An energy-saving self-priming pump abnormal state alarm method and system based on big data

Abstract

The present application relates to the field of energy-saving self-priming pump abnormal state alarm, and provides an energy-saving self-priming pump abnormal state alarm method and system based on big data. The data packets of multiple sensors are acquired in real time, and the arrival time and physical time stamp of the data packets are collected to determine the arrival characteristics of the sensors. Based on these arrival characteristics, abnormality is judged. When the arrival characteristics in a plurality of continuous time detection windows are detected to exhibit a predetermined abnormal mode, a link abnormality flag is triggered to be set. Otherwise, the link abnormality flag is released. If the link abnormality flag is in a set state, the control command is prevented from being issued to the actuator, and an alarm is executed. Thus, the data timing dislocation problem caused by industrial Ethernet transmission delay, data packet cluster delay and differences in uplink strategies of different sensors is solved.

Description

Technical Field

[0001] This application relates to the field of alarm for abnormal conditions of energy-saving self-priming pumps, and more specifically, to a method and system for alarming abnormal conditions of energy-saving self-priming pumps based on big data. Background Technology

[0002] In industrial plants, energy-saving self-priming pumps often employ a low-load intermittent operation strategy at night—"stop for three minutes, run for two minutes"—to reduce energy consumption. Each restart generates transient characteristics such as suction-side backflow, pressure fluctuations, vibration spikes, and nonlinear current changes, requiring multi-source sensor data to distinguish between "normal self-priming" and genuine anomalies. However, high-priority services in industrial Ethernet (such as visual inspection) can preempt bandwidth, causing a 1-2 second delay in low-priority sensor data such as vibration and pressure data. Some data even arrives in clusters due to edge retransmission strategies. Real-time streaming engines merge data based on arrival time, causing current, pressure, and vibration signals that should be synchronized to be misallocated to different processing windows, disrupting the temporal correspondence between features. This temporal disorder amplifies feature indicators, causing the system to misjudge normal self-priming as continuous cavitation, triggering speed reduction or load reduction protection. The reduced speed, in turn, prolongs the gas-liquid mixing time, exacerbating vibration and pressure fluctuations. Meanwhile, the alarm merging mechanism obscures the details of the incident, making it easy for maintenance personnel to misjudge it as a continuous anomaly and execute a shutdown, ultimately disrupting the production cycle and triggering a chain reaction.

[0003] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0004] This application discloses a big data-based alarm method and system for abnormal states of energy-saving self-priming pumps. It aims to solve the technical problem that, in the intermittent operation mode of energy-saving self-priming pumps in industrial plants, due to the data transmission delay and timing misalignment of multi-source sensors and the limitations of existing alarm logic, it is difficult to distinguish between "normal self-priming" and "real abnormality", which leads to misjudgment, misoperation, and affects production efficiency and equipment life.

[0005] The technical solution of this application is as follows:

[0006] In a first aspect, this application discloses an alarm method for abnormal status of an energy-saving self-priming pump based on big data. The energy-saving self-priming pump includes an actuator, and the method includes:

[0007] It acquires data packets from multiple sensors in real time and collects the arrival time and physical timestamps within the data packets.

[0008] Arrival characteristics of multiple sensors are determined based on arrival time and physical timestamp; these arrival characteristics include arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between physical timestamp and arrival time.

[0009] Anomaly detection is performed based on arrival characteristics. If an arrival characteristic exhibiting a predetermined abnormal pattern is detected within multiple consecutive time detection windows, the link anomaly flag is set. If no arrival characteristic exhibiting a predetermined abnormal pattern is detected within multiple consecutive time detection windows, the link anomaly flag is released.

[0010] If the link anomaly flag is set, control commands will be prevented from being sent to the actuator, and an alarm will be triggered.

[0011] Furthermore, this application also discloses an alarm method for abnormal status of an energy-saving self-priming pump based on big data, wherein multiple sensors include a current sensor, a pressure sensor, and a vibration sensor; and anomaly judgment is made based on arrival characteristics, including: if the current sensor data packet arrives on time, while the pressure sensor data packet or the vibration sensor data packet arrives late and in batches, then it is judged that a predetermined abnormal mode has occurred.

[0012] Based on this, this application further proposes that the method also includes: in response to the link anomaly flag being set, switching the anomaly judgment logic to a robust logic mode; wherein, in the robust logic mode, using statistical information of a single sensor within a predetermined time window for anomaly judgment, instead of using the arrival time relationship between data packets from different sensors; wherein, the statistical information includes energy integral, short-term average surge duration, and instantaneous peak frequency.

[0013] In some preferred embodiments, anomaly judgment is made based on arrival characteristics, including: when the following conditions are met within a time detection window, a predetermined abnormal mode is determined to have occurred: the number of current sensor data packets arriving meets the expected arrival frequency; and at least one of the pressure sensor or vibration sensor has a continuous period of no data arrival exceeding a first preset threshold; and after the continuous period of no data arrival, a batch of data packets arrives, with the in-cluster arrival count greater than a second preset threshold and the in-cluster arrival duration less than a third preset threshold; and the median difference between the physical timestamp and the actual arrival time of the batch of data packets is greater than a fourth preset threshold.

[0014] Furthermore, if the link anomaly flag is set, control commands are prevented from being sent to the actuator, and an alarm is executed. This includes: checking the link anomaly flag and the physical timestamp span of the data packets in the current data window before generating automatic control commands; if the link anomaly flag is set or the physical timestamp span exceeds a preset span threshold, control commands are prevented from being sent to the actuator, and alarm information is generated to execute the alarm; wherein, the alarm information includes the link anomaly flag status and arrival characteristics.

[0015] As a technological improvement, the method involves acquiring data packets from multiple sensors in real time, collecting the arrival time and physical timestamps within the data packets, and also includes: setting a silence period threshold based on the expected data arrival interval of the sensors; determining the start of a new data cluster when the arrival interval of the next data packet exceeds the silence period threshold after the arrival of a data packet; determining that the data cluster is still continuing when the arrival interval of consecutive data packets within the data cluster is less than the expected arrival interval multiplied by a reduction factor; and using the data cluster for arrival feature determination only when the number of data packets contained in the identified data cluster is greater than or equal to a preset minimum cluster size threshold.

[0016] As a further improvement, the method acquires data packets from multiple sensors in real time and collects the arrival time and physical timestamps within the data packets. It also includes: calculating the mean and standard deviation of the physical timestamp differences of consecutive data packets within a data cluster; calculating the mean and standard deviation of the arrival time intervals of consecutive data packets within a data cluster; calculating the packet order consistency index within a cluster based on the consistency between the physical timestamp order and the system arrival order; and calculating the cross-stream synchronization degree of the cluster based on the proportion of data clusters appearing from other sensors within the same time period.

[0017] Based on this, this application also proposes that the method further includes: if the link anomaly flag is not in the set state, then determining a disturbance vector based on the standard deviation of the physical timestamp difference of consecutive data packets in the data cluster, the standard deviation of the arrival time interval, the consistency index, and the cluster cross-flow synchronization degree; and allowing or preventing the issuance of control commands to the actuator based on the disturbance vector.

[0018] Building upon the above, this application further proposes determining a disturbance vector based on the fluctuation of the physical timestamp difference of consecutive data packets within a cluster, the fluctuation of the arrival time interval, the consistency index, and the cluster cross-flow synchronization. This includes: increasing the host-side read / write pause evidence score when both the standard deviation of the physical timestamp difference and the standard deviation of the arrival time interval are below the corresponding thresholds and the packet order consistency index within the cluster is above the corresponding thresholds; increasing the link-layer transmission instability evidence score when the standard deviation of the arrival time interval is significantly greater than the standard deviation of the physical timestamp difference, or when there are high-variability arrival intervals and out-of-order arrivals within the cluster; increasing the management flow concurrency evidence score when the proportion of data clusters appearing in the same time window for the same type of sensor exceeds the threshold; and combining the link-layer transmission instability evidence score, the host-side read / write pause evidence score, and the management flow concurrency evidence score into a disturbance vector.

