Multi-working-condition power coupling coefficient adaptive early warning method and device and storage medium

Abstract

The application provides a multi-condition power coupling coefficient adaptive early warning method and device and a storage medium. The method comprises the following steps: calculating a power coupling coefficient sequence, determining a rated power reference value, and dividing power data into multiple condition intervals; performing condition labeling on the cleaned power coupling coefficient sequence to obtain a power coupling coefficient sub-sequence; calculating the mean and quartile deviation of each condition, and obtaining the optimal threshold coefficient corresponding to each condition; determining the threshold band of the condition, and performing out-of-bound detection on the power coupling coefficient sub-sequence of each condition to obtain an abnormal point set; generating a correlation analysis result between the load level and the power coupling coefficient feature; generating multi-level early warning information according to the length of the continuous sequence of abnormal points, identifying extreme abnormal events, and classifying them into preset early warning levels; and outputting the multi-level early warning information corresponding to each condition and the correlation analysis result. The application can improve the early warning accuracy and stability, reduce false positives and false negatives, and enhance cross-condition risk insight.

Description

Technical Field

[0001] This application relates to the field of industrial equipment condition monitoring technology, and in particular to a method, device and storage medium for adaptive early warning of multi-condition power coupling coefficient. Background Technology

[0002] During operation, the coupling relationship between torque, speed, and power of power equipment changes with factors such as load level, operating condition switching, and component wear. The power coupling coefficient can be used to characterize the changing state of this coupling relationship. In industrial settings, torque, speed, and power data are typically collected to calculate a power coupling coefficient sequence. Based on rated power, operating conditions are categorized into full load, overload, and other conditions. Anomalies in the power coupling coefficient within each condition are detected and warned of. In existing technologies, a common approach is to construct a threshold band using interquartile ranges within a single operating condition. Upper and lower thresholds are determined by fixing threshold coefficients and interquartile ranges, and whether the power coupling coefficient exceeds these limits is used as the criterion for anomaly judgment, displayed using curves and threshold lines.

[0003] However, the existing solutions still have the following shortcomings: First, the threshold parameters lack adaptability to different operating conditions. Fixed threshold coefficients are difficult to match the differences in the dispersion of power coupling coefficients under different operating conditions, which can easily lead to misjudgment under operating conditions with large dispersion or missed judgment under operating conditions with small dispersion. Second, there is a lack of operating condition correlation analysis. Thresholds are often calculated and anomalies are determined independently for a single operating condition. There is a lack of cross-operating condition comparison and integration, making it difficult to form correlation information between load level and power coupling coefficient characteristics. Third, the early warning granularity is insufficient. Usually, binary results of abnormal and normal are output, without hierarchical characterization of continuous anomalies and extreme anomalies, which is not conducive to differentiated decision-making in operation and maintenance. Fourth, the optimization mechanism is insufficient. Threshold parameters mostly rely on manual debugging and are difficult to adaptively update with changes in operating data. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method, device and storage medium for adaptive early warning of power coupling coefficient under multiple operating conditions, in order to solve the problems of poor threshold operating condition adaptability, lack of cross-operating condition correlation analysis and imprecise early warning classification in the prior art.

[0005] A first aspect of this application provides a multi-condition power coupling coefficient adaptive early warning method, comprising: acquiring torque data, speed data, and power data during the operation of a power equipment; calculating a power coupling coefficient sequence based on the torque data, speed data, and power data; removing invalid values ​​from the power coupling coefficient sequence to obtain a cleaned power coupling coefficient sequence; determining a rated power reference value based on the power data; dividing the power data into multiple condition intervals according to the rated power reference value; labeling the cleaned power coupling coefficient sequence according to the multiple condition intervals to obtain a power coupling coefficient sub-sequence corresponding to each condition; calculating the mean and interquartile range of each condition for the power coupling coefficient sub-sequence corresponding to each condition; and using a threshold coefficient as the parameter to be optimized, employing swarm intelligence optimization within a preset range. The algorithm independently optimizes the threshold coefficient for each operating condition, obtaining the optimal threshold coefficient for each condition. Based on the mean, interquartile range, and optimal threshold coefficient for each operating condition, the threshold band for that condition is determined. Out-of-bounds detection is then performed on the power coupling coefficient subsequence for each operating condition according to the threshold band, obtaining the set of outliers for each condition. Cross-condition integrated analysis is then performed on the optimal threshold coefficient, power coupling coefficient statistical characteristics, and outlier set for each operating condition, generating correlation analysis results between load level and power coupling coefficient characteristics. Based on the outlier set for each operating condition, continuous sequences of outliers are identified. Multi-level early warning information is generated based on the length of the continuous sequence of outliers. Extreme thresholds are constructed based on the threshold band to identify extreme abnormal events and categorize them into preset early warning levels. Finally, the multi-level early warning information and correlation analysis results for each operating condition are output.

[0006] A second aspect of this application provides a multi-condition power coupling coefficient adaptive early warning device, comprising: an acquisition module configured to acquire torque data, speed data, and power data during the operation of a power equipment, calculate a power coupling coefficient sequence based on the torque data, speed data, and power data, and remove invalid values ​​from the power coupling coefficient sequence to obtain a cleaned power coupling coefficient sequence; a determination module configured to determine a rated power reference value based on the power data, divide the power data into multiple condition intervals according to the rated power reference value, and label the cleaned power coupling coefficient sequence according to the multiple condition intervals to obtain a power coupling coefficient sub-sequence corresponding to each condition; and an optimization module configured to calculate the mean and interquartile range of each condition for the power coupling coefficient sub-sequence corresponding to each condition, and use a threshold coefficient as the parameter to be optimized, and employ swarm intelligence optimization within a preset range. The algorithm independently optimizes the threshold coefficient for each operating condition to obtain the optimal threshold coefficient for each condition. The detection module is configured to determine the threshold band for each operating condition based on the mean, interquartile range, and optimal threshold coefficient, and to perform out-of-bounds detection on the power coupling coefficient subsequence for each operating condition based on the threshold band to obtain the set of anomalies for each operating condition. The analysis module is configured to perform cross-operating condition integrated analysis on the optimal threshold coefficient, power coupling coefficient statistical characteristics, and anomaly set for each operating condition to generate correlation analysis results between load level and power coupling coefficient characteristics. The output module is configured to identify continuous sequences of anomalies based on the anomaly set for each operating condition, generate multi-level early warning information based on the length of the continuous sequence of anomalies, construct extreme thresholds based on the threshold band to identify extreme abnormal events and classify them into preset early warning levels, and output multi-level early warning information and correlation analysis results for each operating condition.

[0007] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0008] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects:

[0009] By acquiring torque, speed, and power data during the operation of power equipment, and calculating a power coupling coefficient sequence based on these data, invalid values ​​are removed from the power coupling coefficient sequence to obtain a cleaned power coupling coefficient sequence. A rated power reference value is determined based on the power data, and the power data is divided into multiple operating condition intervals according to this reference value. The cleaned power coupling coefficient sequence is then labeled according to these operating condition intervals to obtain a power coupling coefficient subsequence corresponding to each operating condition. For each operating condition's power coupling coefficient subsequence, the mean and interquartile range of each operating condition are calculated. Using a threshold coefficient as the parameter to be optimized, a swarm intelligence optimization algorithm is employed within a preset range to independently optimize the threshold coefficient for each operating condition. This method obtains the optimal threshold coefficient for each operating condition; determines the threshold band for each operating condition based on the mean, interquartile range, and optimal threshold coefficient; performs boundary detection on the power coupling coefficient subsequence for each operating condition according to the threshold band, and obtains the set of anomalies for each operating condition; performs cross-operating condition integrated analysis on the optimal threshold coefficient, power coupling coefficient statistical characteristics, and anomaly set for each operating condition, generating correlation analysis results between load level and power coupling coefficient characteristics; identifies continuous sequences of anomalies based on the anomaly set for each operating condition, generates multi-level early warning information based on the length of the continuous sequences of anomalies, constructs extreme thresholds based on the threshold band to identify extreme abnormal events and classify them into preset early warning levels; and outputs multi-level early warning information and correlation analysis results for each operating condition. This application can improve the accuracy and stability of early warnings, reduce false alarms and missed alarms, and enhance cross-operating condition risk insight. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating the adaptive early warning method for multi-condition power coupling coefficient provided in an embodiment of this application.

[0012] Figure 2 This is a schematic diagram of the structure of the multi-condition power coupling coefficient adaptive early warning device provided in the embodiments of this application;

[0013] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0014] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0015] During the operation of power equipment, abnormal fluctuations in the power coupling coefficient (K value) directly reflect potential faults such as equipment wear and load imbalance. Existing technologies suffer from three major problems:

[0016] 1. Poor threshold adaptability: The warning threshold is calculated using a fixed interquartile range (IQR) coefficient (e.g., k1=1.5), which cannot adapt to the difference in the dispersion of K value under different operating conditions, resulting in missed alarms when fully loaded and false alarms when overloaded.

