A pump room automatic monitoring and early warning method and system

Abstract

The present application relates to the field of monitoring and early warning technology, and discloses a method and system for automated monitoring and early warning of a pump room. The method includes: collecting hydraulic pulsation, cavitation acoustic emission, and bearing temperature rise data through multiple sensors; obtaining a pump efficiency attenuation vector based on characteristics optimized by an environmental adaptive algorithm; predicting the health status of the equipment using a hydraulic model; using a fault propagation network for deep learning classification; and combining water inflow-pump efficiency coupling judgment to achieve a four-level graded early warning. The present application solves the technical problem in the prior art of being unable to accurately identify gradual failures and multi-parameter coupling failure modes of pump room equipment under complex underground working conditions, thereby improving the accuracy of fault prediction and the timeliness of early warning in underground drainage pump rooms.

Description

Technical Field

[0001] The present application relates to the field of monitoring and early warning technology, and in particular to a method and system for automated monitoring and early warning of a pump room. Background Art

[0002] Existing automated monitoring and early warning technologies for pump rooms primarily rely on single-sensor data and fixed thresholds. These systems deploy water level sensors, pressure sensors, temperature sensors, and other devices within the pump room to collect operating parameters. Alarm signals are triggered when these parameters exceed preset thresholds. Traditional systems typically utilize a programmable logic controller (PLC) or distributed control system (DCS) as the core control system, employing simple logical judgments and comparison operations to implement equipment status monitoring and alarm functions. Existing monitoring and early warning methods primarily include equipment performance evaluation based on empirical formulas, fault identification based on statistical analysis, and diagnostic decision-making based on expert systems. These methods have, to a certain extent, achieved automated monitoring and basic early warning functions for pump room equipment.

[0003] However, the traditional single-sensor monitoring method cannot fully reflect the complex operating status of the equipment, especially in the complex environment of the mine. Equipment failures are often manifested as coupled changes in multiple parameters, and the threshold judgment of a single parameter is prone to false alarms and missed alarms. Secondly, the alarm mechanism with a fixed threshold lacks the ability to adapt to environmental changes. Changes in environmental factors such as underground temperature, humidity, and water quality will affect sensor accuracy and equipment performance, but the existing system cannot dynamically adjust the judgment criteria to adapt to environmental changes. In addition, traditional methods mainly rely on post-event alarms rather than predictive maintenance. They can only alarm when the equipment has already obviously failed. They cannot identify the gradual failure and potential risks of the equipment in advance, resulting in passive maintenance and irreversible equipment damage.

[0004] Based on the above analysis, it can be seen that the problem with existing technologies is that they lack multidimensional data fusion and intelligent analysis capabilities, and are unable to accurately identify and predict pump room equipment failure modes under complex underground working conditions. Specifically, existing technologies lack hydraulic-mechanical coupled monitoring methods tailored to the specific underground environment, and are unable to effectively capture the characteristic information of typical underground faults such as cavitation, wear, and lubrication failure. They lack environmentally adaptive variable selection and feature optimization algorithms, and are unable to dynamically adjust the weight and importance of monitoring parameters based on changes in the underground environment. They lack deep learning-based fault pattern recognition capabilities, and are unable to learn and identify complex fault evolution patterns and multi-fault coupling patterns. They also lack early warning mechanisms that couple water inflow with equipment performance analysis, and are unable to comprehensively consider the impact of changing hydrological conditions on equipment operating status, resulting in insufficient early warning accuracy and timeliness. Summary of the Invention

[0005] The present application provides a pump room automated monitoring and early warning method and system, which is used to solve the technical problem in the existing technology that it is impossible to accurately identify the gradual failure and multi-parameter coupling failure mode of pump room equipment under complex working conditions underground, thereby improving the accuracy of underground drainage pump room fault prediction and the timeliness of early warning.

[0006] In a first aspect, the present application provides a pump room automated monitoring and early warning method, the pump room automated monitoring and early warning method comprising: performing multi-dimensional data acquisition and processing on an underground drainage pump room through a hydraulic vibration sensor, an impeller cavitation acoustic sensor, and a bearing temperature rise sensor to obtain hydraulic pulsation amplitude, cavitation acoustic emission intensity, and bearing temperature rise data;

[0007] Performing feature optimization processing on the hydraulic pulsation amplitude, cavitation acoustic emission intensity, and bearing temperature rise data according to an adaptive variable selection algorithm for downhole environment to obtain a pump efficiency attenuation feature vector;

[0008] The pump efficiency attenuation characteristic vector is processed by a hydraulic attenuation model to perform fault mode modeling to obtain water pump health status prediction data;

[0009] Performing deep learning training on the water pump health status prediction data based on the downhole water pump fault propagation network to obtain a fault identification and classification result;

[0010] The fault identification and classification results are used to perform graded early warning processing on the pump room operation status through water inflow-pump efficiency coupling judgment, and a four-level early warning output signal is obtained.

[0011] In a second aspect, the present application provides a pump room automation monitoring and early warning system, the pump room automation monitoring and early warning system comprising:

[0012] The acquisition module is used to collect and process multi-dimensional data of the underground drainage pump room through hydraulic vibration sensors, impeller cavitation acoustic sensors and bearing temperature rise sensors to obtain hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data;

[0013] an optimization module for performing feature optimization processing on the hydraulic pulsation amplitude, cavitation acoustic emission intensity, and bearing temperature rise data according to an adaptive variable selection algorithm for downhole environment to obtain a pump efficiency attenuation feature vector;

[0014] a modeling module, configured to perform fault mode modeling processing on the pump efficiency attenuation characteristic vector through a hydraulic attenuation model to obtain pump health status prediction data;

[0015] A training module is used to perform deep learning training on the water pump health status prediction data according to the downhole water pump fault propagation network to obtain a fault identification and classification result;

[0016] The early warning module is used to perform graded early warning processing on the pump room operation status based on the fault identification and classification results through water inflow-pump efficiency coupling judgment, and obtain a four-level early warning output signal.

