaly trend prediction system based on time series data analysis of claim 2, wherein the interval sensing module comprises: the time offset mapping sub-module is used for executing time difference operation between adjacent data frames according to time stamp sequences arranged in sequence in the operation data of the storage cabinet, obtaining non-uniform sampling intervals, carrying out numerical comparison on the non-uniform sampling intervals and preset standard sampling step sizes, calculating distortion degree of the intervals relative to the standard step sizes through a non-linear mapping algorithm, quantifying local expansion and contraction characteristics on a time axis, and generating dynamic offset; the gradient ratio calculation sub-module extracts temperature values at adjacent moments in the operation data of the storage cabinet, calculates absolute differences, calculates instantaneous variation amplitude of the temperature gradient according to the corresponding time differences, calls dynamic offset as a denominator adjustment item, performs ratio operation between the variation amplitude of the temperature gradient and the dynamic offset, and generates a time-temperature difference differential ratio sequence; And the jump interval locking sub-module invokes a preset jump judging reference value, performs point-by-point scanning and numerical judgment on the time-temperature difference ratio sequence, screens data point indexes with numerical values larger than the jump judging reference value, merges continuous or adjacent overrun index points into independent time windows, and performs abnormal identification on the coverage range of the time windows in the original data stream to generate abnormal jump fragments.
5. The system for predicting abnormal trend of storage cabinet based on time series data analysis according to claim 4, wherein the jump judging reference value is set by selecting a time-temperature difference value sequence generated in a steady-state operation period of the storage cabinet to form a reference sample set, calculating an arithmetic average value and a standard deviation of the reference sample set, and adding the arithmetic average value and the standard deviation of a preset multiple to obtain the jump judging reference value.
6. The system for predicting a cabinet anomaly trend based on temporal data analysis of claim 5, wherein the path reconstruction module comprises: The rate sequence construction submodule calculates differential vectors of state vectors at adjacent moments based on the operation data of the storage cabinet, counts the number of dimensions with amplitude exceeding a preset noise margin in the differential vectors, and combines time intervals of corresponding sampling points to execute ratio operation to generate a jump rate sequence; the stability index calculation sub-module is used for obtaining the jump rate sequence, calculating an arithmetic mean value and a statistical variance, extracting a voltage fluctuation amplitude value and a current fluctuation amplitude value corresponding to the operation data of the storage cabinet, and calculating to obtain a state stability index; And the path evolution reconstruction sub-module is used for calling the state stability index, carrying out numerical comparison with a confidence coefficient threshold value, eliminating discrete disturbance points with the state stability index lower than the confidence coefficient threshold value, extracting time sequence indexes and state characteristics of the rest nodes, establishing a node topological chain with a time sequence progressive relation, and generating a fault evolution path.
7. The storage cabinet anomaly trend prediction system based on temporal data analysis of claim 6, wherein the symptom activation module comprises: The segmentation statistics calculation sub-module extracts a current instantaneous value sequence comprising a time index according to the operation data of the storage cabinet, divides the current instantaneous value sequence into a plurality of continuous and mutually non-overlapping time segments according to a preset statistics period step length, performs accumulation and variance operation on current sampling points in each time segment, calculates an arithmetic mean value of current amplitude values in the segments and a standard deviation value reflecting fluctuation dispersion respectively, and generates a segmentation current statistics feature set; The variation coefficient generation submodule calls the segmented current statistical feature set, sequentially reads a standard deviation value and an arithmetic mean value corresponding to each statistical period, carries out ratio calculation of the standard deviation value and the arithmetic mean value, quantifies the relative fluctuation intensity of a current signal under various load levels, eliminates the influence of the current reference value on fluctuation amplitude evaluation, constructs a one-dimensional vector of the calculated ratio values according to time sequence, and generates a current fluctuation variation coefficient sequence; And the trend deviation recognition sub-module is used for executing sliding window scanning based on the current fluctuation variation coefficient sequence, recognizing the variation form of the numerical value, judging whether coefficient values of a plurality of continuous periods keep the deviation state in the same direction relative to a historical reference, simultaneously calculating the differential variation rate of the deviation amplitude, screening a section with a non-negative variation rate, confirming the section as a non-fallback situation, and carrying out logic setting marking on the continuous section meeting the condition to generate a trend activation signal.
8. The abnormal trend prediction system of the storage cabinet based on time series data analysis of claim 7, wherein the process of judging whether the coefficient values of the continuous multiple periods keep the deviation state in the same direction relative to the historical reference is specifically that the current fluctuation variation coefficient in the steady running state in the historical preset period is extracted and the arithmetic average value is calculated as the historical reference; The process for calculating the differential change rate of the deviation amplitude comprises the steps of obtaining an absolute value sequence of a difference value between a current fluctuation variation coefficient and a historical reference as a deviation amplitude sequence, calculating a differential value between a current moment value and a previous moment value in the deviation amplitude sequence, and taking the differential value as the differential change rate of the deviation amplitude; the process of screening the interval with the non-negative change rate and confirming the non-return situation comprises the steps of traversing the differential change rate, extracting a time interval corresponding to a data point with the numerical value larger than or equal to zero, and confirming the time interval as the non-return situation with the deviation amplitude without convergence characteristics.