[0017] 2. Lack of correlation between operating conditions: The K-value characteristics and early warning parameters under multiple operating conditions are not integrated and analyzed, making it impossible to uncover the intrinsic correlation between load levels and abnormal risks, which is detrimental to overall operation and maintenance decisions. In other words, isolated analysis of single operating condition data fails to uncover the intrinsic correlation between load levels and K-values, making it impossible to predict faults across operating conditions.

[0018] 3. Insufficient precision in early warning: It only judges the binary state of "abnormal / normal" without distinguishing the severity of the fault, which leads to excessive or delayed operation and maintenance response.

[0019] Currently, the mainstream solution for power coupling coefficient anomaly early warning in the industrial field is the "fixed threshold + single-condition detection" mode:

[0020] Data acquisition: Obtain raw data on the torque, speed, and power of the equipment during operation;

[0021] Feature calculation: The power coupling coefficient K is calculated using K = torque × speed / power;

[0022] Operating conditions are divided based on the rated power P_rated, using a fixed ratio (e.g., full load [0.8P_rated-1.0P_rated]).

[0023] Threshold calculation: The warning threshold is calculated using a fixed coefficient k1 (usually 1.5) combined with the interquartile range (IQR). The formula is: threshold increment ΔK = k1 × IQR (IQR = Q3 - Q1, where Q1 is the 25th quartile and Q3 is the 75th quartile); upper threshold = K_mean + ΔK, lower threshold = K_mean - ΔK;

[0024] Anomaly detection: Anomalies are only detected by whether the "K value exceeds the upper / lower" range, without tiered warnings;

[0025] Visualization: The K-value curve for a single working condition and a fixed threshold are displayed without cross-working condition comparisons or warning level markings.

[0026] However, the aforementioned existing technical solutions still have the following drawbacks:

[0027] 1. Threshold lacks adaptability to different operating conditions: A fixed k1 cannot match the variation in the dispersion of K values ​​under different operating conditions (e.g., the K value dispersion is large under overload conditions, and a fixed k1=1.5 will result in an overly narrow threshold, causing a large amount of normal data to be misjudged as abnormal).

[0028] 2. Limited warning levels: The warnings are based solely on a binary "abnormal / normal" approach, which fails to differentiate between minor (occasional anomalies), moderate (3-5 consecutive anomalies), and severe (6 or more consecutive or extreme anomalies) risks. This makes it difficult for maintenance personnel to accurately allocate resources and to clearly distinguish the degree of risk.

[0029] 3. Lack of correlation analysis of working conditions: The characteristics of K-values, threshold parameters and early warning effects of different working conditions were not compared, and the correlation between load level and abnormal risk could not be revealed;

[0030] 4. Lack of optimization mechanism: There is no adaptive optimization algorithm, and the threshold parameters need to be manually adjusted repeatedly, which is inefficient and has poor adaptability.

[0031] In view of the problems existing in the prior art, this application proposes an adaptive early warning method and system for multi-condition power coupling coefficient based on a hierarchical integration strategy. This application belongs to the field of industrial equipment condition monitoring and fault early warning technology, specifically involving the abnormal detection of power coupling coefficients in multi-condition power equipment (motors, engines, etc.), and is particularly suitable for real-time early warning and health management under alternating full-load and overload operation scenarios. The main contents of the technical solution of this application are summarized below with examples, specifically including the following:

[0032] 1. "Layered-Integrated" Hybrid Comparison Strategy:

[0033] Layered approach: Independently optimize threshold parameters (IQR coefficient k1) for full load and overload conditions to adapt to the distribution characteristics of K values ​​for each condition;

[0034] Integration phase: cross-condition comparison and optimization of parameters (k1), number of anomalies, exploration of the correlation between load level and mean K value, and multi-level early warning statistics.

[0035] 2. PSO-based k1 adaptive optimization mechanism:

[0036] This solution selects the Particle Swarm Optimization (PSO) algorithm in the hierarchical optimization stage, rather than other optimization algorithms such as Genetic Algorithm (GA), Simulated Annealing (SA), or Gradient Descent (GD). The core reason is that the characteristics of PSO are perfectly matched with the core requirement of this solution, "multi-condition k1 parameter optimization"—it not only meets the requirements of industrial scenarios for convergence speed, robustness, and engineering implementation difficulty, but also accurately adapts to the technical characteristics of "single parameter, narrow range, and independent optimization of multiple conditions".

[0037] The optimization range of k1 is set to [1.0, 3.0] (covering the operating conditions of most power equipment); the fitness function is the proportion of outliers, and the optimal k1 is solved for each operating condition by the particle swarm optimization algorithm (PSO). The parameters are configured as follows: population size 30, maximum iteration 100 times, inertia weight w=0.4-0.9, and learning factor c1=c2=2.0.

[0038] 3. A multi-level early warning mechanism integrating "continuous anomalies + extreme thresholds":

[0039] Minor warning (Level 1): A single anomaly or two consecutive anomalies;

[0040] Moderate alert (Level 2): ​​3-5 consecutive anomalies;

[0041] Severe warning (Level 3): Six or more consecutive anomalies or extreme anomalies exceeding twice the threshold (K_mean±2×ΔK).

[0042] 4. Robust data preprocessing workflow:

[0043] After calculating the K value, inf and NaN values ​​are automatically removed to ensure data validity.

[0044] Based on the rated power P_rated = max (power) × 0.9, an adaptive multi-condition range is defined (full load [0.85P_rated, 1.0P_rated], overload [1.05P_rated, ∞)).

[0045] 5. The repaired multi-dimensional visualization system:

[0046] Displays K-value curves, threshold bands, anomaly point markers, load-K-value relationships, and multi-level early warning statistics for various operating conditions.

[0047] The technical solution of this application has the following advantages:

[0048] Strong adaptability to working conditions: No manual parameter adjustment is required; it can automatically adapt to the switching between full load and overload working conditions, covering the differences in K-value distribution of different power equipment.

[0049] Refined early warning decision-making: The three-level early warning clearly defines the severity of the anomaly, enabling maintenance personnel to take targeted "observation-inspection-shutdown maintenance" measures;

[0050] High robustness in data processing: Invalid data is automatically removed, and the operating condition classification is adaptively adjusted based on the actual operating power of the equipment, avoiding the subjectivity of experience thresholds;

[0051] The optimization process is traceable: During the PSO optimization process, the number of iterations and the current optimal k1 value are output in real time, which makes it easy for technicians to verify the optimization effect.

[0052] The technical solution of this application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0053] Figure 1 This is a flowchart illustrating the adaptive early warning method for multi-condition power coupling coefficient provided in an embodiment of this application. Figure 1 As shown, the multi-condition power coupling coefficient adaptive early warning method may specifically include:

[0054] S101: Obtain torque data, speed data, and power data during the operation of the power equipment, calculate the power coupling coefficient sequence based on the torque data, speed data, and power data, and remove invalid values ​​from the power coupling coefficient sequence to obtain the cleaned power coupling coefficient sequence.

[0055] S102, determine the rated power reference value based on the power data, divide the power data into multiple operating condition intervals according to the rated power reference value, and label the cleaned power coupling coefficient sequence according to the multiple operating condition intervals to obtain the power coupling coefficient subsequence corresponding to each operating condition.

[0056] S103, For the power coupling coefficient subsequence corresponding to each working condition, calculate the mean and interquartile range of each working condition, and use the threshold coefficient as the parameter to be optimized. Within the preset range, the swarm intelligence optimization algorithm is used to independently optimize the threshold coefficient for each working condition to obtain the optimal threshold coefficient corresponding to each working condition.

[0057] S104. Based on the mean, interquartile range and optimal threshold coefficient of each working condition, the threshold band of the working condition is determined, and the power coupling coefficient subsequence corresponding to each working condition is checked for out-of-bounds based on the threshold band to obtain the set of abnormal points corresponding to each working condition.

[0058] S105, perform cross-condition integrated analysis on the optimal threshold coefficient, power coupling coefficient statistical characteristics and outlier set corresponding to each working condition, and generate correlation analysis results between load level and power coupling coefficient characteristics;

[0059] S106: Identify continuous sequences of abnormal points based on the set of abnormal points corresponding to each working condition; generate multi-level early warning information based on the length of the continuous sequence of abnormal points; construct extreme thresholds based on threshold bands to identify extreme abnormal events and classify them into preset early warning levels; output multi-level early warning information and correlation analysis results corresponding to each working condition.

[0060] In some embodiments, a power coupling coefficient sequence is calculated based on torque data, speed data, and power data. Invalid values ​​are removed from the power coupling coefficient sequence to obtain a cleaned power coupling coefficient sequence, including:

[0061] Based on torque data, speed data, and power data, an initial sequence of power coupling coefficients corresponding to each sampling time is generated according to the preset power coupling coefficient calculation relationship.

[0062] Perform a validity check on the initial sequence of power coupling coefficients, and mark infinite values ​​and null values ​​caused by zero, missing or abnormal power data as invalid values;

[0063] Invalid values ​​are removed from the initial power coupling coefficient sequence to obtain the cleaned power coupling coefficient sequence.

