A storage cabinet abnormal trend prediction system based on time series data analysis

Abstract

The application relates to the technical field of anomaly prediction, in particular to a storage cabinet abnormal trend prediction system based on time series data analysis, which comprises a state monitoring module, an interval sensing module, a path reconstruction module, a symptom activation module and an evolution prediction module.In the application, a state vector containing temperature, voltage, current and door lock state is constructed, time stamp information is combined to form a time series data sequence, a dynamic expression mode of state change is established, the ratio of time interval to state change amplitude is used to analyze the jump characteristic, and a jump rate statistical index is combined to distinguish a short-time disturbance and a trend evolution path, an evolution activation signal is identified based on trend retention and non-falling characteristics, a neural network structure with long-time dependent learning ability is introduced to capture the aperiodic thermal abnormal trend in the state sequence, and the accuracy and timeliness of abnormal identification are improved through multi-dimensional parameter collaborative processing and path construction logic.

Description

Technical Field

[0001] This invention relates to the field of anomaly prediction technology, and in particular to a storage cabinet anomaly trend prediction system based on time-series data analysis. Background Technology

[0002] The field of anomaly prediction technology involves modeling, analyzing, and predicting trends of potential abnormal states that may occur during the operation of systems or equipment. Core aspects include behavioral modeling based on historical operational data, the application of time series analysis methods, the identification and evaluation of anomaly patterns, and the fusion and processing of multi-dimensional data. This field systematically integrates statistical analysis, machine learning, and time series modeling methods to continuously observe and analyze the operating parameters of equipment or systems, aiming to predict the evolution trend of anomalies before failures occur. This field is widely used in industrial equipment, network systems, logistics management, and intelligent operation and maintenance scenarios. Its research focus is on building stable and generalizable anomaly detection and prediction models, identifying potential anomaly risks through quantitative indicators, and providing early warnings.

[0003] Traditional storage cabinet anomaly trend prediction systems rely on historical operational data to assess future state trends and identify potential anomalies. These systems typically collect operational parameters such as temperature, current, voltage, and door lock status at different times, using fixed threshold methods based on sliding windows or linear regression prediction models to assess trends and identify potential risks such as temperature increases, voltage instability, or switch failures. However, these methods depend on pre-defined statistical indicators or simple mathematical models, limiting their ability to identify nonlinear behavior, long-term dependencies, or complex change patterns, and making it difficult to fully reveal the path of anomalies. Furthermore, traditional systems often use univariate analysis, mean fluctuation range settings, and anomaly count accumulation as judgment criteria, lacking comprehensive utilization of the correlations between multiple operational data points, significantly limiting their predictive capabilities. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a storage cabinet anomaly trend prediction system based on time-series data analysis.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a storage cabinet anomaly trend prediction system based on time-series data analysis, the system comprising:

[0006] The status monitoring module collects real-time monitoring data during the operation of the storage cabinet, associates it with timestamps to construct a status vector, arranges the status vectors in sequence, and generates storage cabinet operation data.

[0007] The interval sensing module calculates the time interval mapping as a dynamic offset based on the storage cabinet operation data, calculates the temperature gradient change amplitude, performs the dynamic offset to change amplitude ratio calculation, filters intervals exceeding the preset benchmark, and marks abnormal jump segments.

[0008] The path reconstruction module calculates the difference between adjacent state vectors based on the storage cabinet operation data, constructs a jump rate sequence by combining time intervals, calculates the state stability index, removes high-frequency disturbance intervals from the abnormal jump segments, and connects nodes to construct a fault evolution path.

[0009] The symptom activation module calculates the mean and standard deviation of the instantaneous current value of the storage cabinet operation data segment by segment, calculates the ratio to generate the current fluctuation variation coefficient, identifies the interval where the current fluctuation variation coefficient deviates continuously in the same direction and does not fall back, and generates a trend activation signal.

[0010] The evolution prediction module responds to the trend activation signal, inputs the fault evolution path into the long short-term memory network, calculates the future state vector evolution direction, and outputs the abnormal trend prediction result.

[0011] As a further aspect of the present invention, the storage cabinet operation data includes a timestamped state matrix, a multi-dimensional sensor parameter sequence, and a cabinet operation monitoring log; the abnormal jump segment includes a non-uniform sampling interval index, a gradient offset ratio exceeding the limit interval, and a local thermal fluctuation timestamp set; the fault evolution path includes a trend cumulative state node chain, a filtered state transition sequence, and a cumulative mutation trajectory; the trend activation signal includes a continuous deviation trigger identifier, a non-fallback state criterion, and a coefficient of variation threshold exceeding the limit marker; and the abnormal trend prediction result includes a future time state vector sequence, a thermal anomaly confidence interval, and a predicted value for the duration of abnormal temperature rise.

[0012] As a further aspect of the present invention, the status monitoring module includes:

[0013] The multi-dimensional parameter acquisition submodule collects real-time monitoring data during the operation of the storage cabinet. The monitoring data includes temperature, voltage, current and door lock status. It calls the multi-channel signal acquisition card to synchronously sample and digitize each analog quantity, reads the level status of discrete switch signals, and integrates all sensor values ​​acquired at the same sampling trigger moment in parallel and maps them with channel identifiers to generate a multi-dimensional sensor parameter set.