[0017] In the third aspect, a pump room automation monitoring and early warning device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory to enable the pump room automation monitoring and early warning device to execute the above-mentioned pump room automation monitoring and early warning method.

[0018] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium is run on the computer, the computer executes the above-mentioned pump room automatic monitoring and early warning method.

[0019] The technical solution provided in this application utilizes a hydraulic vibration sensor, an impeller cavitation acoustic sensor, and a bearing temperature rise sensor to collect multidimensional data from underground drainage pump rooms. This enables simultaneous monitoring of hydraulic pulsation amplitude, cavitation acoustic emission intensity, and bearing temperature rise data. Compared to the existing single-sensor monitoring method, this method can comprehensively capture the multiple fault characteristics of underground pump room equipment under complex hydrogeological conditions, effectively avoiding the information loss and misjudgment problems caused by single-parameter monitoring. The downhole environment adaptive variable selection algorithm dynamically adjusts sensor parameter weights based on real-time changes in downhole temperature, humidity, and pH. This overcomes the limitations of traditional fixed-weight methods that are unable to adapt to environmental changes, ensures accurate extraction of key fault characteristics even in harsh downhole environments, and significantly improves the environmental adaptability and accuracy of feature extraction. The hydraulic attenuation model establishes pump efficiency attenuation laws based on the cavitation loss coefficient and the impeller wear loss coefficient. This model accurately describes the impact of downhole-specific fault modes such as cavitation, wear, and lubrication failure on equipment performance, providing a solid theoretical foundation for fault prediction and possessing greater physical significance and prediction accuracy than traditional empirical formulas.

[0020] A fault propagation network for underground water pumps employs deep learning techniques to train and process pump health prediction data. The network is capable of automatically learning and identifying complex fault patterns and their evolution patterns, overcoming the technical bottleneck of traditional rule-based expert systems, which are unable to handle complex nonlinear relationships. It accurately classifies and identifies normal conditions, minor faults, moderate faults, and severe faults, achieving significantly higher classification accuracy than traditional methods. A coupled water inflow-pump efficiency judgment mechanism innovatively correlates water inflow conditions with equipment performance, comprehensively considering the impact of changing underground hydrological conditions on equipment operation. This shift from passive, post-event alarms to proactive, predictive, and early warning systems is achieved. Four levels of early warning output signals provide a hierarchical management strategy for varying degrees of fault risk, making maintenance decisions more scientific and rational. Specifically for the specific application of underground coal mine drainage, the algorithm model fully considers the unique characteristics of the underground environment and the complexity of water pump failures. By integrating multi-sensor data fusion, environmentally adaptive processing, and deep learning analysis, it significantly improves the accuracy of fault identification and the timeliness of early warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 This is a schematic diagram of an embodiment of the pump room automatic monitoring and early warning method in the embodiment of the present application;

[0023] Figure 2 This is a schematic diagram of an embodiment of the pump room automatic monitoring and early warning system in the embodiment of the present application;

[0024] Figure 3 It is a schematic block diagram of the structure of the pump room automatic monitoring and early warning equipment in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The embodiments of the present application provide a method and system for automated monitoring and early warning of a pump room. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or that are inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the pump room automatic monitoring and early warning method includes:

[0027] Step S101: multi-dimensional data acquisition and processing is performed on the underground drainage pump room through a hydraulic vibration sensor, an impeller cavitation acoustic sensor, and a bearing temperature rise sensor to obtain hydraulic pulsation amplitude, cavitation acoustic emission intensity, and bearing temperature rise data;

[0028] Step S102: performing feature optimization processing on the hydraulic pulsation amplitude, cavitation acoustic emission intensity, and bearing temperature rise data according to a downhole environment adaptive variable selection algorithm to obtain a pump efficiency attenuation feature vector;

[0029] Step S103: Perform fault mode modeling on the pump efficiency attenuation feature vector using a hydraulic attenuation model to obtain pump health status prediction data;

[0030] Step S104: performing deep learning training on the water pump health status prediction data according to the downhole water pump fault propagation network to obtain a fault identification and classification result;

[0031] Step S105: Perform graded early warning processing on the pump room operation status based on the fault identification and classification results through water inflow-pump efficiency coupling judgment, and obtain a four-level early warning output signal.

[0032] It is understandable that the execution subject of this application can be a pump room automatic monitoring and early warning system, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0033] Specifically, multi-dimensional data acquisition is achieved through a dedicated sensor network in the underground drainage pumphouse. The hydraulic vibration sensor integrates a piezoelectric vibration detection unit and a water pressure pulsation measurement unit to simultaneously monitor mechanical vibration and hydraulic pulsation signals induced by water flow. The piezoelectric vibration detection unit converts mechanical vibration into an electrical signal, while the water pressure pulsation measurement unit directly measures changes in water pressure. After fusing these data, frequency domain analysis is used to extract the hydraulic pulsation amplitude, which reflects the intensity changes in the interaction between the pump impeller and the water flow. The impeller cavitation acoustic sensor uses broadband acoustic emission technology to capture high-frequency acoustic emission signals generated by impeller cavitation. When cavitation bubbles form and implode on the surface of the pump impeller, they generate acoustic signals in a specific frequency band. The sensor converts these acoustic signals into digital signals and calculates the acoustic emission intensity value, which directly reflects the severity of impeller cavitation. The bearing temperature rise sensor uses a thermistor to measure bearing seat temperature changes. When the bearing is worn or poorly lubricated, additional heat is generated, causing the temperature to rise. The sensor continuously records temperature data and calculates the temperature rise rate.