[0064] Specifically, the data acquisition module samples key operating parameters of the power equipment at a fixed sampling period, obtaining torque, speed, and power data sequences that correspond one-to-one with the sampling time. To ensure consistency of the three types of data on the time axis, the acquisition module adopts a unified timestamp alignment strategy for the three types of data. That is, at each sampling time, the torque value, speed value, and power value are recorded simultaneously, and the three are combined into the original data set of the same sampling point. In some implementations, if there is a sampling delay between different sensing channels, the data is aligned and compensated before the acquisition module outputs, so that the torque data, speed data, and power data have a corresponding relationship at the same sampling point, avoiding misalignment in the subsequent calculation of the power coupling coefficient.

[0065] After obtaining the raw data set, the calculation module generates an initial sequence of power coupling coefficients corresponding to each sampling time according to a preset power coupling coefficient calculation relationship. Specifically, for each sampling point, the calculation module reads the torque, speed, and power values, and obtains the initial value of the power coupling coefficient for that sampling point based on the preset calculation relationship, thus forming an initial sequence of power coupling coefficients arranged in chronological order. The preset power coupling coefficient calculation relationship at least reflects the product of torque and speed, as well as the ratio to power, so that the power coupling coefficient reflects the coupling state between torque, speed, and power. In some implementations, when the dimensions or units of measurement of torque, speed, and power are inconsistent, the calculation module can introduce a preset conversion factor to match the units of torque, speed, or power before performing the power coupling coefficient calculation, to ensure that data from different acquisition channels can be used for the same calculation relationship.

[0066] Because data collected in industrial settings may be affected by factors such as sensor disconnection, communication jitter, instantaneous sampling distortion, or zero power, the initial sequence of power coupling coefficients may contain infinite or null values. Therefore, this embodiment performs validity checks and invalid value marking after the initial sequence of power coupling coefficients is generated. The triggering conditions for validity checks include at least: power data being zero or close to zero causing an abnormal denominator; missing power data preventing the calculation of a valid ratio for that sampling point; and abnormal sudden changes or out-of-bounds power data leading to unreliable calculation results. During validity checks, the verification module assesses the availability of the power value at each sampling point. If the power value at a sampling point is determined to be zero, missing, or in an abnormal state, the initial value of the power coupling coefficient corresponding to that sampling point is marked as invalid. Similarly, if the initial value of the power coupling coefficient at a sampling point is infinite or unrepresentable, it is also marked as invalid. Here, "null value" can include unassigned values ​​due to missing sensor data, parsing failure, or sampling point removal; "infinite value" can include unusable values ​​due to division by zero or numerical overflow.

[0067] Furthermore, after marking invalid values, the cleaning module removes invalid values ​​from the initial power coupling coefficient sequence, resulting in a cleaned power coupling coefficient sequence. Specifically, based on the invalid value marking results, the cleaning module deletes the sampling points marked as invalid from the initial power coupling coefficient sequence and reconstructs the power coupling coefficient sequence in chronological order for the remaining valid sampling points. In some implementations, to maintain consistency with subsequent operating condition labeling processing, the cleaning module simultaneously maintains a valid sampling point index set while removing invalid values. This allows subsequent steps to establish a consistent mapping relationship between the power data sequence and the power coupling coefficient sequence based on this index set, avoiding alignment errors caused by inconsistencies between the power sequence length and the power coupling coefficient sequence length. Through the above removal process, this embodiment achieves robust cleaning of the power coupling coefficient sequence, ensuring that the input data used for subsequent calculations of the mean, interquartile range, and threshold band determination based on operating conditions are all valid samples.

[0068] The above process will be further illustrated below with an example. In a set of sampled data, the power signal at some sampling times is temporarily missing due to communication jitter, and the power value at a few sampling points is close to zero. After the calculation module generates the initial sequence of power coupling coefficients, the verification module identifies that the initial power coupling coefficient values ​​corresponding to the above sampling points are in an infinite or null state and marks them as invalid values; the cleaning module removes the sampling points corresponding to these invalid values ​​and retains the remaining sampling points to form a cleaned power coupling coefficient sequence. After this processing, the cleaned power coupling coefficient sequence is used for subsequent determination of rated power reference values, division of operating condition intervals, and labeling of operating conditions, so that the subsequent optimization of threshold coefficients for different operating conditions and the multi-level early warning judgment process are not affected by invalid samples.

[0069] Through the power coupling coefficient calculation and invalid value removal process in this embodiment, the validity verification and cleaning of the power coupling coefficient sequence are completed without introducing manual intervention. This avoids infinite values ​​and null values ​​caused by zero, missing or abnormal power data from entering the subsequent statistical calculation and threshold determination process, thereby providing a stable and consistent data input basis for multi-condition adaptive threshold band determination and multi-level early warning output.

[0070] In some embodiments, a rated power reference value is determined based on power data, and the power data is divided into multiple operating condition ranges according to the rated power reference value, including:

[0071] Determine the maximum power value in the power data, and convert the maximum power value based on the preset rated coefficient to obtain the rated power reference value;

[0072] Based on the rated power reference value, multiple operating condition ranges are set, including at least a full-load operating condition range and an overload operating condition range; wherein, the minimum power range of the full-load operating condition range is the product of the first proportional coefficient and the rated power reference value, and the maximum power range is the product of the second proportional coefficient and the rated power reference value; the power range of the overload operating condition range is the power not less than the product of the third proportional coefficient and the rated power reference value.

[0073] Specifically, the rated power reference value is used as a normalized benchmark for operating condition division, and it is derived from the maximum power value in the power data sequence. In some examples, the processing module first traverses the power data within a preset statistical window to determine the maximum power value; this statistical window can be the full sampling interval of a single operation task or a sliding window interval that satisfies data stability. Subsequently, the processing module converts the maximum power value based on a preset rated coefficient to obtain the rated power reference value. The preset rated coefficient is used to suppress the influence of instantaneous spikes in the maximum power value on the rated power reference value; in this embodiment, it can be taken as 0.9, so that the rated power reference value is consistent with the actual sustainable operating power level of the equipment. Thus, the rated power reference value does not depend on the equipment nameplate parameters or manual input, and can adaptively update with changes in equipment, load, and sampling period.

[0074] Furthermore, after determining the rated power reference value, the processing module sets multiple operating condition ranges based on this rated power reference value. This embodiment sets at least a full-load operating condition range and an overload operating condition range. The minimum power range of the full-load operating condition range is determined by the product of a first proportional coefficient and the rated power reference value, and the maximum power range is determined by the product of a second proportional coefficient and the rated power reference value. The power range of the overload operating condition range is a power not less than the product of a third proportional coefficient and the rated power reference value.

[0075] For example, the first proportionality coefficient can be 0.85, the second proportionality coefficient can be 1.0, and the third proportionality coefficient can be 1.05. Furthermore, to avoid frequent switching of operating condition tags due to power fluctuations near the critical point, this embodiment retains an interval between the upper limit of full load and the lower limit of overload. That is, when the power is between 1.0 times the rated power reference value and 1.05 times the rated power reference value, these sampling points are not included in the full load or overload range. Subsequent operating condition tagging steps ignore or process these points separately according to a preset strategy to ensure the stability of the operating condition tag sequence.

[0076] The above process is illustrated below with an example. Assume that the maximum power value of a power device acquired in one sampling period is 120kW. The processing module, based on a preset rated coefficient of 0.9, calculates a rated power reference value of 108kW. Based on this rated power reference value, the power range for the full-load operating condition can be determined as [0.85×108kW, 1.0×108kW], i.e., [91.8kW, 108kW]; the power range for the overload operating condition can be determined as a power not less than 1.05×108kW, i.e., a power not less than 113.4kW. Therefore, when the power of a sampling point falls between 91.8kW and 108kW, it can be determined as a full-load condition; when the power of a sampling point is not less than 113.4kW, it can be determined as an overload condition; when the power is between 108kW and 113.4kW, the sampling point falls into the interval zone and can be excluded from the statistical calculation of full-load and overload according to a preset strategy, thereby avoiding the alternating fluctuations in operating conditions caused by critical fluctuations. This example demonstrates that the rated power reference value and operating condition range are adaptively determined by actual power data, eliminating the need for manual pre-configuration of rated power parameters for different devices or scenarios.

[0077] The rated power reference value determination and multi-condition interval division process in this embodiment can reduce the subjectivity caused by the reliance on manual setting of rated power and condition boundaries, improve the consistency and reproducibility of condition division rules under different equipment and different load levels, reduce the frequent switching of condition labels caused by critical power fluctuations, and thus provide a stable condition input basis for subsequent independent optimization of threshold coefficients and multi-level early warning judgment based on subsequences of sub-condition power coupling coefficients.

[0078] In some embodiments, the cleaned power coupling coefficient sequence is labeled according to multiple operating condition intervals to obtain a power coupling coefficient subsequence corresponding to each operating condition, including:

[0079] A condition determination mask is generated for each of the multiple operating condition intervals. The operating condition determination mask is used to indicate whether the sampling point in the power data falls into the corresponding operating condition interval.

