A sewage pump intelligent maintenance method based on fault prediction

Abstract

The application discloses a kind of intelligent maintenance methods of sewage pump based on fault prediction, it is related to sewage pump management technical field, comprising: step one, the state jump signal that appears in the process of running is collected to sewage pump control system;Each class state jump signal is marked as state jump event, and state jump event sequence is established according to actual occurrence time sequence;Step two, for each state jump event identified in step one, set the response window of fixed length, collect the transient response process of multiple operating parameters in the window, extract the behavior characteristic index reflecting the stability and recovery capability of sewage pump system, form structured response vector set.The application constructs the response atlas model with state jump event as core, and the behavior characteristic index under various operating states is bound to atlas node in a structured manner, while retaining the time sequence relationship of operating event, the system stability performance and evolution path characteristics are fused.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sewage pump management, in particular to a sewage pump intelligent maintenance method based on fault prediction. BACKGROUND

[0002] In a sewage treatment system, as one of the key power equipment, the running state of the sewage pump has an important influence on the whole hydraulic conveying and sewage treatment efficiency. The sewage pump often appears state jump phenomena such as frequent start-stop, flow fluctuation and vibration aggravation in the running process due to factors such as water quality change, mechanical wear and electrical fluctuation. These jump behaviors reflect the change trend of the equipment state and the running stability to some extent. How to accurately identify these jump events and carry out subsequent analysis is the basis for realizing efficient operation and maintenance.

[0003] The existing sewage pump monitoring method generally uses a distributed sensor system to collect parameters such as current, voltage, flow, pressure and temperature in real time, and identifies and warns abnormal states through threshold judgment, trend analysis and other means. This kind of method has good adaptability in engineering practice, and can realize timely response to part of the sudden abnormality, especially suitable for single-point rapid detection of key monitoring indicators. However, in the face of multi-variable linkage abnormality and systematic degradation trend under complex operating environment, the traditional method still has certain limitations in the continuity of event identification, modeling ability of response behavior and foresight of fault warning.

[0004] With the development of intelligent sensing and graph modeling technology, graph-based behavior modeling and cross-cycle path analysis have gradually become an effective means to improve prediction ability. Taking state jump events as graph nodes and behavior characteristics as node attributes to construct a response graph can not only preserve the temporal relationship between events, but also mine the system stability change law from the overall structure evolution, providing data foundation and modeling support for intelligent maintenance oriented to path characteristics. The system degradation identification method based on graph evolution and response deviation is becoming a research hotspot in device predictive maintenance. Therefore, the present application provides a sewage pump intelligent maintenance method based on fault prediction. SUMMARY

[0005] The present application aims to provide a sewage pump intelligent maintenance method based on fault prediction to solve the problems mentioned in the background.

[0006] The present application can be realized by the following technical scheme: a sewage pump intelligent maintenance method based on fault prediction, comprising the following steps:

[0007] Step one, collect the state jump signals appearing in the running process of the sewage pump control system;

[0008] Mark each type of state jump signal as a state jump event, and establish a state jump event sequence according to the actual time sequence;

[0009] Step two, for each state jump event identified in step one, set a fixed length response window, collect the transient response process of multiple operating parameters in the window, extract the behavior characteristic index reflecting the stability and recovery ability of the sewage pump system, and form a structured response vector set;

[0010] Step three, the state jump event obtained in step one is taken as a graph node, and a graph structure connection relationship is constructed; the behavior characteristic index extracted in step two is bound to the corresponding graph node to form a response graph model, and the behavior characteristic index of the graph node is evaluated in multiple dimensions, and the node exceeding the abnormal judgment threshold is marked as an abnormal node;

[0011] Step four, the historical operation records of various state jump events in the response graph model are counted, the recovery failure rate is calculated, and the minimum risk path generation value from the current node to the historical fault node is calculated combined with the number of abnormal nodes and the characteristic degradation degree;

[0012] Step five, based on the preset degradation judgment condition, it is judged whether the system has a stability degradation trend;

[0013] Step six, if it is judged that there is stability degradation, a maintenance suggestion strategy is generated combined with the graph path characteristic offset direction and the historical evolution mode.

[0014] Further technical improvements of the present application are that in step two, based on the response curves of the same type of state jump events in the historical multi-period, a plurality of response behavior morphological clusters are formed, and each response morphological cluster is assigned a unique identifier;

[0015] After each jump event occurs, the change curves of multiple operating parameters in the response window are collected, the response morphological cluster to which it belongs is judged, and the offset degree from the cluster center is calculated;

[0016] The type identification of the jump event, the identification of the belonging cluster, the behavior index group and the cluster offset value are combined to form a structured response vector, which is used as the subsequent graph node feature.

[0017] Further technical improvements of the present application are that the index value acquisition step includes:

[0018] In step two, for the operating parameters with overshoot peak value in the response window and forming a monotone falling section, the overshoot peak value in the response curve is identified as the starting point, and the time point of first entering the steady state tolerance interval is determined as the ending point. The time point corresponding to the overshoot peak value and the time point of first entering the steady state tolerance interval are defined as the "stable falling region";

[0019] Based on the response curve corresponding to the stable falling region in the same state jump event in the historical period, a "standard falling behavior envelope band" is constructed, and the intersection overlap ratio between the response curve corresponding to the current stable falling region and the envelope and the envelope penetration rate are calculated as "stable trend consistency indicators";

[0020] Meanwhile, the trend reversal point and response inflection point in the response curve of the operating parameter are extracted, the time length and amplitude change value from the first trend turning to entering the steady state are calculated, and the "response inertia deviation index" is obtained;

[0021] The stable trend consistency index and the response inertia deviation index are written into the behavior characteristic index set of the corresponding state jump event node in the response atlas model as structured indexes.

[0022] Further technical improvement of the application is that in step four, the evaluation process of the feature degradation degree includes:

[0023] Z1, historical stable index band construction:

[0024] For each type of state jump event node, the system statistics its behavior characteristic index set in the historical normal period, and based on the historical mean value of each behavior characteristic index in the behavior characteristic index set And the standard deviation , the stability interval of each behavior characteristic index is constructed ;

[0025] Z2, current period index deviation calculation:

[0026] For the behavior characteristic index of the target node in the current operation period , the deviation of the target node is calculated ;

[0027] Z3, degradation level determination rule:

[0028] If the behavior characteristic index falls outside the stability interval, it is marked as "abnormal index";

[0029] If the deviation > The preset deviation threshold ( Based on the standard deviation ), it is marked as "serious degradation";

[0030] For each node, the number of abnormal indexes and the average deviation are counted to form the feature degradation degree value of the node .