[0034] The downhole environment adaptive variable selection algorithm first establishes a water inflow condition classification matrix, classifying downhole conditions into different levels based on real-time water inflow. Each level corresponds to a specific range of pump operating parameters. The algorithm then extracts key inflection point parameters from the pump head-flow characteristic curve, including the optimal efficiency point flow rate, shutoff head, maximum flow rate, and the efficiency drop inflection point. These parameters describe the pump's performance characteristics under different operating conditions. A principal component analysis algorithm calculates the correlation between each sensor parameter and pump efficiency degradation, generating a weight matrix. Each weight coefficient represents the degree of influence of the corresponding sensor parameter on the change in pump efficiency. Because changes in temperature, humidity, and water quality in the downhole environment can affect sensor accuracy, the algorithm dynamically modifies the weight matrix, taking into account the effects of temperature changes on the sensitivity of piezoelectric sensors, the effects of humidity changes on the performance of electronic components, and the effects of pH changes on the response characteristics of acoustic sensors. The modified weight coefficients are used to select the most representative characteristic variables to form a pump efficiency degradation feature vector.

[0035] The hydraulic attenuation model establishes pump efficiency degradation based on pump similarity laws and impeller wear theory. The model decomposes pump efficiency degradation into two main components: cavitation loss and wear loss. The cavitation loss coefficient is calculated using the cavitation coefficient, which reflects the relationship between pump inlet conditions and impeller geometry. A lower cavitation coefficient indicates increased cavitation. The wear loss coefficient is calculated based on water corrosivity and particle concentration. Higher particle concentrations in the water increase impeller erosion wear, and deviations from neutral pH accelerate metal corrosion. Polynomial fitting is used to describe the degradation of the pump head-flow characteristic curve over time, with the fitting coefficient reflecting the changing shape of the characteristic curve. Bearing water lubrication theory analyzes the changes in bearing lubrication behavior in high-humidity underground environments. When water enters the bearing cavity, it dilutes the grease and reduces lubrication effectiveness, leading to increased bearing temperature and increased vibration.

[0036] The downhole pump fault propagation network utilizes a multi-layer neural network architecture to handle complex fault pattern recognition tasks. The input layer receives multidimensional data from the pump efficiency attenuation feature vector, including key parameters such as the impeller cavitation acoustic emission intensity, bearing temperature rise rate, and hydraulic vibration main frequency. The hidden layer uses the ReLU activation function for nonlinear transformation. The ReLU function resets negative inputs to zero while maintaining positive values, effectively avoiding the vanishing gradient problem and accelerating network convergence. The network adjusts connection weights using a backpropagation algorithm. The cross-entropy loss function is used to calculate the error for classification tasks, and the mean squared error loss function is used for regression tasks. A weighted combination of the two forms a total loss function to guide network training. The four nodes in the output layer correspond to normal state, minor fault, moderate fault, and severe fault, respectively. Each node outputs a probability value for the corresponding state.

[0037] A multidimensional judgment matrix is ​​established for the coupled water inflow and pump efficiency judgment. The rows represent different water inflow levels, the columns represent different pump efficiency ranges, and the matrix element values ​​indicate the warning level under the corresponding operating conditions. A level 1 warning is triggered when the water inflow is low and the pump efficiency is above the set threshold, indicating that the pump is operating under ideal conditions and in good condition. A level 2 warning is triggered when the water inflow increases or the pump efficiency drops to a moderate range, indicating that equipment performance is beginning to deteriorate and requiring attention. A level 3 warning is triggered when the water inflow reaches a high level and the pump efficiency drops significantly, indicating a significant risk of equipment failure and requiring prompt maintenance. A level 4 warning is triggered when the water inflow exceeds the design limit or the pump efficiency drops below the dangerous threshold, indicating a serious threat of equipment failure and requiring immediate emergency measures. The coupling weight calculation takes into account the impact of water inflow changes on pump efficiency. A sharp increase in water inflow requires a higher warning level, even if the pump efficiency has not yet significantly decreased, because high water inflow conditions increase the equipment load and the risk of failure.

[0038] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0039] The hydraulic vibration sensor collects and processes signals from the piezoelectric vibration detection unit and the water pressure pulsation measurement unit at the water pump inlet to obtain hydraulic excitation spectrum data in the 0.1Hz-5kHz frequency band.

[0040] Perform amplitude extraction processing on hydraulic excitation spectrum data to obtain hydraulic pulsation amplitude;

[0041] The impeller cavitation acoustic sensor is used to perform broadband acoustic emission detection in the 20kHz-200kHz frequency range of the impeller housing to obtain the acoustic characteristic signal of bubble burst.

[0042] The acoustic emission intensity of the bubble burst acoustic characteristic signal is calculated and processed to obtain the cavitation acoustic emission intensity;

[0043] The bearing seat temperature change is monitored and processed with a resolution of 0.1°C based on the bearing temperature rise sensor to obtain the bearing temperature rise data.