[0080] Based on the operating condition determination mask, each sampling point of the power data is written into an operating condition label, and the sampling point index corresponding to each operating condition is extracted from the cleaned power coupling coefficient sequence based on the operating condition label.

[0081] Based on the sampling point index, construct the power coupling coefficient subsequence corresponding to each operating condition, and perform statistical feature calculation on the power coupling coefficient subsequence corresponding to each operating condition to obtain the statistical features of the power coupling coefficient corresponding to each operating condition.

[0082] Specifically, the operating condition labeling process establishes an operating condition determination mask primarily based on power data. Specifically, the processing module generates a corresponding operating condition determination mask for each type of operating condition interval. This mask indicates whether a sampling point in the power data falls within that operating condition interval. Referring to the aforementioned example, when the full-load operating condition interval is defined by power values ​​between [0.85 × rated power reference value, 1.0 × rated power reference value], and the overload operating condition interval is defined by power values ​​not less than 1.05 × rated power reference value, the processing module performs interval determination on the power value of each sampling point: if the power value falls within the full-load operating condition interval, the sampling point is marked as a hit in the full-load operating condition determination mask; if the power value meets the requirements of the overload operating condition interval, the sampling point is marked as a hit in the overload operating condition determination mask.

[0083] To avoid the same sampling point hitting multiple operating conditions at the same time, this embodiment introduces an interval band in the operating condition interval design stage, so that the upper limit of the full-load operating condition and the lower limit of the overload operating condition do not overlap, thereby ensuring that the mask hit is mutually exclusive; for sampling points whose power values ​​fall into the interval band, the full-load or overload labels can be not written according to the preset strategy, or a preset "unclassified" label can be written for subsequent selective processing.

[0084] Furthermore, after generating the operating condition determination mask, the processing module writes operating condition labels to each sampling point of the power data based on the operating condition determination mask, and extracts the sampling point index corresponding to each operating condition from the cleaned power coupling coefficient sequence based on the operating condition labels. In some examples, the processing module maintains an operating condition label for each sampling point, which is used to characterize the operating condition category to which the sampling point belongs; when a sampling point is matched in the full-load operating condition determination mask, the operating condition label of the sampling point is written as the full-load label; when a sampling point is matched in the overload operating condition determination mask, the operating condition label of the sampling point is written as the overload label.

[0085] Subsequently, the processing module extracts a set of sampling point indices that satisfy the corresponding operating condition label from the time series. This set of sampling point indices is used to establish a mapping relationship of "operating condition → sampling point set → power coupling coefficient subsequence". To ensure the consistency of the index mapping, this embodiment has synchronously maintained a set of valid sampling point indices during the preceding invalid value removal step. Therefore, when generating operating condition labels, this embodiment can filter the power data based on the valid sampling point indices, ensuring that the operating condition labels and the cleaned power coupling coefficient sequence are consistent in length and order, avoiding mislabeling of operating conditions due to "alignment deviation between the power sequence and the power coupling coefficient sequence".

[0086] Furthermore, after obtaining the sampling point indexes corresponding to each operating condition, the processing module constructs a power coupling coefficient subsequence corresponding to each operating condition based on the sampling point indexes. In some examples, the processing module extracts power coupling coefficient samples belonging to the same operating condition from the cleaned power coupling coefficient sequence according to the index, and forms a power coupling coefficient subsequence for that operating condition according to the sampling time order; similarly, corresponding power coupling coefficient subsequences are constructed for different operating conditions.

[0087] Furthermore, to support the subsequent determination of the threshold band and independent optimization of the threshold coefficients for each operating condition, the processing module performs statistical feature calculations on the power coupling coefficient subsequences corresponding to each operating condition, obtaining the statistical features of the power coupling coefficients for each operating condition. The statistical features include at least the mean and interquartile range of the power coupling coefficients, with the interquartile range determined by the difference between the upper and lower quartiles.

[0088] In some implementations, parameters such as sample size and quantile boundaries can also be calculated for subsequent construction of threshold bands and evaluation metrics such as the proportion of outliers. The above statistical feature calculations are consistent with the description of "calling functions to obtain parameters such as the mean of the power coupling coefficients and the interquartile range between the two operating conditions," and their output serves as input data for subsequent hierarchical optimization threshold coefficients and out-of-bounds detection.

[0089] The following example illustrates the specific process of operating condition labeling. Continuing with the previous example, assuming a rated power reference value of 108kW, the full-load operating condition range is [91.8kW, 108kW], and the overload operating condition range is not less than 113.4kW. For a certain segment of continuous sampling data, if the power values ​​at sampling points t1 to t100 are all between 95kW and 105kW, then the full-load operating condition determination mask is hit at all sampling points in that segment. The processing module writes the corresponding sampling points into the full-load label and uses t1 to t100 as the full-load sampling point index set. If the power values ​​at sampling points t150 to t180 are between 115kW and 120kW, then the overload operating condition determination mask is hit at all sampling points in that segment. The processing module writes the corresponding sampling points into the overload label and uses t150 to t180 as the overload sampling point index set.

[0090] Subsequently, the processing module extracts power coupling coefficient samples corresponding to t1 to t100 from the cleaned power coupling coefficient sequence to form a full-load power coupling coefficient subsequence, and extracts power coupling coefficient samples corresponding to t150 to t180 to form an overload power coupling coefficient subsequence. It then calculates the mean and interquartile range of the power coupling coefficients for both subsequences to obtain the statistical feature outputs for the two operating conditions, which can be used for subsequent threshold coefficient optimization and anomaly detection.

[0091] Through the working condition labeling and statistical feature calculation process of this embodiment, the cleaned power coupling coefficient sequence can be divided into sub-sequences based on the working condition judgment results of power data, and sampling point indexes and statistical features corresponding to each working condition can be formed. This allows subsequent independent optimization of threshold coefficients by working condition, threshold band construction and boundary detection to be based on the same working condition division results, reducing statistical errors caused by mislabeling of working conditions and data alignment deviations, thereby improving the stability and consistency of the multi-working condition early warning processing link.

[0092] In some embodiments, using a threshold coefficient as the parameter to be optimized, a swarm intelligence optimization algorithm is used within a preset range to independently optimize the threshold coefficient for each working condition, thereby obtaining the optimal threshold coefficient for each working condition, including:

[0093] For any power coupling coefficient subsequence corresponding to any operating condition, a search range for the threshold coefficient is set, and a population containing multiple candidate threshold coefficients is initialized within the search range;

[0094] A fitness function is constructed based on the statistical characteristics of the power coupling coefficient corresponding to the candidate threshold coefficient and the operating condition. The fitness value is calculated for each candidate threshold coefficient in the population. The fitness function is used to characterize the anomaly detection evaluation quantity corresponding to the candidate threshold coefficient.

[0095] Within a preset number of iterations, the velocity and position of each individual in the population are updated based on the fitness value of each candidate threshold coefficient, and constraints are imposed on the updated velocity and position to keep the candidate threshold coefficient within the search range and ensure that the velocity meets the preset limit condition.

[0096] During the iterative update process, the individual optimal threshold coefficient and the global optimal threshold coefficient are maintained, and when the termination condition is met, the global optimal threshold coefficient is output as the optimal threshold coefficient corresponding to the working condition.

[0097] Specifically, the threshold coefficient is used to adjust the width of the threshold band constructed based on the interquartile range. Its distance between the upper and lower thresholds is determined together with the dispersion of the power coupling coefficient under different operating conditions. To enable the threshold band to adaptively adjust according to differences in operating conditions, this embodiment independently optimizes the threshold coefficient for full-load and overload conditions, rather than optimizing a single threshold coefficient across all samples.

[0098] In some examples, the processing module sets a search range for threshold coefficients for a power coupling coefficient subsequence corresponding to any operating condition, and initializes a group containing multiple candidate threshold coefficients within the search range. For example, the search range for threshold coefficients can be set to 1.0 to 3.0 to cover the threshold bandwidth adjustment requirements under common power equipment operating conditions; the group size can be set to 30, so that the candidate threshold coefficients form a discrete distribution within the search range, thereby avoiding optimization getting stuck in a local region due to single-point search.

[0099] Furthermore, after population initialization, this embodiment constructs a fitness function based on the statistical characteristics of the candidate threshold coefficients and the power coupling coefficients corresponding to the operating conditions, and calculates the fitness value for each candidate threshold coefficient in the population. The fitness function is used to characterize the anomaly detection evaluation quantity corresponding to the candidate threshold coefficient, and its input includes at least the mean power coupling coefficient, the interquartile range, and the candidate threshold coefficient itself for that operating condition.

[0100] In some implementations, the anomaly detection evaluation metric can be expressed in terms of anomaly percentage, early warning accuracy, or consistency with manually labeled anomaly segments. The core of this approach is to map the "threshold band determined by the candidate threshold coefficient" to a "set of anomalies," and then evaluate this set of anomalies, thereby enabling the fitness calculation to reflect the impact of the candidate threshold coefficient on the anomaly detection results. To ensure that the fitness function can be consistently invoked under different operating conditions, this embodiment uses the same fitness calculation logic for both full-load and overload conditions, but the input power coupling coefficient statistical characteristics and sample sequences are derived from the power coupling coefficient subsequences of the corresponding operating conditions.