[0031] Further technical improvements of the present application are as follows: after the response graph model is constructed, state jump event paths with completely consistent structures in multiple operation cycles are aggregated to construct cross-cycle state jump event path clusters, and the behavior characteristic index set of state jump event nodes in the cross-cycle state jump event path clusters in different cycles is subjected to time sequence analysis to determine whether there is a trend that the behavior characteristic index value deviates from its historical stable statistical interval in each cycle;

[0032] If the deviation enlarges in continuous multiple cycles, it is determined that the cross-cycle state jump event path cluster has a degradation trend, and the trend is taken as an enhanced criterion of system stability degradation risk and is incorporated into the graph model.

[0033] Further technical improvements of the present application are as follows: the method for obtaining the minimum risk path cost value comprises:

[0034] S1, based on the response graph model constructed in the current cycle, the nodes with current active or running state in the graph are screened out , and the node set that has been marked as a “known fault node” in all historical cycles is retrieved ;

[0035] S2, for each edge in the response graph model (from node to node ), the path weight of the edge is defined as , and the risk degree of the terminal node is calculated as follows:

[0036] ;

[0037] In the formula, is the abnormal marking value of node , which is set to 1 if the node is identified as an abnormal node, otherwise it is 0;

[0038] is the characteristic degradation degree value of node ;

[0039] is a Boolean value indicating whether the cross-cycle state jump event path cluster to which node belongs has a degradation trend, which is set to 1 if it has, otherwise it is 0;

[0040] and are the risk weight coefficients of the corresponding items, respectively;

[0041] S3, minimum path cost value calculation:

[0042] Based on the path weight of edge in step S2, , take the current node as the starting point, search all paths leading to the historical failure node using the shortest path search algorithm in the weighted graph, find the minimum value of the path total generation value, denoted as:

[0043] ; in the formula, represents a feasible path from the current node to the historical failure node ;

[0044] The minimum value of the path total generation value is taken as the risk generation value of the current operating state, and is written into the response graph model for subsequent stability degradation judgment and intelligent maintenance strategy generation.

[0045] Further technical improvements of the application are that: in multiple historical operating cycles, state jump event paths with consistent structures are extracted, and their corresponding cycle behavior index matrices are constructed as path behavior trajectory sets to form a standard path pool;

[0046] In the current operating cycle, the cycle behavior index matrix corresponding to the path to be evaluated is extracted, matched with the path behavior trajectory in the standard path pool, and the trajectory matching deviation degree is calculated.

[0047] If the behavior trajectory of the current path has a trend of continuously deviating from the central region of the main trajectory cluster in multiple consecutive cycles, and the deviation direction is consistent, it is determined that the path has a self-evolution feature deviation risk;

[0048] The trajectory deviation trend score of the path is written into the corresponding state jump event path structure in the response graph model as an additional weighting factor in the path risk generation value calculation formula, so as to enhance the early identification ability of the system evolution degradation trend.

[0049] Compared with the prior art, the application has the following beneficial effects:

[0050] The application binds the behavior characteristic indexes in various operating states to the graph nodes in a structured manner by constructing a response graph model with state jump events as the core, retains the timing relationship of operating events, and integrates system stability performance and evolution path characteristics, thereby realizing global degradation trend recognition ability under multi-dimensional data driving.

[0051] And, the "stable falling area" identification and "response behavior clustering" mechanism proposed by the present application can extract representative dynamic behavior templates in multiple operation cycles, and based on the historical templates, a new event is quickly attributed and a deviation degree is calculated, and then a "cluster deviation-behavior index" double-factor driven structured response vector is constructed, which improves the recognition accuracy of the system to the abnormal response behavior; in addition, through the cycle-by-cycle trend analysis of the behavior index on the path scale, the tracking of the degradation trend across cycles is realized, which helps to identify the state path with implicit degradation risk in advance.

[0052] On the other hand, the present application introduces "minimum risk path generation value" as a fault prediction index, and combines path structure, node abnormal state, index degradation degree, trajectory deviation trend and other multi-factor modeling to effectively construct a quantitative risk mapping between system operation state and historical fault state. Through the index, continuous quantitative evaluation of the current health state of the system can be realized, and personalized intelligent maintenance strategies can be pushed accordingly, including inspection suggestions, operation strategy adjustment suggestions and control linkage execution schemes, so as to realize the intelligent maintenance transformation from passive response to active prediction, and improve the safety, reliability and maintenance economy of equipment operation. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to facilitate the understanding of those skilled in the art, the present application will be further described below with reference to the accompanying drawings.

[0054] Figure 1 The method logic diagram of the present application is shown. DETAILED DESCRIPTION

[0055] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the specific implementation, structure, features and effects according to the present application are described in detail as follows in combination with the preferred embodiments and the drawings.

[0056] Embodiment 1: Please refer to Figure 1 As shown in the drawings, the present application provides a sewage pump intelligent maintenance method based on fault prediction, which comprises the following steps:

[0057] Step 1, trigger event identification:

[0058] Collect the state jump signals of the sewage pump control system during operation, including but not limited to pump start, pump stop, water level mutation, flow sudden rise or drop, valve opening and closing and other significant working condition changes during operation;

[0059] Specifically, the significant working condition change refers to the sudden jump of the system operation state caused by some key operations or external conditions during the operation of the sewage pump control system, which is usually manifested as:

[0060] Key parameters change dramatically in a short time, for example:

[0061] The rate of increase or decrease of monitoring parameters such as current, voltage, speed, flow, liquid level, pump cavity pressure exceeds the preset first speed threshold;

[0062] The rate of increase of temperature rise exceeds the corresponding second speed threshold, or cannot recover to steady state within the set time.

[0063] The operation mode changes suddenly, for example:

[0064] Suddenly start from the shutdown state;

[0065] Forced to stop or restart in continuous operation;

[0066] The control logic triggers an emergency bypass or switch pump group switching.

[0067] The system stability is disturbed, for example:

[0068] The response time after a certain start is prolonged by more than the preset prolongation threshold (such as from the normal 5 seconds to more than 8 seconds);

[0069] The steady state process that should be quickly established appears overshoot, oscillation or lag recovery, etc.

[0070] These working condition changes are not characterized by the continuity of time, but are discrete jump events, which can be identified and captured through state quantity monitoring of the control system, device log recording or abnormal event detection model.

