Generator set starting condition detection method and system based on multi-moment sequence

Abstract

The invention provides a generator set starting condition detection method and system based on a multi-moment sequence, and belongs to the technical field of power system monitoring and diagnosis. The method comprises the following steps: firstly, acquiring rotating speed and power time sequence data of a generator set, and constructing a candidate time node set of a static state and a high-speed running state; and for the high-speed operation state candidate nodes, screening static state nodes meeting the time difference requirement, determining a power rising node, and judging whether power abnormal reset exists or not. And judging conditions of full load reaching, load stabilization and load reduction in sequence according to the priority, and determining the starting completion time and type. And reselecting a power rising node, determining a starting time period and generating a starting event record. Repeated records are removed, and the starting event record with the shortest duration time in the records with the same completion node is reserved. And detecting whether the starting process is high in power for a long time, and if yes, updating the starting completion node. And finally, outputting a starting working condition record containing the multi-dimensional information.

Description

Technical Field

[0001] This invention belongs to the field of power system monitoring and diagnosis technology, and more specifically relates to a method and system for detecting generator set startup conditions based on multi-time series. Background Technology

[0002] As core equipment in the power system, the stability of generator sets during startup directly affects the safe operation of the power grid and the reliability of power supply. The startup phase involves several critical stages, including speed increase, voltage establishment, and grid synchronization. Abnormalities in any of these stages can lead to equipment failure or grid fluctuations. Therefore, accurate monitoring of generator set startup conditions and real-time identification of key nodes and status changes during startup have become core tasks of power system operation monitoring. This plays an irreplaceable role in timely detection of potential risks, optimization of startup control strategies, and extension of equipment lifespan. Currently, mainstream generator set startup condition detection methods in the industry have significant technical limitations. Threshold detection methods based on a single parameter rely solely on changes in the value of a single physical quantity such as power or speed to determine the status, ignoring the correlation and dynamic coupling characteristics between multiple parameters during startup. For example, judging startup completion solely by setting a speed threshold may lead to misjudgments due to grid load fluctuations or mechanical losses, failing to reflect the actual operating status of the unit. This method has extremely poor adaptability to complex operating conditions and struggles to meet the detection needs of different generator set models or under varying operating environments. Both time-window-based statistical analysis methods and expert-rule-based judgment methods have significant drawbacks. The former uses a sliding time window for parameter statistical analysis, but its fixed window length cannot match the dynamic change rates at different stages of startup, easily leading to missed detections or delayed identification of critical nodes. The latter relies excessively on the accumulated experience of operators, and its preset rules lack adaptability. When units age, are upgraded, or the operating environment changes, the rule base requires frequent manual adjustments, increasing maintenance costs and potentially leading to judgment errors due to experience bias. These problems are even more pronounced in new power systems with a high proportion of renewable energy integration, severely hindering the intelligent development of the power grid. It is evident that as the power system's requirements for operational flexibility and safety continue to increase, traditional methods for detecting startup conditions can no longer meet actual needs. Summary of the Invention

[0003] To address the above problems, the present invention aims to provide a generator set starting condition detection method and system based on multi-time series. Through collaborative analysis of speed and power time series data, it realizes automatic identification of the generator set starting process, intelligent judgment of the starting completion status, and accurate detection of the starting condition.

[0004] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, embodiments of this application provide a generator set startup condition detection method based on a multi-time sequence, including: Collect time-series data of generator speed and power, and perform outlier processing on the power time-series data; By traversing the rotational speed time series data, time nodes that meet the rotational speed threshold conditions are extracted, and a set of candidate time nodes for stationary state and a set of candidate time nodes for high-speed running state are constructed. For each time node in the candidate time node set for high-speed operation, select the stationary state time node that meets the time difference requirement, check the power data during the startup process to determine the power rise time node, and determine whether there is an abnormal power reset. The conditions for full load reaching the target load, the conditions for load stabilization, and the conditions for load decrease are judged in order of priority to determine the corresponding start-up completion time node and start-up type. A better power rise time point is selected, and the final start-up time period is determined, generating a start-up event record; Remove duplicate startup event records and filter and retain startup events with the same startup completion time. Detect whether there is a prolonged period of high power consumption during the startup process, and update the startup completion node accordingly; The output includes startup start time, end time, key time nodes, and startup type startup status records.

[0005] In an optional implementation, the acquisition of generator set speed time series data and power time series data, and the outlier processing of the power time series data, includes: The generator set's speed, power, and timestamp are obtained every ten minutes through the time-series data interface, generating time-series data of speed RPM_SEQ and power time-series data of POWER_SEQ. For the power time series data POWER_SEQ, data points with power values ​​greater than a preset upper limit threshold are marked as invalid values.

[0006] In an optional implementation, the step of extracting time nodes that meet the speed threshold condition by traversing the speed time series data and constructing a set of candidate time nodes for the stationary state and a set of candidate time nodes for the high-speed running state includes: Traverse the RPM_SEQ time series data. For any data point RPM_SEQ[i], if RPM_SEQ[i] ≤ RPM_LOW_THRESHOLD, then the corresponding time node is recorded as a candidate time node for the stationary state. After the traversal is completed, generate a set of candidate time nodes for the stationary state T0_CANDIDATES based on the candidate time nodes for the stationary state. Here, RPM_LOW_THRESHOLD is the low speed threshold, and i is the index position of the RPM_SEQ time series data. Traverse the RPM_SEQ time series data. For any data point RPM_SEQ[i], if RPM_SEQ[i] ≥ RPM_HIGH_THRESHOLD, then the corresponding time node is recorded as a candidate time node for high-speed operation. After traversal, generate a set of candidate time nodes for high-speed operation based on the candidate time nodes for high-speed operation. Among them, RPM_HIGH_THRESHOLD is the high speed threshold.