[0101] Within a preset number of iterations, the processing module updates the velocity and position of each individual in the swarm based on the fitness value of each candidate threshold coefficient, and imposes constraints on the updated velocity and position to keep the candidate threshold coefficients within the search range and ensure that the velocity meets the preset limit condition. This embodiment employs particle swarm optimization as an implementation of a swarm intelligence optimization algorithm: each individual carries current position and velocity parameters. The current position parameter represents the current candidate threshold coefficient, and the velocity parameter represents the search step size for the next round. In each iteration, the individual updates its velocity based on a combination of inertia, individual optimal guidance, and global optimal guidance, and updates its position based on the updated velocity, allowing the swarm to gradually converge towards regions with better fitness within the search space.

[0102] For example, the preset number of iterations can be set to 100, and the inertia weight can be decayed between 0.4 and 0.9 according to a preset rule, making the early search more global and the later search more convergent; the learning factor can adopt the same configuration to balance individual experience and group experience; the speed limiting condition can adopt a limiting strategy with an absolute value not greater than 0.4 to suppress search oscillations caused by excessive step size, and after the position is updated, the candidate threshold coefficients that exceed the search range are boundary-reset or truncated so that the candidate threshold coefficients are always within the search range of 1.0 to 3.0.

[0103] During the iterative update process, the processing module maintains the individual optimal threshold coefficient and the global optimal threshold coefficient, and outputs the global optimal threshold coefficient as the optimal threshold coefficient corresponding to the working condition when the termination condition is met. Specifically, the processing module records the candidate threshold coefficient corresponding to the optimal fitness in its historical iterations for each individual as the individual optimal threshold coefficient, and selects the individual with the best fitness from the individual optimal values ​​of all individuals after each iteration as the global optimal threshold coefficient; the termination condition may include reaching a preset number of iterations, or the global optimal fitness no longer significantly improving in consecutive iterations. In some implementations, the processing module may also record the iteration number and the current global optimal threshold coefficient during the iteration process to form a traceable optimization process record, which is convenient for technicians to verify whether the optimization has converged and reproduce the threshold coefficient output.

[0104] The following example illustrates the independent optimization under different operating conditions. In a set of full-load operating condition data, the processing module uses the power coupling coefficient subsequence and its mean and interquartile range as input, optimizing within the range of 1.0 to 3.0 using a population size of 30 and 100 iterations. The final output of the optimal threshold coefficient for the full-load condition is 1.006. In a set of overload operating condition data, the processing module uses the power coupling coefficient subsequence and its statistical characteristics for the overload condition as input, independently optimizing using the same population size and iteration count. The final output of the optimal threshold coefficient for the overload condition is 1.000. This demonstrates that different optimal threshold coefficients can be obtained for full-load and overload conditions, and this output is not dependent on manual experience but is driven by the power coupling coefficient distribution under the operating condition and the anomaly detection evaluation metrics.

[0105] Through the threshold coefficient independent optimization process for each working condition in this embodiment, a swarm intelligence optimization algorithm can be used within a preset search range to determine the threshold coefficient that matches the power coupling coefficient distribution for different working conditions. The convergence stability of the optimization is improved by boundary constraints and speed limits in the iterative process, so that the subsequent threshold band construction and out-of-bounds detection can be performed based on the optimal threshold coefficient corresponding to each working condition. This reduces the risk of misjudgment and missed judgment introduced by fixed threshold coefficients in multi-working-condition switching scenarios, and improves the adaptive capability of the warning parameters under different working conditions.

[0106] In some embodiments, a threshold band for each operating condition is determined based on the mean, interquartile range, and optimal threshold coefficient. Out-of-bounds detection is then performed on the power coupling coefficient subsequence corresponding to each operating condition based on the threshold band to obtain an outlier set for each operating condition, including:

[0107] For any power coupling coefficient subsequence corresponding to any operating condition, the threshold increment is calculated based on the interquartile range and the optimal threshold coefficient corresponding to the operating condition.

[0108] The upper and lower thresholds are determined based on the mean and threshold increment corresponding to the working conditions, and the threshold band is limited by the lower and upper thresholds.

[0109] The power coupling coefficient of each sampling point in the power coupling coefficient subsequence corresponding to the operating condition is compared with the threshold band. When the power coupling coefficient is less than the lower threshold or greater than the upper threshold, the corresponding sampling point is identified as an anomaly.

[0110] Record the sampling point index of the anomaly points to form the anomaly point set corresponding to each working condition.

[0111] Specifically, the threshold band is defined by both an upper threshold and a lower threshold, and the distance between the upper and lower thresholds is determined by the threshold increment. In some examples, for any power coupling coefficient subsequence corresponding to any operating condition, the processing module calculates the threshold increment based on the interquartile range and the optimal threshold coefficient corresponding to that operating condition. The interquartile range is used to characterize the dispersion of the power coupling coefficient distribution under that operating condition, and the optimal threshold coefficient is used to scale the dispersion. Therefore, the threshold increment can be obtained by combining the optimal threshold coefficient and the interquartile range. In the embodiments of this application, the interquartile range is determined by the difference between the upper and lower quartiles of the power coupling coefficient sample under that operating condition, and the threshold increment is obtained by multiplying the optimal threshold coefficient and the interquartile range, so that the threshold increment changes in the same direction as the operating condition dispersion.

[0112] Furthermore, after obtaining the threshold increment, the processing module determines the upper and lower thresholds based on the mean and threshold increment corresponding to the operating condition, and then defines the threshold band using the lower and upper thresholds. For example, the upper threshold can be obtained by adding the mean and the threshold increment, and the lower threshold can be obtained by subtracting the mean and the threshold increment, thus forming a threshold band centered on the mean and with the threshold increment as half its width. Since the mean and interquartile range are both obtained statistically from the power coupling coefficient subsequence corresponding to the operating condition, and the threshold coefficient is the optimal threshold coefficient obtained independently for the operating condition, both the upper and lower thresholds match the power coupling coefficient distribution corresponding to the operating condition, avoiding the situation where the threshold band is too wide or too narrow due to using the same fixed coefficient for different operating conditions.

[0113] Furthermore, after the threshold band is determined, the processing module compares the power coupling coefficient of each sampling point in the power coupling coefficient subsequence corresponding to the operating condition with the threshold band: when the power coupling coefficient is less than the lower threshold or greater than the upper threshold, the corresponding sampling point is identified as an anomaly; when the power coupling coefficient is between the lower and upper thresholds, the corresponding sampling point is identified as a normal point. To ensure that the anomaly points can be used for subsequent continuous anomaly sequence splitting, this embodiment records the sampling point index of the anomaly points to form an anomaly point set corresponding to the operating condition. The sampling point index can be a position index in the time series or an index identifier bound to a timestamp. The processing module organizes the anomaly point indexes into an index list or index set in chronological order and stores them in association with the corresponding operating condition label, thereby achieving a consistent output of "operating condition - threshold band - anomaly point set".

[0114] The following examples illustrate threshold band construction and boundary detection. For the full-load condition, the processing module obtains an optimal threshold coefficient of 1.006 during the hierarchical optimization phase and statistically calculates the mean and interquartile range on the full-load power coupling coefficient subsequence. Based on this, the processing module calculates the full-load threshold increment, generates the upper and lower full-load thresholds, and compares each point on the full-load power coupling coefficient subsequence to obtain a set of full-load anomalies. For the overload condition, the processing module obtains an optimal threshold coefficient of 1.000 during the hierarchical optimization phase and statistically calculates the mean and interquartile range on the overload power coupling coefficient subsequence. Based on this, the processing module generates an overload threshold band and performs boundary detection to obtain a set of overload anomalies. Since the mean, interquartile range, and optimal threshold coefficient for full-load and overload conditions are generated by the corresponding operating condition data, the threshold band parameters and anomaly sets for the two conditions are independent and comparable, providing data support for subsequent cross-condition integrated analysis and multi-level early warning statistics.

[0115] Through the threshold band determination and out-of-bounds detection process in this embodiment, the optimal threshold coefficient obtained by optimizing under different working conditions can be coupled with the statistical characteristics of the power coupling coefficient of that working condition to generate an adaptive threshold band for the working condition. The set of abnormal points is output in the form of sampling point index, so that the subsequent continuous abnormal sequence identification and early warning level classification have a clear input basis. This improves the stability and consistency of the multi-working-condition abnormal detection process under different discrete working conditions, and reduces the misjudgment and missed judgment introduced by the fixed threshold bandwidth in the working condition switching scenario.

[0116] In some embodiments, cross-condition integrated analysis is performed on the optimal threshold coefficient, power coupling coefficient statistical characteristics, and outlier set corresponding to each operating condition to generate correlation analysis results between load level and power coupling coefficient characteristics, including:

[0117] Based on the statistical characteristics of the optimal threshold coefficient and power coupling coefficient corresponding to each working condition, and the number of outliers extracted from the outlier set, a multi-dimensional statistical data structure for cross-working condition comparison is constructed.