[0071] Each type of state jump signal is marked as a state jump event, and a state jump event sequence is established according to the actual time sequence, which is used as an event index for subsequent data organization.

[0072] Step two, event response data collection and behavior feature extraction:

[0073] For each state jump event identified in step one, a fixed length response window is set, and the transient response process of multiple operating parameters in the response window is collected;

[0074] The operating parameters include but are not limited to current, flow, pressure, temperature, vibration;

[0075] Based on the transient response process, behavior feature indicators reflecting the stability and recovery ability of the sewage pump system are extracted, and the behavior feature indicators at least include:

[0076] Steady state establishment time: the time length required for the system operating parameters to gradually transition from the initial disturbance state to the stable operation state after a certain state jump event (such as pump start, valve action, etc.) occurs;

[0077] The acquisition method is based on the dynamic change curve of the target parameter (such as flow, current, temperature, etc.) in the response process;

[0078] The tolerance interval of stable operation is set (such as ±5% of the steady-state value);

[0079] From the event trigger time point, the time length experienced by the measured parameter when it first enters the tolerance zone and continuously maintains for a set time (such as 2 seconds).

[0080] Maximum overshoot amplitude: the maximum deviation of the target operating parameter from its steady-state value after a certain state jump event occurs; the acquisition method is to obtain the steady-state value (such as steady-state flow, steady-state current);

[0081] Find the maximum or minimum value of the parameter in the response process;

[0082] Maximum overshoot = | extreme value - steady-state value |.

[0083] Response recovery rate: refers to the change speed per unit time of the system parameter from the peak value (or valley value) after the jump to the steady-state tolerance interval; the acquisition method is to obtain the time period from the maximum deviation point to the stable interval in the response curve;

[0084] Calculate the average decline / rise rate of the parameter amplitude per unit time;

[0085] Recovery rate = (maximum deviation value - steady-state threshold value) / recovery time.

[0086] Response fluctuation residual amplitude: the maximum amplitude of the residual fluctuation of the parameter within a set time window after the system response tends to be stable; the acquisition method is to obtain the maximum amplitude of the residual fluctuation of the parameter within a set time window (such as 5 seconds) after the system enters the steady-state threshold interval;

[0087] Statistical parameter fluctuation amplitude;

[0088] That is, half of the difference between the maximum value and the minimum value, or the absolute difference.

[0089] Each state jump event and its corresponding behavior characteristic index form an event response characteristic group, forming a structured response vector set;

[0090] Step three, response atlas construction:

[0091] The state jump event obtained in step one is taken as the atlas node, and the graph structure connection relationship is constructed according to the actual operation sequence of the event;

[0092] The behavior characteristic index extracted in step two is bound to the corresponding graph node to form a response graph model including event nodes, event connection edges, and node characteristic attributes, which is used to reflect the state evolution path and dynamic stability performance of the sewage pump in different operation periods;

[0093] The behavior characteristic index of each state jump event node in the response graph model is multi-dimensionally evaluated to obtain an index value, and if the index value exceeds a preset abnormality judgment threshold, the node is marked as an abnormal node;

[0094] The index value obtaining step includes:

[0095] In step two, for the operation parameter that has an overshoot peak value and forms a monotonous falling section in the response window, the overshoot peak value in the response curve is identified as a starting point, and a time point at which the operation parameter first enters a steady state tolerance interval is determined as an ending point, and a "stable falling region" is defined between the time point corresponding to the overshoot peak value and the time point at which the operation parameter first enters the steady state tolerance interval;

[0096] Based on the response curve corresponding to the stable falling region in the same type of state jump event in the historical period, a "standard falling behavior envelope band" is constructed, and an intersection overlap ratio and an envelope penetration rate between the response curve corresponding to the current stable falling region and the envelope band are calculated as a "steady trend consistency index";

[0097] Meanwhile, a trend reversal point and a response inflection point in the response curve of the operation parameter are extracted, and a time length and a magnitude change value from the first trend turning to entering the steady state are calculated to obtain a "response inertia deviation index";

[0098] The steady trend consistency index and the response inertia deviation index are written into the behavior characteristic index set of the corresponding state jump event node in the response graph model as structured indexes.

[0099] Specifically, in step two, for the operation parameter (such as current, flow, pressure, etc.) that has an overshoot peak value and forms a monotonous falling section in the response window, a "stable falling region" in the response process is extracted, the stable falling region refers to a time period in the response curve, from a time point corresponding to the overshoot peak value to a time point at which the operation parameter first enters and continuously remains in a steady state tolerance interval, which is used to describe the dynamic behavior of the system from an abnormal response state to a steady state transition;

[0100] Based on the response curve corresponding to the stable falling area of the same type of state jump event in multiple historical periods, a "standard falling behavior envelope band" is constructed, which represents the typical falling behavior range of the system under normal stable conditions in history. The system compares the response curve corresponding to the stable falling area of the current state jump event with the standard falling behavior envelope band, calculates the intersection overlap ratio and envelope penetration rate, which are used as indicators to evaluate whether the current behavior deviates from the stable trend, defined as "stable trend consistency index";

[0101] At the same time, the system performs derivative analysis on the response curve formed by the response parameter time series, extracts the trend reversal point and response inflection point, and calculates the time length and amplitude change from the first trend turning point to the final entering of the stable state interval, and compares it with the historical normal sample mean, to obtain an index for measuring the response inertia change of the system, defined as "response inertia deviation index";

[0102] The stable trend consistency index and the response inertia deviation index constitute a structured response vector, which is used for multi-dimensional evaluation in this embodiment, as part of the behavior characteristic index set of the corresponding node of the current state jump event in the response graph model, for subsequent graph modeling, abnormal node identification and stability degradation evaluation.

[0103] Step four, recovery failure rate modeling and path risk assessment:

[0104] The historical running records of each type of state jump event in the response graph model are counted, and the recovery failure rate of the event type in the past running period is calculated, which refers to the proportion of event response characteristics that fail to recover to the stable threshold within a preset time window;

[0105] Combined with the number of abnormal nodes marked in the current graph path and the feature degradation degree, the minimum risk path value of the current state node to the historical known fault node is calculated, which is used as a quantitative index of the stability degradation degree of the current system;

[0106] In step four, the evaluation process of the feature degradation degree includes:

[0107] Z1, construction of historical stability index band:

[0108] For each type of state jump event node, the system counts its behavior characteristic index set in the historical normal period, and based on the historical mean and standard deviation of each behavior characteristic index in the behavior characteristic index set, the stability interval of each behavior characteristic index is constructed ;

[0109] Z2, calculation of current period index deviation:

[0110] behavior characteristic index of the target node in the current running cycle , calculate its deviation from the historical mean ;

[0111] Z3, deterioration level determination rule:

[0112] If the behavior characteristic index falls outside the stability interval, it is marked as an "abnormal index";

[0113] If the deviation > preset deviation threshold ( based on standard deviation ), it is marked as "severe deterioration";

[0114] For each node, count the number of abnormal indexes and the average deviation to form the characteristic deterioration degree value of the node .