[0019] Secondly, this application also discloses an energy-saving self-priming pump abnormal status alarm system based on big data. The energy-saving self-priming pump includes an actuator. The system includes: an acquisition module for acquiring data packets from multiple sensors in real time and collecting the arrival time and physical timestamp within the data packets; a feature determination module for determining the arrival features of multiple sensors based on the arrival time and physical timestamp; the arrival features include arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between the physical timestamp and the arrival time; an anomaly judgment module for judging anomalies based on the arrival features, triggering a link anomaly flag to be set in response to the detection of an arrival feature exhibiting a predetermined abnormal pattern within multiple consecutive time detection windows; and releasing the link anomaly flag in response to the failure to detect an arrival feature exhibiting a predetermined abnormal pattern within multiple consecutive time detection windows; and an alarm module for preventing the issuance of control commands to the actuator and executing an alarm if the link anomaly flag is set.

[0020] Beneficial effects

[0021] This application discloses a big data-based alarm method for abnormal states of energy-saving self-priming pumps. It acquires data packets from multiple sensors in real time, collecting the arrival time and physical timestamp of these packets to determine sensor arrival characteristics, including arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between the physical timestamp and the arrival time. Based on these arrival characteristics, anomaly detection is performed. When arrival characteristics exhibit a predetermined abnormal pattern within multiple consecutive time detection windows, a link anomaly flag is set; otherwise, the flag is released. If the link anomaly flag is set, control commands are prevented from being sent to the actuator, and an alarm is triggered.

[0022] This method effectively solves the data timing misalignment problems caused by industrial Ethernet transmission delays, delayed packet clustering, and differences in uplink strategies among different sensors in existing technologies. Through in-depth analysis of packet arrival characteristics, this application can accurately distinguish between "normal self-priming" and "true anomalies," avoiding misjudgments that may be caused by traditional real-time stream processing engines based on arrival time merging strategies. For example, when current sensor data arrives on time while pressure or vibration sensor data arrives late in clusters, this application can identify this predetermined anomaly pattern and promptly prevent the issuance of erroneous protective automatic speed reduction or load reduction control commands, thereby avoiding negative impacts such as prolonging the pump's intake mixing state and lengthening vibration peaks and pressure fluctuation time curves. Simultaneously, through timely and accurate alarms, maintenance personnel can obtain fine-grained event evolution information, avoiding misjudging persistent anomalies based on experience and performing shutdown inspections, ultimately reducing the risk of pump unit shutdown and impacting upstream production cycles. Therefore, this application significantly improves the reliability, stability, and production efficiency of energy-saving self-priming pumps. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the steps of the energy-saving self-priming pump abnormal status alarm method based on big data disclosed in an embodiment of the present invention;

[0024] Figure 2 This is a schematic diagram of the structure of the energy-saving self-priming pump abnormal status alarm system based on big data disclosed in an embodiment of the present invention. Detailed Implementation

[0025] The implementation details of the technical solution in this embodiment are described in detail below:

[0026] This application proposes a method for alarming abnormal states of an energy-saving self-priming pump based on big data, wherein the energy-saving self-priming pump includes an actuator. For example... Figure 1 As shown, the method includes:

[0027] S101, acquire data packets from multiple sensors in real time, and collect the arrival time and physical timestamp inside the data packets;

[0028] S102, based on the arrival time and physical timestamp, determine the arrival characteristics of the plurality of sensors; the arrival characteristics include arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between the physical timestamp and the arrival time;

[0029] S103, perform anomaly judgment based on the arrival feature; in response to the arrival feature being detected to exhibit a predetermined abnormal mode within multiple consecutive time detection windows, trigger the link abnormal flag to be set; in response to the arrival feature not being detected to exhibit a predetermined abnormal mode within multiple consecutive time detection windows, release the link abnormal flag.

[0030] S104, if the link abnormality flag is set, then control commands are prevented from being sent to the actuator, and an alarm is executed.

[0031] This application aims to effectively distinguish between genuine equipment anomalies and "false anomalies" caused by data transmission delays, packaging strategies, etc., through refined analysis of the arrival time characteristics of sensor data, thereby avoiding misjudgments and unnecessary protective shutdowns, and improving the reliability and production efficiency of energy-saving self-priming pumps.

[0032] In this embodiment, the "energy-saving self-priming pump" refers to a type of pump that possesses self-priming capability and can achieve energy savings through optimized operating strategies. Its operating state is affected by various physical parameters, such as current, pressure, and vibration. To monitor these parameters, the energy-saving self-priming pump is typically equipped with multiple sensors. These sensors collect data in real time and transmit it in the form of data packets. A "data packet" is the basic unit of data collected by the sensors, encapsulated and transmitted over the network. Each data packet contains actual measurement data and a "physical timestamp," which records the precise physical moment the data was collected at the sensor. Furthermore, when a data packet arrives at the receiving end, the system records its "arrival time," i.e., the time when the data packet is received by the receiving end. "Arrival characteristics" are one of the core concepts of this method. They characterize the transmission characteristics of the data stream by analyzing the arrival time and physical timestamp of the data packets. These characteristics include: "Arrival count": the number of data packets arriving within a specific time window; "Arrival interval": the time interval between two consecutive data packets arriving at the receiving end; and "Intra-cluster arrival count": the number of data packets contained within a data cluster. "Intra-cluster arrival time": The total time from the arrival of the first data packet to the arrival of the last data packet in a data cluster. "Difference between physical timestamp and arrival time": The difference between the time a data packet is collected at the sensor and the time it arrives at the receiver, reflecting the data transmission delay. "Actuator" refers to the component in the energy-saving self-priming pump system responsible for executing control commands, such as motor controllers and valve controllers, which adjust the pump's operating state according to the received commands. "Link anomaly flag" is a status indicator used to mark whether there is an anomaly in the data transmission link. When the flag is set, it indicates that there is an anomaly in the link; when the flag is released, it indicates that the link is normal. "Predetermined anomaly pattern" refers to the combination of data arrival characteristics identified through pre-set rules or models. These combinations are usually associated with abnormal behavior in the data transmission link, such as packet delay, out-of-order delivery, and clustered arrival.

[0033] This embodiment provides a method for alarming abnormal states of an energy-saving self-priming pump based on big data. The specific implementation method is as follows:

[0034] First, it's necessary to acquire data packets from multiple sensors in real time, and collect the arrival time and physical timestamp within each packet. For example, a network interface can be configured in the data acquisition module to listen for and receive data packets from various sensors on the energy-saving self-priming pump (such as current sensors, pressure sensors, vibration sensors, etc.). When a data packet arrives, the system immediately records the current system time as the arrival time. Simultaneously, the header or payload of the data packet is parsed to extract the physical timestamp contained within. The physical timestamp is typically added by the sensors when the data is generated, reflecting the actual moment the data was collected in the physical world.

[0035] Secondly, based on the arrival time and physical timestamp, the arrival characteristics of the multiple sensors are determined. For example, a sliding time window can be set, within which statistical analysis is performed on the data packets of each sensor. The arrival count can be obtained by simply accumulating the number of data packets within the window. The arrival interval can be obtained by calculating the difference in arrival times of two consecutively arriving data packets. The intra-cluster arrival count and intra-cluster arrival duration require first identifying the data clusters. For example, when the arrival interval of consecutive data packets is less than a certain preset threshold, these data packets can be considered to belong to the same data cluster. The difference between the physical timestamp and the arrival time can be calculated by subtracting the physical timestamp from the arrival time of each data packet. The calculation of these characteristics can be completed in real time by the data processing unit, and the results can be stored for subsequent analysis.

[0036] Next, anomaly detection is performed based on the arrival characteristics. For example, a series of rules can be preset or a machine learning model can be trained to identify predetermined anomaly patterns. When, within multiple consecutive time detection windows, analysis of the aforementioned arrival characteristics (such as arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between physical timestamp and arrival time) reveals features consistent with a predetermined anomaly pattern—for example, a sudden and significant increase in the arrival interval of a sensor, or a large number of data packets arriving in clusters within a short period, with a generally large difference between their physical timestamps and arrival times—this may indicate congestion or delay in the data transmission link. Once this predetermined anomaly pattern is detected, the system will trigger the link anomaly flag to be set. Conversely, if no arrival characteristics exhibiting a predetermined anomaly pattern are detected within multiple consecutive time detection windows, the system will release the link anomaly flag, indicating that the link has returned to normal.