[0014] The temporal vector construction submodule, based on the sampling trigger time of the multidimensional sensing parameter set, calls the synchronous clock source to obtain the corresponding time information and converts it into a unified format timestamp, constructs a feature space including temperature dimension, voltage dimension, current dimension and door lock state dimension, maps each value in the multidimensional sensing parameter set to the corresponding coordinate axis position in the feature space, and embeds the timestamp as a time sequence index dimension to construct a time-stamped state vector;

[0015] The time-series data generation submodule cyclically acquires the time-stamped state vectors within a continuous monitoring cycle and stores them in a first-in-first-out queue buffer. It reads the timestamp indices of all time-stamped state vectors in the buffer, reorders and aligns all vectors based on the monotonically increasing characteristic of the timestamp values, removes conflicting items with duplicate timestamps, and concatenates and matrixes the verified ordered vector sequence according to the time dimension to generate storage cabinet operation data.

[0016] As a further aspect of the present invention, the process of constructing a feature space including temperature, voltage, current, and door lock status dimensions, and mapping each value in the multidimensional sensing parameter set to the corresponding coordinate axis position in the feature space, specifically involves: establishing a multidimensional coordinate system with temperature parameter as the first dimension, voltage parameter as the second dimension, current parameter as the third dimension, and door lock status parameter as the fourth dimension; obtaining the historical extreme value range of each parameter in the multidimensional sensing parameter set; performing linear normalization operation on each value collected at the current moment using the historical extreme value range; and projecting the normalized value onto the corresponding coordinate axis interval of the multidimensional coordinate system.

[0017] The process of reordering and aligning all vectors based on the monotonically increasing characteristic of timestamp values, and eliminating conflicting items with duplicate timestamps, specifically involves: performing bubble sort based on timestamp values ​​on the time-stamped state vectors in the first-in-first-out queue buffer to establish a monotonically non-decreasing temporal order of the time-stamped state vectors; traversing the sorted vector sequence to identify multiple duplicate vector groups with identical timestamp values; for each duplicate vector group, calculating the Euclidean distance between each vector in the group and the vector at the previous time step in the feature space, selecting the time-stamped state vector with the smallest Euclidean distance as valid data to retain, and deleting the remaining time-stamped state vectors in the group from the vector sequence.

[0018] As a further aspect of the present invention, the interval sensing module includes:

[0019] The time offset mapping submodule performs time difference calculation between adjacent data frames based on the timestamp sequence arranged in the storage cabinet operation data to obtain a non-uniform sampling interval. It then compares the non-uniform sampling interval with a preset standard sampling step size and calculates the degree of distortion of the interval relative to the standard step size through a non-linear mapping algorithm. This quantifies the local scaling characteristics on the time axis and generates a dynamic time offset.

[0020] The gradient ratio calculation submodule extracts the temperature values ​​of adjacent moments in the storage cabinet's operating data and calculates the absolute difference. Based on the corresponding time difference, it calculates the instantaneous change amplitude of the temperature gradient, calls the dynamic time offset as the denominator adjustment term, performs the ratio calculation between the temperature gradient change amplitude and the dynamic time offset, and generates a time-temperature difference ratio sequence.

[0021] The jump interval locking submodule calls the preset jump judgment benchmark value, performs point-by-point scanning and numerical discrimination on the time-temperature difference ratio sequence, filters the data point index with a value greater than the jump judgment benchmark value, merges consecutive or adjacent out-of-limit index points into an independent time window, and marks the range covered by the time window in the original data stream as abnormal, generating abnormal jump segments.

[0022] As a further aspect of the present invention, the process of setting the jump judgment benchmark value specifically involves selecting a time-temperature difference ratio sequence generated by the storage cabinet during the steady-state operation cycle to form a reference sample set, calculating the arithmetic mean and standard deviation of the reference sample set, and adding the arithmetic mean and the standard deviation of a preset multiple as the jump judgment benchmark value.

[0023] As a further aspect of the present invention, the path reconstruction module includes:

[0024] The rate sequence construction submodule calculates the difference vector of the state vector at adjacent time points based on the storage cabinet operation data, counts the number of dimensions in the difference vector whose amplitude exceeds the preset noise tolerance, and performs a ratio operation in combination with the time interval of the corresponding sampling points to generate a jump rate sequence.

[0025] The stability index calculation submodule acquires the jump rate sequence and calculates the arithmetic mean and statistical variance, extracts the corresponding voltage fluctuation amplitude and current fluctuation amplitude from the storage cabinet operation data, and calculates and obtains the state stability index.

[0026] The path evolution reconstruction submodule calls the state stability index, compares it with the confidence threshold, removes discrete disturbance points whose state stability index is lower than the confidence threshold, extracts the temporal index and state features of the remaining nodes, establishes a node topology chain with temporal progression relationship, and generates a fault evolution path.

[0027] As a further aspect of the present invention, the symptom activation module includes:

[0028] The segmented statistical calculation submodule extracts the instantaneous current value sequence including time index from the storage cabinet operation data, divides the instantaneous current value sequence into multiple continuous and non-overlapping time segments according to the preset statistical period step size, accumulates and performs variance calculation on the current sampling points in each time segment, calculates the arithmetic mean of the current amplitude and the standard deviation reflecting the fluctuation dispersion value in the segment respectively, and generates a segmented current statistical feature set.

[0029] The coefficient of variation generation submodule calls the segmented current statistical feature set, reads the standard deviation and arithmetic mean corresponding to each statistical period in sequence, performs the ratio calculation of standard deviation and arithmetic mean, quantifies the relative fluctuation intensity of current signal under various load levels, eliminates the influence of current reference value on fluctuation amplitude evaluation, constructs a one-dimensional vector according to time order of the calculated ratio values, and generates current fluctuation coefficient of variation sequence.

