Industrial chain node anomaly detection method and system based on power time sequence characteristics

Abstract

The invention discloses an industrial chain node anomaly detection method and system based on electric power time sequence characteristics, and belongs to the technical field of industrial chain data processing, and the method comprises the steps: obtaining electric power data, social and economic data of different industries, and upstream and downstream associated data in the industries in real time; performing season adjustment on the index time sequence in the primary selection index pool by using a time sequence model; performing association analysis to obtain an industrial node association rule, analyzing a time difference earlier stage or lagging stage of indexes of upstream and downstream nodes according to the industrial node association rule, and dividing into consistent, first and lagging index groups; calculating a diffusion index and a synthesis index, and analyzing supply and demand conduction indexes of upstream and downstream of the industry to form a prosperity sequence; comparing the distribution difference between the current sequence and the reference business sequence through KS test to judge whether abnormity exists or not; and tracing the source link with the abnormal condition according to the abnormal test result, and quantifying the influence degree of the source link on the whole industry chain.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial chain data processing, and in particular to a method and system for detecting anomalies of industrial chain nodes based on power timing characteristics. Background Art

[0002] The current methods for detecting abnormalities in industrial chain nodes generally rely on lagged data of relevant indicators in the industrial chain. The existing technologies cannot effectively utilize real-time power data, a key production factor, and lack targeted processing methods for seasonal fluctuations, noise interference and other problems in power data, resulting in the analysis results deviating from the actual supply and demand status. The existing algorithm model uses static correlation coefficients to calculate the correlation between upstream and downstream nodes in time series correlation analysis, ignoring the time difference effect in the transmission process of the industrial chain. For example, there is a significant time delay between the changes in electricity consumption in the upstream raw material production and the downstream manufacturing links, and the static model cannot capture such dynamic characteristics.

[0003] In addition, the fixed threshold warning mechanism is difficult to adapt to the volatility characteristics of different industries. It is easy to produce false alarms in high-volatility industries such as chemical industry, and may miss alarms in low-volatility industries such as machinery manufacturing. Although some studies have tried to introduce machine learning models to improve prediction accuracy, its black box characteristics lead to poor interpretability of the transmission path and high computational complexity, making it difficult to meet the engineering needs of real-time monitoring of large-scale industrial chains.

[0004] The technical solution of prior art document 1 (CN118332025A) deals with big data in the field of industrial economic knowledge graph construction and analysis, and uses interpolation filling models and industry big models to mine potential correlations in industrial economic data. However, its deep mining model based on random mask prediction and intelligent agent analysis method with preset thinking chain have technical problems in reliability and accuracy.

[0005] The technical solution of prior art document 2 (CN118297147A) processes big data in the field of industrial network knowledge graph construction, and uses an undirected graph matrix decomposition method to mine closely related sub-industry clusters within the industry. However, the matrix decomposition method based on target diagonal area adjustment has technical problems in reliability and accuracy. Summary of the invention

[0006] In view of the technical problems existing in the prior art, the present invention provides a method and system for detecting anomalies in industrial chain nodes based on power timing characteristics.

[0007] The present invention adopts the following technical solution.

[0008] In a first aspect, the present invention provides a method for detecting abnormalities in industry chain nodes based on power time series characteristics, comprising the following steps:

[0009] Combined with the characteristics of the industrial chain and the upstream and downstream associations within the industry, real-time obtain power data, socio-economic data, and upstream and downstream association data within the industry for different industries. Each type of data contains multiple indicators. After preprocessing the collected data, construct a preliminary selection indicator pool;

[0010] Use a time series model to perform seasonal adjustment on the indicator time series in the preliminary selection indicator pool;

[0011] Conduct correlation analysis based on the adjusted indicator time series to obtain industrial node association rules. According to the industrial node association rules, analyze the lead or lag period of the time difference of the indicators of upstream and downstream nodes, and divide them into consistent, leading, and lagging indicator groups;

[0012] Calculate the diffusion index and the composite index, and analyze the supply and demand transmission index of the upstream and downstream of the industry to form a prosperity sequence. The prosperity sequence is a multi-dimensional time series including the diffusion index, the composite index, and the supply and demand transmission index;

[0013] Take the prosperity sequence as the current sequence, and compare the distribution difference between the current sequence and the benchmark prosperity sequence through the KS test to determine whether there is an anomaly. If there is an anomaly, obtain an anomaly test result including the anomaly indicator group and the anomaly classification;

[0014] Trace back to the source link where the anomaly occurs according to the anomaly test result, quantify its impact on the overall industrial chain, and issue corresponding warnings according to its impact degree.

[0015] Optionally, the obtaining of industrial node association rules by conducting correlation analysis based on the adjusted indicator time series includes:

[0016] Construct a spatio-temporal two-dimensional data set to record the production status of each industrial node at each time point;

[0017] Through the frequent pattern mining algorithm, count the conduction paths and time intervals between nodes.