[0007] In an optional implementation, for each time node in the candidate time node set for high-speed operation, the step of filtering out the stationary state time nodes that meet the time difference requirements, checking the power data during the startup process to determine the power rise time node, and determining whether there is an abnormal power reset includes: Perform the following operations on each time node t1 in the set of candidate time nodes T1_CANDIDATES for high-speed operation status: In any time node t0 in T0_CANDIDATES, if t0 is before t1 and (t1 - t0) ≤ MAX_TIME_DIFF, then t0 is the relevant static state time node; where MAX_TIME_DIFF is the maximum time difference threshold. Check the power time series data POWER_SEQ within the time period [t0, t1]. If there is a data point POWER_SEQ[j] > POWER_ANOMALY_THRESHOLD, then there is a power anomaly in that time period, and the current t0 is invalid; where j is the index position of the power time series data POWER_SEQ, and POWER_ANOMALY_THRESHOLD is the abnormal power threshold. In the power time series data after the stationary state time node t0, the time node of the first data point that satisfies the conditions POWER_SEQ[k]>0 and POWER_SEQ[k-1]≤0 is taken as the power rise time node t2; where k is the index position of the power time series data POWER_SEQ; Check if there is a situation in the power time series data after the power rise time node t2 where the power first rises above the power reset threshold POWER_RESET_THRESHOLD and then falls below 0. If so, the current t2 is invalid.

[0008] In an optional implementation, the step of sequentially determining the full load condition, load stability condition, and load decrease condition according to priority, and determining the corresponding startup completion time node and startup type, includes: The condition for reaching full load is the first priority, the condition for load stability is the second priority, and the condition for load decrease is the third priority. The conditions for reaching full load include: Check if there are any data points in the power time series data after the power rise time node t2 that satisfy POWER_SEQ[m] ≥ FULL_LOAD, where FULL_LOAD is the full load power and m is the index position of the power time series data POWER_SEQ; If so, then the time point corresponding to that data point will be taken as the startup completion time node t3, and the startup type will be marked as CONDITION_A; If not, then perform a load stability condition check; The load stability condition judgment includes: Check if there are consecutive STABILITY_WINDOW data points in the power time series data after the power rise time node t2 where POWER_SEQ[n] ≥ LOAD_THRESHOLD; where LOAD_THRESHOLD = FULL_LOAD × LOAD_THRESHOLD_RATIO, LOAD_THRESHOLD_RATIO is set to 0.95 by default, and STABILITY_WINDOW is set to 30 by default; If so, the time point corresponding to the first data point reaching LOAD_THRESHOLD will be taken as the startup completion time node t3, and the startup type will be marked as CONDITION_B; If not, then execute the load reduction condition judgment; The conditions for determining load reduction include: Check if there are any cases in the power time series data after the power rise time node t2 where the power first reaches LOAD_THRESHOLD and then falls below that threshold; If so, the time point corresponding to the first data point that drops below LOAD_THRESHOLD will be taken as the startup completion time node t3, and the startup type will be marked as CONDITION_C.

[0009] In an optional implementation, the step involves reselecting a better power rise time node, determining the final startup time period, and generating a startup event record. After determining t3, based on the power time series data, select the power rise data point closest to t3 within the time period [t0, t3], and take the time point corresponding to this data point as the power rise time node t2; Generate a startup event record based on the current t0, t1, t2, and t3, and determine the startup time period [START_TIME, END_TIME], where START_TIME = t0 and END_TIME = t3.

[0010] In an optional implementation, the step of removing duplicate startup event records and filtering and retaining startup events with the same startup completion time node includes: Remove duplicate startup event records; For multiple startup event records with the same t3, retain the startup event record with the shortest duration; where duration = END_TIME - START_TIME.

[0011] In an optional implementation, detecting whether the startup process involves prolonged high-power conditions and updating the startup completion node accordingly includes: In the startup event log, check if there is a time period in the startup time period [START_TIME, END_TIME] that lasts for more than 24 hours and has a power greater than the high power threshold HIGH_POWER_THRESHOLD; If so, update t3 to the time point corresponding to the maximum power within 24 hours after the start of this time period, and update END_TIME synchronously.

[0012] In an optional implementation, the output includes a startup status record containing startup start time, end time, key time nodes, and startup type, including: Generate a startup status record containing the following fields: Startup start time START_TIME; Startup end time END_TIME; Key time points t0, t1, t2, t3; Startup type CONDITION_TYPE; Among them, CONDITION_TYPE is CONDITION_A, CONDITION_B, or CONDITION_C.

[0013] Secondly, embodiments of this application also provide a generator set startup condition detection system based on a multi-time sequence, including: The data acquisition and preprocessing module is used to collect the generator set's speed time series data and power time series data, and to process outliers in the power time series data. The key time node construction module is used to extract time nodes that meet the speed threshold conditions by traversing the speed time series data, and to construct a set of candidate time nodes for the stationary state and a set of candidate time nodes for the high-speed running state. The startup process detection module is used to filter out the stationary state time nodes that meet the time difference requirements for each time node in the candidate time node set of high-speed operation state, check the power data during the startup process to determine the power rise time node, and determine whether there is an abnormal power reset. The startup completion condition judgment module is used to judge the full load reaching condition, load stabilization condition, and load decrease condition in order of priority, and determine the corresponding startup completion time node and startup type. The timing optimization module is used to reselect a better power rise time point, determine the final start-up time period, and generate a start-up event record. The result deduplication and filtering module is used to remove duplicate startup event records and filter and retain startup events with the same startup completion time node. The special case handling module is used to detect whether there is a prolonged high power situation during the startup process and update the startup completion node accordingly; The startup condition data output module is used to output startup condition records that include startup start time, end time, key time nodes, and startup type.

[0014] As can be seen from the above technical solutions, the present invention has the following advantages: The generator set startup condition detection method based on multi-time series provided in this application establishes a multi-stage time node identification mechanism by fusing dual-parameter time series data of speed and power, so as to accurately locate the startup start time, speed establishment time, power establishment time, and startup completion time; it constructs an intelligent discrimination method for three startup completion states to accurately identify different operating conditions such as full-load startup completion, partial-load stable operation, and power drop after startup; and it adopts adaptive threshold setting and multi-condition logic judgment to significantly improve detection accuracy and reliability.