[0030] The trend deviation identification submodule performs a sliding window scan based on the current fluctuation variation coefficient sequence to identify the change pattern of the values, determine whether the coefficient values ​​of multiple consecutive cycles deviate in the same direction relative to the historical benchmark, calculate the differential rate of change of the deviation magnitude, filter the intervals with non-negative rates of change to confirm the non-decline trend, and logically set the continuous intervals that meet the conditions to generate a trend activation signal.

[0031] As a further aspect of the present invention, the process of determining whether the coefficient values ​​of multiple consecutive cycles deviate in the same direction relative to the historical benchmark specifically involves: extracting the coefficient of variation of current fluctuations under stable operation within a preset historical time period and calculating the arithmetic mean as the historical benchmark; calculating the difference between the coefficient of variation of current fluctuations and the historical benchmark at each moment within the coverage of the sliding window; verifying whether the difference corresponding to each moment within the sliding window is greater than zero or less than zero; if the verification result is true, it is determined to be a deviation state that maintains the same direction.

[0032] The process of calculating the differential change rate of the deviation amplitude specifically involves obtaining the absolute value sequence of the difference between the current fluctuation variation coefficient and the historical benchmark as the deviation amplitude sequence; calculating the difference between the current value and the previous value in the deviation amplitude sequence; and using the difference value as the differential change rate of the deviation amplitude.

[0033] The process of screening intervals with non-negative rates of change to confirm non-declining trends specifically involves traversing the differential rates of change and extracting the time intervals corresponding to data points with values ​​greater than or equal to zero; and confirming the time intervals as non-declining trends where the deviation amplitude has not shown convergence characteristics.

[0034] As a further aspect of the present invention, the evolution prediction module includes:

[0035] The sequence feature loading submodule calls the node feature data arranged in time sequence included in the fault evolution path, constructs a multidimensional matrix including time step and state variables according to the dimensional requirements of the preset long short-term memory network input layer, performs normalization mapping on the state values ​​in the matrix to eliminate the difference in units, and converts the mapped data into a tensor structure that conforms to the network input format to generate the evolution path feature tensor.

[0036] The evolution direction deduction submodule inputs the evolution path feature tensor into a pre-set long short-term memory network model, performs logical operations of forget gate, input gate and output gate in the hidden layer, updates the memory unit state, performs time backpropagation calculation logic for the time dimension dependency, solves the gradient distribution of the loss function with respect to the state variable, determines the change trajectory of each component of the state in the future time domain according to the inverse or tangent logic of gradient descent, and generates the state vector evolution direction.

[0037] The prediction result generation submodule performs numerical extrapolation calculations for future times based on the evolution direction of the state vector, generates a state vector sequence for future time points by combining a preset prediction step size, calculates the probability confidence interval corresponding to the predicted trajectory and the expected duration of the abnormal temperature rise state, encapsulates the prediction data of time sequence state, probability range and duration, and generates an abnormal trend prediction result.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0039] In this invention, a dynamic expression of state changes is established by constructing a state vector containing temperature, voltage, current, and door lock status, and combining it with timestamp information to form a time-series data sequence. The ratio of time interval to state change amplitude is used to analyze jump characteristics, and jump rate statistics are combined to distinguish between short-term disturbances and trend evolution paths. At the same time, the fluctuation of current data is segmented to extract variation features. Evolution activation signals are identified based on trend maintenance and non-fallback characteristics. On this basis, a neural network structure with long-term dependent learning capability is introduced to capture non-periodic thermal anomaly trends in the state sequence. Through multi-dimensional parameter collaborative processing and path construction logic, the accuracy and timeliness of anomaly identification are improved, effectively overcoming the technical bottlenecks of traditional methods in handling complex change patterns and anomaly path perception. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a system flowchart of the present invention;

[0042] Figure 2 This is a schematic diagram of the system framework of the present invention;

[0043] Figure 3 This is a flowchart of the status monitoring module of the present invention;

[0044] Figure 4 This is a flowchart of the interval sensing module of the present invention;

[0045] Figure 5 This is a flowchart of the path reconstruction module of the present invention;

[0046] Figure 6 This is a flowchart of the symptom activation module of the present invention;

[0047] Figure 7 This is a flowchart of the evolution prediction module of the present invention. Detailed Implementation

[0048] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0049] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0050] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0051] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0052] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0053] Please see Figure 1An anomaly trend prediction system for storage cabinets based on time-series data analysis is provided. The system includes a status monitoring module, an interval sensing module, a path reconstruction module, a symptom activation module, and an evolution prediction module.

[0054] The status monitoring module collects real-time monitoring data during the operation of the storage cabinet. The monitoring data includes temperature, voltage, current and door lock status. It associates the corresponding timestamps to construct a status vector. The status vectors of multiple consecutive moments are arranged in chronological order to generate storage cabinet operation data.

[0055] The interval sensing module calculates the actual time interval based on the timestamps of adjacent data points in the storage cabinet operation data and maps it to a dynamic offset. It calculates the temperature gradient change amplitude within the corresponding interval, performs a two-way difference ratio calculation between the dynamic offset and the change amplitude, and filters out the intervals whose calculation results exceed the preset benchmark to mark abnormal jump segments.

[0056] The path reconstruction module calculates the difference between state vectors at adjacent times based on the storage cabinet operation data, constructs a jump rate sequence by combining the time interval, generates a state stability index based on the mean and variance of the jump rate sequence, removes high-frequency disturbance intervals from abnormal jump segments based on the state stability index, and connects nodes with continuous small-amplitude mutation characteristics to construct a fault evolution path.