[0044] Specifically, the hydraulic vibration sensor is deployed at a key location at the pump's water inlet, integrating two independent measurement units for coordinated monitoring. The piezoelectric vibration detection unit is made of piezoelectric ceramic. When the pump generates mechanical vibrations, the piezoelectric material deforms under stress. This deformation alters the charge distribution within the material, generating a corresponding voltage signal whose amplitude is proportional to the vibration intensity. The water pressure pulsation measurement unit uses a strain gauge pressure sensor. The elastic diaphragm within the sensor undergoes minute displacements due to changes in water pressure. This displacement is converted by the strain gauge into a resistance change, ultimately outputting a voltage signal. The two units synchronously collect data, then amplify and filter it through a signal conditioning circuit. The sampling frequency is set to 10,000 times per second, ensuring full signal capture within the 0.1 Hz to 5 kHz frequency range. A digital signal processor performs a fast Fourier transform on the acquired time-domain signal, converting it into a frequency-domain spectrum. The spectrum contains the amplitude and phase information for each frequency component, with the amplitude information reflecting the vibration energy distribution at that frequency. The amplitude extraction process of hydraulic excitation spectrum data adopts a method that combines peak detection algorithm and power spectrum density analysis. The peak detection algorithm identifies local maximum points in the spectrum data. These peak points correspond to the main excitation frequency components during the operation of the pump, including the impeller pass frequency, shaft frequency and each order harmonic frequency. Power spectrum density analysis calculates the energy density distribution in each frequency band, and obtains the total energy value in a specific frequency band through integration operation. The hydraulic pulsation amplitude is obtained by weighted averaging the amplitudes corresponding to the main excitation frequencies. The weight coefficient is determined according to the degree of influence of each frequency component on the hydraulic performance of the pump. The impeller pass frequency has the highest weight because it directly reflects the interaction strength between the impeller and the water flow. The weight of the shaft frequency and its harmonics is relatively low, but its contribution to the overall vibration level still needs to be considered.

[0045] The impeller cavitation acoustic sensor is mounted on the impeller casing and uses broadband acoustic emission technology to monitor the acoustic signals generated by cavitation. During cavitation, dissolved gases in water precipitate in low-pressure areas, forming bubbles. These bubbles, transported by the water flow to high-pressure areas, instantly burst, releasing a large amount of energy. This bursting process generates high-frequency acoustic waves ranging from 20,000 to 200,000 Hz. The acoustic sensor uses a piezoelectric ceramic transducer to convert acoustic vibrations into electrical signals. The transducer's surface is coated with a special coating to enhance its sensitivity to high-frequency signals. The signal conditioning circuit comprises a preamplifier, a bandpass filter, and an automatic gain control module. The preamplifier amplifies the weak acoustic emission signal to a processable range, while the bandpass filter filters out-of-band interference noise. The automatic gain control module dynamically adjusts the amplification factor based on signal strength to prevent signal saturation. The digitized acoustic emission signal is then analyzed using a time-frequency analysis algorithm to extract characteristic parameters of the bubble burst, including the frequency, duration, and energy release of the burst event.

[0046] The acoustic emission intensity of the acoustic characteristic signal of bubble collapse is calculated using a combination of the energy integration method and the event counting method. The energy integration method calculates the total energy of the signal within the time window by integrating the square of the acoustic emission signal amplitude over time. The integral result reflects the total energy released by the cavitation bubble collapse. The event counting method counts the number of acoustic emission events exceeding a preset threshold. Each bubble collapse corresponds to an acoustic emission event, and the number of events is positively correlated with the cavitation intensity. The acoustic emission intensity comprehensively considers the energy density and event frequency. The final cavitation acoustic emission intensity value is obtained through weighted calculation. This value quantitatively describes the severity of the cavitation phenomenon on the impeller surface. The larger the value, the more destructive the cavitation is to the impeller.

[0047] The bearing temperature-rise sensor uses a platinum resistance thermometer to achieve high-precision temperature measurement. The resistance value of the platinum resistance material is linearly related to temperature, and the resistance change corresponding to each degree Celsius change in temperature is fixed and highly repeatable. The sensor is enclosed in a waterproof and explosion-proof housing and directly contacts the outer surface of the bearing seat, sensing temperature changes within the bearing through heat conduction. The measurement circuit uses a four-wire connection method to eliminate the impact of lead resistance on measurement accuracy. A constant current source provides a stable excitation current, and the voltage measurement circuit detects the voltage change across the platinum resistance and converts it into a digital temperature value. The temperature data acquisition cycle is set to ten times per second, and a sliding average filter algorithm is used during continuous monitoring to eliminate the impact of short-term temperature fluctuations. The bearing temperature-rise data is calculated by taking the difference between the current temperature value and the reference temperature value. The reference temperature value is the steady-state temperature under normal operating conditions of the bearing. The temperature-rise data is updated in real time and stored in a circular buffer.

[0048] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0049] Based on the water inflow classification matrix, the underground water inflow conditions are divided into four levels: low flow, medium-low flow, medium-high flow, and high flow, and the water inflow condition classification labels are obtained;

[0050] According to the classification label of water inrush conditions, the key inflection point parameters of the pump head-flow characteristic curve are extracted and processed to obtain the best efficiency point flow, shut-off head, maximum flow and efficiency drop inflection point parameters;

[0051] The hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data are input into the principal component analysis algorithm for correlation weight calculation and processing to obtain the sensor parameter weight matrix;

[0052] Dynamically correct the sensor parameter weight matrix for changes in downhole temperature, humidity, and pH value to obtain the environmental adaptive weight coefficient.

[0053] The core characteristic variables are screened and optimized based on the environmental adaptive weight coefficient to obtain the pump efficiency attenuation characteristic vector.