[0015] This application ensures the validity of the raw data through targeted data preprocessing mechanisms. Specifically, a preset upper limit threshold is set to mark invalid values ​​for the power time series data. Simultaneously, during the startup process detection, an abnormal power threshold is used to screen for power anomaly time periods, eliminating quiescent time nodes with power anomalies. Furthermore, a power reset threshold is used to determine power reset anomalies and exclude invalid power increase nodes. These operations effectively filter noise, false alarms, or abnormal fluctuations during data acquisition, providing high-quality basic data support for subsequent operating condition detection.

[0016] This application achieves precise differentiation between stationary and high-speed operating states through explicit threshold rules. The method constructs a candidate time node set by traversing the speed time series data based on low and high speed thresholds, ensuring objective identification of stationary and high-speed operating states. This state division method based on quantified thresholds avoids subjective judgment errors and provides clear state boundaries for subsequent time node correlation and analysis during the startup process.

[0017] This application uses multi-dimensional inspection and screening to accurately locate key time nodes in the startup process. When associating nodes in the static and high-speed states, it filters valid time pairs using the maximum time difference threshold; it identifies power increase nodes through power sequence characteristics; and it determines startup completion nodes by combining priority conditions such as full load, stable load, and load decrease. This comprehensive capture of key nodes enables a refined breakdown of the startup process, clearly reflecting the complete timing characteristics of the unit from standstill to stable operation.

[0018] This application improves the accuracy and uniqueness of the detection results through multiple logical optimizations. On the one hand, it eliminates interference cases by removing periods of abnormal power and invalid power increase nodes; on the other hand, it removes duplicate events by using the rule of "retaining the shortest record among those with the same completion time node," thus avoiding redundancy; simultaneously, it dynamically updates the startup completion node for prolonged high-power situations, further adapting to complex scenarios in actual operation. These optimization measures ensure that the output startup event records have high reliability and representativeness.

[0019] The startup log generated by this application contains multi-dimensional information such as startup start time, end time, key nodes, and startup type. This information not only fully presents the temporal characteristics of the startup process, but also distinguishes different modes of stable operation of the unit through startup type (such as full load achievement, stable load achievement, load reduction termination, etc.), providing detailed data support for generator set operation and maintenance analysis, performance evaluation, and fault diagnosis, and has strong practical application value. Attached Figure Description

[0020] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating the generator set startup condition detection method based on multi-time series provided in this application.

[0022] Figure 2 The first startup condition recognition diagram provided for this application.

[0023] Figure 3 The second startup condition recognition diagram provided for this application.

[0024] Figure 4 A schematic diagram of the generator set start-up condition detection system based on multi-time series provided in this application. Detailed Implementation

[0025] The various embodiments of this disclosure will be described more fully in the following detailed description of the specific steps of the generator set start-up condition detection method based on multi-time series. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0026] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions, operations, or elements, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a particular feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see Figure 1 The diagram shows a flowchart of a generator set startup condition detection method based on a multi-time series in a specific embodiment. The method includes: S1: Collect the generator set's speed time series data and power time series data, and perform outlier processing on the power time series data.

[0029] In a specific implementation, firstly, the generator set's operating data is acquired every 10 minutes via a time-series data interface, specifically including speed, power, and corresponding timestamps. The acquired speed data is arranged sequentially to form a speed time-series data RPM_SEQ; similarly, the power data is arranged sequentially to form a power time-series data POWER_SEQ, and the corresponding timestamps are combined to form a timestamp sequence TIME_SEQ. Each data point in these three sequences corresponds one-to-one, recording the generator set's operating status at different time points. For example, in a practical application scenario, the generator set's control system reads speed, power, and timestamp information from the data acquisition module at set 10-minute intervals, generating RPM_SEQ=[r1,r2,r3,…,rn], POWER_SEQ=[p1,p2,p3,…,pn], and TIME_SEQ=[t1,t2,t3,…,tn], where ri, pi, and ti represent the speed, power, and timestamp at the i-th time point, respectively.

[0030] Then, for the generated power time series data POWER_SEQ, a power upper limit threshold POWER_UPPER_LIMIT is set. Each power data point in POWER_SEQ is iterated over; if the power value of any data point is found to be greater than POWER_UPPER_LIMIT, it is marked as invalid. For example, when analyzing POWER_SEQ, if the k-th data point pk is found to be greater than POWER_UPPER_LIMIT, then pk is marked as invalid. Subsequent data processing will filter out these invalid power data points to prevent outliers from interfering with subsequent startup process detection.

[0031] S2: Extract time nodes that meet the speed threshold conditions by traversing the speed time series data, and construct a set of candidate time nodes for stationary state and a set of candidate time nodes for high-speed running state.

[0032] In a specific implementation, on the one hand, the rotational speed time series data RPM_SEQ is traversed. For each rotational speed data point RPM_SEQ[i], it is compared with a preset low rotational speed threshold RPM_LOW_THRESHOLD. If RPM_SEQ[i] ≤ RPM_LOW_THRESHOLD, the generator set is considered to be in a stationary or low-speed state at that time point, and its corresponding timestamp TIME_SEQ[i] is extracted as a candidate time node for a stationary state. After the traversal is completed, all the extracted candidate time nodes for a stationary state are combined into a set T0_CANDIDATES. For example, when analyzing RPM_SEQ, if RPM_SEQ[5] ≤ RPM_LOW_THRESHOLD is found, then TIME_SEQ[5] is added to T0_CANDIDATES. The time nodes in this set indicate that the generator set may be in a stationary state before startup.

[0033] On the other hand, the process also iterates through RPM_SEQ, comparing each RPM_SEQ[i] with the preset high-speed threshold RPM_HIGH_THRESHOLD. If RPM_SEQ[i] ≥ RPM_HIGH_THRESHOLD, the generator set is considered to be in high-speed operation, and the corresponding timestamp TIME_SEQ[i] is extracted as a candidate time node for high-speed operation. After the iteration is complete, these candidate time nodes for high-speed operation are grouped into a set T1_CANDIDATES. For example, when RPM_SEQ

[20] ≥ RPM_HIGH_THRESHOLD, TIME_SEQ

[20] is added to T1_CANDIDATES. The time nodes in the set indicate that the generator set may have completed startup and entered the high-speed operation stage.