9. The storage cabinet anomaly trend prediction system based on temporal data analysis of claim 8, wherein the evolution prediction module comprises: The sequence feature loading submodule invokes node feature data which are arranged according to time sequence and are included in the fault evolution path, constructs a multidimensional matrix which comprises time step and state variable according to the dimension requirement of a preset long-short-term memory network input layer, performs normalized mapping on state values in the matrix, eliminates dimension differences, converts the mapped data into tensor structures which accord with network input formats, and generates evolution path feature tensors; The evolution direction deduction submodule inputs the evolution path characteristic tensor into a preset long-short-term memory network model, logic operation of a forgetting gate, an input gate and an output gate is executed in an implicit layer, the state of a memory unit is updated, time back propagation calculation logic is executed aiming at the dependency relationship of a time dimension, gradient distribution of a loss function relative to a state variable is solved, each component change track of the state in a future time domain is determined according to the reverse direction of gradient descent or tangent logic, and the state vector evolution direction is generated; And the prediction result generation sub-module is used for executing numerical extrapolation operation at a future moment based on the evolution direction of the state vector, generating a state vector sequence at a future time point by combining a preset prediction step length, calculating a probability confidence interval corresponding to a prediction track and a predicted duration of an abnormal temperature rise state, and packaging the prediction data of a time sequence state, a probability range and the duration to generate an abnormal trend prediction result.

Description

Storage cabinet abnormal trend prediction system based on time sequence data analysis Technical Field The invention relates to the technical field of anomaly prediction, in particular to a storage cabinet anomaly trend prediction system based on time sequence data analysis. Background The technical field of anomaly prediction relates to modeling analysis and trend judgment on possible abnormal states of a system or equipment in the running process, and core matters comprise behavior modeling based on historical running data, application of a time sequence analysis method, recognition and evaluation of an anomaly mode and fusion processing of multidimensional data. The field systematically integrates methods such as statistical analysis, machine learning, time sequence modeling and the like, and continuously observes and analyzes the operation parameters of equipment or a system so as to predict the abnormal evolution trend before failure occurs. The field is widely applied to scenes such as industrial equipment, network systems, logistics management, intelligent operation and maintenance and the like, and the research focus is on constructing a stable and generalizable abnormality detection and prediction model, identifying potential abnormality risks through quantitative indexes and early warning in advance. The conventional storage cabinet abnormal trend prediction system is a system for evaluating the future state change trend of the storage cabinet based on the historical operation data of the storage cabinet so as to identify possible abnormality. The system generally utilizes a fixed threshold judgment method or a linear regression prediction model based on a sliding window to carry out trend judgment by periodically collecting the operating parameters such as temperature, current, voltage, door lock state and the like of the storage cabinet at different time points, so as to identify possible risk states such as temperature rise, unstable voltage, failure of a switch and the like. These methods rely on a set statistical index or a simple mathematical model, and have limited ability to identify nonlinear behavior, long-term dependence, or complex change patterns, and it is difficult to sufficiently reveal an abnormal development path. In the traditional system, the model construction mostly adopts modes of univariate analysis, mean fluctuation range setting, abnormal count accumulation and the like as judgment basis, and the comprehensive utilization of the association relation among multiple operation data is lacked, so that the prediction capability is obviously restricted. Disclosure of Invention The invention aims to solve the defects in the prior art, and provides a storage cabinet abnormal trend prediction system based on time sequence data analysis. In order to achieve the above purpose, the present invention adopts the following technical scheme, and the system for predicting abnormal trend of a storage cabinet based on time series data analysis comprises: the state monitoring module is used for collecting real-time monitoring data in the operation process of the storage cabinet, constructing state vectors by associating time stamps, and arranging the state vectors in sequence to generate the operation data of the storage cabinet; The interval sensing module calculates the time interval mapping into dynamic offset according to the operation data of the storage cabinet, calculates the variation amplitude of the temperature gradient, executes the ratio operation of the dynamic offset and the variation amplitude, screens the interval exceeding a preset reference, and marks the abnormal jump segment; the path reconstruction module calculates adjacent state vector difference values according to the operation data of the storage cabinet, constructs a jump rate sequence in combination with a time interval, calculates a state stability index, eliminates a high-frequency disturbance interval from the abnormal jump segment, and constructs a fault evolution path by connecting nodes; The symptom activating module is used for calculating the average value and the standard deviation of the instantaneous value of the operation data current of the storage cabinet in a segmented mode, calculating the ratio to generate a current fluctuation variation coefficient, and identifying a section where the current fluctuation variation coefficient continuously deviates in the same direction and does not fall back to generate a trend activating signal; And the evolution prediction module responds to the trend activation signal, inputs the fault evolution path into a long-term and short-term memory network, calculates the evolution direction of a future state vector and outputs an abnormal trend prediction result. As a further scheme of the invention, the storage cabinet operation data comprises a state matrix with a timestamp, a multidimensional sensing parameter sequence and a cabinet operation monitoring log, the abno