[0037] Finally, if the link anomaly flag is set, control commands will be prevented from being sent to the actuator, and an alarm will be triggered. For example, when the link anomaly flag is set, the system will intercept all automatic control commands attempted to be sent to the energy-saving self-priming pump actuator to prevent incorrect control commands from being issued due to misjudgment caused by data transmission anomalies. At the same time, the system will generate alarm information and issue alerts to maintenance personnel through various means (such as audible and visual alarms, SMS notifications, emails, and interface pop-ups) to inform them that there is a data link anomaly and that inspection and handling are required.

[0038] This embodiment of the energy-saving self-priming pump abnormal status alarm method based on big data effectively identifies abnormal states in the data transmission link by performing refined analysis of the arrival time and physical timestamp of sensor data packets. Specifically, during the operation of the energy-saving self-priming pump, multiple sensors (such as current sensors, pressure sensors, and vibration sensors) collect data in real time and generate data packets. During transmission, these data packets may arrive delayed, out of order, or in clusters due to factors such as network congestion, priority scheduling, or the packetization strategy of edge collectors. Traditional methods often rely solely on data content or simple arrival time windows for anomaly judgment, easily misjudging "false anomalies" caused by transmission problems as "real anomalies" of the equipment itself, thereby triggering unnecessary protective shutdowns or load reductions.

[0039] This embodiment collects the arrival time and physical timestamps within data packets, and calculates a series of arrival characteristics based on this time information, including arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between the physical timestamp and the arrival time. These characteristics can comprehensively and meticulously reflect the transmission behavior of the data stream. For example, when data packets are delayed, the difference between the physical timestamp and the arrival time will increase significantly; when data packets arrive in clusters, the intra-cluster arrival count and intra-cluster arrival duration will exhibit specific patterns.

[0040] By continuously monitoring and identifying anomalies in these arrival characteristics, the system can accurately identify link anomalies and trigger a link anomaly flag when these characteristics exhibit a predetermined abnormal pattern within multiple consecutive time detection windows. Once the link anomaly flag is set, the system immediately prevents the issuance of control commands to the actuator, thus avoiding erroneous control operations on the energy-saving self-priming pump due to misjudgments caused by transmission anomalies. Simultaneously, the system will issue an alarm to promptly notify maintenance personnel for intervention. When the link returns to normal, the link anomaly flag will be released, and the system will resume issuing normal control commands.

[0041] The core innovation of this method lies in its use of the health of the data transmission link as a crucial basis for judging the abnormal state of the energy-saving self-priming pump. By distinguishing between "false anomalies" caused by transmission problems and "real anomalies" of the equipment itself, it avoids misjudgments caused by data timing misalignment in traditional methods. Compared to existing technologies, this embodiment can more accurately identify the abnormal state of the energy-saving self-priming pump, reducing unnecessary downtime and maintenance, and improving the operational reliability and production efficiency of the equipment. For example, during the re-priming stage of the energy-saving self-priming pump, if the data from pressure and vibration sensors arrives in clusters with delays due to network congestion, traditional methods may misjudge it as cavitation or unstable flow. However, this embodiment, by analyzing the arrival characteristics of data packets, can identify that this is a link anomaly rather than an equipment malfunction, thereby avoiding erroneous protective speed reduction or shutdown and ensuring production continuity.

[0042] Furthermore, this application also proposes the above-mentioned alarm method for abnormal status of energy-saving self-priming pump based on big data, wherein the multiple sensors include a current sensor, a pressure sensor, and a vibration sensor; the abnormal judgment based on the arrival characteristics includes: if the current sensor data packet arrives on time, while the pressure sensor data packet or the vibration sensor data packet arrives late and in batches, then it is judged that a predetermined abnormal mode has occurred.

[0043] Specifically, the aforementioned sensors are configured as current sensors, pressure sensors, and vibration sensors. The current sensor monitors the motor current of the energy-saving self-priming pump, reflecting its operating load status; the pressure sensor monitors the inlet and outlet pressures of the pump, reflecting the fluid transport status; and the vibration sensor monitors the vibration of the pump body, reflecting the smoothness of mechanical operation. These sensors work together to comprehensively reflect the operating status of the energy-saving self-priming pump. "On-time arrival" means that data packets arrive at the receiving end within the expected transmission period or within an acceptable delay range, indicating that the data link is generally healthy. "Delayed and concentrated batch arrival" means that the arrival time of data packets is significantly later than expected, and multiple data packets arrive in a concentrated manner within a short period of time. This is usually caused by network congestion, data buffering, recovery after link interruption, or sensor malfunction leading to data accumulation and subsequent one-time transmission.

[0044] This application's solution, by introducing specific types of sensors and defining specific anomaly patterns, can more accurately identify potential problems with energy-saving self-priming pumps. When current sensor data packets arrive on time, it indicates that the pump's motor is still powered normally and attempting to operate. However, if pressure sensor or vibration sensor data packets arrive late and in concentrated batches, it strongly indicates an anomaly. For example, an abnormal arrival pattern of pressure sensor data packets may mean fluid pipeline blockage, cavitation, or abnormal pumping medium, leading to obstructed or accumulated pressure data transmission; an abnormal arrival pattern of vibration sensor data packets may indicate faults such as bearing wear, impeller imbalance, or loose mechanical components, which may cause unstable data transmission or erupt in concentrated bursts under certain conditions. Through this combined judgment, this application can effectively distinguish between data delays caused by network fluctuations and actual equipment operational anomalies, thereby avoiding misjudgments and specifically identifying anomalies in physical parameters crucial to the operation of the energy-saving self-priming pump.

[0045] In some preferred embodiments, it is assumed that an energy-saving self-priming pump is operating, equipped with a current sensor, a pressure sensor, and a vibration sensor. Under normal operating conditions, data packets from these three sensors arrive on time at the expected frequency and time intervals. However, at some point, if the pump inlet becomes partially blocked, increasing the pump's operating load while the motor continues to run, the current sensor data packets may still arrive on time, reflecting continuous motor operation. However, due to the change in fluid dynamics caused by the blockage, the pressure sensor data may experience a delay during transmission and be sent in a concentrated batch within a short period. Simultaneously, the blockage may cause abnormal pump vibration, resulting in a similar delay and batch arrival of vibration sensor data packets. In this situation, the method of this application will detect that the current sensor data packets arrive on time, while the pressure sensor data packets or vibration sensor data packets (or both) exhibit delayed and concentrated batch arrival characteristics. Based on a preset abnormal mode, the system will determine that a predetermined abnormal mode has occurred and trigger the link abnormality flag to be set, thereby preventing the issuance of control commands to the actuator and executing an alarm. This promptly reminds the operator to check the pump inlet or related pipelines, preventing the equipment from continuing to operate under abnormal conditions and causing further damage.

[0046] This application further proposes an optimization scheme, namely, switching the anomaly judgment logic to a robust logic mode in response to when the link anomaly flag is set, to improve the accuracy and reliability of anomaly judgment. The method further includes: switching the anomaly judgment logic to a robust logic mode in response to when the link anomaly flag is set; wherein, in the robust logic mode, statistical information from a single sensor within a predetermined time window is used for anomaly judgment, instead of using the arrival time relationship between data packets from different sensors; wherein the statistical information includes energy integral, short-term average surge duration, and instantaneous peak frequency.