[0115] The method for obtaining the minimum risk path value includes:

[0116] S1, based on the response graph model constructed in the current cycle, filter out the nodes in the graph that are currently active or in running state , and retrieve the node set that has been marked as "known fault node" in all historical cycles ;

[0117] S2, for each edge in the response graph model from node , define its path weight as the risk degree of its terminal node, and calculate as follows:

[0118] ;

[0119] where, is the abnormal index value of node , if identified as an abnormal node, set to 1, otherwise 0;

[0120] is the characteristic deterioration degree value of node ;

[0121] and

[0122] S3, minimum path value calculation:

[0123] Based on the path weight of edge in step S2 , the shortest path search algorithm in the weighted graph is used to search all paths leading to the historical failure node , and the minimum path total generation value is found, denoted as:

[0124] ; in the formula, represents a feasible path from the current node to the historical failure node ;

[0125] The minimum path total generation value is taken as the risk generation value of the current running state, and is written into the response graph model for subsequent stability degradation judgment and intelligent maintenance strategy generation.

[0126] Step five, running stability degradation judgment:

[0127] Determine whether the current sewage pump running state has a stability degradation trend by combining at least one of the following judgment conditions:

[0128] The response feature index of a certain type of state jump event continuously deviates from its historical cluster center in multiple consecutive running periods;

[0129] The recovery failure rate of a certain type of event shows a continuous upward trend in the sliding time window, and breaks through the preset statistical threshold;

[0130] The minimum risk path generation value from the current node to the historical failure node decreases, and there are multiple consecutive weak recovery nodes in the path;

[0131] If any of the judgment conditions is met, it is considered that the system is in a state of stability degradation;

[0132] Step six, intelligent maintenance strategy generation and linkage control execution:

[0133] After determining that the system has stability degradation, based on the feature deviation direction in the current graph path and the evolution mode of the historical degradation path, a maintenance suggestion strategy containing the following contents is generated:

[0134] Recommended key part inspection items;

[0135] Suggested maintenance execution time window;

[0136] Control parameter operation strategy adjustment suggestion;

[0137] Running load limiting operation suggestion;

[0138] ​And through the linkage interface with the control system, maintenance suggestion push, early warning signal release or maintenance work order generation are realized to achieve early intervention and fine intelligent maintenance of potential faults.

[0139] Embodiment 2: A sewage pump intelligent maintenance method based on fault prediction, comprising the following steps:

[0140] Step one, trigger event identification:

[0141] Collect the state jump signal of the sewage pump control system during operation;

[0142] Mark each type of state jump signal as a state jump event, and establish a state jump event sequence according to the actual time sequence as the event index for subsequent data organization.

[0143] Step two, event response data collection and behavior feature extraction:

[0144] For each state jump event identified in step one, set a fixed length response window, and collect the transient response process of multiple operating parameters in the response window;

[0145] The operating parameters include but are not limited to current, flow, pressure, temperature and vibration;

[0146] Based on the transient response process, the behavior characteristic index reflecting the stability and recovery ability of the sewage pump system is extracted, and the behavior characteristic index at least includes steady state establishment time, maximum overshoot amplitude, response recovery rate and response fluctuation residual amplitude;

[0147] Each state jump event and its corresponding behavior characteristic index form an event response feature group, forming a structured response vector set;

[0148] Compared with embodiment 1, embodiment 2 is based on the response curve of the same type of state jump event in the historical multi-period to form a plurality of response behavior form clusters, and each response form cluster is assigned a unique identifier;

[0149] After each jump event occurs, the change curve of multiple operating parameters in the response window is collected, the response form cluster to which it belongs is judged, and the offset degree from the cluster center is calculated;

[0150] The type identification of the jump event, the identification of the belonging cluster, the behavior index group and the cluster offset value are combined to form a structured response vector, which is used as the subsequent graph node feature.

[0151] Specifically, it includes:

[0152] One, based on the running data in the historical multi-period, the response parameter change curve corresponding to the same type of state jump event (such as "start event", "shutdown event", "load surge event", etc.) is obtained, and the response parameters can include pump current, voltage, flow, liquid level, rotating speed, vibration intensity, etc.

[0153] For each state jump event, a fixed length response time window (for example, 10 seconds) is set from the occurrence time, and the time series of each response parameter in the window is extracted as the original response curve set; for these response curve sets:

[0154] First, normalization processing (such as Z-score or Min-Max standardization) is performed;

[0155] Then, the difference between different response curves is calculated by using DTW (Dynamic Time Warping) distance, Euclidean distance, or similarity measurement method of Fourier transform coefficient;

[0156] Further, K-Means, DBSCAN or hierarchical clustering algorithm is used to cluster the response curves with similar dynamic behavior patterns into a cluster, and a plurality of response behavior pattern clusters (each cluster represents a typical response behavior pattern) are obtained;

[0157] Each response behavior pattern cluster is assigned a unique cluster identifier, and the curve in the cluster with the smallest average distance to all other curves is selected as the pattern center template of the cluster, which is used for subsequent matching judgment.

[0158] Second, after a new state jump event occurs, the system immediately collects the change curves of each monitoring parameter in the response window of the event to form the response sample of the current event. Then the following steps are performed:

[0159] The current response sample is processed using the same normalization method as the historical curve;

[0160] The similarity (such as DTW distance) between the sample and all historical behavior cluster center templates is calculated;

[0161] The sample is attributed to the behavior pattern cluster with the smallest distance, and the corresponding attribution cluster identifier is obtained;

[0162] At the same time, the offset value between the response sample and the cluster center template is recorded, and the offset value can be defined as: DTW distance value, or Euclidean distance of response key features (such as difference square sum of steady state establishment time, maximum overshoot amplitude, recovery rate, residual fluctuation amplitude, etc. Behavior index), or set threshold to judge whether the response is "abnormal response behavior" or "drift behavior".