[0057] The symptom activation module performs segmented statistical analysis on the instantaneous current values ​​in the storage cabinet's operating data and calculates the mean and standard deviation. It then calculates the ratio of these two values ​​to generate the current fluctuation variation coefficient, identifies the intervals in which the current fluctuation variation coefficient maintains the same direction of deviation over multiple consecutive time periods and the deviation magnitude shows a non-decreasing trend, and generates a trend activation signal.

[0058] The evolution prediction module, in response to the trend activation signal, inputs the fault evolution path into a pre-set long short-term memory network, calls the time backpropagation algorithm to calculate the evolution direction of the state vector at future moments, and outputs the abnormal trend prediction result.

[0059] The storage cabinet operation data includes a timestamped state matrix, a multi-dimensional sensor parameter sequence, and a cabinet operation monitoring log. Abnormal jump segments include a non-uniform sampling interval index, a gradient offset ratio exceeding the limit interval, and a local thermal fluctuation timestamp set. The fault evolution path includes a trend cumulative state node chain, a filtered state transition sequence, and a cumulative mutation trajectory. The trend activation signals include a continuous deviation trigger identifier, a non-fallback state criterion, and a coefficient of variation threshold exceeding the limit marker. The abnormal trend prediction results include a future time state vector sequence, a thermal anomaly confidence interval, and a predicted value for the duration of abnormal temperature rise.

[0060] Please see Figure 2 and Figure 3 The status monitoring module includes:

[0061] The time offset mapping submodule performs time difference calculation between adjacent data frames based on the timestamp sequence arranged in order in the storage cabinet operation data, obtains the non-uniform sampling interval, compares the non-uniform sampling interval with the preset standard sampling step size, calculates the degree of distortion of the interval relative to the standard step size through a non-linear mapping algorithm, quantifies the local scaling characteristics on the time axis, and generates dynamic time offset.

[0062] For example, the timestamp of frame 1 is milliseconds, the second frame is milliseconds, the 3rd frame is milliseconds, then the actual time intervals are respectively milliseconds and Milliseconds. These non-uniform sampling intervals are compared with a preset standard sampling step size (e.g., set to milliseconds). Numerical comparisons were performed using milliseconds. An exponential mapping algorithm was employed to quantify the local scaling characteristics on the time axis; the calculation formula is: dynamic time offset equals the natural constant. The exponent is calculated by dividing the actual time interval by the standard sampling step size. The specific data generated based on this logic is shown in Table 1.

[0063] Table 1. Calculation of Sampling Interval and Dynamic Offset

[0064] ;

[0065] As shown in Table 1, the deviation of the time interval is mapped to a nonlinear dynamic time offset through exponential operation, which is used for subsequent gradient correction.

[0066] The gradient ratio calculation submodule extracts the temperature values ​​of adjacent moments in the storage cabinet operation data and calculates the absolute difference. Based on the corresponding time difference, it calculates the instantaneous change amplitude of the temperature gradient, calls the dynamic time offset as the denominator adjustment term, performs the ratio calculation between the temperature gradient change amplitude and the dynamic time offset, eliminates the influence of uneven sampling interval on the gradient amplitude, and generates a time-temperature difference ratio sequence.

[0067] For example, time Temperature normalization value ,time Temperature normalization value Calculate the absolute difference as Based on the corresponding time difference (e.g.) (seconds), calculate the instantaneous change amplitude of the temperature gradient as Every second. The corresponding dynamic time offset from Table 1 is used as the denominator adjustment term. If the dynamic time offset at the corresponding moment is... Then the ratio operation is performed: The corrected ratio is approximately Perform this operation on all data points in the sequence to generate a time-temperature difference ratio sequence.

[0068] The jump interval locking submodule calls the preset jump judgment benchmark value, performs point-by-point scanning and numerical discrimination on the time-temperature difference ratio sequence, filters the data point index with a value greater than the jump judgment benchmark value, merges consecutive or adjacent out-of-limit index points into independent time windows, and marks the range covered by the time window in the original data stream as abnormal, generating abnormal jump segments.

[0069] The specific process of setting the jump judgment benchmark value is as follows: select the time temperature difference ratio sequence generated by the storage cabinet during the steady-state operation cycle to form a reference sample set, calculate the arithmetic mean and standard deviation of the reference sample set, and add the arithmetic mean and the standard deviation of the preset multiple as the jump judgment benchmark value.

[0070] Selecting storage cabinets in the past The time-temperature difference ratio sequence generated within the hourly steady-state operating cycle constitutes a reference sample set, and the arithmetic mean of this reference sample set is calculated as follows: The standard deviation is The arithmetic mean and Add up the standard deviations by a factor of 1, that is This serves as the benchmark value for jump determination. The current time-temperature difference ratio sequence is scanned point by point; if the ratio calculation result at a certain point is... (greater than) If a data point index is found to be out of bounds (e.g., index), then the data point index for that point will be filtered. This will filter out consecutive or adjacent out-of-bounds index points (e.g., indexes). to The data is merged into an independent time window, and the range covered by the time window is marked as abnormal in the original data stream to generate abnormal jump segments.

[0071] Please see Figure 2 and Figure 5 The path reconstruction module includes:

[0072] The rate sequence construction submodule calculates the difference vector of the state vector at adjacent time points based on the storage cabinet operation data, counts the number of dimensions in the difference vector whose amplitude exceeds the preset noise tolerance, and performs a ratio calculation in combination with the time interval of the corresponding sampling points to generate a jump rate sequence.