[0018] Optionally, the analyzing of the lead or lag period of the time difference of the indicators of upstream and downstream nodes according to the industrial node association rules and dividing them into consistent, leading, and lagging indicator groups includes:

[0019] Based on the set of node transmission time intervals obtained by the industrial node association rules, the optimal transmission interval is determined by calculating the time difference correlation coefficient. Taking the time series of industrial production nodes as the benchmark indicator, the sequence of the selected indicator group is aligned with the benchmark sequence according to different time offsets, and the ratio of the covariance of the two to the product of their respective standard deviations is calculated respectively to obtain the absolute value of the time difference correlation coefficient under each offset; the offset that maximizes the absolute value is selected as the optimal transmission interval. When the offset is zero, the selected indicator group is divided into a consistent indicator group, when the offset is negative, the selected indicator group is divided into a leading indicator group, and when the offset is positive, the selected indicator group is divided into a lagging indicator group.

[0020] Optionally, comparing the difference between the current sequence and the benchmark prosperity sequence distribution by using the KS test to determine whether there is an abnormality includes:

[0021] Take the prosperity sequence as the current sequence and obtain the historical prosperity sequence as the benchmark prosperity sequence;

[0022] The null hypothesis is that the current sequence distribution is consistent with the benchmark sequence distribution;

[0023] Traverse all time points and calculate the maximum absolute difference between the current sequence and the benchmark sequence at each time point as the test statistic;

[0024] When the statistic exceeds the preset critical value or the P value is less than the significance level, it is judged that there is an abnormality;

[0025] The abnormality is classified into at least one of supply-demand imbalance, conduction blockage, or overcapacity / insufficient capacity according to the abnormal indicator type.

[0026] Optionally, the process of determining the abnormality also includes dynamically adjusting the significance level according to industry characteristics and seasonal fluctuations, including:

[0027] Divide the historical volatility over the same period by the annual average volatility to get the volatility adjustment factor;

[0028] Multiplying the basic significance level by the volatility adjustment coefficient generates a dynamically adjusted significance level, where a higher sensitivity threshold is adopted for industries whose historical volatility during the same period is higher than the annual average, and vice versa.

[0029] Optionally, the link of tracing the source of the abnormal situation according to the abnormality inspection result includes:

[0030] Narrow the scope of investigation based on the type of abnormal indicators in the abnormal test results;

[0031] According to the time difference correlation coefficient of the abnormal indicator group, the time and node where the abnormal situation occurs are located.

[0032] Optionally, the abnormal index type in the abnormal inspection result for narrowing down the investigation scope includes:

[0033] If the leading indicator group is abnormal, the reasons for the abnormality include upstream raw material supply or policy changes;

[0034] If the lagging indicator group is abnormal, the reasons for the abnormality include downstream demand side or ineffective conduction in the intermediate link;

[0035] If the coincident indicator group is abnormal, the reasons for the abnormality include technical interruption in the current production link of the industrial chain or overall supply - demand synchronization imbalance.

[0036] Optionally, the quantification of the impact degree of the abnormal situation on the entire industrial chain includes:

[0037] Calculate the weight of node electricity consumption;

[0038] Superimpose the impact weight of the node related to the time difference correlation coefficient;

[0039] Sum up the impact weights of each abnormal node to obtain the total impact degree of the abnormal situation on the entire industrial chain, and select local optimization or overall optimization according to the size of the total impact degree.

[0040] In the second aspect, the present invention provides an industrial chain node anomaly detection system based on power time - series characteristics. Based on the industrial chain node anomaly detection method described in the first aspect of the present invention, the system includes:

[0041] Data acquisition and pre - processing module: used to obtain power, social economy and industrial chain related data in real - time, clean outliers and construct a primary selection index pool;

[0042] Seasonal adjustment module: used to eliminate the seasonal factors of data through a time - series model and generate a stationary time series;

[0043] Association analysis module: used to mine the association rules of industrial chain nodes and divide them into coincident, leading and lagging indicator groups;

[0044] Prosperity index calculation module: used to calculate the diffusion index, composite index and supply - demand conduction index, and generate a multi - dimensional prosperity sequence;

[0045] Anomaly detection module: used to identify the abnormal indicator group and abnormal type by comparing the current sequence with the benchmark distribution through the KS test;

[0046] Impact assessment and early - warning module: used to trace the source of the anomaly and quantify the impact degree, and trigger a hierarchical early - warning signal.

[0047] Optionally, the association analysis module includes:

[0048] FP-Growth improved algorithm engine, used to generate a timestamped conduction path map;

[0049] Neo4j graph database, storing the industrial association topology in the node relationship model.