[0034] S3: For each time node in the candidate time node set of high-speed operation state, filter out the stationary state time nodes that meet the time difference requirements, check the power data during the startup process to determine the power rise time node, and determine whether there is a power abnormal reset situation.

[0035] In a specific implementation, for each time node t1 in T1_CANDIDATES, the following sub-steps are executed sequentially: S3.1 Valid t0 node selection: Look for a time node t0 in T0_CANDIDATES that satisfies two conditions: First, t0 must be before t1, i.e., t0 < t1; Second, the time difference between t1 and t0 (t1 - t0) cannot exceed the preset maximum time difference threshold MAX_TIME_DIFF. Filter out all t0 nodes that meet the conditions, and these nodes may be the starting points of the generator set startup process. For example, for t1 = TIME_SEQ

[100] in T1_CANDIDATES, look for t0 in T0_CANDIDATES such that t0 < TIME_SEQ

[100] and (TIME_SEQ

[100] - t0) ≤ MAX_TIME_DIFF. The t0 nodes that meet the conditions will be regarded as possible startup start times.

[0036] S3.2 Abnormal power check: For each t0 node that has been filtered out, check the power time series data POWER_SEQ from t0 to t1. Specifically, check whether there is any case where each power data point POWER_SEQ[j] in this time period is greater than the abnormal power threshold POWER_ANOMALY_THRESHOLD. If any data point satisfies POWER_SEQ[j] > POWER_ANOMALY_THRESHOLD, it is considered that there is a power anomaly in the startup process corresponding to this t0 node, skip the current t0 node, and no longer use it as a valid startup start node for subsequent processing. For example, in the time period [t0, t1], if it is found that POWER_SEQ

[15] > POWER_ANOMALY_THRESHOLD, discard the current t0 node and continue to check the next t0 node.

[0037] S3.3 Determination of the power rising time node t2: After determining the valid t0 nodes, start looking for the starting point of the power rise in the power time series data after t0. The specific method is to find the first time node that satisfies POWER_SEQ[k] > 0 and POWER_SEQ[k - 1] ≤ 0, and use it as the power rising time node t2. Here, k is searched sequentially backward from the index position corresponding to t0. For example, assume that the index position corresponding to t0 is i, then start checking POWER_SEQ[i + 1], POWER_SEQ[i + 2], etc. in sequence until the first k that meets the conditions is found. At this time, TIME_SEQ[k] is t2. This t2 represents the moment when the generator set power starts to rise and may be a key turning point in the startup process.

[0038] S3.4 Power reset detection: After determining node t2, check the power time series data after t2. Determine if there is a situation where the power first rises above the power reset threshold POWER_RESET_THRESHOLD and then drops below 0. If this situation exists, it is considered that the startup process corresponding to the current t2 node is incomplete or abnormal, and this t2 node is marked as invalid and no further processing is performed. For example, if in the power data after t2, it is found that POWER_SEQ

[30] ≥POWER_RESET_THRESHOLD, while the subsequent POWER_SEQ

[35] ≤0, it is considered that a power reset situation has occurred, and the current t2 node is invalid.

[0039] S4: Perform the full load condition judgment, load stability condition judgment, and load decrease condition judgment in order of priority to determine the corresponding start-up completion time node and start-up type.

[0040] In a specific implementation, the following judgments are performed sequentially in descending order of priority: 1. Conditions for determining full load: Starting from power rise time point t2, check the subsequent power time series data POWER_SEQ to see if there is a data point that satisfies POWER_SEQ[m]≥FULL_LOAD (full load power). If such a data point exists, the corresponding timestamp TIME_SEQ[m] is set as the start-up completion time point t3, and the start-up type is marked as "CONDITION_A". For example, when checking the power data after t2, if POWER_SEQ

[100] ≥FULL_LOAD is found, then t3=TIME_SEQ

[100] , and the start-up type is "CONDITION_A". This indicates that the generator set has successfully reached full load operation and completed the start-up process.

[0041] 2. Determining load stability conditions: If the full load condition is not met, the power data after t2 is checked. LOAD_THRESHOLD = FULL_LOAD × LOAD_THRESHOLD_RATIO is calculated; by default, LOAD_THRESHOLD_RATIO is 0.95. Then, it checks if there are consecutive STABILITY_WINDOW (default 30) data points where the power value of each data point is greater than or equal to LOAD_THRESHOLD. If such consecutive data points exist, the time point when LOAD_THRESHOLD is first reached is taken as the startup completion time node t3, and the startup type is marked as "CONDITION_B". For example, in the power data after t2, if the power value of 30 consecutive data points starting from a certain starting point is ≥ LOAD_THRESHOLD, then t3 is the timestamp corresponding to the first data point that reaches LOAD_THRESHOLD, and the startup type is "CONDITION_B". This indicates that although the generator set has not reached full load, it has stabilized at a load level close to full load, and the startup process can be considered complete.

[0042] 3. Determining load reduction conditions: If neither of the first two conditions is met, the power data after t2 is further examined. It is determined whether there is a situation where the power first reaches LOAD_THRESHOLD and then drops below that threshold. If this situation exists, the time point when the power first drops below LOAD_THRESHOLD is taken as the start-up completion time node t3, and the start-up type is marked as "CONDITION_C". For example, if the power data shows that the power reaches LOAD_THRESHOLD from a certain time point, but subsequently drops below LOAD_THRESHOLD, then t3 is the time point of the first drop, and the start-up type is "CONDITION_C". This situation may indicate that the generator set experienced load fluctuations or a fault during startup, resulting in its inability to stably maintain the expected load level.

[0043] S5: Select a better power rise time node, determine the final start-up time period, and generate a start-up event record.

[0044] In a specific implementation, after determining node t3, the power time series data from t0 to t3 is re-examined. The power rise point closest to t3 is searched, i.e., the time node satisfying POWER_SEQ[k]>0 and POWER_SEQ[k-1]≤0 is searched again, but this time the node closest to t3 is selected as the final determined power rise time node t2. For example, within the time period [t0, t3], there may be multiple power rise points; in this case, the one closest to t3 is selected to more accurately reflect the power change characteristics during startup.