[0047] Specifically, when the system detects that the link anomaly flag is set, it indicates a preliminary judgment that an anomaly exists at the link level. At this point, to avoid interference or misjudgment that may arise from the complex arrival time relationships between different sensors, the anomaly judgment logic is switched to a robust logic mode. In this mode, anomaly judgment no longer relies on the arrival time relationships between data packets from different sensors, but focuses on in-depth analysis of the data stream of a single sensor. The robust logic mode refers to a more focused and interference-resistant anomaly judgment strategy, the core of which lies in performing internal characteristic analysis on each independent sensor data stream. Specifically, the system calculates and analyzes the statistical information of the data from a single sensor within a predetermined time window. The predetermined time window can be flexibly configured according to the actual application scenario and sensor data characteristics; for example, it can be several seconds, tens of seconds, or longer to ensure that sufficient data samples are captured for statistical analysis. The statistical information includes energy integral, short-term mean surge duration, and instantaneous peak frequency. The energy integral can be understood as the accumulated energy of the sensor data signal within the predetermined time window, reflecting the overall activity or intensity of the sensor data over a period of time. The duration of a short-term mean spike refers to the length of time during which the sensor data's mean is significantly higher than normal and persists for a short period. Its purpose is to identify whether there are persistently high abnormal values ​​in the data stream. The frequency of instantaneous peaks refers to the frequency of instantaneous peaks in the sensor data within a predetermined time window. Its purpose is to detect frequent and drastic instantaneous fluctuations in the data stream. Through these statistical information, the inherent behavioral patterns of a single sensor data stream can be characterized from different dimensions, thereby more accurately determining whether it is in an abnormal state.

[0048] This application's solution effectively addresses the issue of decreased accuracy in judgments that might arise from relying on the arrival time relationships between different sensor data packets when a link anomaly already exists. This is achieved by switching to robust logic mode after the link anomaly flag is set. When the link anomaly flag is set, it indicates a potential problem with link transmission. In this case, the arrival time relationships between different sensor data packets may be subject to widespread interference, making it difficult to accurately reflect the actual equipment operating status. By switching to robust logic mode, the system shifts its focus to the intrinsic statistical characteristics of individual sensor data streams, such as energy integral, duration of short-term mean spikes, and instantaneous peak frequencies. These statistical information are independent of external interference in link transmission and directly reflect the abnormal behavior of the physical quantities monitored by the sensors, such as abnormal current fluctuations, sustained pressure increases, or drastic vibration changes. This approach allows anomaly judgment to focus more on genuine physical anomalies, reducing the ambiguity of link transmission anomalies on the judgment results, thereby improving the ability to identify genuine equipment anomalies in complex anomaly scenarios.

[0049] In some preferred embodiments, it is assumed that during the operation of the energy-saving self-priming pump, due to network congestion or sensor malfunction, pressure sensor data packets or vibration sensor data packets arrive in batches with delays, while current sensor data packets arrive on time. Based on the aforementioned judgment logic, the system will detect a predetermined abnormal pattern and trigger a link anomaly flag. At this point, if the system continues to rely entirely on the arrival time relationship between different sensor data packets for judgment, it may be difficult to further distinguish between a simple link problem and a more serious fault in the device itself due to the continuous instability of the link status. As a specific implementation, when the link anomaly flag is set, the system immediately switches to a robust logic mode. For example, for a pressure sensor, the system continuously calculates the energy integral, short-term average spike duration, and instantaneous peak frequency of its data stream within a preset 10-second time window. If, in this mode, the energy integral of the pressure sensor is detected to continuously and abnormally increase within a short period, or the short-term average spike duration exceeds a preset threshold, even if the data packet arrival pattern remains chaotic, the system can independently determine that the pressure system may have a physical anomaly of blockage or overload. Similarly, for vibration sensors, if their instantaneous peak frequency increases significantly in robust logic mode, it can be determined that there may be a mechanical fault in the pump body. In this way, even in the context of link anomalies, the system can provide deeper and more accurate anomaly diagnosis by analyzing the intrinsic data characteristics of individual sensors, thereby guiding more precise maintenance and intervention measures.

[0050] To further improve the accuracy and robustness of anomaly detection, this application proposes a more refined anomaly detection logic. Specifically, this application further proposes the above-mentioned anomaly detection based on the arrival characteristics, including: when the following conditions are met within a time detection window, a predetermined anomaly mode is determined to have occurred: the number of arriving data packets from the current sensor meets the expected arrival frequency; and at least one of the pressure sensor or vibration sensor experiences a continuous period of no data arrival exceeding a first preset threshold; and after the continuous period of no data arrival, a batch of arriving data packets appears, with an in-cluster arrival count greater than a second preset threshold and an in-cluster arrival duration less than a third preset threshold; and the median difference between the physical timestamp and the actual arrival time of the batch of arriving data packets is greater than a fourth preset threshold.

[0051] Specifically, "the number of current sensor data packets arriving meets the expected arrival frequency" means that within a set time detection window, the actual number of current sensor data packets arriving is counted and compared with the theoretical or historical average arrival frequency of the sensor. If the actual number of arrivals is within a reasonable fluctuation range of the expected frequency, the current sensor's data link is considered to be working normally, providing a stable reference benchmark for judging abnormalities in other sensor links. "At least one of the pressure sensor or vibration sensor has a continuous period of no data arrival exceeding a first preset threshold" means that the system continuously monitors the data stream of the pressure sensor or vibration sensor. Once it is found that a sensor has no data packets arriving for a period of time, and the duration of this lack of data arrival exceeds the preset first threshold, it indicates that the sensor's data link may be interrupted or severely delayed. The first preset threshold can be set according to factors such as the sensor's normal data transmission cycle and network latency tolerance; for example, it can be set to several times the normal data packet interval. In practical applications, the phrase "after a period of no data arrivals, a batch of data packets arrives, with an intra-cluster arrival count greater than the second preset threshold and an intra-cluster arrival duration less than the third preset threshold" refers to a situation where, after a period of no data arrivals, the system suddenly receives a large number of data packets from the sensor, which arrive densely within a very short time. Here, "intra-cluster arrival count" refers to the total number of data packets in the batch; a count greater than the second preset threshold indicates an abnormally concentrated data volume. "Intra-cluster arrival duration" refers to the time span from the first to the last data packet in the batch; a count less than the third preset threshold indicates that these data packets arrived very rapidly. The second and third preset thresholds can be empirically set based on the sensor's normal data transmission rate and network transmission characteristics. Furthermore, the phrase "the median difference between the physical timestamp and the actual arrival time of the batch of arriving data packets is greater than the fourth preset threshold" refers to analyzing this batch of concentrated arrival data packets and calculating the difference between the physical timestamp recorded within each data packet and the actual arrival time of the data packet received by the system. If the median of these differences exceeds the fourth preset threshold, it indicates that the data packet experienced significant delay or queuing during transmission, resulting in a large deviation between the time recorded internally and the actual reception time. The fourth preset threshold can be set according to the system's time synchronization requirements and the normal range of network latency.

[0052] This application's solution constructs a multi-condition, high-confidence anomaly detection model by combining four dimensions: the stability of the current sensor data stream, prolonged interruptions of specific sensor data streams, concentrated bursts of data after interruption, and significant differences between the internal timestamps of data packets and their actual arrival times. This combined detection mechanism can effectively distinguish between normal network fluctuations and genuine link anomalies. For example, when network congestion causes data packets to queue at a node and are sent all at once after the congestion eases, the above characteristics will appear. Through comprehensive analysis of these characteristics, anomaly patterns caused by network congestion, sensor failures, or unstable communication links can be identified more accurately.

[0053] In some preferred embodiments, assuming that within a time detection window, the system detects 100 arriving data packets from the current sensor, which matches the expected frequency of 10 data packets per second (i.e., 100 packets within 10 seconds). Simultaneously, the system detects that no data packets have arrived from the pressure sensor in the past 5 seconds (a first preset threshold of 3 seconds). Then, in the following 0.5 seconds (a third preset threshold of 1 second), the system receives 50 data packets from the pressure sensor (a second preset threshold of 20), and the intra-cluster arrival count and intra-cluster arrival duration of these data packets both meet the conditions. Further analysis of these 50 data packets reveals that the median difference between their physical timestamps and actual arrival times is 2 seconds (a fourth preset threshold of 0.5 seconds). Since all four conditions are met, the system determines that the pressure sensor's data link has entered a predetermined abnormal mode and triggers corresponding alarms and control commands to prevent further disruption.

[0054] This application proposes a more robust alarm mechanism by introducing a check on the physical timestamp span of data packets in the steps of blocking control commands and executing alarms, and providing more detailed alarm information, thereby further improving the system's anomaly detection capability and the effectiveness of alarms.