[0050] Compared with the prior art, the method and system for detecting anomalies in industrial chain nodes based on power time series features provided by the present invention have the following beneficial effects:

[0051] 1. Aiming at the problems of the reliability and accuracy of the analysis method in using big data in the field of industrial economic knowledge graph in the prior art, the present invention uses a time series model to eliminate the seasonal and noise interference of power data, achieving an improvement in the reliability of the time series data benchmark; by dynamically dividing the leading / lagging index groups through the cross-correlation coefficient and determining the optimal conduction interval in combination with the sliding window mechanism, the accuracy of identifying the time difference between upstream and downstream nodes is improved.

[0052] 2. Aiming at the problems of the reliability and accuracy of the analysis method in using big data in the field of industrial network knowledge graph in the prior art, the present invention uses a frequent pattern mining algorithm to mine spatio-temporal frequent conduction paths, achieving an improvement in the reliability of constructing dynamic association rules; through the composite calculation of the diffusion index and the synthetic index and dynamically allocating index weights in combination with the entropy weight method, a multi-dimensional prosperity sequence including the supply-demand conduction index is generated, achieving an improvement in the accuracy of analyzing the supply-demand conduction state of the industrial chain.

[0053] 3. Aiming at the problems of the reliability and accuracy of the analysis method in using fixed threshold anomaly detection in the prior art, the present invention uses a dynamic significance level adjustment mechanism of KS test to generate an adaptive threshold based on the characteristics of industry fluctuations, achieving an improvement in the reliability of anomaly determination; through the reverse tracing of the cross-correlation coefficient and the superposition algorithm of electricity consumption weights, the abnormal source node is located and the total influence degree is quantified, achieving an improvement in the accuracy of abnormal source location and impact assessment. Description of the Drawings

[0054] Figure 1 is the flowchart of the method provided by the embodiment of the present invention. Detailed Embodiments

[0055] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the spirit of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0056] The present invention provides a method for detecting anomalies in industrial chain nodes based on power time series features in Embodiment 1, asFigure 1 As shown in the figure, it includes the following steps:

[0057] Step 1: Combining the characteristics of the industrial chain and the upstream and downstream relationships within the industry, obtain real-time power data, socio-economic data, and upstream and downstream relationship data within the industry for different industries. Each type of data contains multiple indicators. After preprocessing the collected data, construct a preliminary selected indicator pool;

[0058] In a further preferred but non-limiting implementation, the industries in Step 1 include steel, textile, chemical industry, automobile manufacturing, and mechanical equipment. Specifically, the multiple indicators included in the power data, socio-economic data, and upstream and downstream relationship data within each industry are shown in the following table:

[0059] Table 1 Indicators of power data, socio-economic data, and upstream and downstream relationship data for each industry

[0060]

[0061]

[0062] Step 2: Use a time series model to perform seasonal adjustment on the indicator time series in the preliminary selected indicator pool to eliminate data noise, seasonality, and trend interference, and improve the reliability of subsequent analysis.

[0063] Preferably, Step 2 includes:

[0064] Use the X12-ARIMA model to perform seasonal adjustment on the indicator time series;

[0065] Specifically, the X12-ARIMA model based on moving average analyzes the original series through multiple moving averages, adjusts the influence of holidays, etc., and automatically adjusts extreme values such as "single-point outliers (outliers occurring at a certain moment)", "level shift (instantaneously changing to a new level and remaining at the new level)", and "temporary change (instantaneously changing to a new level but gradually returning to the original level)" to obtain more accurate results at each step. Specific steps:

[0066] Let Y t represent a monthly time series without singular values. According to the characteristics of monthly data, decompose Y t into a trend-cycle term TC t , a seasonal term S t , and an irregular component I t . Since the seasonal, trend, and cycle terms of industrial fluctuations are independent of each other, an additive model is adopted. The original series can be expressed by the following formula:

[0067] Y t = TC t + St +I t

[0068] The specific calculation is divided into three stages:

[0069] (1) Initial estimate of seasonal adjustment

[0070] ① Calculate the initial estimate of the trend-cycle component by centering the 12-term moving average:

[0071]

[0072] ② Calculate the initial estimate of the SI term:

[0073]

[0074] ③ Calculate the initial estimate of the seasonal factor S by 3×3 moving average:

[0075]

[0076] ④ Eliminate the initial estimate of the seasonal adjustment result:

[0077]

[0078] ⑤ Initial estimate of the seasonal adjustment result:

[0079]

[0080] (2) Calculate the tentative trend-cycle component and the final seasonal factor

[0081] ① Calculate the tentative trend-cycle component using the Henderson moving average formula:

[0082]

[0083] ② Calculate the tentative SI:

[0084]

[0085] ③ Calculate the tentative seasonal factor by 3×5 term moving average:

[0086]

[0087] ④ Calculate the final seasonal factor:

[0088]

[0089] (3) Calculate the final trend-cycle component and irregular component

[0090] ① Calculate the final trend-cycle component using the Henderson moving average formula:

[0091]

[0092] ② Calculate the final irregular elements:

[0093]

[0094] After calculating the seasonal factor trend cycle elements and irregular elements a seasonally adjusted series can be obtained:

[0095]