[0045] Then, the final determined node t0 is used as the startup start time START_TIME, and node t3 is used as the startup end time END_TIME. This defines a complete startup time period [START_TIME, END_TIME], covering the entire process of the generator set from a standstill to the final startup completion. Simultaneously, startup event records are generated based on the current t0, t1, t2, and t3.

[0046] S6: Remove duplicate startup event records and filter and retain startup events with the same startup completion time node.

[0047] In a specific implementation, when generating a startup event record, multiple records may have the same START_TIME and END_TIME combination. For example, due to the frequency of data acquisition or the accuracy of the algorithm detection, multiple seemingly different startup events may be detected, but in reality, their start and end times are the same. In this case, by comparing the START_TIME and END_TIME of different startup event records, duplicate records are removed, and only one unique record is retained.

[0048] For multiple startup event records with the same t3 node (i.e., the same start-end time), calculate the duration (END_TIME - START_TIME) of each record. Select the record with the shortest duration and discard the others. For example, if there are three startup event records with the same t3 but durations of 1 hour, 1.5 hours, and 2 hours respectively, retain the record with the shortest duration of 1 hour. This filters out event records that are more likely to represent the actual startup process.

[0049] S7: Detect whether there is a prolonged period of high power during the startup process and update the startup completion node accordingly.

[0050] In a specific implementation, within the defined startup time period [START_TIME, END_TIME], it is checked whether there exists a continuous time period exceeding 24 hours in duration, during which the power values ​​are all greater than the high power threshold HIGH_POWER_THRESHOLD. For example, by iterating through the power data within the startup time period, it is checked whether there are any instances where power data points exceeding HIGH_POWER_THRESHOLD exist for more than 24 consecutive hours. This situation may indicate that the generator set is operating at high load for an extended period after startup, potentially impacting equipment lifespan or operational safety.

[0051] If a prolonged high-power period is detected, node t3 is updated to the time point within 24 hours after the start of that period when the power reaches its maximum value, and END_TIME is updated synchronously. For example, assuming the prolonged high-power period starts from TIME_SEQ

[200] , and the index position corresponding to the maximum power value is found to be 250 between TIME_SEQ

[200] and TIME_SEQ

[200] + 24 hours, then t3 is updated to TIME_SEQ

[250] , and END_TIME is also updated accordingly to TIME_SEQ

[250] . This can more accurately reflect the actual operating status of the generator set after startup.

[0052] S8: Output includes startup start time, end time, key time nodes, and startup type startup status records.

[0053] In a specific implementation, a startup status record is generated, which includes the following fields: Start-up time: START_TIME indicates the time point at which the generator set startup process begins.

[0054] Start-up end time: also known as END_TIME, indicates the point in time when the generator set startup process ends.

[0055] Key time nodes: including t0 (candidate node for stationary state), t1 (candidate node for high-speed operation state), t2 (power increase node), and t3 (startup completion node), these nodes record the key moments in the startup process.

[0056] Startup type: CONDITION_TYPE, which is the type determined by the steps to determine the startup completion conditions. It can be "CONDITION_A", "CONDITION_B" or "CONDITION_C", reflecting the characteristics and results of the startup process.

[0057] In this embodiment, by collecting, preprocessing, and analyzing the generator set's speed and power timing data, the generator set's startup process can be accurately detected, and its startup completion time and type can be determined. Through screening key candidate time points, checking abnormal power, determining power rise points, and prioritizing startup completion conditions, the startup status of the generator set under different operating conditions can be comprehensively and meticulously identified. Furthermore, through steps such as time point optimization, result deduplication and optimization, and post-processing for special cases, the accuracy and reliability of the detection results are further improved, providing strong data support for generator set operation management, fault analysis, and performance evaluation. This helps to promptly detect abnormalities during the generator set startup process and ensure the safe and stable operation of the generator set.

[0058] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to fully illustrate the specific implementation process of this embodiment, another generator set start-up condition detection method based on multi-time sequence is provided, including the following steps: Step 1: Time series data acquisition and preprocessing steps.

[0059] 1.1 Data Acquisition: The system acquires unit operation data within the time range of 00:00:00 on January 1, 2021 to 00:00:00 on December 31, 2021. The RPM_SEQ time series data contains 52,560 data points, the POWER_SEQ power time series data contains 52,560 data points, and the TIME_SEQ timestamp sequence corresponds one-to-one with the RPM and power data. All data are acquired at 10-minute intervals.

[0060] 1.2 Outlier Handling: The power upper limit threshold POWER_UPPER_LIMIT was set to 1000MW, and outlier checks were performed on the power time series data POWER_SEQ. The checks revealed that none of the power data points exceeded the 1000MW upper limit threshold; therefore, all data points were marked as valid values, and no outlier removal was necessary.

[0061] Step 2: Construction of the set of key candidate time nodes.

[0062] 2.1 Construction of the candidate set of static states: The static speed threshold RPM_LOW_THRESHOLD was set to 5 rpm. The speed time series data RPM_SEQ was traversed to extract all time nodes that met the condition RPM_SEQ[i]≤5 rpm. By traversing 52560 speed data points, a candidate time node set T0_CANDIDATES for static state was successfully constructed. This candidate set contains all time points when the unit is in a static or extremely low speed state.

[0063] 2.2 Construction of the candidate set for high-speed operation: The high-speed operating speed threshold RPM_HIGH_THRESHOLD was set to 1500 rpm. The speed time series data RPM_SEQ was traversed, and all time nodes that met the condition RPM_SEQ[i]≥1500 rpm were extracted. By completely traversing the speed data, a candidate time node set T1_CANDIDATES for high-speed operation was successfully constructed. This candidate set covers all time points when the unit is in a high-speed stable operating state.

[0064] Step 3: Startup process detects the main loop steps.