[0055] If the aforementioned link anomaly flag is set, control commands will be prevented from being issued to the aforementioned actuator, and an alarm will be executed. Specifically, before generating an automatic control command, the link anomaly flag and the physical timestamp span of the data packets in the current data window will be checked. If the link anomaly flag is set or the physical timestamp span exceeds a preset span threshold, control commands will be prevented from being issued to the actuator, and alarm information will be generated to execute the alarm. The alarm information includes the set status of the link anomaly flag and arrival characteristics.

[0056] Specifically, a crucial pre-check is performed before the system issues any automatic control commands to the actuators. This check aims to assess the current state of the system and the validity of the data it relies on. Specifically, checking the link anomaly flag and the physical timestamp span of data packets within the current data window means the system first verifies the status of the link anomaly flag to determine if there are link problems caused by continuous abnormal arrival patterns. Simultaneously, the system calculates the maximum time difference between the physical timestamps of all received data packets within the current data window, i.e., the physical timestamp span. This physical timestamp span reflects the time range in which data packets are generated at the sensor; an excessively large span may indicate severe data delays, out-of-order delivery, or partial data loss, even if the arrival pattern has not yet triggered the link anomaly flag.

[0057] Furthermore, if the link anomaly flag is set or the physical timestamp span exceeds a preset span threshold, control commands will be prevented from being sent to the actuator, and an alarm message will be generated to execute the alarm. This means that if the link anomaly flag is set, or even if the link anomaly flag is not set but the calculated physical timestamp span exceeds the preset span threshold, the system will take immediate action. At this time, any upcoming automatic control commands will be prevented from being sent to the actuator to avoid operation based on potentially inaccurate or outdated data. Simultaneously, the system will generate an alarm message and execute the corresponding alarm operation. The preset span threshold can be set according to the actual application scenario and data real-time requirements; for example, it can be set to several times the expected update cycle of the sensor data.

[0058] The alarm information includes the link anomaly flag status and arrival characteristics. To provide more comprehensive diagnostic information, the generated alarm information not only includes the current status of the link anomaly flag (set or not set), but also details the arrival characteristics that caused the alarm. These arrival characteristics can include arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between the physical timestamp and the arrival time. This information helps maintenance personnel quickly locate the root cause of the problem and determine whether it is a link layer transmission issue, a sensor data generation problem, or another system anomaly.

[0059] This application's solution addresses the limitation of traditional solutions that may not fully cover data quality issues by introducing additional checks on the physical timestamp span into the logic for blocking control commands and executing alarms. Specifically, even if the link anomaly flag has not yet been set due to continuous anomaly patterns, if the physical timestamp span of the sensor data packets is too large, indicating potential severe data delays or timing discrepancies, the system can promptly identify and prevent the issuance of control commands based on such unreliable data. This avoids the risk of erroneous operations caused by data lag or internal timing issues. Furthermore, by including the link anomaly flag status and specific arrival characteristics in the alarm information, this solution provides maintenance personnel with richer and more accurate fault diagnosis data, thereby accelerating the problem localization and resolution process.

[0060] In some preferred embodiments, it is assumed that the current sensor, pressure sensor, and vibration sensor of the energy-saving self-priming pump are functioning normally, and that the arrival pattern of its data packets does not exhibit a predetermined abnormal pattern sufficient to trigger the setting of the link anomaly flag within multiple consecutive time detection windows. However, due to some instantaneous network jitter or internal clock drift of the sensors, although the number of data packets arriving is normal within a certain data window, the physical timestamp span within it is abnormally large. For example, in a system with an expected data update cycle of 1 second, the physical timestamp span of the data packets in the current data window reaches 5 seconds, far exceeding the preset span threshold (e.g., 2 seconds). In this case, even if the link anomaly flag is not set, the solution of this application will detect the abnormal physical timestamp span, thereby preventing the issuance of control commands to the actuator and generating alarm information. The alarm message will clearly indicate that the link anomaly flag is not set, but the physical timestamp span is abnormal. It will also include arrival characteristics such as arrival count and arrival interval within the current data window, enabling maintenance personnel to quickly determine that the problem may be due to data timeliness or internal timing, rather than a simple link interruption or congestion. This allows them to take targeted troubleshooting measures, such as checking sensor clock synchronization or data caching mechanisms.

[0061] Furthermore, this application proposes to improve the accuracy and reliability of data processing by introducing a data cluster identification and filtering mechanism when acquiring data packets from multiple sensors in real time and collecting the arrival time and physical timestamps within the data packets. This ensures that the arrival features used for anomaly detection are based on valid and meaningful data clusters.

[0062] The above-mentioned method of acquiring data packets from multiple sensors in real time and collecting the arrival time and physical timestamps within the data packets also includes: setting a silence period threshold based on the expected data arrival interval of the sensors; determining the start of a new data cluster when the arrival interval of the next data packet exceeds the silence period threshold after the arrival of a data packet; determining that the data cluster is still continuing when the arrival interval of consecutive data packets within the data cluster is less than the expected arrival interval multiplied by a reduction factor; and using the data cluster for the determination of the arrival characteristics only when the number of data packets contained in the identified data cluster is greater than or equal to a preset minimum cluster size threshold.

[0063] Specifically, the expected data arrival interval of the sensor refers to the expected time interval between two consecutively transmitted data packets under normal operating conditions. For example, if a sensor is designed to transmit a data packet every 100 milliseconds, its expected data arrival interval is 100 milliseconds. The silence period threshold is a time length set based on the expected data arrival interval, and its purpose is to distinguish the start of different data clusters. When a data packet arrives, if no subsequent data packet arrives within a time exceeding the silence period threshold, the previous data cluster is considered to have ended, and the next arriving data packet will mark the start of a new data cluster. For example, the silence period threshold can be set to several times the expected data arrival interval to allow for some network jitter while effectively distinguishing long data interruptions. The reduction factor is a multiplier less than 1, used to adjust the conditions for determining whether a data cluster continues. Within the data cluster, if the arrival interval of consecutive data packets is less than the expected data arrival interval multiplied by the reduction factor, it indicates that data packets are arriving densely at a faster rate than expected, which is usually a characteristic of the data cluster, therefore it is determined that the data cluster is still continuing. The reduction factor is introduced to more accurately capture dense arrival patterns within data clusters. For example, when network congestion causes packets to arrive in bulk, the intervals between them may be significantly smaller than expected.

[0064] In practical applications, the minimum cluster size threshold is a preset integer value. Its purpose is to filter out small, statistically insignificant data clusters that may be formed by network noise or sporadic delayed data packets. Only when the number of data packets contained in an identified data cluster reaches or exceeds the minimum cluster size threshold is the data cluster considered valid and meaningful, and used for subsequent arrival feature determination. For example, if the minimum cluster size threshold is set to 5, a cluster containing only 3 data packets will be ignored, thus avoiding the analysis of incomplete or unstable data streams and improving the reliability of anomaly detection.

[0065] This application's solution effectively addresses the problem of inaccurate arrival characteristic judgment caused by irregular data flow in basic solutions by introducing a data cluster identification and filtering mechanism. Specifically, firstly, by setting a silence period threshold, the system can intelligently identify the starting points of different data clusters in the data flow, effectively segmenting long-interrupted data flows and avoiding mixing unrelated data packets in one cluster. Secondly, within a data cluster, by comparing the arrival interval of consecutive data packets with the expected arrival interval adjusted by a reduction factor, the continuity of the data cluster can be accurately determined, capturing the true pattern of dense data packet arrivals. This is crucial for identifying predetermined abnormal patterns such as "delayed and concentrated arrival of pressure sensor data packets or vibration sensor data packets." Finally, by setting a minimum cluster size threshold, the system can filter out small, unrepresentative data clusters generated by accidental factors or network noise, ensuring that only sufficiently complete and stable data clusters are used for subsequent arrival characteristic determination, thereby significantly improving the accuracy and reliability of features such as "intra-cluster arrival count" and "intra-cluster arrival duration." It is precisely because of this refined data cluster processing that subsequent anomaly detection can be based on more realistic and stable data patterns, effectively reducing the false alarm rate.