[0118] Based on multiple operating condition ranges, load level indicators are determined for each operating condition, and the load level indicators are correlated with the average power coupling coefficient corresponding to each operating condition.

[0119] The correlation coefficient is calculated based on the average load level index and power coupling coefficient corresponding to each operating condition, and is used as part of the correlation analysis results.

[0120] The load change trend is determined based on the magnitude of the average power coupling coefficient under different operating conditions, and the trend determination result and the correlation coefficient are output as the correlation analysis result.

[0121] Specifically, to facilitate cross-operating condition comparisons, this embodiment first constructs a multi-dimensional statistical data structure for cross-operating condition comparisons. Specifically, the processing module extracts the number of outliers based on the statistical characteristics of the optimal threshold coefficient and power coupling coefficient corresponding to each operating condition, as well as the outlier set. The number of outliers is then combined with statistical characteristics such as the optimal threshold coefficient, the mean of the power coupling coefficient, and the interquartile range for the corresponding operating condition to form statistical entries organized by operating condition. Each statistical entry includes at least an operating condition label, the optimal threshold coefficient, the mean of the power coupling coefficient, the interquartile range, and the number of outliers, thereby achieving a data structure output that allows direct comparison of the same field under different operating conditions. Furthermore, this embodiment can also record evaluation quantities such as threshold parameters or warning accuracy in the statistical entries for subsequent comparison of the sources of differences between threshold parameters and outlier distributions under different operating conditions, but this part is not a mandatory limitation.

[0122] Furthermore, after constructing the multidimensional statistical data structure, this embodiment determines load level indicators for each operating condition based on multiple operating condition intervals, and correlates the load level indicators with the average power coupling coefficient corresponding to each operating condition. The load level indicators are used to normalize the power levels of different operating conditions; they originate from the operating condition intervals themselves, avoiding the introduction of external parameters unrelated to the operating conditions.

[0123] For example, in a specific case, the load level index for full-load conditions can be taken as 0.925 times the per-unit value of the representative value of the full-load range, and the load level index for overload conditions can be taken as 1.05 times the per-unit value of the initial representative value of the overload range. Here, the per-unit value is normalized based on the rated power reference value. The processing module associates 0.925 and 1.05 with full-load and overload conditions respectively, and establishes a mapping relationship between them and the average power coupling coefficient of the corresponding conditions, forming paired data for correlation calculation.

[0124] Furthermore, after obtaining paired data of load level indicators and the mean power coupling coefficient, this embodiment calculates a correlation coefficient based on the load level indicators and the mean power coupling coefficient corresponding to each operating condition, as part of the correlation analysis results. For example, the processing module uses the load level indicators of each operating condition as the independent variable sequence and the mean power coupling coefficient of each operating condition as the dependent variable sequence, and outputs the correlation coefficient according to a preset correlation measurement rule to characterize the linear consistency between changes in load level and changes in the mean power coupling coefficient.

[0125] In the specific implementation, since this embodiment only includes two types of working conditions: full load and overload, the processing module can directly calculate the linear correlation metric and output the correlation coefficient based on the two sets of data points. When the number of working conditions is expanded, the processing module can perform correlation calculation on data points of multiple working conditions according to the same rule, thereby maintaining the scalability of the correlation analysis.

[0126] In addition to the correlation coefficient, this embodiment also generates a load change trend determination result based on the magnitude relationship of the average power coupling coefficients corresponding to different operating conditions, and outputs the trend determination result and the correlation coefficient together as the correlation analysis result. For example, the processing module compares the average power coupling coefficient under full load condition with the average power coupling coefficient under overload condition: when the average power coupling coefficient under overload condition is greater than that under full load condition, it outputs a trend determination that the load increases and the average power coupling coefficient increases; when the average power coupling coefficient under overload condition is less than that under full load condition, it outputs a trend determination that the load increases and the average power coupling coefficient decreases. This trend determination and the correlation coefficient together constitute the correlation analysis result, so that the output includes both calculable correlation metrics and directly interpretable trend descriptions, and both can be linked with fields such as the number of outliers and the optimal threshold coefficient in the multidimensional statistical data structure for analysis, to locate the correspondence between changes in abnormal risk and changes in threshold parameters under different operating conditions.

[0127] The above integrated analysis process is illustrated below with specific examples. In this embodiment, the optimal threshold coefficient for the full-load condition is 1.006 and the optimal threshold coefficient for the overload condition is 1.000 during the hierarchical optimization stage. After threshold band out-of-bounds detection, the abnormal point sets for the two conditions are obtained respectively. The processing module first extracts the number of abnormal points under full load and overload conditions, and constructs cross-condition statistical entries together with statistical features such as the mean and interquartile range of the power coupling coefficients for the two conditions.

[0128] Subsequently, the processing module determines the load level indicators for the two operating conditions based on 0.925 times the per-unit value for full load and 1.05 times the per-unit value for overload, and forms paired data with the corresponding average power coupling coefficient. Based on this paired data, a correlation coefficient is output, and the average power coupling coefficients of the two operating conditions are further compared to generate a load change trend judgment result. Finally, the correlation coefficient and trend judgment result are output together as the correlation analysis result. This example shows that the input of the correlation analysis comes entirely from the previous sub-operating condition statistics and detection results, and the load level indicators are directly derived from the operating condition intervals, making it easy to reuse in different equipment and different operating tasks.

[0129] Through the cross-condition integrated analysis process in this embodiment, the optimal threshold coefficient, power coupling coefficient statistical characteristics, and number of outliers output by the hierarchical links can be uniformly organized and compared on the condition dimension. Based on the load level index derived from the condition interval, the correlation measurement and trend judgment results between the load level and the power coupling coefficient characteristics are output, thereby providing maintenance personnel with a structured analysis basis across conditions and improving the interpretability and comparability of the power coupling coefficient change law and anomaly distribution differences under different load levels.

[0130] In some embodiments, a continuous sequence of abnormal points is identified based on the set of abnormal points corresponding to each working condition; multi-level early warning information is generated according to the length of the continuous sequence of abnormal points; and extreme thresholds are constructed based on threshold bands to identify extreme abnormal events and classify them into preset early warning levels, including:

[0131] For any set of abnormal points corresponding to any working condition, the sampling point indices of the abnormal points are sorted, and the set of abnormal points is split into one or more consecutive sequences of abnormal points based on whether adjacent sampling point indices are consecutive.

[0132] The sequence length of each continuous sequence of anomalies is obtained, and the warning level corresponding to the continuous sequence of anomalies is determined according to the correspondence between the sequence length and the preset length interval.

[0133] Extreme thresholds are constructed based on the upper and lower thresholds corresponding to the threshold band. The extreme thresholds are obtained by amplifying the threshold increments corresponding to the threshold band by a preset multiple.

[0134] The power coupling coefficient of each sampling point in the power coupling coefficient subsequence corresponding to the operating condition is compared with the extreme threshold. When the power coupling coefficient is greater than the extreme upper threshold or less than the extreme lower threshold, the corresponding sampling point is identified as an extreme abnormal event, and the extreme abnormal event is classified into the warning level corresponding to the severe warning.

[0135] The number of consecutive abnormal point sequences and the number of extreme abnormal events corresponding to each warning level are summarized to generate multi-level warning information for each operating condition.

[0136] Specifically, the continuous sequence of outliers is used to characterize the continuous clustering characteristics of outliers in a time series. In this embodiment, for any set of outliers corresponding to any working condition, the sampling point indices of the outliers are sorted, and the set of outliers is split into one or more continuous sequences of outliers based on whether adjacent sampling point indices are consecutive.

[0137] In some examples, the processing module sorts the sampling point indices in the anomaly set in ascending order and iterates through adjacent indices in turn: when the difference between two adjacent anomaly indices is 1, they are determined to be consecutive anomalies and included in the same consecutive anomaly sequence; when the difference between adjacent indices is greater than 1, the anomaly sequence is determined to be broken, the current consecutive anomaly sequence is ended and a new consecutive anomaly sequence is started.

[0138] To adapt to scenarios where the actual sampling interval is fixed, this embodiment uses "whether the index is continuous" as the continuity criterion. In some implementations, the correspondence between the timestamp difference and the sampling period can also be used as the continuity criterion to accommodate the situation where there are missing sampling points, but this does not change the core idea of ​​"splitting the sequence according to the continuity of adjacent points" in this embodiment.

[0139] Furthermore, after obtaining one or more consecutive sequences of anomalies, this embodiment acquires the sequence length of each consecutive sequence and determines the warning level corresponding to the consecutive sequence of anomalies based on the correspondence between the sequence length and a preset length range. For example, based on specific level definitions, consecutive sequences of anomalies with a sequence length of 1 or 2 correspond to a minor warning, consecutive sequences of anomalies with a sequence length of 3 to 5 correspond to a moderate warning, and consecutive sequences of anomalies with a sequence length of not less than 6 correspond to a severe warning. After determining the length of each consecutive sequence of anomalies, the processing module can write a corresponding warning level identifier for the sequence and include the start index, end index, and sequence length of the sequence as part of the sequence-level warning record for subsequent statistics on the number of warnings at each level and for marking the range of anomalies in the visualization output.