[0065] The first testing process (started in February 2021): 3.1 Filtering of valid t0 nodes: Select time node t1=2021-02-1503:40:00 from T1_CANDIDATES as the starting point for high-speed operation, and set the maximum time difference threshold MAX_TIME_DIFF to 24 hours. Search for a valid t0 node in the candidate set of static states before t1, and find t0=2021-02-1502:40:00. Verify the time difference condition: (03:40:00-02:40:00)=1 hour ≤ 24 hours, which meets the filtering condition.

[0066] 3.2 Abnormal Power Check: An abnormal power threshold, POWER_ANOMALY_THRESHOLD, was set to 1000MW. Anomaly checks were performed on power time-series data within the time period [2021-02-15 02:40:00, 2021-02-15 03:40:00]. Point-by-point checks revealed that all power values ​​within this time period were less than or equal to 1000MW, and no abnormal power data points were found. This confirmed the anomaly detection.

[0067] 3.3 Determination of power rise time node t2: In the power time-series data after time node t0 = 2021-02-15 02:40:00, the first time node that satisfies the conditions POWER_SEQ[k] > 0 and POWER_SEQ[k-1] ≤ 0 is searched in chronological order. After searching, the power rise time node t2 = 2021-02-15 05:50:00 is determined. This time point marks the critical moment when the unit power starts to rise from zero.

[0068] 3.4 Power Reset Detection: The power reset threshold POWER_RESET_THRESHOLD was set to 50MW. The power data after t2 (2021-02-15 05:50:00) was checked for a power reset phenomenon where the power first rose above 50MW and then fell below 0. The check revealed that the power data after t2 showed a continuous upward trend, and no power reset occurred.

[0069] The second testing process (started in November 2021): Repeat steps 3.1-3.4.

[0070] Following the same detection procedure, t1=2021-11-2120:00:00 was selected from T1_CANDIDATES. A valid t0=2021-11-2119:10:00 was found before it, with a time difference of 50 minutes, satisfying the 24-hour constraint. The abnormal power check passed; there was no abnormal power exceeding 1000MW within this time period. The power rise time node was determined to be t2=2021-11-2122:00:00, and no power reset phenomenon was found after t2.

[0071] Step 4: Start the completion condition judgment step.

[0072] First startup condition check: 4.1 Full load condition judgment (priority 1): The full-load power threshold FULL_LOAD was set to 718MW. The power time-series data after t2 (2021-02-15 05:50:00) was checked to see if there were any points in the time data that reached or exceeded 718MW. It was found that the power first reached 718MW at 10:50:00 on 2021-02-16, meeting the full-load condition. Therefore, t3 was set to 2021-02-16 10:50:00, and the start-up type was marked as "CONDITION_A" (full load reached).

[0073] Second startup condition judgment: 4.2. Criteria for reaching full load: Similarly, examining the power data after t2 = 2021-11-21 22:00:00, it was found that the power reached a full load level of 718MW at 18:30:00 on 2021-11-22. Therefore, t3 was set to 2021-11-22 18:30:00, and the start-up type was marked as "CONDITION_A" (reaching full load). Since both starts met the highest priority full load condition, there was no need to perform load stabilization and load reduction condition checks.

[0074] Step 5: Optimize and adjust the time nodes.

[0075] First-time startup optimization: 5.1 Reselecting node t2: After determining t3 = 2021-02-16 10:50:00, the power increase point closest to t3 was selected again within the time period [2021-02-15 02:40:00, 2021-02-16 10:50:00] as the final t2. After re-screening and confirmation, the original t2 = 2021-02-15 05:50:00 remains the optimal choice and is kept unchanged.

[0076] 5.2 Determining the start-up time period: Based on the optimized time nodes, set the startup start time START_TIME = t0 = 2021-02-15 02:40:00 and the startup end time END_TIME = t3 = 2021-02-16 10:50:00 to determine the complete startup time period.

[0077] Second startup optimization: The same time node optimization process was executed, and it was confirmed that t2 = 2021-11-21 22:00:00 was the optimal power increase point. START_TIME = 2021-11-21 19:10:00 and END_TIME = 2021-11-22 18:30:00 were set.

[0078] Step 6: Result deduplication and optimization steps.

[0079] 6.1 Basic Deduplication: The detected startup events are deduplicated based on the (START_TIME, END_TIME) time combination. The first startup time combination is (2021-02-15 02:40:00, 2021-02-16 10:50:00), and the second startup time combination is (2021-11-21 19:10:00, 2021-11-22 18:30:00). The two time combinations are completely different, and there are no duplicate records, so they need to be removed.

[0080] 6.2 Intelligent Filtering: Check for multiple startup events with the same t3 time point to select the record with the shortest duration for retention. The check revealed that the t3 time points for the two startup events were 2021-02-16 10:50:00 and 2021-11-22 18:30:00, respectively. Since the time points are different, no intelligent filtering is required.

[0081] Step 7: Post-processing steps for special cases.

[0082] 7.1 Long-term high-power testing: Set the high-power threshold HIGH_POWER_THRESHOLD to 600MW and check for any abnormal situations during startup that lasted longer than 24 hours and had a power output greater than 600MW. The first startup lasted 32 hours and 10 minutes, exceeding 24 hours, but the power output gradually increased from 0MW to 718MW, which is a normal process and does not constitute an abnormally long-term high-power situation. The second startup lasted 23 hours and 20 minutes, less than 24 hours, and required no special handling.

[0083] 7.2, t3 node update: Since no abnormal prolonged high-power conditions were detected during either startup, node t3 does not need to be updated. The t3 time for the first startup remains 2021-02-16 10:50:00, and the t3 time for the second startup remains 2021-11-22 18:30:00; the corresponding END_TIME also remains unchanged.

[0084] Step 8: Start the process of outputting operating condition data.

[0085] Based on the detection and optimization results, a startup status record containing complete field information is generated. The first startup status record includes: startup start time 2021-02-15 02:40:00, startup end time 2021-02-16 10:50:00, key time nodes t0, t1, t2, and t3 are 02:40:00, 03:40:00, 05:50:00, and 10:50:00 respectively, and the startup type is CONDITION_A (reaching full load). The second startup log includes: startup start time 2021-11-21 19:10:00, startup end time 2021-11-22 18:30:00, key time nodes t0, t1, t2, and t3 are 19:10:00, 20:00:00, 22:00:00, and 18:30:00 respectively, and the startup type is CONDITION_A (reaching full load).