[0163] Thirdly, the following elements are combined to form a structured response vector, which is used as a feature attribute of an event node in the response graph:

[0164] Event type identification (such as "start event", "valve jump event", etc., Event_Type_ID);

[0165] Home cluster identification;

[0166] Behavioral characteristic index set, for example: steady-state establishment time (time from event triggering to key parameter entering stable range), maximum overshoot amplitude (maximum deviation from steady-state target), recovery rate (parameter return to stable range speed, such as slope), residual fluctuation amplitude (average / peak value of small oscillation after steady state);

[0167] In-cluster offset value (difference quantization value from the center of the attributed behavior pattern);

[0168] Thirdly, after each new state jump event occurs, the operating parameter curve in its response window is collected, matched according to the preset behavior pattern template, the attributed response pattern cluster is judged, and the offset degree from the cluster center is calculated to obtain the drift distance or deviation value;

[0169] Fourthly, the type identification, home cluster identification, behavioral characteristic index set and in-cluster offset value of the above state jump event are integrated into a structured response vector, which is used as a feature attribute of a subsequent response graph node.

[0170] Step three, response graph construction:

[0171] The state jump event obtained in step one is taken as a graph node, and the graph structure connection relationship is constructed according to the actual operation sequence of the event;

[0172] The behavioral characteristic index extracted in step two is bound to the corresponding graph node to form a response graph model containing event nodes, event connection edges and node feature attributes, which is used to reflect the state evolution path and dynamic stability performance of the sewage pump under different operation periods;

[0173] The behavioral characteristic index of each state jump event node in the response graph model is evaluated in multiple dimensions and its index value is obtained. If the index value exceeds the preset abnormal judgment threshold, the node is marked as an abnormal node;

[0174] Step four, recovery failure rate modeling and path risk assessment:

[0175] The historical operation records of each type of state jump event in the response graph model are counted, and the recovery failure rate of the event type in the past operation period is calculated. The recovery failure rate refers to the proportion of event response characteristics that fail to recover to the stable threshold within the preset time window.

[0176] The minimum risk path generation value of the current node to the historical fault node is calculated as a quantitative indicator of the degradation of the current system stability, in combination with the number of abnormal nodes marked in the current graph path and the degree of feature degradation.

[0177] Step five, running stability degradation judgment:

[0178] Determine whether the current sewage pump running state has a stability degradation trend by combining at least one of the following determination conditions:

[0179] The response feature index of a certain type of state jump event continuously deviates from its historical cluster center in multiple consecutive running periods;

[0180] The recovery failure rate of a certain type of event shows a continuous upward trend within a sliding time window, and breaks through a preset statistical threshold;

[0181] The minimum risk path generation value of the current node to the historical fault node decreases, and there are multiple consecutive weak recovery nodes in the path;

[0182] If any of the determination conditions is met, it is considered that the system is in a state of stability degradation;

[0183] Step six, intelligent maintenance strategy generation and linkage control execution:

[0184] After determining that the system has stability degradation, based on the feature deviation direction in the current graph path and the evolution mode of the historical degradation path, a maintenance recommendation strategy is generated, which includes the following content:

[0185] Recommended key part inspection items;

[0186] Suggested maintenance execution time window;

[0187] Control parameter operation strategy adjustment suggestion;

[0188] Limiting operation suggestion of running load;

[0189] And through the linkage interface with the control system, maintenance suggestion pushing, early warning signal publishing or maintenance work order generation are implemented to realize early intervention and fine intelligent maintenance of potential faults.

[0190] Implementation:3: An intelligent maintenance method for sewage pumps based on fault prediction, comprising the following steps:

[0191] Step one, trigger event identification:

[0192] Collect state jump signals that occur in the running process of the sewage pump control system;

[0193] Each type of state jump signal is marked as a state jump event, and a state jump event sequence is established according to the actual time sequence, as an event index for subsequent data organization.

[0194] Step two, event response data collection and behavior characteristic extraction:

[0195] For each state jump event identified in step one, a fixed-length response window is set, and the transient response process of multiple operating parameters in the response window is collected;

[0196] The operating parameters include but are not limited to current, flow, pressure, temperature, vibration;

[0197] Based on the transient response process, the behavior characteristic index reflecting the stability and recovery ability of the sewage pump system is extracted, and the behavior characteristic index at least includes the steady-state establishment time, the maximum overshoot amplitude, the response recovery rate and the response fluctuation residual amplitude;

[0198] Each state jump event and its corresponding behavior characteristic index form an event response characteristic group, forming a structured response vector set;

[0199] Step three, response atlas construction:

[0200] The state jump events obtained in step one are taken as atlas nodes, and the connection relationship of the graph structure is constructed according to the actual operation sequence of the events;

[0201] The behavior characteristic index extracted in step two is bound to the corresponding atlas node, forming a response atlas model containing event nodes, event connection edges and node characteristic attributes, which is used to reflect the state evolution path and dynamic stability performance of the sewage pump under different operation periods;

[0202] The behavior characteristic index of each state jump event node in the response atlas model is evaluated in multiple dimensions and its index value is obtained, and if the index value exceeds the preset abnormal judgment threshold, the node is marked as an abnormal node;

[0203] The index value acquisition step includes:

[0204] In step two, for the operating parameters that have overshoot peak value and form a monotone falling section in the response window, the overshoot peak value in the response curve is identified as the starting point, and the time point of first entering the steady-state tolerance interval is determined as the ending point. The time point corresponding to the overshoot peak value and the time point of first entering the steady-state tolerance interval are defined as the "stable falling region";

[0205] Based on the response curves corresponding to the stable falling region in the same state jump event in the historical period, a "standard falling behavior envelope band" is constructed, and the intersection overlap ratio and envelope penetration rate between the response curve corresponding to the current stable falling region and the envelope band are calculated as the "stable trend consistency index";

[0206] At the same time, the trend reversal point and response inflection point in the response curve of the operating parameter are extracted, and the time length and amplitude change value from the first trend turning point to the entering of the steady state are calculated to obtain the "response inertia deviation index";

[0207] The stable trend consistency index and the response inertia deviation index are written into the response atlas model as the behavior characteristic index set of the corresponding state jump event node.