[0140] In addition to continuous abnormal sequences, this embodiment further constructs extreme thresholds based on threshold bands to identify extreme abnormal events. For example, the processing module constructs extreme thresholds based on the upper and lower thresholds corresponding to the threshold bands. The extreme thresholds are obtained by amplifying the threshold increments corresponding to the threshold bands by a preset factor.

[0141] For example, the preset multiplier can be 2, thus constructing the extreme upper threshold as the mean plus twice the threshold increment, and the extreme lower threshold as the mean minus twice the threshold increment. This makes the extreme thresholds have a wider out-of-bounds distance requirement compared to the threshold band, used to filter out outliers with larger amplitudes. Since the threshold increment is determined by the optimal threshold coefficient and the interquartile range, the extreme thresholds still maintain an adaptive property consistent with the operating condition distribution characteristics. That is, full load and overload are respectively constructed based on their respective mean, interquartile range, and optimal threshold coefficient to construct the corresponding extreme thresholds.

[0142] Furthermore, after the extreme thresholds are constructed, this embodiment compares the power coupling coefficient of each sampling point in the power coupling coefficient subsequence corresponding to the operating condition with the extreme thresholds. When the power coupling coefficient is greater than the upper extreme threshold or less than the lower extreme threshold, the corresponding sampling point is identified as an extreme abnormal event, and the extreme abnormal event is classified into the warning level corresponding to the severe warning. In other words, extreme abnormal events do not need to meet the continuous length condition to be directly classified into the severe warning, so that the warning system covers both "persistent risks" and "sudden high-amplitude risks".

[0143] In some implementations, the processing module can record the sampling point index of extreme abnormal events separately as an extreme abnormal index set, and merge it with the continuous sequence judgment result during subsequent statistics to avoid the same sampling point being counted in both ordinary abnormal sequences and extreme abnormal events, thus avoiding duplicate statistics. For example, when a sampling point belongs to both the continuous sequence of abnormal points and meets the extreme threshold condition, it can be classified as an extreme abnormal event first, and the sampling point can be removed from the sequence or used only for labeling without counting in the statistics of ordinary abnormal sequences, so as to maintain the uniqueness of the warning count.

[0144] The following example illustrates the multi-level warning and extreme anomaly determination. For full-load conditions, the processing module obtains a set of full-load anomaly points after boundary detection. These anomaly points are then indexed and split into multiple consecutive sequences. If a consecutive sequence of length 2 exists, it is classified as a minor warning; if a consecutive sequence of length 4 exists, it is classified as a moderate warning; and if a consecutive sequence of length 6 exists, it is classified as a severe warning.

[0145] Meanwhile, the processing module constructs an extreme threshold with a 2x threshold increment based on the threshold band for full-load conditions, and identifies sampling points exceeding the extreme threshold in the full-load power coupling coefficient subsequence, directly including them as extreme abnormal events in the severe warning. For overload conditions, the processing module also performs continuous sequence splitting and grading based on the overload abnormal point set, and constructs extreme thresholds based on the overload threshold band to identify extreme abnormal events.

[0146] This example shows that both full load and overload follow the same multi-level warning rules, but their threshold ranges and extreme thresholds are generated by the optimal threshold coefficients and statistical characteristics of their respective working conditions. Therefore, the classification judgment maintains a consistent framework under different working conditions and has adaptive differences.

[0147] Furthermore, after completing the continuous sequence classification and extreme anomaly identification, this embodiment summarizes the number of continuous anomaly sequences and the number of extreme anomaly events corresponding to each warning level, generating multi-level warning information for each operating condition. Specifically, the processing module counts the number of continuous sequences corresponding to minor warnings, moderate warnings, and severe warnings, and adds the number of extreme anomaly events to form a supplementary count for severe warnings. At the same time, the sequence-level warning records and extreme anomaly index sets of each level can be encapsulated together into an operating condition warning result object for subsequent visualization modules to perform level labeling and statistical display.

[0148] Through the continuous sequence identification of anomalies, multi-level early warning generation, and extreme threshold determination process in this embodiment, it is possible to further output hierarchical early warning information for operation and maintenance based on the out-of-bounds detection results of various operating conditions. This enables the early warning output to distinguish between short-term sporadic anomalies and persistent anomalies, and supplements the identification channel for high-amplitude sudden anomalies, thereby improving the interpretability and handling direction of the early warning information and reducing the risk of over-response or delayed response caused by relying solely on binary anomaly determination.

[0149] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0150] Figure 2 This is a schematic diagram of the structure of the multi-condition power coupling coefficient adaptive early warning device provided in the embodiments of this application. Figure 2 As shown, the multi-condition power coupling coefficient adaptive early warning device includes:

[0151] The acquisition module 201 is configured to acquire torque data, speed data and power data during the operation of the power equipment, calculate the power coupling coefficient sequence based on the torque data, speed data and power data, and remove invalid values ​​from the power coupling coefficient sequence to obtain a cleaned power coupling coefficient sequence.

[0152] The determination module 202 is configured to determine the rated power reference value based on the power data, divide the power data into multiple operating condition intervals according to the rated power reference value, and label the cleaned power coupling coefficient sequence according to the multiple operating condition intervals to obtain the power coupling coefficient subsequence corresponding to each operating condition.

[0153] The optimization module 203 is configured to calculate the mean and interquartile range of each operating condition for the power coupling coefficient subsequence corresponding to each operating condition, and use the threshold coefficient as the parameter to be optimized. Within a preset range, the swarm intelligence optimization algorithm is used to independently optimize the threshold coefficient for each operating condition to obtain the optimal threshold coefficient corresponding to each operating condition.

[0154] The detection module 204 is configured to determine the threshold band of each operating condition based on the mean, interquartile range and optimal threshold coefficient of each operating condition, and to perform out-of-bounds detection on the power coupling coefficient subsequence corresponding to each operating condition according to the threshold band, so as to obtain the set of abnormal points corresponding to each operating condition.

[0155] Analysis module 205 is configured to perform cross-condition integrated analysis on the optimal threshold coefficient, power coupling coefficient statistical characteristics and outlier set corresponding to each operating condition, and generate correlation analysis results between load level and power coupling coefficient characteristics;

[0156] The output module 206 is configured to identify a continuous sequence of abnormal points based on the set of abnormal points corresponding to each working condition, generate multi-level early warning information according to the length of the continuous sequence of abnormal points, construct extreme thresholds based on threshold bands to identify extreme abnormal events and classify them into preset early warning levels; and output multi-level early warning information and correlation analysis results corresponding to each working condition.

[0157] It should be understood that the sequence number of each step in the above embodiments 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 this application.

[0158] Figure 3 This is a schematic diagram of the electronic device 3 provided in an embodiment of this application. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various device embodiments described above.

[0159] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or different components.

[0160] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0161] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 3. The memory 302 can also include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.

[0162] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0163] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which may be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0164] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A multi-condition power coupling coefficient adaptive early warning method, characterized in that, include: The torque, speed and power data of the power equipment during operation are obtained, and the power coupling coefficient sequence is calculated based on the torque data, speed data and power data. Invalid values ​​are removed from the power coupling coefficient sequence to obtain a cleaned power coupling coefficient sequence. Based on the power data, a rated power reference value is determined, and the power data is divided into multiple operating condition intervals according to the rated power reference value. The cleaned power coupling coefficient sequence is labeled according to the multiple operating condition intervals to obtain the power coupling coefficient subsequence corresponding to each operating condition. Statistical feature calculation is performed on the power coupling coefficient subsequence to obtain the statistical features of the power coupling coefficient. For each operating condition, the mean and interquartile range of the power coupling coefficient subsequence are calculated. The threshold coefficient is used as the parameter to be optimized. Within a preset range, the swarm intelligence optimization algorithm is used to independently optimize the threshold coefficient for each operating condition to obtain the optimal threshold coefficient for each operating condition. The threshold band of each working condition is determined based on the mean, interquartile range and the optimal threshold coefficient, and the power coupling coefficient subsequence corresponding to each working condition is checked for out-of-bounds based on the threshold band to obtain the set of outliers corresponding to each working condition. A cross-operating-condition integrated analysis is performed on the optimal threshold coefficient, power coupling coefficient statistical characteristics, and outlier set corresponding to each operating condition to generate a correlation analysis result between load level and power coupling coefficient characteristics. Specifically, based on the optimal threshold coefficient, power coupling coefficient statistical characteristics, and outlier set corresponding to each operating condition, the number of outliers is extracted to construct a multi-dimensional statistical data structure for cross-operating-condition comparison. A load level index is determined for each operating condition based on the multiple operating condition intervals, and the load level index is correlated with the average power coupling coefficient corresponding to each operating condition. A correlation coefficient is calculated based on the load level index and the average power coupling coefficient corresponding to each operating condition, and is used as part of the correlation analysis result. A load change trend determination result is generated based on the magnitude relationship of the average power coupling coefficient corresponding to different operating conditions, and the trend determination result and the correlation coefficient are output together as the correlation analysis result. Based on the set of abnormal points corresponding to each working condition, a continuous sequence of abnormal points is identified. Multi-level early warning information is generated according to the length of the continuous sequence of abnormal points. An extreme threshold is constructed based on the threshold band to identify extreme abnormal events and classify them into a preset early warning level. The multi-level early warning information corresponding to each working condition and the correlation analysis results are output.