[0054] Specifically, the process of establishing a water inflow classification matrix first determines the classification criteria based on the underground hydrogeological conditions and pump design parameters. The matrix uses a four-row, four-column structure, corresponding to four levels: low flow, medium-low flow, medium-high flow, and high flow. The low flow level corresponds to normal underground seepage conditions, with water inflow typically ranging from 0 to 50 cubic meters per hour. During this period, the pump operates at partial load, with relatively low efficiency but minimal burden on the equipment. The medium-low flow level corresponds to mild water inflow conditions, with water inflow ranging from 50 to 100 cubic meters per hour. The pump begins to enter a more ideal operating range, with gradually increasing efficiency. The medium-high flow level corresponds to moderate water inflow conditions, with water inflow ranging from 100 to 200 cubic meters per hour. The pump operates near the design operating condition, achieving high efficiency, but with significantly increased equipment load. The high flow level corresponds to heavy or sudden water inflow conditions, with water inflow exceeding 200 cubic meters per hour. The pump operates under overload conditions, with reduced efficiency and a greater risk of equipment failure. Grading processing is achieved by comparing real-time water inflow data with preset thresholds. When the water inflow exceeds the upper threshold of a certain level, it automatically switches to a higher level. Each level corresponds to a specific water inflow condition classification label for subsequent parameter extraction and analysis processing.

[0055] The pump head-flow characteristic curve describes the pump's head output capacity at different flow rates, and the curve shape reflects the pump's hydraulic performance characteristics. The key inflection point parameter extraction process first selects the corresponding characteristic curve data based on the water inflow condition classification label. The operating characteristics of the pump under different water inflow conditions vary and need to be addressed separately. The optimal efficiency point flow rate is determined by the peak of the efficiency curve. This point corresponds to the flow rate at which the pump's efficiency is highest and represents the ideal operating point for pump design. The shutoff head refers to the pump's maximum head output at zero flow. It is obtained by measuring the intercept of the characteristic curve on the vertical axis and reflects the pump's maximum head capacity. The maximum flow rate refers to the pump's maximum flow output at zero head. It is obtained by measuring the intercept of the characteristic curve on the horizontal axis and reflects the pump's maximum displacement capacity. The efficiency decline inflection point is the turning point where the efficiency curve begins to significantly decline from its peak. It is determined by analyzing the change in the curve's slope and marks the point where the pump begins to deviate from its optimal operating range. The inflection point parameters are extracted by combining numerical differentiation and curve fitting. First, cubic spline fitting is performed on the characteristic curve to obtain a smooth continuous function. Then, the precise position of each inflection point is determined by calculating the first-order derivative and the second-order derivative.

[0056] When processing multidimensional sensor data, the principal component analysis algorithm first normalizes the data to eliminate the influence of different physical dimensions. Data on hydraulic pulsation amplitude, cavitation acoustic emission intensity, and bearing temperature rise are converted into standardized data with a mean of zero and a variance of one. The covariance matrix calculation describes the linear correlation between the sensor parameters. The matrix element values ​​reflect the degree of correlation between the two parameters, with positive values ​​indicating positive correlation and negative values ​​indicating negative correlation. Eigenvalue decomposition decomposes the covariance matrix into eigenvalues ​​and eigenvectors. The eigenvalues ​​reflect the variance contribution of the corresponding principal component, while the eigenvectors reflect the weight coefficient of each sensor parameter on that principal component. The weight matrix is ​​composed of eigenvectors arranged by eigenvalue size. The first few principal components generally contain the main information of the data, while the subsequent components are mainly noise. Correlation weights are calculated by the correlation coefficient between each sensor parameter and pump efficiency degradation. A larger absolute value of the correlation coefficient indicates a more significant impact of the parameter on the change in pump efficiency, and the weight coefficient is set accordingly.

[0057] The dynamic correction of the sensor parameter weight matrix considers the impact of downhole environmental factors on sensor performance. The correction process is based on real-time monitoring data of environmental parameters. Downhole temperature changes primarily affect the sensitivity of piezoelectric sensors and the operating characteristics of electronic components. As temperature rises, the piezoelectric constant of the piezoelectric material decreases, resulting in a decrease in the sensor output signal amplitude. The correction factor is calculated using the temperature coefficient and the difference between the current temperature and the reference temperature. Downhole humidity changes affect the insulation performance of electronic components and signal transmission quality. High humidity environments are prone to leakage current and signal interference. The correction factor is determined based on the relative humidity value and humidity sensitivity parameter. Changes in water pH affect the response characteristics of acoustic sensors and the degree of corrosion of metal components. Acidic or alkaline environments can alter the sensor surface properties and affect sound wave propagation. The correction factor is calculated based on the pH deviation from neutral and the material corrosion constant. Dynamic correction processing multiplies the original weight coefficient with the various environmental correction factors to obtain the corrected weight value. This correction process is performed in real time to ensure that the weight matrix accurately reflects the sensor reliability under current environmental conditions.

[0058] Core feature variable selection is optimized based on environmentally adaptive weight coefficients. The screening process aims to extract the most representative and reliable feature parameters from multidimensional sensor data. A weight threshold is set to determine the importance of feature variables. Variables with weight coefficients above the threshold are selected as core features, while variables with weight coefficients below the threshold are considered redundant information or noise interference. Feature variables include key parameters such as impeller cavitation acoustic emission intensity, bearing temperature rise rate, hydraulic vibration main frequency offset, real-time pump efficiency, head deviation, and flow fluctuation coefficient. Each parameter reflects the operating status and health of the pump from different perspectives. The screening optimization uses a stepwise regression method. The feature variable with the highest weight is first selected as the basis. Then, additional feature variables are gradually added and the improvement in prediction accuracy is evaluated. Addition of additional variables is stopped when no significant improvement in prediction results is achieved. The pump efficiency attenuation feature vector is composed of the selected core feature variables arranged by weight. The vector length is determined by balancing the required prediction accuracy with computational complexity and typically contains eight to twelve feature variables.