[0066] In some preferred embodiments, a specific example is given below. Assume a pressure sensor is designed to send a data packet every 100 milliseconds, meaning its expected data arrival interval is 100 milliseconds. The system can set a silence period threshold of 500 milliseconds (i.e., 5 times the expected interval), a reduction factor of 0.8, and a minimum cluster size threshold of 5. Specifically, when the system begins receiving data packets from the pressure sensor:

[0067] 1. If data packet A arrives at time T0, data packet B arrives at time T0 + 120 milliseconds, data packet C arrives at time T0 + 230 milliseconds, data packet D arrives at time T0 + 340 milliseconds, and data packet E arrives at time T0 + 450 milliseconds, then the arrival intervals of these consecutive data packets (120ms, 110ms, 110ms, 110ms) are all less than the silence period threshold of 500 milliseconds, and are all not less than the expected arrival interval of 100 milliseconds multiplied by the reduction factor of 0.8 (i.e., 80 milliseconds). Therefore, these data packets are determined to belong to the same data cluster.

[0068] 2. If data packet F arrives at time T0 + 1000 milliseconds, and the time since the arrival of data packet E (T0 + 450 milliseconds) has exceeded the silence period threshold of 500 milliseconds (1000 - 450 = 550 milliseconds > 500 milliseconds), then the system will determine that data packet F marks the beginning of a new data cluster.

[0069] 3. Within a data cluster, if data packet G arrives 20 milliseconds after data packet F, and data packet H arrives 30 milliseconds after data packet G, then both arrival intervals (20ms and 30ms) are less than the expected arrival interval of 100 milliseconds multiplied by a reduction factor of 0.8 (i.e., 80 milliseconds). The system will determine that the data cluster is still in progress, indicating that data packets are arriving in rapid succession.

[0070] 4. In the above process, if an identified data cluster ultimately contains only 3 data packets, since its number is less than the preset minimum cluster size threshold of 5, the data cluster will be ignored by the system and will not be used for subsequent arrival feature determination. Only when a data cluster contains 5 or more data packets, such as a cluster containing data packets A, B, C, D, and E, will it be considered a valid cluster and used to calculate its "intra-cluster arrival count," "intra-cluster arrival time," and other arrival features.

[0071] In this way, the system can effectively identify and filter meaningful data clusters, avoiding the analysis of scattered or incomplete data, thus providing a more accurate and reliable data foundation for subsequent anomaly detection.

[0072] To more comprehensively assess the quality and stability of the data stream and provide richer and more refined diagnostic information for subsequent anomaly detection, this application further proposes a scheme for in-depth analysis of identified data clusters. Specifically, this application further proposes that the steps of acquiring data packets from multiple sensors in real time and collecting the arrival time and internal physical timestamps of the data packets also include: calculating the mean and standard deviation of the physical timestamp differences of consecutive data packets within a data cluster; calculating the mean and standard deviation of the arrival time intervals of consecutive data packets within a data cluster; calculating a packet order consistency index within a cluster based on the consistency between the physical timestamp order and the system arrival order; and calculating the cross-stream synchronization degree of the cluster based on the proportion of data clusters appearing in other sensors within the same time period.

[0073] Specifically, calculating the mean and standard deviation of the physical timestamp differences among consecutive data packets within a data cluster involves calculating the difference in internal physical timestamps for each consecutively arriving data packet pair within the cluster, and then performing statistical analysis on these differences to obtain their mean and standard deviation. The mean of the physical timestamp differences reflects the average rate of data packet generation, while the standard deviation reflects the volatility of the generation rate. For example, a large standard deviation may indicate instability in the sensor's internal clock or jitter in data generation.

[0074] The calculation of the mean and standard deviation of the arrival time intervals of consecutive data packets within a data cluster involves calculating the arrival time interval at the system receiver for each consecutive pair of data packets within the cluster, and then performing statistical analysis on these intervals to obtain their mean and standard deviation. The mean arrival time interval reflects the average transmission rate of the data packets on the transmission link, while the standard deviation reflects the fluctuation of the transmission rate or network jitter. For example, a high standard deviation may indicate network congestion or transmission instability.

[0075] In practical applications, the intra-cluster packet order consistency index, based on the consistency between the physical timestamp order and the system arrival order, compares the physical timestamp order of data packets generated at the sensor end with their actual arrival order when received by the system at the receiver end. If these two orders are inconsistent, it indicates that the data packets may have been out of order during transmission. The intra-cluster packet order consistency index can quantify the degree of this out-of-order transmission, for example, by calculating the proportion of out-of-order data packets or the number of out-of-order pairs. Its purpose is to assess the orderliness of the data transmission link; out-of-order transmission can lead to complexity or errors in data processing.

[0076] Furthermore, based on the proportion of data clusters appearing in other sensors within the same time period, cluster cross-stream synchronization is calculated. This refers to whether, within a specific time window, when a data cluster appears in the data stream of one sensor, data clusters also appear simultaneously or approximately simultaneously in the data streams of other related sensors. Cluster cross-stream synchronization quantifies the synchronicity between data streams from different sensors at the cluster level. Its purpose is to identify systemic problems that may affect the transmission of data from multiple sensors, such as congestion of the shared transmission medium or performance bottlenecks of upstream processing nodes.

[0077] The solution presented in this application, through the aforementioned multi-dimensional and refined statistical analysis of identified data clusters, can more comprehensively characterize the internal characteristics of data streams and the health status of transmission links. Specifically, the mean and standard deviation of physical timestamp differences can reveal the stability of the sensor data generation end; the mean and standard deviation of arrival time intervals can reflect the real-time performance and jitter of the data transmission link; the intra-cluster packet order consistency index directly quantifies the orderliness of data transmission, helping to detect out-of-order packet problems; and the inter-cluster flow synchronization degree assesses the coordination between multiple sensor data streams at a macroscopic level, helping to identify common problems affecting the entire system. It is precisely because of these additional diagnostic information that this application can gain a deeper understanding of abnormal patterns in data streams, thereby overcoming the limitations of relying solely on data cluster identification.

[0078] In some preferred embodiments, it is assumed that within a certain time detection window, current sensor data packets, pressure sensor data packets, and vibration sensor data packets are all identified as forming data clusters. The present application's scheme further analyzes these data clusters. Specifically, for the pressure sensor data cluster, the standard deviation of the physical timestamp difference between consecutive data packets within the cluster is calculated to be 0.5 ms, while the standard deviation of the arrival time interval is 5 ms. Simultaneously, the intra-cluster packet order consistency index shows that 10% of the data packets are out of order. Furthermore, within the same time period, current sensor data clusters and vibration sensor data clusters are also observed to appear simultaneously, and their inter-current synchronization reaches 80%. These detailed statistical information, such as the arrival time interval standard deviation being significantly greater than the physical timestamp difference standard deviation, and the presence of a certain proportion of out-of-order packets, clearly indicate severe jitter and out-of-order problems in the transmission link, rather than merely a problem with sensor data generation. This refined diagnostic capability enables the system to more accurately identify abnormal patterns of unstable link-layer transmission and provides a direct basis for subsequent decision-making.

[0079] Furthermore, this application also proposes that the above method further includes: if the link anomaly flag is not in the set state, then a disturbance vector is determined based on the standard deviation of the physical timestamp difference of consecutive data packets within the data cluster, the standard deviation of the arrival time interval, the consistency index, and the cluster cross-flow synchronization degree; and control commands are allowed or prevented from being issued to the actuator based on the disturbance vector.

[0080] Specifically, when the link anomaly flag is not set, the system uses various statistical information calculated from the above methods to assess the health of the data link. This statistical information includes: the standard deviation of the physical timestamp differences of consecutive data packets within a cluster, reflecting the stability of the timestamps within the data packets; the standard deviation of the arrival time interval, reflecting the regularity of data packet arrival times; a consistency index, which measures the degree of matching between the physical timestamp order of the data packets and the system's receiving order; and cluster-to-stream synchronization, which assesses the temporal synchronization of different sensor data clusters.

[0081] Based on this statistical information, the system determines a "disturbance vector." This disturbance vector can be understood as a set of values, each corresponding to one or more statistical indicators, collectively characterizing the stability and reliability level of the current data link. For example, the disturbance vector can be a multi-dimensional vector, with its components corresponding to the aforementioned standard deviations, consistency indicators, and synchronization degrees, or a comprehensive score obtained by weighting or mapping these indicators. After determining the disturbance vector, the system will decide whether to allow the issuance of control commands to the actuators based on the value of the disturbance vector. This means that even if the severity of triggering a link anomaly flag is not reached, if the disturbance vector indicates a certain degree of instability or anomaly, the system can take preventative measures, such as temporarily blocking or delaying the issuance of control commands, to avoid potential risks.