[0208] Specifically, in step two, for the operating parameters (such as current, flow, pressure, etc.) that have overshoot peak values and form a monotonous falling section in the response window, the "stable falling region" in the response process is extracted, which refers to the time period from the time point corresponding to the overshoot peak value to the time point when the operating parameter first enters and continuously remains in the steady state tolerance interval, used to depict the dynamic behavior of the system from the abnormal response state to the steady state;

[0209] Based on the response curves corresponding to the stable falling region in the same state jump event in the historical period, a "standard falling behavior envelope band" is constructed, and the intersection overlap ratio and envelope penetration rate between the response curve corresponding to the current stable falling region and the envelope band are calculated as the "stable trend consistency index";

[0210] At the same time, the system performs derivative analysis on the response curve formed by the response parameter time sequence, extracts the trend reversal point and response inflection point, and calculates the time length and amplitude change value from the first trend turning point to the entering of the steady state interval, and compares it with the historical normal sample mean to obtain the index for measuring the response inertia change of the system, defined as the "response inertia deviation index";

[0211] The stable trend consistency index and the response inertia deviation index constitute a structured response vector, which is part of the behavior characteristic index set of the corresponding node in the response atlas model of the current state jump event, used for subsequent atlas modeling, abnormal node identification and stability degradation evaluation.

[0212] Step four, recovery failure rate modeling and path risk assessment:

[0213] The historical running records of each type of state jump event in the response graph model are counted, and the recovery failure rate of the event type in the past running period is calculated, which refers to the proportion of the event response feature failing to recover to the stable threshold within the preset time window;

[0214] In combination with the number of abnormal nodes marked in the current graph path and the feature degradation degree thereof, the minimum risk path generation value of the current state node to the historical known fault node is calculated, serving as a quantitative index of the degradation degree of the current system stability;

[0215] The evaluation process of the feature degradation degree includes:

[0216] Z1, historical stability index band construction:

[0217] For each type of state jump event node, the system counts the behavior feature index set thereof in the historical normal period, and based on the historical mean value of each behavior feature index in the behavior feature index set and the standard deviation , the stability interval of each behavior feature index is constructed ;

[0218] Z2, current period index deviation calculation:

[0219] For the behavior feature index of the target node in the current running period , the deviation thereof relative to the historical mean value is calculated ;

[0220] Z3, degradation level determination rule:

[0221] If the behavior feature index falls outside the stability interval, it is marked as an “abnormal index”;

[0222] If the deviation is greater than the preset deviation threshold ( which is set based on the standard deviation ), it is marked as “severe degradation”;

[0223] For each node, the number of abnormal indexes and the average deviation thereof are counted to constitute the feature degradation degree value of the node .

[0224] After the response graph model is constructed, the state jump event paths with the same structure in multiple running periods are aggregated to construct a cross-period state jump event path cluster, and the behavior feature index sets of the state jump event nodes in the cross-period state jump event path cluster in different periods are analyzed in time sequence to determine whether there is a trend that the behavior feature index value deviates from the historical stable statistical interval in each period.

[0225] If the deviation widens in multiple consecutive cycles, it is determined that the path cluster of cross-cycle state transition events has a deterioration trend, and this trend is incorporated into the graph model as an enhanced criterion for the risk of system stability degradation.

[0226] Specifically, "a state transition event path with completely consistent structure" means that the types of state transition event nodes, the order of event connections, and the graph structure topology contained in each path are consistent. For example, "startup event structured response vector → structured response vector load surge event structured response vector → structured response vector vibration fluctuation event". If this path repeats in multiple cycles, it is defined as the same cross-cycle state transition event path cluster.

[0227] For each state transition event node in the path cluster, the system statistically analyzes its corresponding set of behavioral characteristic indicators in different operating cycles, including steady-state establishment time, maximum overshoot amplitude, recovery rate, and residual fluctuation amplitude.

[0228] All indicators are organized according to event node number and period number, forming a two-dimensional time series indicator matrix: = The k-th index value of the j-th node in the i-th period; where:

[0229] i∈[1,n], is the cycle number;

[0230] j∈[1,m], is the node number of the state transition event in the path;

[0231] k∈[1,k], where k is the type number of the behavioral characteristic index.

[0232] Specifically, the two-dimensional time series index matrix can also be understood as a periodic behavior index matrix that constitutes the state transition event path in cross-cycle operation, which is used for subsequent path trajectory matching and offset analysis.

[0233] Performing trend analysis on the above indicator matrix includes the following steps:

[0234] Constructing a historical stable statistical interval: Determining the mean of each behavioral characteristic indicator over a historical period T. with standard deviation Calculate the stability interval: ;

[0235] Calculate the deviation of behavioral characteristic indicators: Within the current sliding window, calculate the deviation of the k-th indicator value from the center value in the i-th period. ;

[0236] Conditions for identifying degradation trends: If any of the following conditions are met, the path cluster is determined to have a degradation trend:

[0237] deviation degree of a certain behavior characteristic index in consecutive L periods continuously increasing;

[0238] at least P% of the behavior characteristic indexes exceed their stable intervals in consecutive L periods;

[0239] the linear fitting slope of the index deviation degree sequence exceeds the set slope threshold, indicating that the behavior characteristic index is rapidly deteriorating.

[0240] Once a path cluster satisfying any of the above deterioration conditions is identified, the system marks the cross-period state jump event path cluster as a deterioration path cluster and writes its marking information into the response graph model for use as a stability degradation risk identification basis in subsequent fault prediction and intelligent maintenance strategy generation steps.

[0241] The method for obtaining the minimum risk path cost value comprises the following steps:

[0242] S1. Based on the response graph model constructed in the current period, filter out the nodes in the graph that are currently active or in an operating state , and retrieve the set of nodes that have been marked as "known fault nodes" in all historical periods ;

[0243] S2. For each edge in the response graph model (from node to node ), define its path weight as the risk degree of its terminating node, calculated as follows:

[0244] ;

[0245] wherein is the abnormality marking value of node , which is set to 1 if it is identified as an abnormal node, otherwise it is 0;

[0246] is the characteristic deterioration degree value of node ;

[0247] is a Boolean value indicating whether the cross-period state jump event path cluster to which node belongs has a deterioration trend, which is set to 1 if it does, otherwise it is 0;

[0248] and

[0249] S3. Minimum path cost value calculation:

[0250] based on the edgePath weight of From the current node , search all paths leading to the historical failure node using the shortest path search algorithm in the weighted graph, find the minimum path total generation value, denoted as:

[0251] ; In the formula, represents a feasible path from the current node to the historical failure node ;

[0252] The minimum path total generation value is taken as the risk generation value of the current operating state, and is written into the response graph model for subsequent stability degradation judgment and intelligent maintenance strategy generation.

[0253] In this embodiment, the warning level is set according to the size of , for example:

[0254] <1.0: low risk;

[0255] 1.0≤ <2.5: medium risk;

[0256] ≥2.5: high risk, triggering maintenance strategy generation.