[0059] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0060] The cavitation coefficient is calculated based on the cavitation acoustic emission intensity in the pump efficiency attenuation characteristic vector and the impeller cavitation acoustic sensor data to obtain the cavitation loss coefficient.

[0061] The wear loss coefficient is calculated based on the water pH value and particle concentration parameters in the pump efficiency attenuation characteristic vector to obtain the impeller wear loss coefficient;

[0062] The cavitation loss coefficient and impeller wear loss coefficient are input into the water pump similarity law to model the pump efficiency attenuation law and obtain the time-varying attenuation data of the pump efficiency.

[0063] The degradation parameters of the pump head-flow characteristic curve are obtained by performing polynomial fitting calculation on the time-varying attenuation data of the pump efficiency.

[0064] Based on the degradation parameters of the pump head-flow characteristic curve and the bearing water lubrication theory, a comprehensive health status assessment is performed to obtain the pump health status prediction data.

[0065] Specifically, the cavitation coefficient is calculated based on a quantitative analysis of the hydraulic conditions at the impeller inlet and cavitation acoustic emission data. The cavitation coefficient is a key indicator of a pump's anti-cavitation performance. The calculation process first determines the intensity and frequency of cavitation using real-time data from the impeller cavitation acoustic sensor. The cavitation acoustic emission intensity is derived by comprehensively evaluating the energy integral and frequency of acoustic emission events. The cavitation coefficient is defined as the ratio of the effective NPSH at the impeller inlet to the dynamic pressure head. The effective NPSH is calculated using inlet pressure, steam pressure, and velocity head. Cavitation begins when the effective NPSH falls below a critical value. The cavitation loss coefficient is established by the ratio of the cavitation coefficient to the critical cavitation coefficient. When the actual cavitation coefficient falls below the critical value, the cavitation loss coefficient increases dramatically. The more severe the cavitation, the larger the loss coefficient. The calculation process also considers the impact of water quality on cavitation characteristics. Dissolved gas content, water temperature, and impurity concentration can alter the cavitation characteristics of water. These factors are incorporated into the calculation via correction coefficients.

[0066] The wear loss coefficient is calculated based on a quantitative assessment of water quality analysis data and particle wear theory. The complex and variable downhole water quality causes varying degrees of wear damage to the pump impeller. The pH value reflects the water's acidity or alkalinity. Acidic environments accelerate electrochemical corrosion of metal materials, while alkaline environments can damage the passive film on metal surfaces. The further the pH deviates from neutral, the more corrosive the impeller material becomes. Particle concentration parameters, including number density, particle size distribution, and hardness, are measured using a turbidity meter and a particle counter. Hard particles exert the most significant erosion wear on the impeller surface. The wear loss coefficient is calculated using a modified form of Archard's wear law, accounting for the combined effects of multiple factors, including particle hardness, impact velocity, contact time, and material properties. The synergistic effect of corrosive wear is described by the coupled relationship between the corrosion and wear factors. When both the pH and particle concentration are elevated, material loss is more severe than if both factors were acting alone. The calculation also considers the cumulative effect of operating time. Wear loss increases nonlinearly with operating time, with a slow initial increase and a gradual deterioration.

[0067] The pump similarity law describes the performance relationships of geometrically similar pumps under different operating conditions, providing a theoretical basis for modeling pump efficiency degradation. Similarity laws encompass three fundamental relationships: flow rate similarity, head similarity, and power similarity. These relationships can deviate under the influence of cavitation and wear, requiring correction. Pump efficiency degradation modeling incorporates the cavitation loss coefficient and wear loss coefficient as correction factors into the similarity law. The modified efficiency relationship can describe the performance degradation process of pumps under actual operating conditions. Cavitation loss primarily affects the pump's flow coefficient and head coefficient. Cavitation reduces the effective flow area of ​​the impeller and increases pressure pulsation, thereby reducing the pump's hydraulic efficiency. Wear loss primarily affects the impeller's geometry and surface roughness. Wear leads to reduced blade thickness, altered blade angle, and increased surface roughness, all of which reduce the impeller's hydraulic performance and volumetric efficiency. The modeling process utilizes time series analysis to represent efficiency degradation as a function of time. The function parameters are determined through regression analysis of historical operating data. The time-varying pump efficiency degradation data includes efficiency predictions and confidence intervals at different times, providing a quantitative basis for equipment maintenance decisions.

[0068] Polynomial fitting is used to describe the temporal degradation of a pump's head-flow characteristic curve. The fitting process is based on time-varying pump efficiency degradation data and hydraulic theory. Characteristic curve degradation is primarily manifested by an overall downward shift, slope change, and shape distortion. These changes reflect the degradation of impeller geometry and hydraulic performance. The fitting method utilizes a piecewise polynomial approach, dividing the characteristic curve into low-flow, medium-flow, and high-flow segments. A cubic polynomial is used to describe the relationship between head and flow in each segment. The temporal variation of the polynomial coefficients is determined through time series analysis. The constant term reflects the overall position change of the curve, the linear term coefficient reflects the slope change, and the higher-order term coefficient reflects the curvature change. The quality of the fit is assessed using coefficient of determination and residual analysis. The coefficient of determination reflects the degree to which the fitted curve explains the actual data, while residual analysis verifies the distribution characteristics and systematic biases of the fitting error. Degradation parameters include key indicators such as the maximum head decay rate, the offset of the best efficiency point, the shrinkage of the high-efficiency range, and the curve distortion index. These parameters quantitatively describe the degree of characteristic curve degradation and its development trend.