[0096] Step 3: Conduct correlation analysis on the adjusted indicator time series in Step 2 to obtain industrial node association rules, analyze the lead or lag period of the time difference of the indicators of upstream and downstream nodes according to the industrial node association rules, and divide them into consistent, leading, and lagging indicator groups;

[0097] Preferably, in the said Step 3, obtaining the industrial node association rules by conducting correlation analysis on the adjusted indicator time series in Step 2 includes:

[0098] (1) Construct a spatio-temporal two-dimensional data set to record the production status of each industrial node at each time point

[0099] Based on the spatio-temporal two-dimensional data set established from the power data indicator time series seasonally adjusted in Step 2, screen out the spatio-temporal two-dimensional frequent sets, and the patterns presented during the calculation are shown in the following table:

[0100] Timeline / Industry Node <![CDATA[A1]]> <![CDATA[A2]]> <![CDATA[A3]]> ... <![CDATA[A N > 1 B I I ... I 2 B B I ... I 3 I B B ... I 4 I I B ... B

[0101] (2) Through the frequent pattern mining algorithm, count the conduction paths and time intervals between nodes

[0102] where, {A1, A2, A3,..., A N}(is the downstream node after starting to run from computing node P1. First, it reads data starting from time axis 1, judges the production status value of downstream node A1 (A1 = B), connects to the next industrial node A2. The time interval between A1 and A2 is 1 time unit, and it connects 1 time between A1 and A2. Then it judges the production status value of node A2 (A2 = I). At this time, it will continue to judge node A3. The time interval between A2 and A3 is 1 time unit, and it connects 1 time among A1, A2 and A3. Then it judges the power consumption status value of node A3 (A3 = I). If the condition is met, it judges the non-production status branch; otherwise, it selects the production branch flow and connects to the next computing node P2. Starting from time axis 2, the above process occurs again. At this time, the connection times between A1 and A2 are updated to 2 times, and the connection times among A1, A2 and A3 are also updated to 2 times. The connection times and interval times between nodes will be recorded in the computing node for the screening of important rules. Until all tuples in the dataset are searched, the algorithm stops running. Then the connection form of the association rule can be expressed as:

[0103] A1(B)(l = 0) → A N (B)(l = (N - 1)!)

[0104] A1(B)(l = 0) ∧ A2(I)(l = 1) → A N (B)(l = (N - 1)!)

[0105] A1(B)(l = 0) ∧ A2(I)(l = 1) ∧ A3(I)(t = 2) → A N (B)(l = (N - 1)!)

[0106] According to the above steps, the set of conduction time intervals l between nodes can be obtained.

[0107] Preferably, in step 3, according to the industrial node association rule, analyze the time difference leading period or lag period of the indicators of upstream and downstream nodes, and divide the consistent, leading and lagging indicator groups, including:

[0108] Based on the set of conduction time intervals l between nodes obtained from the industrial node association rule, determine the optimal conduction time interval l between nodes for each indicator by calculating the time difference correlation coefficient, and classify the indicator group based on this optimal l.

[0109] Specifically, for a certain industrial production node y = {y1, y2, y3,..., y n}} as the reference indicator; x = {x1, x2, x3,..., x n}} as the selected indicator, that is, the indicators in the indicator pool; r l is the time difference correlation coefficient, then:

[0110]

[0111] Select the l that makes |r l | the largest as the final lag period of this indicator, and classify the indicator group based on this l; where t represents time, n represents the number of data, l represents the leading or lag period, and when l = 0, it represents synchronization, then x is classified as a coincident indicator; when l < 0, it represents leading, then x is classified as a leading indicator; when l > 0, it represents lagging, then x is classified as a lagging indicator.

[0112] Step 4, based on the leading, coincident, and lagging indicator groups divided in Step 3, calculate the diffusion index and the composite index, and analyze the supply - demand transmission index of the upstream and downstream of the industry to form a prosperity sequence, where the prosperity sequence is a multi - dimensional time series including the diffusion index, the composite index, and the supply - demand transmission index;

[0113] Preferably, the specific content of Step 4 includes:

[0114] After screening the electricity consumption sequences of the leading, coincident, and lagging upstream and downstream nodes of the target node through the time - difference correlation analysis method, then respectively using the calculation methods of the diffusion index and the composite index, and selecting the time series of the indicators described in the table of Step 1 as the index items of the index for calculation, the supply - demand transmission index of the upstream and downstream of the industry can be evaluated. The specific calculation steps are as follows:

[0115] (1) Calculate the diffusion index

[0116] Calculate the ratio DI(t) of the number of expanding (rising) indicators in the indicator group (leading, coincident, lagging) at time t to the number of indicators adopted within the group:

[0117]

[0118] The judgment of whether it expands or not requires setting a specific time point in advance. The ideal setting is to compare the current value with the previous value. To avoid deviation, a 3 - month comparison interval is considered.