[0257] Step five, running stability degradation judgment:

[0258] Determine whether the current sewage pump operating state has a stability degradation trend by combining at least one of the following judgment conditions:

[0259] The response feature index of a certain type of state jump event continuously deviates from its historical cluster center in multiple consecutive running cycles;

[0260] The recovery failure rate of a certain type of event shows a continuous upward trend within a sliding time window, and breaks through a preset statistical threshold;

[0261] The minimum risk path generation value from the current node to the historical failure node decreases, and there are multiple consecutive weak recovery nodes in the path;

[0262] If any of the judgment conditions is met, it is considered that the system is in a stability degradation state;

[0263] Step six, intelligent maintenance strategy generation and linkage control execution:

[0264] After determining that the system has stability degradation, based on the feature offset direction in the current atlas path and the evolution mode of the historical degraded path, a maintenance recommendation strategy is generated, which includes the following contents:

[0265] Recommended key part inspection items to be performed;

[0266] Recommended maintenance execution time window;

[0267] Control parameter operation strategy adjustment suggestion;

[0268] Limiting operation suggestion of running load;

[0269] And through the linkage interface with the control system, maintenance suggestion pushing, early warning signal issuing or maintenance work order generation are realized to achieve early intervention and fine intelligent maintenance of potential faults.

[0270] Embodiment 4: A sewage pump intelligent maintenance method based on fault prediction, comprising the following steps:

[0271] Step one, trigger event identification:

[0272] Collect the state jump signal of the sewage pump control system during operation;

[0273] Mark each type of state jump signal as a state jump event, and establish a state jump event sequence according to the actual time sequence, as an event index for subsequent data organization.

[0274] Step two, event response data collection and behavior feature extraction:

[0275] For each state jump event identified in step one, set a fixed response window to collect the transient response process of multiple operating parameters in the response window;

[0276] The operating parameters include but are not limited to current, flow, pressure, temperature and vibration;

[0277] Based on the transient response process, the behavior characteristic indexes reflecting the stability and recovery ability of the sewage pump system are extracted, including at least steady state establishment time, maximum overshoot amplitude, response recovery rate and response fluctuation residual amplitude;

[0278] Each state jump event and its corresponding behavior characteristic index form an event response feature group, forming a structured response vector set;

[0279] Step three, response atlas construction:

[0280] The state jump events obtained in step one are taken as atlas nodes, and the graph structure connection relationship is constructed according to the actual operation sequence of the events;

[0281] Bind the behavior characteristic indexes extracted in step two to the corresponding graph nodes to form a response graph model containing event nodes, event connection edges, and node characteristic attributes, which is used to reflect the state evolution path and dynamic stability performance of the sewage pump under different operation cycles;

[0282] Perform multi-dimensional evaluation on the behavior characteristic indexes of each state jump event node in the response graph model and obtain the index values thereof, and if the index values exceed the preset abnormality determination threshold, mark the node as an abnormal node;

[0283] Step four, recovery failure rate modeling and path risk assessment:

[0284] Statistically analyze the historical operation records of each type of state jump event in the response graph model, calculate the recovery failure rate of the event type in the past operation cycle, and the recovery failure rate refers to the proportion of event response characteristics that fail to recover to the stable threshold within the preset time window;

[0285] Combine the number of abnormal nodes marked in the current graph path and the characteristic degradation degree thereof, calculate the minimum risk path generation value from the current state node to the historical known fault node, and use the minimum risk path generation value as a quantitative index of the degradation degree of the current system stability;

[0286] Compared with embodiment 3, embodiment 4 extracts state jump event paths with consistent structures in multiple historical operation cycles, constructs the corresponding period behavior index matrix as a path behavior trajectory set, and forms a standard path pool;

[0287] In the current operation cycle, extract the period behavior index matrix corresponding to the path to be evaluated, match it with the path behavior trajectory in the standard path pool, and calculate the trajectory matching deviation degree;

[0288] If the behavior trajectory of the current path has a consistent deviation direction and a trend of continuously deviating from the central region of the standard path pool main trajectory cluster in multiple consecutive cycles, it is determined that the path has a self-evolution type characteristic deviation risk;

[0289] The trajectory deviation trend score of the path is written into the corresponding state jump event path structure in the response graph model as a new index dimension, and is used as an additional weighting factor in the path risk generation value calculation formula, so as to enhance the early identification ability of the system evolution degradation trend.

[0290] Step five, operation stability degradation judgment:

[0291] Determine whether the current sewage pump operation state has a stability degradation trend by comprehensively considering at least one of the following conditions:

[0292] The response characteristic indexes of a certain type of state jump event continuously deviate from the historical clustering center in multiple consecutive operation cycles;

[0293] The recovery failure rate of a certain type of event shows a continuous upward trend in a sliding time window, and breaks through a preset statistical threshold;

[0294] The minimum risk path value of the current node to the historical failure node decreases, and there are multiple consecutive weak recovery nodes in the path;

[0295] If any of the determination conditions is met, it is considered that the system is in a state of stability degradation;

[0296] Step six, intelligent maintenance strategy generation and linkage control execution:

[0297] After determining that the system has stability degradation, based on the characteristic offset direction in the current graph path and the evolution mode of the historical degraded path, a maintenance suggestion strategy is generated, which includes the following contents:

[0298] Recommended key part inspection items for execution;

[0299] Suggested maintenance execution time window;

[0300] Control parameter operation strategy adjustment suggestion;

[0301] Limiting operation suggestion of running load;

[0302] And through the linkage interface with the control system, maintenance suggestion pushing, early warning signal issuing or maintenance work order generation are realized to achieve early intervention and fine intelligent maintenance of potential faults.

[0303] The above formulas are all dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the latest real situation, and the preset parameters and threshold values in the formula are set by the person skilled in the art according to the actual situation.

[0304] The above is only a preferred embodiment of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the technical solution of the present application, and any simple modification, equivalent change and modification of the above embodiment according to the technical essence of the present application are still within the scope of the technical solution of the present application.