[0073] For example, the vector at the current time is The vector at the previous time step was Then the absolute value of the difference vector is The magnitude of the statistical difference vector exceeds the preset noise tolerance (e.g., ...). The number of dimensions, here is Each dimension. Combined with the time interval of the corresponding sampling points (e.g.) (seconds), number of execution dimensions Divide by time interval The ratio operation yields the rate value. Perform this operation on all neighboring points to generate a sequence of jump rates.

[0074] The stability index calculation submodule acquires the jump rate sequence and calculates its arithmetic mean and statistical variance. It then extracts the corresponding voltage and current fluctuation amplitudes from the storage cabinet's operating data, using the following formula:

[0075] ;

[0076] Calculate and obtain state stability indicators;

[0077] in, This represents an index of state stability. This represents the sequence window length, which is obtained by setting a preset number of sampling points. The normalized value of the jump rate at time i is obtained by calling the jump rate sequence and dividing it by a preset maximum rate benchmark value. The normalized mean of the jump rate sequence is calculated within a window. Obtaining the arithmetic mean The normalized variance representing the jump rate sequence is calculated within a window. The statistical variance was obtained. This represents the prevention and elimination of zero minute quantities, obtained through a preset non-zero minimum constant. The normalized value of the voltage fluctuation at time i is obtained by calculating the absolute value of the voltage difference and dividing it by the rated voltage. The normalized value of the current fluctuation at time i is obtained by calculating the absolute value of the current difference and dividing it by the rated current. The power normalization reference factor is obtained by setting a standardization coefficient per unit power. The time-weighted factor is obtained through a system-preset weighting of the time-related effects. The normalized value of the time interval at time i is obtained by calculating the difference between adjacent timestamps. This represents the normalized value of the average time interval, calculated by considering all values ​​within the window. The average value is obtained, and the confidence threshold represents the criterion for path selection. It is obtained by analyzing the statistical distribution of the state stability index under historical normal conditions and setting quantile boundaries.

[0078] Set sequence window length for The maximum rate reference value is , prevent the removal of tiny amounts of for Rated voltage is Volts, rated current is Ampere, power normalization reference factor for Time weighting factor for Suppose at a certain moment... The parameters are as follows: normalized value of the jump rate. Normalized mean of the jump rate sequence within the window Normalized variance ;time voltage fluctuation normalized value Normalized value of current fluctuation ;time normalized value of time interval Normalized value of average time interval within the window The specific operational logic is as follows:

[0079] First, calculate the rate fluctuation term: .

[0080] Then calculate the time-weighted term: .

[0081] Finally, combine and calculate the single-point indicators: .

[0082] All within the window The final state stability index is obtained by summing and averaging the individual point indices at each time point. If all parameters remain consistent within the window, the result is... .

[0083] The path evolution reconstruction submodule calls the state stability index, compares it with the confidence threshold, removes discrete disturbance points whose state stability index is lower than the confidence threshold, extracts the temporal index and state features of the remaining nodes, establishes a node topology chain with temporal progression relationship, and generates a fault evolution path.

[0084] Set the confidence threshold as Based on the above calculation results ,because If the node is determined to be a low-intensity discrete disturbance point or background noise, it is discarded; if the index calculated at a certain moment is (greater than) If the node's temporal index and state features are extracted, the remaining nodes that meet the threshold conditions are connected in chronological order to establish a node topology chain with a temporal progression relationship, thus generating a fault evolution path.

[0085] Please see Figure 2 and Figure 6 The symptom activation module includes:

[0086] The segmented statistical calculation submodule extracts the instantaneous current value sequence, including the time index, from the storage cabinet operation data. Based on the preset statistical period step size, the instantaneous current value sequence is divided into multiple continuous and non-overlapping time segments. The current sampling points in each time segment are accumulated and variance is calculated. The arithmetic mean of the current amplitude and the standard deviation reflecting the fluctuation dispersion are calculated respectively to generate a segmented current statistical feature set.

[0087] Based on the preset statistical period step size (e.g.) The instantaneous current value sequence is divided into multiple continuous and non-overlapping time segments (seconds). Statistical operations are performed on the current sampling points within each time segment. For example, in the first... Captured within a 1-second segment Given a number of current points, calculate their arithmetic mean. Ampere, standard deviation value is Ampere. Perform this operation on all segments to generate a segmented current statistical feature set.

[0088] The coefficient of variation generation submodule calls the segmented current statistical feature set, reads the standard deviation and arithmetic mean values ​​corresponding to each statistical period in sequence, performs the ratio calculation of the standard deviation and arithmetic mean, quantifies the relative fluctuation intensity of the current signal under various load levels, eliminates the influence of the current reference value on the evaluation of fluctuation amplitude, constructs a one-dimensional vector according to the time sequence of the calculated ratio values, and generates the current fluctuation coefficient of variation sequence.

[0089] Based on the aforementioned data, the calculation logic is as follows: The result was obtained. The calculated ratio values ​​were arranged in chronological order to construct a one-dimensional vector, generating a sequence of current fluctuation variation coefficients. Some calculation results are shown in Table 2.

[0090] Table 2. Statistics and Coefficient of Variation of Current Segments

[0091] ;

[0092] The trend deviation identification submodule performs a sliding window scan based on the current fluctuation coefficient of variation sequence to identify the change pattern of the values, determine whether the coefficient values ​​of multiple consecutive periods deviate in the same direction relative to the historical benchmark, calculate the differential rate of change of the deviation magnitude, filter the intervals with non-negative rates of change to confirm the non-fallback trend, and set the logical bits of the consecutive intervals that meet the conditions to generate a trend activation signal.