[0086] like Figure 2 , Figure 3 As shown, the key characteristics of the two start-up processes are as follows, verified by the dual-axis timing diagram: the speed change rapidly increases from a stationary state (≤5rpm) to a high-speed operating state (≥1500rpm); the power build-up starts from 0MW and eventually reaches full load of 718MW; the four key time nodes t0, t1, t2, and t3 accurately identify the different stages of the start-up process; the start-up type meets the full load achievement condition and is correctly identified as CONDITION_A type.

[0087] As can be seen, this embodiment accurately identified the two unit start-up conditions throughout 2021, accurately located key time nodes, correctly judged the start-up type, and executed the algorithm stably, providing reliable data support for power plant operation analysis and equipment maintenance.

[0088] like Figure 4 As shown, the following are embodiments of the generator set starting condition detection system based on multiple time sequences provided in this disclosure. This system and the generator set starting condition detection method based on multiple time sequences in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the generator set starting condition detection system based on multiple time sequences, please refer to the embodiments of the generator set starting condition detection method based on multiple time sequences.

[0089] A generator set start-up condition detection system based on multi-time series includes: The data acquisition and preprocessing module is used to collect the generator set's speed time series data and power time series data, and to process outliers in the power time series data.

[0090] The key time node construction module is used to extract time nodes that meet the speed threshold conditions by traversing the speed time series data, and to construct a set of candidate time nodes for stationary state and a set of candidate time nodes for high-speed running state.

[0091] The startup process detection module is used to filter out the stationary state time nodes that meet the time difference requirements for each time node in the candidate time node set of high-speed operation state, check the power data during the startup process to determine the power rise time node, and determine whether there is an abnormal power reset.

[0092] The startup completion condition judgment module is used to judge the full load reaching condition, load stabilization condition, and load decrease condition in sequence according to priority, and determine the corresponding startup completion time node and startup type.

[0093] The timing optimization module is used to reselect a better power rise time point, determine the final start-up time period, and generate a start-up event record.

[0094] The result deduplication and filtering module is used to remove duplicate startup event records and filter and retain startup events with the same startup completion time node.

[0095] The special case handling module is used to detect whether there is a prolonged high power condition during the startup process and update the startup completion node accordingly.

[0096] The startup condition data output module is used to output startup condition records that include startup start time, end time, key time nodes, and startup type.

[0097] The generator set startup condition detection system provided in this embodiment systematically processes the generator set speed and power time-series data. First, it filters outliers to ensure data reliability. Then, it accurately classifies the stationary and high-speed operating states based on quantification thresholds. Combining time difference constraints and power characteristics, it captures key nodes such as power increase and startup completion. The startup type is determined by priority condition judgment. With the help of optimization logic such as duplicate event elimination and dynamic adjustment of long-term high-power scenarios, the final output is a comprehensive record containing complete time-series nodes and startup modes. This not only achieves refined detection of startup conditions but also provides high-quality data support for unit operation and maintenance decisions, effectively improving the accuracy, comprehensiveness, and practical application value of the detection.

[0098] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting the starting operating condition of a generator set based on a multi-time series, characterized in that, include: Collect time-series data of generator speed and power, and perform outlier processing on the power time-series data; By traversing the rotational speed time series data, time nodes that meet the rotational speed threshold conditions are extracted, and a set of candidate time nodes for stationary state and a set of candidate time nodes for high-speed running state are constructed. For each time node in the candidate time node set for high-speed operation, select the stationary state time node that meets the time difference requirement, check the power data during the startup process to determine the power rise time node, and determine whether there is an abnormal power reset. The conditions for full load reaching the target load, the conditions for load stabilization, and the conditions for load decrease are judged in order of priority to determine the corresponding start-up completion time node and start-up type. A better power rise time point is selected, and the final start-up time period is determined, generating a start-up event record; Remove duplicate startup event records and filter and retain startup events with the same startup completion time. Detect whether there is a prolonged period of high power consumption during the startup process, and update the startup completion node accordingly; The output includes startup start time, end time, key time nodes, and startup type startup status records.

2. The generator set start-up condition detection method based on multi-time series according to claim 1, characterized in that, The process involves collecting time-series data on the generator set's rotational speed and power, and processing outliers in the power time-series data, including: The generator set's speed, power, and timestamp are obtained every ten minutes through the time-series data interface, generating time-series data of speed RPM_SEQ and power time-series data of POWER_SEQ. For the power time series data POWER_SEQ, data points with power values ​​greater than a preset upper limit threshold are marked as invalid values.

3. The generator set start-up condition detection method based on multi-time series according to claim 2, characterized in that, The step of extracting time nodes that meet the speed threshold conditions by traversing the speed time series data, and constructing a set of candidate time nodes for the stationary state and a set of candidate time nodes for the high-speed running state, includes: Traverse the RPM_SEQ time series data. For any data point RPM_SEQ[i], if RPM_SEQ[i] ≤ RPM_LOW_THRESHOLD, then the corresponding time node is recorded as a candidate time node for the stationary state. After the traversal is completed, generate a set of candidate time nodes for the stationary state T0_CANDIDATES based on the candidate time nodes for the stationary state. Here, RPM_LOW_THRESHOLD is the low speed threshold, and i is the index position of the RPM_SEQ time series data. Traverse the RPM_SEQ time series data. For any data point RPM_SEQ[i], if RPM_SEQ[i] ≥ RPM_HIGH_THRESHOLD, then the corresponding time node is recorded as a candidate time node for high-speed operation. After traversal, generate a set of candidate time nodes for high-speed operation based on the candidate time nodes for high-speed operation. Among them, RPM_HIGH_THRESHOLD is the high speed threshold.