[0069] Bearing water lubrication theory analyzes the lubrication state and performance degradation patterns of bearings in high-humidity underground environments, providing theoretical support for health assessment. Water in underground environments easily enters the bearing cavity, diluting the grease and altering its lubrication properties. Under water lubrication, bearing friction coefficients, load capacity, and service life can all be significantly altered. Lubrication assessment uses a comprehensive approach based on bearing temperature rise data, vibration characteristics, and grease analysis results. Under normal lubrication conditions, bearing temperature rise is stable and vibration levels are low. Under poor lubrication conditions, temperature rise accelerates and vibration intensifies. Water lubrication theory considers the impact of water film thickness, viscosity variations, and boundary lubrication on bearing performance. Insufficient water film thickness can lead to direct metal contact and increased wear. Comprehensive health assessment combines weighted factors from the head-flow characteristic curve degradation parameters and bearing lubrication state parameters, with weights determined based on the degree of impact of each parameter on overall equipment performance. The assessment results are presented as a health index, ranging from zero to one hundred. Higher values ​​indicate better equipment health. When the health index falls below a set threshold, maintenance recommendations are triggered. Pump health prediction data includes information such as current health, predicted health for the next week, identification of key degradation factors, and maintenance recommendations.

[0070] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0071] The pump health status prediction data is input into the input layer of the multi-layer neural network for data reception and processing, and the input data corresponding to the impeller cavitation acoustic emission intensity, bearing temperature rise rate, and hydraulic vibration main frequency are obtained;

[0072] Based on the input data, the ReLU activation function is calculated and processed on the hidden layer neuron nodes to obtain the nonlinear feature mapping result;

[0073] According to the nonlinear feature mapping results, the fault status of the four nodes in the output layer are classified to obtain the classification probabilities of normal state, minor fault, moderate fault and severe fault;

[0074] Input the classification probability into the cross entropy loss function and the mean square error loss function for weighted combination loss calculation to obtain the network training loss value;

[0075] The network parameters are back-propagated and optimized based on the network training loss value to obtain the fault identification and classification results.

[0076] Specifically, the input layer of the multi-layer neural network is designed to receive and preprocess pump health prediction data. The input layer contains twelve neuron nodes, each corresponding to a different characteristic parameter. The pump health prediction data first undergoes data normalization, converting parameters with different physical dimensions and numerical ranges into standardized values ​​with a mean of zero and a standard deviation of one, ensuring that each characteristic parameter has the same weighting during network training. Impeller cavitation acoustic emission intensity data reflects the severity of cavitation on the impeller surface. The input processing compares the acoustic emission intensity values ​​with historical baseline values ​​and calculates the relative rate of change. This rate of change is then logarithmically transformed to reduce the dynamic range of the data for easier network processing. Bearing temperature rise rate data is obtained by numerically differentiating the bearing temperature time series. The differentiation process uses a central difference scheme to improve computational accuracy. The temperature rise rate reflects the changing trend of the bearing health status, with positive values ​​indicating a temperature increase and negative values ​​indicating a temperature decrease. The hydraulic vibration main frequency data is extracted from the vibration time domain signal using a fast Fourier transform. The main frequency value reflects the primary excitation characteristics of the pump operation, and the main frequency offset is determined by calculating the difference from the design main frequency. The input layer performs a linear weighted combination of each feature parameter, and the weight coefficient is automatically adjusted and optimized during the network training process.

[0077] Calculating the ReLU activation function for hidden layer neurons is a key step in implementing nonlinear transformations in the network. The ReLU function is defined as outputting the same value as the input when the input value is greater than zero, and outputting zero when the input value is less than or equal to zero. The activation function calculation process first applies a linear transformation to the input data. Each hidden layer neuron receives a weighted signal from the input layer. The weighted sum is calculated through matrix multiplication, and the elements of the weight matrix are continuously updated during training. The linear transformation result, after adding a bias term, is then input into the ReLU activation function for nonlinear processing. The advantages of the ReLU function are its computational simplicity and its ability to effectively avoid the vanishing gradient problem. Its asymmetry helps break the network's symmetry and enhance learning ability. The first hidden layer contains sixty-four neurons, the second hidden layer contains thirty-two neurons, and the third hidden layer contains sixteen neurons. The number of layers and nodes balances the network's expressive power and computational complexity. The result of the nonlinear feature mapping is a vector of hidden layer neuron activation values. These activation values ​​encode high-level features of the input data and provide information for classification decisions in the output layer.

[0078] The four nodes in the output layer correspond to the four health states of the pump. Node design is based on the practical needs of fault diagnosis and the hierarchical requirements of maintenance decisions. The normal state node indicates that the pump is operating within design parameters and all indicators are normal. The minor fault node indicates that initial fault signs have appeared but have not yet affected normal operation. The moderate fault node indicates that a fault has manifested and maintenance plans are required. The severe fault node indicates that the fault is severe and requires immediate shutdown for repair. Fault state classification uses the Softmax activation function to convert the output values ​​of the hidden layer into a probability distribution. The Softmax function ensures that the sum of the probability values ​​of the four output nodes is equal to one and that each probability value lies between zero and one. The probability calculation process first performs an exponential transformation on the input value of each node. The exponential value of each node is then divided by the sum of all the exponential values ​​to obtain a normalized probability. The classification probability reflects the network's confidence in each fault state. A higher probability value indicates a greater likelihood of the corresponding state. Classification decisions are made by comparing the probability values, and the state with the highest probability value is typically selected as the network's classification result.