[0093] The process of determining whether the coefficient values ​​of multiple consecutive cycles deviate in the same direction relative to the historical benchmark is as follows: extract the coefficient of variation of current fluctuations under stable operation within a preset historical time period and calculate the arithmetic mean as the historical benchmark; calculate the difference between the coefficient of variation of current fluctuations and the historical benchmark at each moment within the coverage of the sliding window; verify whether the difference corresponding to each moment within the sliding window is greater than zero or less than zero; if the verification result is true, it is determined that the deviation is maintained in the same direction.

[0094] The specific process for calculating the differential rate of change of the deviation amplitude is as follows: obtain the absolute value sequence of the difference between the coefficient of variation of the current fluctuation and the historical benchmark as the deviation amplitude sequence; calculate the difference between the current value and the previous value in the deviation amplitude sequence; and use the difference value as the differential rate of change of the deviation amplitude.

[0095] The process of screening intervals with non-negative rates of change to confirm non-declining trends involves traversing the differential rates of change and extracting the time intervals corresponding to data points with values ​​greater than or equal to zero; and confirming the time intervals as non-declining trends where the deviation magnitude has not shown convergence characteristics.

[0096] Extract the coefficient of variation of current fluctuations during a historical preset time period (e.g., the past 24 hours) under stable operating conditions and calculate their arithmetic mean. The result is... This is used as a historical baseline. The difference between the coefficient at each moment within the sliding window coverage area and the historical baseline is calculated point-by-point. Based on the data in Table 2, the difference for segment 1 is... The difference in segment 2 is The difference in segment 3 is The difference in segment 4 is The verification revealed that all differences were greater than zero, indicating a deviation in the same direction. Subsequently, the rate of change of the deviation magnitude was calculated: the rate of change of segment 2 relative to segment 1 was... The rate of change of fragment 3 relative to fragment 2 By traversing the data, it was found that all the change rate values ​​were non-negative, confirming that the time interval was a non-declining trend in which the deviation amplitude did not show convergence characteristics. Logical settings were then applied to this continuous interval to generate a trend activation signal.

[0097] Please see Figure 2 and Figure 7 The evolution prediction module includes:

[0098] The sequence feature loading submodule calls the node feature data arranged in time sequence included in the fault evolution path. Based on the dimensional requirements of the preset long short-term memory network input layer, it constructs a multi-dimensional matrix including time step and state variables, performs normalization mapping on the state values ​​in the matrix to eliminate dimensional differences, and converts the mapped data into a tensor structure that conforms to the network input format to generate the evolution path feature tensor.

[0099] Based on the dimensional requirements of the pre-defined Long Short-Term Memory (LSTM) network input layer (time step set to...) The feature dimension is set to Construct a multidimensional matrix including the time step and state variables. Perform a normalization mapping on the state values ​​within the matrix, scaling all data to [a specific scale]. arrive Within the interval, the mapped data is converted into a shape of... The three-dimensional tensor structure generates the evolution path feature tensor.

[0100] The evolution direction deduction submodule inputs the evolution path feature tensor into a pre-set long short-term memory network model, performs logical operations of forget gate, input gate and output gate in the hidden layer, updates the memory unit state, performs time backpropagation calculation logic for the time dimension dependency, solves the gradient distribution of the loss function with respect to the state variable, determines the change trajectory of each component of the state in the future time domain according to the inverse or tangent logic of gradient descent, and generates the state vector evolution direction.

[0101] The Long Short-Term Memory (LSTM) network model architecture consists of one input layer and two stacked LSTM hidden layers (the first layer). One neuron, second layer The network consists of one hidden layer (containing 10 neurons) and a fully connected output layer. Inside the hidden layers, the forget gate uses the Sigmoid function to process the previous state, the input gate combines the Sigmoid and Tanh functions to update the current memory unit, and the output gate generates the output value based on the current state. During model training, the mean squared error (MSE) is used as the loss function, and the Adam optimizer is used to update the weights. During the extrapolation phase, based on the trained network parameters, the gradient distribution of the loss function with respect to the state variables is calculated. The trajectory of each component of the state in the future time domain is determined based on the inverse or tangent logic of gradient descent, generating the direction of state vector evolution.

[0102] The prediction result generation submodule performs numerical extrapolation calculations for future moments based on the evolution direction of the state vector, generates a state vector sequence for future time points by combining the preset prediction step size, calculates the probability confidence interval corresponding to the predicted trajectory and the expected duration of the abnormal temperature rise state, encapsulates the prediction data of time sequence state, probability range and duration, and generates abnormal trend prediction results.

[0103] Based on a preset prediction step size (e.g., future) (minutes), generating a sequence of state vectors for future time points. Calculate the probability confidence interval (e.g., minutes) corresponding to the predicted trajectory. The confidence interval, and the expected duration of the abnormal temperature rise. If the prediction results show that the temperature will continue to rise in the future... If the value continues to rise within minutes and exceeds the warning threshold, calculate the duration of this over-limit state (e.g., ...). (seconds). The predicted data, including time series status, probability range, and duration, are encapsulated to generate anomaly trend prediction results. See Table 3 for specific prediction data examples.