4. The generator set start-up condition detection method based on multi-time series according to claim 3, characterized in that, For each time node in the candidate time node set for high-speed operation, the stationary state time node that meets the time difference requirement is selected, the power data during the startup process is checked to determine the power rise time node, and it is determined whether there is an abnormal power reset, including: Perform the following operations on each time node t1 in the set of candidate time nodes T1_CANDIDATES for high-speed operation status: In any time node t0 of T0_CANDIDATES, if t0 is before t1 and (t1 - t0) ≤ MAX_TIME_DIFF, then t0 is the relevant static state time node; where MAX_TIME_DIFF is the maximum time difference threshold. Check the power time series data POWER_SEQ within the time period [t0, t1]. If there is a data point POWER_SEQ[j] > POWER_ANOMALY_THRESHOLD, then there is a power anomaly in that time period, and the current t0 is invalid; where j is the index position of the power time series data POWER_SEQ, and POWER_ANOMALY_THRESHOLD is the abnormal power threshold. In the power time series data after the stationary state time node t0, the time node of the first data point that satisfies the conditions POWER_SEQ[k] > 0 and POWER_SEQ[k-1] ≤ 0 is taken as the power rise time node t2; where k is the index position of the power time series data POWER_SEQ; Check if there is a situation in the power time series data after the power rise time node t2 where the power first rises above the power reset threshold POWER_RESET_THRESHOLD and then falls below 0. If so, the current t2 is invalid.

5. The generator set start-up condition detection method based on multi-time series according to claim 4, characterized in that, The process involves sequentially determining the conditions for reaching full load, stabilizing load, and decreasing load, according to priority, to identify the corresponding startup completion time and startup type, including: The condition for reaching full load is the first priority, the condition for load stability is the second priority, and the condition for load decrease is the third priority. The conditions for reaching full load include: Check if there are any data points in the power time series data after the power rise time node t2 that satisfy POWER_SEQ[m] ≥ FULL_LOAD, where FULL_LOAD is the full load power and m is the index position of the power time series data POWER_SEQ; If so, then the time point corresponding to that data point will be taken as the startup completion time node t3, and the startup type will be marked as CONDITION_A; If not, then perform a load stability condition check; The load stability condition judgment includes: Check if there are consecutive STABILITY_WINDOW data points in the power time series data after the power rise time node t2 where POWER_SEQ[n] ≥ LOAD_THRESHOLD; where LOAD_THRESHOLD = FULL_LOAD × LOAD_THRESHOLD_RATIO, LOAD_THRESHOLD_RATIO is set to 0.95 by default, and STABILITY_WINDOW is set to 30 by default; If so, the time point corresponding to the first data point reaching LOAD_THRESHOLD will be taken as the startup completion time node t3, and the startup type will be marked as CONDITION_B; If not, then execute the load reduction condition judgment; The conditions for determining load reduction include: Check if there are any cases in the power time series data after the power rise time node t2 where the power first reaches LOAD_THRESHOLD and then falls below that threshold; If so, the time point corresponding to the first data point that drops below LOAD_THRESHOLD will be taken as the startup completion time node t3, and the startup type will be marked as CONDITION_C.

6. The generator set start-up condition detection method based on multi-time series according to claim 5, characterized in that, The process involves reselecting a better power rise time node, determining the final startup time period, and generating a startup event record. After determining t3, based on the power time series data, select the power rise data point closest to t3 within the time period [t0, t3], and take the time point corresponding to this data point as the power rise time node t2; Generate a startup event record based on the current t0, t1, t2, and t3, and determine the startup time period [START_TIME, END_TIME], where START_TIME = t0 and END_TIME = t3.

7. The generator set start-up condition detection method based on multi-time series according to claim 6, characterized in that, The process of removing duplicate startup event records and filtering and retaining startup events with the same startup completion time node includes: Remove duplicate startup event records; For multiple startup event records with the same t3, retain the startup event record with the shortest duration; where duration = END_TIME - START_TIME.

8. The generator set start-up condition detection method based on multi-time series according to claim 7, characterized in that, The detection process for whether there is a prolonged period of high power consumption during startup, and the corresponding update of the startup completion node, includes: In the startup event log, check if there is a time period in the startup time period [START_TIME, END_TIME] that lasts for more than 24 hours and has a power greater than the high power threshold HIGH_POWER_THRESHOLD; If so, update t3 to the time point corresponding to the maximum power within 24 hours after the start of this time period, and update END_TIME synchronously.

9. The generator set start-up condition detection method based on multi-time series according to claim 8, characterized in that, The output includes startup start time, end time, key time nodes, and startup type startup status records, including: Generate a startup status record containing the following fields: Startup start time START_TIME; Startup end time END_TIME; Key time points t0, t1, t2, t3; Startup type CONDITION_TYPE; Among them, CONDITION_TYPE is CONDITION_A, CONDITION_B, or CONDITION_C.

10. A generator set start-up condition detection system based on multi-time series, characterized in that, The system employs the generator set start-up condition detection method based on multiple time sequences as described in any one of claims 1 to 9; The system includes: The data acquisition and preprocessing module is used to collect the generator set's speed time series data and power time series data, and to process outliers in the power time series data. The key time node construction module is used to extract time nodes that meet the speed threshold conditions by traversing the speed time series data, and to construct a set of candidate time nodes for the stationary state and a set of candidate time nodes for the high-speed running state. The startup process detection module is used to filter out the stationary state time nodes that meet the time difference requirements for each time node in the candidate time node set of high-speed operation state, check the power data during the startup process to determine the power rise time node, and determine whether there is an abnormal power reset. The startup completion condition judgment module is used to judge the full load reaching condition, load stabilization condition, and load decrease condition in order of priority, and determine the corresponding startup completion time node and startup type. The timing optimization module is used to reselect a better power rise time point, determine the final start-up time period, and generate a start-up event record. The result deduplication and filtering module is used to remove duplicate startup event records and filter and retain startup events with the same startup completion time node. The special case handling module is used to detect whether there is a prolonged high power situation during the startup process and update the startup completion node accordingly; The startup condition data output module is used to output startup condition records that include startup start time, end time, key time nodes, and startup type.

Images