Claims

1. A method for intelligent maintenance of a sewage pump based on failure prediction, characterized in that, The method comprises the following steps: Step 1, collect the state jump signals of the sewage pump control system during operation; Mark each type of state jump signal as a state jump event, and establish a state jump event sequence according to the actual time sequence; Step 2, for each state jump event identified in step 1, set a fixed length response window, collect the transient response process of multiple operating parameters in the window, extract the behavior characteristic index reflecting the stability and recovery ability of the sewage pump system, and form a structured response vector set; Step 3, the state jump events obtained in step 1 are taken as graph nodes to construct a graph structure connection relationship; The behavior characteristic index extracted in step 2 is bound to the corresponding graph node to form a response graph model, and the behavior characteristic index of the graph node is evaluated in multiple dimensions, and the node exceeding the abnormal judgment threshold is marked as an abnormal node; Step 4, the historical operation records of various state jump events in the response graph model are counted, the recovery failure rate is calculated, and the minimum risk path generation value from the current node to the historical fault node is calculated according to the number of abnormal nodes and the characteristic degradation degree; The method for obtaining the minimum risk path generation value comprises: S1, based on the response graph model constructed in the current cycle, screen out the nodes in the graph that are currently active or in operation , and retrieve the set of nodes that have been marked as "known fault nodes" in all historical cycles ; S2. For each edge in the response graph model define its path weight as the risk level of its terminating node, computed as follows: ； In the formula, is an abnormal flag value of the node , and is set to 1 if the node is identified as an abnormal node, otherwise is 0. characteristic degradation degree value of the node characteristic degradation degree value of the node node Boolean value indicating whether there is a degradation trend in the cluster of cross-cycle state jump events belonging to the node, set to 1 if there is, otherwise 0; respectively the risk weight coefficients of the corresponding items; S3, based on the step S2, the edge path weight , using the shortest path search algorithm in the weighted graph, search all paths leading to the historical failure node with the current node as the starting point, find the minimum value of the path total value, recorded as: ; wherein, represents a feasible path from the current node to the historical failure node ; minimizing the path total cost value the risk cost value as the current running state, and write into the response atlas model; Step 5, judge whether the system has a stability degradation trend based on the preset degradation judgment condition; Step 6, if it is judged that there is a stability degradation, a maintenance suggestion strategy is generated according to the graph path characteristic offset direction and the historical evolution mode.

2. A method for intelligent maintenance of a sewage pump based on failure prediction as claimed in claim 1, wherein, The degradation judgment condition in step 5 comprises: The response characteristic index of a certain type of state jump event continuously deviates from its historical cluster center in multiple consecutive operation cycles; The recovery failure rate of a certain type of event shows a continuous upward trend in the sliding time window, and breaks through the preset statistical threshold; The minimum risk path generation value from the current node to the historical fault node decreases, and there are multiple consecutive weak recovery nodes in the path; If any of the judgment conditions is met, it is considered that the sewage pump control system is in a stable degradation state.

3. A method for intelligent maintenance of a sewage pump based on failure prediction as claimed in claim 1 wherein, In step 2, the response curves of the same type of state jump events in multiple historical cycles are clustered to form multiple response behavior morphological clusters, and each response morphological cluster is assigned a unique identifier; After each jump event occurs, the change curves of multiple operating parameters in the response window are collected, the response morphological cluster to which the jump event belongs is judged, and the offset degree of the jump event from the cluster center is calculated; The type identification of the jump event, the cluster identification, the behavior index group and the cluster offset value are combined to form a structured response vector, which is used as the feature of the subsequent graph node.

4. A method for intelligent maintenance of a sewage pump based on failure prediction as claimed in claim 1 wherein, The index value acquisition step comprises: In step 2, for the operating parameters that have overshoot peaks and form a monotonous falling section in the response window, the overshoot peak in the response curve is identified as the starting point, and the time point when the first time enters the steady state tolerance interval is determined as the ending point. The time point corresponding to the overshoot peak and the time point when the first time enters the steady state tolerance interval are defined as the "stable falling region"; Based on the response curve corresponding to the stable falling region in the same state jump event in the historical period, a "standard falling behavior envelope band" is constructed, and the intersection overlap ratio between the response curve corresponding to the current stable falling region and the envelope is calculated as the "stable trend consistency index"; At the same time, the trend reversal point and response inflection point in the response curve of the operating parameter are extracted, and the time length and amplitude change value from the first trend turning to the entering of the steady state are calculated to obtain the "response inertia deviation index"; The stable trend consistency index and the response inertia deviation index are written into the response graph model as the behavior characteristic index set of the corresponding state jump event node.

5. A method for intelligent maintenance of a sewage pump based on failure prediction as claimed in claim 1 wherein, In step four, the feature degradation degree evaluation process includes: Z1, historical stable index band construction: For each type of state jump event node, the system counts a set of behavior characteristic indexes of its behavior in the historical normal period, and based on the historical mean value of each behavior characteristic index in the set of behavior characteristic indexes and the standard deviation , a stability interval of each behavior characteristic index is constructed ; Z2, current period index deviation calculation: behavioral characteristic of the target node in the current operating cycle , calculating its deviation from the historical mean ; Z3, degradation level determination rule: If the behavior characteristic indicator If it falls outside the stability interval, it is marked as "anomaly indicator"; if the deviation degree > the preset deviation threshold then it is marked as "severe deterioration"; For each node, the number of abnormal indicators and the average deviation degree are counted to form the feature degradation degree value of the node .

6. A method for intelligent maintenance of a sewage pump based on failure prediction as claimed in claim 1 wherein, After the response graph model is constructed, the state jump event paths with completely consistent structure in multiple operation periods are aggregated to construct a cross-period state jump event path cluster, and the behavior characteristic index set of the state jump event nodes in the cross-period state jump event path cluster in different periods is analyzed in time sequence to determine whether there is a trend of deviation of the behavior characteristic index value from its historical stable statistical interval in each period in the path cluster; If the deviation expands in multiple consecutive periods, it is determined that the cross-period state jump event path cluster has a degradation trend, and the trend is taken as an enhanced criterion for system stability degradation risk and is included in the graph model.

7. A method for intelligent maintenance of a sewage pump based on failure prediction as claimed in claim 1 wherein, In multiple historical operation periods, state jump event paths with consistent structure are extracted, and their corresponding period behavior index matrices are constructed as path behavior trajectory sets to form a standard path pool; In the current operation period, the period behavior index matrix corresponding to the path to be evaluated is extracted, matched with the path behavior trajectory in the standard path pool, and the trajectory matching deviation degree is calculated; If the behavior trajectory of the current path has a consistent trend of continuously deviating from the center region of the standard path pool main trajectory cluster in multiple consecutive periods, it is determined that the path has a self-evolution feature deviation risk; The trajectory deviation trend score of the path is written into the corresponding state jump event path structure in the response graph model as a new index dimension, and is taken as an additional weighted factor in the path risk value calculation formula to enhance the ability to identify the evolutionary degradation trend of the system in advance.

Images