[0119] In specific calculations, the indicator with accelerating growth is assigned a value of 1, the indicator with decelerating growth is assigned a value of 0, and the indicator with neither acceleration nor deceleration is assigned a value of 0.5. The result of the weighted average is the diffusion index.

[0120] (2) Calculate the composite index

[0121] 1) Calculate the symmetric change rate of the indicator and standardize it

[0122] Let the indicator Y ij (t) be the i - th indicator of the j - th indicator group, representing the leading, coincident, and lagging indicator groups respectively, and i is the serial number of the indicator within the group; k j is the number of indicators in the j - th indicator group. First, find the symmetric change rate C of Y ij (t)ij (t):

[0123]

[0124] When the constituent index Y ij (t) has non-positive values, or when the index is a ratio sequence, take the first-order difference:

[0125] C ij (t) = Y ij (t) - Y ij (t - 1), t = 2, 3,..., n

[0126] To prevent the index with a large change range from dominating in the composite index, standardize the symmetric change rate C ij (t) so that its average absolute value is equal to 1, that is, the standardized change rate S ij (t) is obtained

[0127]

[0128] 2) Calculate the standardized average change rate of each index group

[0129] First, find the average change rate R j (t) of the leading, coincident, and lagging index groups:

[0130]

[0131] where w ij is the weight of the i-th index in the j-th index group. In this method, the entropy weight coefficient method and the time-varying weight model are comprehensively used to scientifically assign weights to improve the accuracy of index construction. The entropy weight method is used to judge the dispersion degree of things, and the greater the dispersion degree, the greater the impact on the comprehensive evaluation.

[0132] ① Normalization processing:

[0133] Assume that there are n samples for the evaluation object, and each sample has m indicators, then x ij represents the value of the j-th indicator of the i-th sample. Considering the non-uniformity of the measurement units of each indicator, it is necessary to standardize each indicator before calculating the weight, converting the absolute value of the indicator into a relative value to make each indicator comparable. In addition, positive indicators and negative indicators have different attributes (the larger the value of the positive indicator, the better; the smaller the value of the negative indicator, the better), so different algorithms need to be set to make positive indicators and negative indicators comparable.

[0134] Positive indicator:

[0135] Negative indicator:

[0136] ② Calculation of entropy value:

[0137] Based on the normalization process, calculate the proportion of the i-th sample value under the j-th indicator in this indicator:

[0138] Furthermore, calculate the entropy value of the j-th indicator:

[0139] ③ Determination of weights:

[0140] Calculate the information entropy redundancy d based on the obtained entropy value: j = 1 - e j ; j = 1, …, m; Finally, calculate the weights of each indicator:

[0141] On this basis, taking the consistent indicators as the benchmark, further calculate the exponential normalization factor F j :

[0142]

[0143] From this, the standardized average rate of change V j (t) can be calculated:

[0144]

[0145] Adjust the average rates of change of the leading indicator sequence and the lagging indicator sequence with the amplitude of the average rate of change of the consistent indicator sequence, so as to apply the three indices as a coordinated and consistent system.

[0146] (3) Calculate the initial composite index I j (t)

[0147] Let I j (1) = 100, then

[0148]

[0149] (4) Trend adjustment

[0150] To make the trends of the composite indices obtained from the three indicator groups consistent with the average trend of the sequences adopted in the consistent indicator group, trend adjustment is required. The trend adjustment makes the three composite indices into an integrated system, which is convenient for measuring cyclic fluctuations.

[0151] First, use the compound interest formula to calculate the respective average growth rates of each sequence in the consistent indicator group,

[0152]

[0153] Among them, and are respectively the average values of the first and last cycles of the \(i\)-th indicator in the coincident indicator group, and are respectively the number of months of the first and last cycles of the \(i\)-th indicator in the coincident indicator group, \(k_2\) is the number of coincident indicators, and \(m\) i is the interval between the center of the first cycle and the center of the last cycle.

[0154] Subsequently, the average growth rate of the coincident indicator group, that is, the target trend \(G\) r :

[0155]

[0156] For the initial composite indexes of leading, coincident, and lagging indicators, their average growth rates \(r\) j ′ are obtained respectively using the compound interest formula:

[0157]

[0158] Finally, trend adjustments are made to the standardized average change rates \(V\) j (\(t\)) of the three indicator groups:

[0159] \(V\) j ′(\(t\)) = \(V\) j (\(t\)) + (\(G\) r - \(r\) j ′), \(j = 1, 2, 3\), \(t = 2, 3, \cdots, b\)

[0160] (5) Select the base period and calculate the composite indicator

[0161] Here, the change rates of the indicators for each year are made into a composite index with the base year as 100.

[0162] Let \(I\) j ′(1) = 100, then

[0163]

[0164] The composite index with the base year as 100, that is:

[0165]

[0166] Among them, is the average value of \(I\) j ′(\(t\)) in the base year.