2. The method according to claim 1, characterized in that, The process of calculating a power coupling coefficient sequence based on the torque data, the speed data, and the power data, and then removing invalid values ​​from the power coupling coefficient sequence to obtain a cleaned power coupling coefficient sequence includes: Based on the torque data, the speed data, and the power data, an initial sequence of power coupling coefficients corresponding to each sampling time is generated according to a preset power coupling coefficient calculation relationship. The initial sequence of power coupling coefficients is validated, and infinite values ​​and null values ​​caused by zero, missing or abnormal power data are marked as invalid values. The invalid values ​​are removed from the initial sequence of power coupling coefficients to obtain the cleaned power coupling coefficient sequence.

3. The method according to claim 1, characterized in that, The process of determining a rated power reference value based on the power data and dividing the power data into multiple operating condition ranges according to the rated power reference value includes: The maximum power value in the power data is determined, and the maximum power value is converted based on a preset rated coefficient to obtain the rated power reference value; Based on the rated power reference value, multiple operating condition ranges are set, including at least a full-load operating condition range and an overload operating condition range; wherein, the minimum value of the power range of the full-load operating condition range is the product of a first proportional coefficient and the rated power reference value, and the maximum value is the product of a second proportional coefficient and the rated power reference value; the power range of the overload operating condition range is a power not less than the product of a third proportional coefficient and the rated power reference value.

4. The method according to claim 1, characterized in that, The step of labeling the cleaned power coupling coefficient sequence according to the multiple operating condition intervals to obtain a power coupling coefficient subsequence corresponding to each operating condition includes: A condition determination mask is generated for each of the multiple operating condition intervals. The operating condition determination mask is used to indicate whether the sampling point in the power data falls into the corresponding operating condition interval. Based on the operating condition determination mask, each sampling point of the power data is written with an operating condition label, and the sampling point index corresponding to each operating condition is extracted from the cleaned power coupling coefficient sequence based on the operating condition label. Based on the sampling point index, construct the power coupling coefficient subsequence corresponding to each operating condition, and perform statistical feature calculation on the power coupling coefficient subsequence corresponding to each operating condition to obtain the statistical features of the power coupling coefficient corresponding to each operating condition.

5. The method according to claim 1, characterized in that, The step of using a threshold coefficient as the parameter to be optimized, and employing a swarm intelligence optimization algorithm within a preset range to independently optimize the threshold coefficient for each working condition to obtain the optimal threshold coefficient for each working condition includes: For any power coupling coefficient subsequence corresponding to any operating condition, a search range for the threshold coefficient is set, and a group containing multiple candidate threshold coefficients is initialized within the search range; A fitness function is constructed based on the statistical characteristics of the candidate threshold coefficients and the power coupling coefficients corresponding to the operating conditions, and a fitness value is calculated for each candidate threshold coefficient in the population. The fitness function is used to characterize the anomaly detection evaluation quantity corresponding to the candidate threshold coefficients. Within a preset number of iterations, the velocity and position of each individual in the population are updated according to the fitness value of each candidate threshold coefficient, and constraints are applied to the updated velocity and position to keep the candidate threshold coefficient within the search range, and the velocity satisfies a preset limiting condition. During the iterative update process, the individual optimal threshold coefficient and the global optimal threshold coefficient are maintained, and when the termination condition is met, the global optimal threshold coefficient is output as the optimal threshold coefficient corresponding to the working condition.

6. The method according to claim 1, characterized in that, The threshold band for each operating condition is determined based on the mean, interquartile range, and optimal threshold coefficient. Out-of-bounds detection is then performed on the power coupling coefficient subsequence corresponding to each operating condition based on the threshold band to obtain the set of outliers for each operating condition, including: For any power coupling coefficient subsequence corresponding to any operating condition, the threshold increment is calculated based on the interquartile range and the optimal threshold coefficient corresponding to the operating condition. The upper threshold and the lower threshold are determined based on the mean value and the threshold increment corresponding to the working condition, and the threshold band is defined by the lower threshold and the upper threshold. The power coupling coefficient of each sampling point in the power coupling coefficient subsequence corresponding to the operating condition is compared with the threshold band. When the power coupling coefficient is less than the lower threshold or greater than the upper threshold, the corresponding sampling point is determined as an anomaly. Record the sampling point index of the anomaly points to form an anomaly point set corresponding to each working condition.

7. The method according to claim 1, characterized in that, The process involves identifying a continuous sequence of abnormal points based on the set of abnormal points corresponding to each operating condition, generating multi-level early warning information based on the length of the continuous sequence of abnormal points, and constructing extreme thresholds based on the threshold band to identify extreme abnormal events and classify them into preset early warning levels, including: For any set of abnormal points corresponding to any working condition, the sampling point indices of the abnormal points are sorted, and the set of abnormal points is split into one or more consecutive sequences of abnormal points according to whether adjacent sampling point indices are consecutive. The sequence length of each continuous sequence of anomalies is obtained, and the warning level corresponding to the continuous sequence of anomalies is determined according to the correspondence between the sequence length and the preset length interval. An extreme threshold is constructed based on the upper and lower thresholds corresponding to the threshold band. The extreme thresholds are the extreme upper and lower thresholds obtained by amplifying the threshold increments corresponding to the threshold band by a preset multiple. The power coupling coefficient of each sampling point in the power coupling coefficient subsequence corresponding to the operating condition is compared with the extreme threshold. When the power coupling coefficient is greater than the extreme upper threshold or less than the extreme lower threshold, the corresponding sampling point is determined as an extreme abnormal event, and the extreme abnormal event is classified into the warning level corresponding to the severe warning. The number of consecutive abnormal point sequences and the number of extreme abnormal events corresponding to each warning level are summarized to generate multi-level warning information for each operating condition.

8. A multi-condition power coupling coefficient adaptive early warning device, characterized in that, include: The acquisition module is configured to acquire torque data, speed data, and power data during the operation of the power equipment, calculate a power coupling coefficient sequence based on the torque data, speed data, and power data, and remove invalid values ​​from the power coupling coefficient sequence to obtain a cleaned power coupling coefficient sequence. The determination module is configured to determine a rated power reference value based on the power data, divide the power data into multiple operating condition intervals according to the rated power reference value, label the cleaned power coupling coefficient sequence according to the multiple operating condition intervals to obtain a power coupling coefficient subsequence corresponding to each operating condition, and perform statistical feature calculation on the power coupling coefficient subsequence to obtain statistical features of the power coupling coefficient. The optimization module is configured to calculate the mean and interquartile range of the power coupling coefficient subsequence corresponding to each operating condition, and use the threshold coefficient as the parameter to be optimized. Within a preset range, the swarm intelligence optimization algorithm is used to independently optimize the threshold coefficient for each operating condition to obtain the optimal threshold coefficient corresponding to each operating condition. The detection module is configured to determine the threshold band of each operating condition based on the mean, interquartile range and the optimal threshold coefficient, and to perform out-of-bounds detection on the power coupling coefficient subsequence corresponding to each operating condition according to the threshold band, so as to obtain the set of abnormal points corresponding to each operating condition. The analysis module is configured to perform cross-operating-condition integrated analysis on the optimal threshold coefficient, power coupling coefficient statistical characteristics, and outlier set corresponding to each operating condition, generating correlation analysis results between load level and power coupling coefficient characteristics. Specifically, the analysis module extracts the number of outliers based on the optimal threshold coefficient, power coupling coefficient statistical characteristics, and outlier set corresponding to each operating condition, constructing a multi-dimensional statistical data structure for cross-operating-condition comparison; determines load level indicators for each operating condition based on the multiple operating condition intervals, and correlates the load level indicators with the average power coupling coefficient corresponding to each operating condition; calculates a correlation coefficient based on the load level indicators and the average power coupling coefficient corresponding to each operating condition, as part of the correlation analysis results; generates a load change trend determination result based on the magnitude relationship of the average power coupling coefficient corresponding to different operating conditions, and outputs the trend determination result and the correlation coefficient together as the correlation analysis results. The output module is configured to identify a continuous sequence of abnormal points based on the set of abnormal points corresponding to each working condition, generate multi-level early warning information according to the length of the continuous sequence of abnormal points, construct extreme thresholds based on the threshold band to identify extreme abnormal events and classify them into preset early warning levels; and output the multi-level early warning information corresponding to each working condition and the correlation analysis results.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images