[0104] Table 3. Example of Abnormal Trend Prediction Results

[0105] ;

[0106] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0107] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0108] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0109] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0110] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0111] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0112] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0113] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0114] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0115] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A storage cabinet anomaly trend prediction system based on time-series data analysis, characterized in that, The system includes: The status monitoring module collects real-time monitoring data during the operation of the storage cabinet. The monitoring data includes temperature, voltage, current and door lock status. It associates the corresponding timestamps to construct a status vector. The status vectors of multiple consecutive moments are arranged in chronological order to generate storage cabinet operation data. The interval sensing module calculates the actual time interval based on the timestamps of adjacent data points in the storage cabinet operation data and maps it to a dynamic offset. It calculates the temperature gradient change amplitude within the corresponding interval, performs a two-way difference ratio calculation between the dynamic offset and the change amplitude, and filters out the intervals whose calculation results exceed the preset benchmark to mark abnormal jump segments. The path reconstruction module calculates the difference between state vectors at adjacent times based on the storage cabinet operation data, constructs a jump rate sequence by combining the time interval, obtains the jump rate sequence and calculates the arithmetic mean and statistical variance, extracts the corresponding voltage fluctuation amplitude and current fluctuation amplitude from the storage cabinet operation data, calculates and obtains the state stability index, removes high-frequency disturbance intervals from the abnormal jump segments based on the state stability index, and connects nodes with continuous mutation characteristics to construct a fault evolution path. The symptom activation module performs segmented statistical analysis on the instantaneous current values ​​in the storage cabinet's operating data and calculates the mean and standard deviation. It then calculates the ratio of the two to generate the current fluctuation variation coefficient, identifies the intervals in which the current fluctuation variation coefficient maintains the same direction of deviation over multiple consecutive time periods and the deviation magnitude shows a non-decreasing trend, and generates a trend activation signal. The evolution prediction module responds to the trend activation signal by inputting the fault evolution path into a preset long short-term memory network, calling the time backpropagation algorithm to calculate the evolution direction of the state vector at future moments, and outputting the abnormal trend prediction result.

2. The storage cabinet anomaly trend prediction system based on time-series data analysis according to claim 1, characterized in that, The status monitoring module includes: The multi-dimensional parameter acquisition submodule collects real-time monitoring data during the operation of the storage cabinet. The monitoring data includes temperature, voltage, current and door lock status. It calls the multi-channel signal acquisition card to synchronously sample and digitize each analog quantity, reads the level status of discrete switch signals, and integrates all sensor values ​​acquired at the same sampling trigger moment in parallel and maps them with channel identifiers to generate a multi-dimensional sensor parameter set. The temporal vector construction submodule, based on the sampling trigger time of the multidimensional sensing parameter set, calls the synchronous clock source to obtain the corresponding time information and converts it into a unified format timestamp, constructs a feature space including temperature dimension, voltage dimension, current dimension and door lock state dimension, maps each value in the multidimensional sensing parameter set to the corresponding coordinate axis position in the feature space, and embeds the timestamp as a time sequence index dimension to construct a time-stamped state vector; The time-series data generation submodule cyclically acquires the time-stamped state vectors within a continuous monitoring cycle and stores them in a first-in-first-out queue buffer. It reads the timestamp indices of all time-stamped state vectors in the buffer, reorders and aligns all vectors based on the monotonically increasing characteristic of the timestamp values, removes conflicting items with duplicate timestamps, and concatenates and matrixes the verified ordered vector sequence according to the time dimension to generate storage cabinet operation data.

3. The storage cabinet anomaly trend prediction system based on time-series data analysis according to claim 2, characterized in that, The process of constructing a feature space including temperature, voltage, current, and door lock status dimensions, and mapping each value in the multidimensional sensing parameter set to the corresponding coordinate axis position in the feature space, specifically involves: establishing a multidimensional coordinate system with temperature as the first dimension, voltage as the second dimension, current as the third dimension, and door lock status as the fourth dimension; obtaining the historical extreme value range of each parameter in the multidimensional sensing parameter set; performing linear normalization on each value collected at the current moment using the historical extreme value range; and projecting the normalized value onto the corresponding coordinate axis interval of the multidimensional coordinate system. The process of reordering and aligning all vectors based on the monotonically increasing characteristic of timestamp values, and eliminating conflicting items with duplicate timestamps, specifically involves: performing bubble sort based on timestamp values ​​on the time-stamped state vectors in the first-in-first-out queue buffer to establish a monotonically non-decreasing temporal order of the time-stamped state vectors; traversing the sorted vector sequence to identify multiple duplicate vector groups with identical timestamp values; for each duplicate vector group, calculating the Euclidean distance between each vector in the group and the vector at the previous time step in the feature space, selecting the time-stamped state vector with the smallest Euclidean distance as valid data to retain, and deleting the remaining time-stamped state vectors in the group from the vector sequence.

4. The storage cabinet anomaly trend prediction system based on time-series data analysis according to claim 2, characterized in that, The interval sensing module includes: The time offset mapping submodule performs time difference calculation between adjacent data frames based on the timestamp sequence arranged in the storage cabinet operation data to obtain a non-uniform sampling interval. It then compares the non-uniform sampling interval with a preset standard sampling step size and calculates the degree of distortion of the interval relative to the standard step size through a non-linear mapping algorithm. This quantifies the local scaling characteristics on the time axis and generates a dynamic offset. The gradient ratio calculation submodule extracts the temperature values ​​of adjacent moments in the storage cabinet's operating data and calculates the absolute difference. Based on the corresponding time difference, it calculates the instantaneous change amplitude of the temperature gradient, calls the dynamic offset as the denominator adjustment term, performs the ratio calculation between the temperature gradient change amplitude and the dynamic offset, and generates a time-temperature difference ratio sequence. The jump interval locking submodule calls the preset jump judgment benchmark value, performs point-by-point scanning and numerical discrimination on the time-temperature difference ratio sequence, filters the data point index with a value greater than the jump judgment benchmark value, merges consecutive or adjacent out-of-limit index points into an independent time window, and marks the range covered by the time window in the original data stream as abnormal, generating abnormal jump segments.