[0167] Further preferably, through the comprehensive analysis of the diffusion index and the composite index, the supply - demand conduction index of the upstream and downstream of the industry can be obtained, and its calculation formula is as follows:

[0168] SCI(t) = α × DI(t) + β × CI j (t)

[0169] Wherein, α and β are the weight coefficients of the diffusion index and the composite index respectively, and α + β = 1; The two weight coefficients can be adjusted according to actual needs. Exemplarily, if more attention is paid to the short-term expansion or contraction trend, the weight of the diffusion index can be increased. If more attention is paid to the comprehensive supply and demand conduction state, the weight coefficient of the composite index can be increased.

[0170] Step 5: Take the prosperity sequence determined in Step 4 as the current sequence, and compare the difference between the current sequence and the benchmark prosperity sequence distribution through the KS test to determine whether there is an abnormality. If there is an abnormality, obtain an abnormality test result including the abnormal index group and the abnormal classification.

[0171] Preferably, Step 5 includes:

[0172] Take the prosperity sequence determined in Step 4 as the current sequence F n (x), and obtain the historical prosperity sequence as the benchmark prosperity sequence F(x);

[0173] Adopt the KS test, that is, the Kolmogorov-Smirnov test. The process of the KS test includes:

[0174] (1) Propose the hypothesis H0: F n (x) = F(x)

[0175] (2) Calculate the statistic D n :

[0176] D n = sup x |F n (x) - F(x)|

[0177] The statistic D n is the maximum absolute difference in the cumulative probability of the current sequence and the benchmark prosperity sequence distribution at all time points;

[0178] (3) Judge the abnormality:

[0179] The comparison can be carried out in the following two ways:

[0180] ① If D n exceeds the critical value reject the original hypothesis and consider that there is an abnormality in the current sequence, otherwise accept the original hypothesis;

[0181] The critical value is obtained by referring to the KS test critical value table, where n is the sample size and α is the significance level.

[0182] ②Verify by calculating the P-value. If the P-value < α, where α is set to 0.05, reject the null hypothesis, consider the current sequence to be abnormal, trigger an alarm, otherwise accept the null hypothesis;

[0183] The P-value is obtained by calculating the statistic D n and the sample size n, and statistical software can be directly used for calculation.

[0184] (4) Further classify according to the type of abnormal index

[0185] The types of abnormal indexes include supply-demand imbalance, conduction block, and overcapacity / insufficiency;

[0186] Exemplarily, when the trend of the diffusion index DI deviates from that of the composite index CI, it indicates the contradiction between short-term demand and long-term production capacity, belonging to supply-demand imbalance;

[0187] When the distribution difference between leading indicators (such as electricity consumption in the steel industry) and lagging indicators (such as electricity consumption in automobile manufacturing) is significant, it reflects the failure of industrial chain conduction, belonging to conduction block;

[0188] When the composite index CI continuously deviates from the historical benchmark range, and D n exceeds the critical value it belongs to overcapacity / insufficiency.

[0189] Further preferably, in step 5, to avoid misjudgment, the significance level α is dynamically adjusted according to industry characteristics and seasonal fluctuations, and the formula is:

[0190]

[0191] where α adj is the adjusted significance level, and α base is the basic significance level;

[0192] Exemplarily, in the peak season of the steel industry, α = 0.1 (tolerating higher fluctuations) is adopted, and in the off-season, it is tightened to α = 0.05.

[0193] Step 6, trace back to the source link where the abnormal situation occurs according to the abnormal test result, quantify its impact on the overall industrial chain, and issue corresponding warnings according to its impact degree.

[0194] Preferably, in step 6, tracing back to the source link where the abnormal situation occurs according to the abnormal test result includes:

[0195] (1) Narrow the scope of investigation according to the type of abnormal index in the abnormal test result;

[0196] Specifically, the index classification includes leading, coincident, and lagging;

[0197] For example, if the leading indicator group (such as electricity consumption in steel and chemical industry) is abnormal, it indicates that the problem may be caused by upstream raw material supply or policy changes; if the lagging indicator group (such as electricity consumption in automobile manufacturing and home appliance production) is abnormal, it is necessary to trace back to the downstream demand side or the failure of transmission in the intermediate links; if the consistent indicator group (such as electricity consumption in machinery manufacturing and electronic assembly) is abnormal, it indicates that the problem may be caused by technical interruption of the current production link of the industrial chain or global imbalance of supply and demand.

[0198] Exemplarily, the abnormal problems of the consistent indicator group specifically include:

[0199] Production equipment failure: abnormal loads (such as overheating or overloading) on machine tools and production lines in the processing link, resulting in a sudden drop or increase in power consumption;

[0200] Energy supply fluctuations: regional power restrictions, voltage instability and other grid-level issues directly affect the production capacity of intermediate links;

[0201] Synchronous contraction / expansion of supply and demand: Order demand and raw material supply change synchronously, without upstream and downstream buffering (such as industry-wide order cancellations or synchronous expansion of production under policy stimulus).