[0082] This application's solution addresses the potential risks that traditional solutions, relying solely on binary anomaly flags, might overlook by introducing a disturbance vector assessment when the link anomaly flag is not set. The absence of a link anomaly flag does not guarantee a completely healthy data link. For example, slight fluctuations in packet arrival intervals, slightly increased differences in physical timestamps, or a slight decrease in the synchronization of data streams from different sensors may occur. While these situations may not trigger severe link anomaly alarms, they could indicate network congestion, sensor drift, or excessive system load. By calculating and analyzing the standard deviation of the physical timestamp difference, the standard deviation of the arrival time interval, consistency indicators, and inter-cluster flow synchronization of consecutive data packets within a cluster, the system can more precisely capture these subtle fluctuations and anomalies. These indicators collectively constitute a disturbance vector, providing a multi-dimensional health assessment for the system. When the disturbance vector indicates a certain degree of "disturbance," the system can make decisions based on this disturbance vector, such as temporarily preventing the issuance of control commands, even if the link anomaly flag is not set. This mechanism enables the system to take intervention measures in advance before the problem worsens to the point of triggering a full alarm, thereby improving the reliability of control command issuance and the operational stability of the energy-saving self-priming pump actuator.

[0083] Through the above technical solution, this application introduces a disturbance vector determined based on multi-dimensional statistical information when the link anomaly flag is not set, thereby achieving a more refined assessment of the data link health status and more flexible control command management. This avoids the risks that may arise from blindly issuing control commands when a serious anomaly has not been reached, as in traditional solutions. This solution can effectively identify and respond to potential problems that are insufficient to trigger a full alarm but may still affect system stability, such as minor network congestion, slight fluctuations in sensor data, or decreased synchronization. Therefore, this application significantly improves the reliability of control command issuance for energy-saving self-priming pumps and the overall robustness of the system, ensuring that the energy-saving self-priming pump can operate stably and efficiently under a wider range of operating conditions, thereby further achieving energy-saving goals and extending equipment life.

[0084] As a specific implementation, it is assumed that during the operation of the energy-saving self-priming pump, the link anomaly flag is not set, indicating that no serious link failure has occurred. However, the system continuously monitors the following: the standard deviation of the physical timestamp difference of consecutive data packets within the cluster is slightly higher than the normal range, but does not reach the threshold for triggering a link anomaly; at the same time, the standard deviation of the arrival time interval also shows a slight increasing trend; while the intra-cluster packet sequence consistency index and inter-cluster flow synchronization remain within acceptable ranges.

[0085] In this scenario, the system determines a disturbance vector based on these statistics. For example, a rule can be set: when the standard deviation of the physical timestamp difference and the standard deviation of the arrival time interval both exceed their respective "minor disturbance thresholds," a specific component of the disturbance vector is set to a high value. Upon receiving this high-value component, the system determines that there is a minor but not negligible disturbance in the current data link. Based on this disturbance vector, the system can take preventative measures, such as temporarily blocking the issuance of new automatic control commands to the actuators, or only allowing the issuance of non-critical, low-frequency control commands until these statistical indicators return to normal levels. Simultaneously, the system can generate a low-level warning message to alert maintenance personnel to potential problems in the data link. In this way, even without a serious link interruption, the system can identify and respond to potential instability factors in an early stage, thereby avoiding more serious failures caused by the accumulation of minor disturbances and ensuring the continuous and stable operation of the energy-saving self-priming pump.

[0086] In some embodiments described above, while the method for determining disturbance vectors based on the mean and standard deviation of the physical timestamp differences of consecutive data packets within a data cluster, the mean and standard deviation of the arrival time intervals, the intra-cluster packet order consistency index, and the inter-cluster flow synchronization degree is proposed, in practical applications, simply calculating these statistics may be insufficient to accurately distinguish different types of system disturbances, such as host-side read / write pauses, unstable link-layer transmissions, or management flow concurrency. The lack of a clear judgment logic regarding the correlation between these statistics and specific disturbance types may result in insufficiently refined disturbance vector generation, thereby affecting the accuracy and reliability of control commands issued to the energy-saving self-priming pump actuator. Therefore, this application further proposes a more refined method for determining disturbance vectors. By setting specific judgment conditions, different statistical characteristics are correlated with specific disturbance evidence scores, thereby constructing a disturbance vector that more accurately reflects the system state.

[0087] Specifically, a disturbance vector is determined based on the fluctuation of the physical timestamp difference, the fluctuation of the arrival time interval, the consistency index, and the cross-flow synchronization of consecutive data packets within the cluster. This includes: increasing the host-side read / write pause evidence score when both the standard deviation of the physical timestamp difference and the standard deviation of the arrival time interval are lower than the corresponding thresholds and the packet order consistency index within the cluster is higher than the corresponding thresholds; increasing the link-layer transmission instability evidence score when the standard deviation of the arrival time interval is significantly greater than the standard deviation of the physical timestamp difference, or when there are high-variability arrival intervals and out-of-order packets within the cluster; and increasing the management flow concurrency evidence score when the proportion of data clusters appearing in the same time window for the same type of sensor exceeds the threshold. The link-layer transmission instability evidence score, the host-side read / write pause evidence score, and the management flow concurrency evidence score are combined into the disturbance vector.

[0088] The "standard deviation of physical timestamp difference" refers to the stability of the data packet generation sequence at the source end; a lower standard deviation usually indicates a more stable data generation process. The "standard deviation of arrival time interval" refers to the stability of the data packet arrival sequence during transmission; a lower standard deviation indicates less jitter during data transmission. The "intra-cluster packet order consistency index" measures whether the physical timestamp order of data packets is consistent with the system's receiving order; high consistency indicates no out-of-order transmission. The "host-side read / write pause evidence score" is increased when data transmission is intermittently interrupted due to host-side read / write operations; in this case, both the physical timestamp difference and the standard deviation of the arrival time interval are low, and packet order consistency is high. The "link-layer transmission instability evidence score" is increased when data transmission at the link layer is unstable (e.g., packet loss, retransmission, congestion); in this case, the standard deviation of the arrival time interval is significantly greater than the standard deviation of the physical timestamp difference, or there is high variability in arrival intervals and out-of-order transmission within the cluster. "Management Flow Concurrency Evidence Score" refers to a score that is increased when a large amount of management or control flow data is transmitted concurrently with sensor data, causing sensor data to queue and form data clusters in the network. Specifically, the score is increased when the proportion of data clusters of the same type of sensor appearing within the same time window exceeds a preset threshold. "Disturbance Vector" is a multi-dimensional vector whose components represent different types of disturbance evidence scores. By combining these scores, the disturbance state of the current system can be comprehensively and precisely described.

[0089] This application's solution achieves refined determination of disturbance vectors by setting specific judgment conditions for different types of system disturbances and associating them with corresponding evidence scores. For example, when the standard deviation of the physical timestamp difference and the standard deviation of the arrival time interval are both below the corresponding thresholds, and the intra-cluster packet order consistency index is above the corresponding threshold, this usually indicates that the data is stable in the initial stage of generation and transmission at the source end, but there may be read / write pauses at the host end, resulting in intermittent data transmission. In this case, the evidence score for host-end read / write pauses is increased. When the standard deviation of the arrival time interval is significantly greater than the standard deviation of the physical timestamp difference, or when there are high-variability arrival intervals and out-of-order packets within the cluster, this directly points to instability in the link layer transmission, thus increasing the evidence score for link layer transmission instability. Furthermore, when the proportion of data clusters appearing in the same time window for the same type of sensor exceeds the threshold, it indicates that there may be resource contention or network congestion caused by management flow concurrency. In this case, the evidence score for management flow concurrency is increased. In this way, the system can accurately identify potential disturbance types based on the subtle characteristics of the data flow and quantify them into different components of the disturbance vector, thereby providing a more accurate basis for subsequent decision-making.