5. The storage cabinet anomaly trend prediction system based on time-series data analysis according to claim 4, characterized in that, The process of setting the jump judgment benchmark value is as follows: select the time-temperature difference ratio sequence generated by the storage cabinet during the steady-state operation cycle to form a reference sample set, calculate the arithmetic mean and standard deviation of the reference sample set, and add the arithmetic mean and the standard deviation of the preset multiple as the jump judgment benchmark value.

6. The storage cabinet anomaly trend prediction system based on time-series data analysis according to claim 5, characterized in that, The path reconstruction module includes: The rate sequence construction submodule calculates the difference vector of the state vector at adjacent time points based on the storage cabinet operation data, counts the number of dimensions in the difference vector whose amplitude exceeds the preset noise tolerance, and performs a ratio operation in combination with the time interval of the corresponding sampling points to generate a jump rate sequence. The stability index calculation submodule acquires the jump rate sequence and calculates the arithmetic mean and statistical variance, extracts the corresponding voltage fluctuation amplitude and current fluctuation amplitude from the storage cabinet operation data, and calculates and obtains the state stability index. The path evolution reconstruction submodule calls the state stability index, compares it with the confidence threshold, removes discrete disturbance points whose state stability index is lower than the confidence threshold, extracts the temporal index and state features of the remaining nodes, establishes a node topology chain with temporal progression relationship, and generates a fault evolution path.

7. The storage cabinet anomaly trend prediction system based on time-series data analysis according to claim 6, characterized in that, The symptom activation module includes: The segmented statistical calculation submodule extracts the instantaneous current value sequence including time index from the storage cabinet operation data, divides the instantaneous current value sequence into multiple continuous and non-overlapping time segments according to the preset statistical period step size, accumulates and performs variance calculation on the current sampling points in each time segment, calculates the arithmetic mean of the current amplitude and the standard deviation reflecting the fluctuation dispersion value in the segment respectively, and generates a segmented current statistical feature set. The coefficient of variation generation submodule calls the segmented current statistical feature set, reads the standard deviation and arithmetic mean corresponding to each statistical period in sequence, performs the ratio calculation of standard deviation and arithmetic mean, quantifies the relative fluctuation intensity of current signal under various load levels, eliminates the influence of current reference value on fluctuation amplitude evaluation, constructs a one-dimensional vector according to time order of the calculated ratio values, and generates current fluctuation coefficient of variation sequence. The trend deviation identification submodule performs a sliding window scan based on the current fluctuation variation coefficient sequence to identify the change pattern of the values, determine whether the coefficient values ​​of multiple consecutive cycles deviate in the same direction relative to the historical benchmark, calculate the differential rate of change of the deviation magnitude, filter the intervals with non-negative rates of change to confirm the non-decline trend, and logically set the continuous intervals that meet the conditions to generate a trend activation signal.

8. The storage cabinet anomaly trend prediction system based on time-series data analysis according to claim 7, characterized in that, The process of determining whether the coefficient values ​​of multiple consecutive cycles deviate in the same direction relative to the historical benchmark specifically involves extracting the coefficient of variation of current fluctuations in a stable operating state within a preset historical time period and calculating the arithmetic mean as the historical benchmark; and calculating the difference between the coefficient of variation of current fluctuations and the historical benchmark at each moment within the coverage area of ​​the sliding window. Verify whether the difference at each moment within the sliding window is greater than zero or less than zero; if the verification result is true, it is determined to be a deviation state that maintains the same direction. The process of calculating the differential change rate of the deviation amplitude specifically involves obtaining the absolute value sequence of the difference between the current fluctuation variation coefficient and the historical benchmark as the deviation amplitude sequence; calculating the difference between the current value and the previous value in the deviation amplitude sequence; and using the difference value as the differential change rate of the deviation amplitude. The process of screening intervals with non-negative rates of change to confirm non-declining trends specifically involves traversing the differential rates of change and extracting the time intervals corresponding to data points with values ​​greater than or equal to zero; and confirming the time intervals as non-declining trends where the deviation amplitude has not shown convergence characteristics.

9. The storage cabinet anomaly trend prediction system based on time-series data analysis according to claim 8, characterized in that, The evolution prediction module includes: The sequence feature loading submodule calls the node feature data arranged in time sequence included in the fault evolution path, constructs a multidimensional matrix including time step and state variables according to the dimensional requirements of the preset long short-term memory network input layer, performs normalization mapping on the state values ​​in the matrix to eliminate the difference in units, and converts the mapped data into a tensor structure that conforms to the network input format to generate the evolution path feature tensor. The evolution direction deduction submodule inputs the evolution path feature tensor into a pre-set long short-term memory network model, performs logical operations of forget gate, input gate and output gate in the hidden layer, updates the memory unit state, performs time backpropagation calculation logic for the time dimension dependency, solves the gradient distribution of the loss function with respect to the state variable, determines the change trajectory of each component of the state in the future time domain according to the inverse or tangent logic of gradient descent, and generates the state vector evolution direction. The prediction result generation submodule performs numerical extrapolation calculations for future times based on the evolution direction of the state vector, generates a state vector sequence for future time points by combining a preset prediction step size, calculates the probability confidence interval corresponding to the predicted trajectory and the expected duration of the abnormal temperature rise state, encapsulates the prediction data of time sequence state, probability range and duration, and generates an abnormal trend prediction result.

Images