[0202] (2) Based on the time difference correlation coefficient l of the abnormal indicator group, locate the time and node of the root cause;

[0203] For example, assuming that the power consumption of automobile manufacturing (lagging indicator, l=3) is abnormal at time t, the root cause time is tl=t-3, and check whether the leading indicator (such as steel power consumption) at time t-3 has a sudden change;

[0204] For example, if the electricity consumption of automobile manufacturing is abnormal in June 2024 (l=3), then the electricity consumption data of the steel industry in March 2024 is traced back. If the electricity consumption of steel suddenly increases by 15% during the same period, it can be determined as the root cause node.

[0205] Preferably, in step 6, quantifying the impact of the abnormal situation on the entire industrial chain includes:

[0206] (1) Calculate the node power consumption weight

[0207] Define the power consumption weight w of node i i :

[0208]

[0209] For example, if the steel industry accounts for 30% of the total electricity consumption in the industrial chain, then w 钢铁 =0.3;

[0210] (2) Superimpose the time difference correlation coefficient to correct the influence weight of node i

[0211] The calculation formula of the influence weight of node i is:

[0212] Influence weight i = w i × l i

[0213] where l i is the correlation coefficient of the time difference of node i;

[0214] (3) Calculate the overall influence degree of the industrial chain and take corresponding measures

[0215] If multiple nodes are abnormal simultaneously, the total influence degree is the sum of the influence weights of each node:

[0216] Total influence degree = ∑ Influence weight i

[0217] Select local optimization or overall optimization according to the magnitude of the total influence degree.

[0218] Exemplarily, if the total influence degree > 20%, trigger a red warning, and the entire industrial chain needs to be coordinated and adjusted; if the total influence degree is 10% - 20%, trigger an orange warning, and local optimization is recommended.

[0219] In Embodiment 2 of the present invention, an industrial chain node anomaly detection system based on power time series characteristics is provided. Based on the industrial chain node anomaly detection method described in Embodiment 1, the system includes:

[0220] Data acquisition and preprocessing module: used to obtain power, social economy, and industrial chain correlation data in real time, clean outliers, and construct a preliminary index pool;

[0221] Seasonal adjustment module: used to eliminate the seasonal factors of data through a time series model to generate a stationary time series;

[0222] Correlation analysis module: used to mine the correlation rules of industrial chain nodes and divide into consistent, leading, and lagging index groups;

[0223] Prosperity index calculation module: used to calculate the diffusion index, composite index, and supply - demand conduction index to generate a multi - dimensional prosperity sequence;

[0224] Anomaly detection module: used to identify the abnormal index group and abnormal type by comparing the current sequence with the reference distribution through the KS test;

[0225] Influence evaluation and warning module: used to trace the source of the anomaly and quantify the influence degree, and trigger a graded warning signal.

[0226] Preferably, the correlation analysis module includes:

[0227] FP - Growth improved algorithm engine, used to generate a conduction path map with timestamps;

[0228] The Neo4j graph database stores the industrial association topology in the node relationship model.

[0229] It should be noted that, through the modular system architecture and graph database storage technology, the present invention overcomes the problem of poor interpretability of the black-box model in the prior art, realizes the visual analysis of industrial association rules and the logical verification of the conduction path, and improves the practicability of engineering deployment and debugging efficiency.

[0230] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications or equivalent replacements can still be made to the specific implementation manners of the present invention, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. An abnormal detection method for industrial chain nodes based on power time series characteristics, characterized in that, It includes the following steps: Combining the characteristics of the industrial chain and the upstream and downstream associations within the industry, real-time obtain the power data, socio-economic data, and upstream and downstream association data within the industry of different industries. Each type of data contains multiple indicators. After preprocessing the collected data, construct a preliminary selection index pool; Use a time series model to perform seasonal adjustment on the index time series in the preliminary selection index pool; Based on the adjusted index time series, conduct correlation analysis to obtain industrial node association rules. According to the industrial node association rules, analyze the lead or lag period of the time difference of the indicators of upstream and downstream nodes, and divide them into consistent, leading, and lagging index groups; Calculate the diffusion index and the composite index, and analyze the supply and demand conduction index of the upstream and downstream of the industry to form a prosperity sequence. The prosperity sequence is a multi-dimensional time series containing the diffusion index, the composite index, and the supply and demand conduction index; Take the prosperity sequence as the current sequence, and use the KS test to compare the distribution difference between the current sequence and the benchmark prosperity sequence to determine whether there is an anomaly. If there is an anomaly, obtain an anomaly test result including an abnormal index group and an abnormal classification; Trace back to the source link where the abnormal situation occurs according to the anomaly test result, quantify its impact on the overall industrial chain, and issue corresponding warnings according to its impact degree.

2. The method for detecting anomalies in industrial chain nodes based on power time series characteristics according to claim 1, characterized in that: The obtaining of industrial node association rules based on the adjusted index time series includes: Construct a two-dimensional spatio-temporal dataset to record the production status of each industrial node at each time point; Through the frequent pattern mining algorithm, count the conduction paths and time intervals between nodes.