[0090] In some preferred embodiments, it is assumed that the sensor data acquisition system of the energy-saving self-priming pump experiences a slight delay in reading and sending sensor data packets due to a brief log write operation performed by the host at a certain moment. During this period, the sensor itself continues to generate data at its expected frequency, so the standard deviation of the physical timestamp difference within the data packets remains at a low level. When the host resumes normal read / write operations, these data packets are quickly sent out, forming a data cluster with a relatively low standard deviation of arrival time interval. Furthermore, because the data is not out of order, the packet order consistency index within the cluster remains high. In this case, according to the scheme of this application, the system will detect that both the standard deviation of the physical timestamp difference and the standard deviation of the arrival time interval are below the corresponding thresholds, and the packet order consistency index within the cluster is above the corresponding thresholds, thereby improving the host-side read / write pause evidence score. At this time, the link layer transmission instability evidence score and the management flow concurrency evidence score may remain at a low level. Ultimately, the combined disturbance vector will clearly indicate that the current system mainly experiences disturbances due to host-side read / write pauses, rather than link transmission problems or management flow concurrency problems, enabling the system to take targeted actions or issue alarms.

[0091] This application also discloses a big data-based alarm system for an energy-saving self-priming pump, wherein the energy-saving self-priming pump includes an actuator, such as... Figure 2 As shown, the system includes:

[0092] The acquisition module 201 is used to acquire data packets from multiple sensors in real time, and collect the arrival time and physical timestamp inside the data packets;

[0093] The feature determination module 202 determines the arrival features of the plurality of sensors based on the arrival time and the physical timestamp; the arrival features include arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between the physical timestamp and the arrival time;

[0094] The anomaly detection module 203 is used to perform anomaly detection based on the arrival feature. In response to the detection that the arrival feature exhibits a predetermined abnormal mode within multiple consecutive time detection windows, the link anomaly flag is set; in response to the failure to detect that the arrival feature exhibits a predetermined abnormal mode within multiple consecutive time detection windows, the link anomaly flag is released.

[0095] The alarm module 204 is used to prevent the issuance of control commands to the actuator and to execute an alarm if the link abnormality flag is set.

[0096] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A big data-based energy-saving self-priming pump abnormal state alarm method, the energy-saving self-priming pump comprising an actuator, characterized in that, The method includes: Data packets from multiple sensors are acquired in real time, and the arrival time and physical timestamp of the data packets are collected. Arrival characteristics of the plurality of sensors are determined based on the arrival time and physical timestamp; the arrival characteristics include arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between the physical timestamp and the arrival time; the plurality of sensors include a current sensor, a pressure sensor, and a vibration sensor; Anomaly detection is performed based on the arrival characteristics. In response to the arrival characteristics exhibiting a predetermined abnormal pattern within multiple consecutive time detection windows, a link anomaly flag is set. In response to the arrival characteristics not exhibiting a predetermined abnormal pattern within multiple consecutive time detection windows, the link anomaly flag is released. The anomaly detection based on the arrival characteristics includes: within a time detection window, if the following conditions are met, a predetermined abnormal pattern is determined to have occurred: the number of arriving current sensor data packets meets the expected arrival frequency; and at least one of the pressure sensor or vibration sensor has a continuous period of no data arrival exceeding a first preset threshold; and after the continuous period of no data arrival, a batch of arriving data packets appears, with an in-cluster arrival count greater than a second preset threshold and an in-cluster arrival duration less than a third preset threshold; and the median difference between the physical timestamp and the actual arrival time of the batch of arriving data packets is greater than a fourth preset threshold. If the link anomaly flag is set, control commands will be prevented from being sent to the actuator, and an alarm will be triggered. In response to the link anomaly flag being set, the anomaly judgment logic is switched to robust logic mode; wherein, in robust logic mode, statistical information of a single sensor within a predetermined time window is used for anomaly judgment, instead of using the arrival time relationship between data packets from different sensors; wherein, the statistical information includes energy integral, short-term mean surge duration, and instantaneous peak frequency; The real-time acquisition of data packets from multiple sensors, and the collection of the arrival time and physical timestamp within the data packets, further includes: Set a silence period threshold based on the expected data arrival interval of the sensor; When the interval between the arrival of the next data packet and the arrival of a data packet exceeds the silence period threshold, a new data cluster is determined to have started. Within the data cluster, if the arrival interval of consecutive data packets is less than the expected arrival interval multiplied by a reduction factor, it is determined that the data cluster is still continuing. A data cluster is used to determine the arrival feature only if the number of data packets contained in the identified data cluster is greater than or equal to a preset minimum cluster size threshold.

2. The method for alarming abnormal states of an energy-saving self-priming pump based on big data as described in claim 1, characterized in that, If the link anomaly flag is set, control commands will be prevented from being sent to the actuator, and an alarm will be triggered, including: Before generating automatic control commands, check the link anomaly flag and the physical timestamp span of data packets within the current data window; If the link anomaly flag is set or the physical timestamp span exceeds a preset span threshold, then control commands are prevented from being sent to the actuator, and alarm information is generated to execute the alarm; wherein, the alarm information includes the set state of the link anomaly flag and the arrival characteristics.

3. The energy-saving self-priming pump abnormal status alarm method based on big data according to claim 1, characterized in that, The real-time acquisition of data packets from multiple sensors, and the collection of the arrival time and physical timestamp within the data packets, further includes: Calculate the mean and standard deviation of the physical timestamp differences of consecutive data packets within a data cluster; Calculate the mean and standard deviation of the arrival time intervals of consecutive data packets within a data cluster; Based on the consistency between the physical timestamp order and the system arrival order, calculate the intra-cluster packet order consistency index; The cross-flow synchronization degree of clusters is calculated based on the proportion of data clusters appearing in other sensors within the same time period.

4. The energy-saving self-priming pump abnormal status alarm method based on big data according to claim 3, characterized in that, The method further includes: if the link anomaly flag is not set, then a disturbance vector is determined based on the standard deviation of the physical timestamp difference of consecutive data packets within the data cluster, the standard deviation of the arrival time interval, the consistency index, and the cluster cross-flow synchronization degree. The perturbation vector may or may not allow the issuance of control commands to the actuator.

5. The energy-saving self-priming pump abnormal status alarm method based on big data according to claim 3, characterized in that, Based on the fluctuations in the physical timestamp differences of consecutive data packets within the cluster, the fluctuations in the arrival time intervals, the consistency index, and the cluster cross-flow synchronization, a perturbation vector is determined, including: When the standard deviation of the physical timestamp difference and the standard deviation of the arrival time interval are both lower than the corresponding threshold and the intra-cluster packet order consistency index is higher than the corresponding threshold, the host-side read / write pause evidence score is increased. When the standard deviation of the arrival time interval is significantly greater than the standard deviation of the physical timestamp difference, or when there are highly variable arrival intervals and out-of-order arrivals within the cluster, the link layer transmission instability evidence score is increased. When the proportion of data clusters appearing in the same time window by similar sensors exceeds a threshold, the concurrent evidence score of the management flow is increased. The link layer transmission instability evidence score, the host-side read / write pause evidence score, and the management flow concurrency evidence score are combined into the disturbance vector.

6. A big data-based alarm system for an energy-saving self-priming pump, used to implement the method described in any one of claims 1-5, wherein the energy-saving self-priming pump includes an actuator, characterized in that, The system includes: The acquisition module is used to acquire data packets from multiple sensors in real time, and to collect the arrival time and physical timestamp inside the data packets. The feature determination module determines the arrival features of the multiple sensors based on the arrival time and physical timestamp; the arrival features include arrival count, arrival interval, intra-cluster arrival count, intra-cluster arrival duration, and the difference between the physical timestamp and the arrival time; The anomaly detection module is used to perform anomaly detection based on the arrival feature. In response to the arrival feature being detected to exhibit a predetermined abnormal pattern within multiple consecutive time detection windows, the link anomaly flag is set; in response to the arrival feature not being detected to exhibit a predetermined abnormal pattern within multiple consecutive time detection windows, the link anomaly flag is released. The alarm module is used to prevent the issuance of control commands to the actuator and to execute an alarm if the link abnormality flag is set.

Images