3. The method for detecting anomalies in industrial chain nodes based on power time series characteristics according to claim 2, characterized in that: The analyzing of the lead or lag period of the time difference of the indicators of upstream and downstream nodes according to the industrial node association rules and dividing them into consistent, leading, and lagging index groups includes: Based on the set of node conduction time intervals obtained from the industrial node association rules, determine the optimal conduction interval through the calculation of the time difference correlation coefficient. Take the time series of the industrial production node as the benchmark index, align the sequences of the selected index groups with the benchmark sequence at different time offsets, and calculate the ratio of the covariance of the two to the product of their respective standard deviations respectively to obtain the absolute value of the time difference correlation coefficient at each offset; select the offset that makes the absolute value the largest as the optimal conduction interval. When the offset is zero, the selected index group is divided into the consistent index group. When the offset is negative, the selected index group is divided into the leading index group. When the offset is positive, the selected index group is divided into the lagging index group.

4. The method for detecting anomalies in industrial chain nodes based on power time series characteristics according to claim 3, characterized in that: The comparing of the distribution difference between the current sequence and the benchmark prosperity sequence through the KS test to determine whether there is an anomaly includes: Take the prosperity sequence as the current sequence and obtain the historical prosperity sequence as the benchmark prosperity sequence; Propose the null hypothesis that the distribution of the current sequence is the same as the distribution of the benchmark sequence; Traverse all time points and calculate the maximum absolute difference between the cumulative probabilities of the current sequence and the benchmark sequence at each time point as the test statistic; When the statistic exceeds the preset critical value or the P value is less than the significance level, it is judged that there is an abnormality; The abnormality is classified into at least one of supply-demand imbalance, conduction blockage, or overcapacity / insufficient capacity according to the abnormal indicator type.

5. The method for detecting anomalies of industrial chain nodes based on power time series characteristics according to claim 4 is characterized in that: The process of judging abnormalities also includes dynamically adjusting the significance level according to industry characteristics and seasonal fluctuations, including: Divide the historical volatility over the same period by the annual average volatility to get the volatility adjustment factor; Multiplying the basic significance level by the volatility adjustment coefficient generates a dynamically adjusted significance level, where a higher sensitivity threshold is adopted for industries whose historical volatility during the same period is higher than the annual average, and vice versa.

6. The method for detecting anomalies of industrial chain nodes based on power time series characteristics according to claim 5 is characterized in that: The process of tracing the source of the abnormal situation according to the abnormal inspection result includes: Narrow the scope of investigation based on the type of abnormal indicators in the abnormal test results; According to the time difference correlation coefficient of the abnormal indicator group, the time and node where the abnormal situation occurs are located.

7. The method for detecting anomalies of industrial chain nodes based on power time series characteristics according to claim 6 is characterized in that: The types of abnormal indicators in the abnormal test results narrow the scope of investigation to include: If the leading indicator group is abnormal, the reasons for the abnormality include upstream raw material supply or policy changes; If the lagging indicator group is abnormal, the reasons for the abnormality include transmission failure at the downstream demand side or in the intermediate links; If the consistent indicator group is abnormal, the reasons for the abnormality include technical interruptions in the current production links of the industrial chain or global imbalance in supply and demand synchronization.

8. The method for detecting anomalies of industrial chain nodes based on power time series characteristics according to claim 7 is characterized in that: The impact of the quantitative abnormal situation on the entire industrial chain includes: Calculate the node power consumption weight; The influence weight of the node is calculated by superimposing the time difference correlation coefficient; The influence weights of each abnormal node are added up to obtain the total impact of the abnormal situation on the entire industrial chain, and local optimization or overall optimization is selected according to the size of the total impact.

9. An industrial chain node anomaly detection system based on power time series characteristics, based on the industrial chain node anomaly detection method according to any one of claims 1-8, characterized in that, The system includes: Data collection and preprocessing module: used to obtain power, social economy and industrial chain related data in real time, clean up abnormal values and build a preliminary index pool; Seasonal adjustment module: used to eliminate data seasonal factors through time series models and generate stable time series; Association analysis module: used to mine the association rules of industrial chain nodes and divide them into consistent, leading and lagging indicator groups; Prosperity index calculation module: used to calculate the diffusion index, composite index and supply-demand transmission index, and generate a multi-dimensional prosperity sequence; Anomaly detection module: used to compare the current sequence with the benchmark distribution through KS test and identify abnormal indicator groups and abnormal types; Impact assessment and early warning module: used to trace the source of anomalies and quantify the degree of impact, triggering graded early warning signals.

10. The industrial chain node anomaly detection system based on power time series characteristics according to claim 9 is characterized in that: The association analysis module comprises: FP-Growth improved algorithm engine, used to generate a timestamped conduction path map; Neo4j graph database, storing the industrial association topology in the node relationship model.

Images