[0079] The cross-entropy loss function measures the difference between the classification prediction and the true label. A smaller function value indicates a closer prediction to the true label. The cross-entropy calculation converts the true label into a one-hot encoding. Each sample label is represented as a four-dimensional vector, with the correct category corresponding to a one and all other zeros. The loss is calculated by taking the negative logarithm of the dot product between the true label vector and the predicted probability vector. The loss is smaller when the predicted probability is close to the true label and larger when the prediction is incorrect. The mean squared error loss function measures the numerical accuracy of the network output. The calculation squares the difference between the predicted probability and the true label and then calculates the mean. The weighted combination loss linearly combines the cross-entropy loss and the mean squared error loss with a set weight ratio. The weight ratio is determined based on the importance of classification accuracy and numerical accuracy. Typically, the cross-entropy loss is weighted at 0.7, and the mean squared error loss is weighted at 0.3. This combination loss function takes into account both classification accuracy and prediction stability. The network training loss is calculated as the average loss across all training samples.

[0080] Backpropagation optimization iteratively updates network parameters based on the gradient descent algorithm, with the goal of minimizing the network training loss. Gradient calculation uses the chain rule, starting from the output layer and propagating layer by layer toward the input layer. The gradient of each layer is determined by multiplying the error signal of the current layer by the output of the previous layer. The weight update formula subtracts the product of the learning rate and the gradient from the current weight value. The learning rate controls the step size of the parameter update. Excessive learning rates can lead to unstable training, while too small learning rates can prolong training time. The optimization process uses the Adam optimizer combined with a momentum mechanism and adaptive learning rate adjustment. The Adam optimizer achieves adaptive parameter updates by maintaining first-order and second-order moment estimates of the gradient.

[0081] The training process is set to a maximum of 500 iterations. Each training round consists of four steps: forward propagation, loss calculation, backpropagation, and parameter update. An early stopping mechanism monitors the loss on the validation set and terminates training early to prevent overfitting if the validation loss does not decrease for 10 consecutive rounds.

[0082] The maximum number of iterations was determined using a progressive testing approach. We ran the model for 100 epochs to observe the downward trend in loss and recorded the epoch number at which the loss stabilized. We then tested the model for 200, 300, 400, and 500 epochs, comparing validation set accuracy and model stability. By plotting the relationship between the number of training epochs and validation accuracy, we found that validation accuracy stabilized around 350 epochs. Taking into account the variability of different data batches, we ultimately set the maximum number of iterations to 500.

[0083] The number of consecutive epochs for the early stopping mechanism was determined through group comparison experiments. The same training data was trained using 5, 8, 10, 12, and 15 epochs of early stopping. The final number of training epochs, validation accuracy, and model generalization performance were recorded for each setting. The experimental results showed that the 5-epoch setting resulted in insufficient training and low accuracy; the 15-epoch setting was prone to overfitting, resulting in decreased performance on the test set; and the 10-epoch setting achieved the optimal balance between sufficient training and preventing overfitting.

[0084] Establish a parameter adjustment experiment record, and record key information such as training date, parameter configuration, dataset version, hardware environment, training time, final accuracy, verification loss, and whether overfitting occurs for each parameter adjustment. The learning rate tuning adopts a gradually narrowing range method. First, select 10 geometric progression points between 0.001 and 0.1 for coarse adjustment. After determining the optimal range, fine-tune within this range. Record the convergence speed and final performance after each adjustment. The selection of batch size needs to consider hardware memory limitations and training stability. By testing four batch sizes of 16, 32, 64, and 128, record the memory usage, single-round training time, gradient update stability, and final model performance under each setting. The network structure tuning adopts the control variable method. First, fix the number of hidden layers and adjust the number of neurons in each layer. Then, fix the number of neurons and adjust the number of layers. The total number of network parameters, training time, and performance indicators are fully recorded for each structural change.

[0085] During training, the model state and training log are saved every 10 epochs, including key metrics such as the current epoch number, training loss, validation loss, learning rate change, and gradient norm. If the training loss continues to decrease but the validation loss begins to increase, an overfitting detection program is immediately initiated, comparing the difference in training and validation loss over the last 10 epochs. If the difference continues to increase, training is terminated early. If gradient explosion or vanishing occurs during training, the learning rate or network initialization method is adjusted to address this. The reason, method, and effect of each adjustment are recorded in detail.

[0086] The original dataset was randomly split into training, validation, and test sets in a ratio of 7:2:1, ensuring consistent distribution of fault categories across the three subsets. All features were normalized before training, and the mean and standard deviation of the training set were calculated. The training, validation, and test sets were transformed using the same normalization parameters. The learning rate started at 0.1 and was automatically reduced to half whenever the validation loss did not improve within five epochs, stopping when the learning rate fell below 0.0001. After each parameter adjustment, a full evaluation process was run on the validation set, calculating accuracy, precision, recall, and F1 score. These metrics were compared with the previous best results. A new configuration was adopted only when it performed at least as well as the previous best across all key metrics and showed significant improvement in at least one metric.

[0087] The fault identification and classification results are obtained by predicting the test data through the trained network. The results include the classification label and corresponding confidence score of each test sample.

[0088] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0089] Based on the fault identification and classification results and real-time pump efficiency data, a coupling judgment matrix is ​​constructed and processed to obtain the water inflow-pump efficiency coupling judgment matrix;

[0090] According to the water inflow-pump efficiency coupling judgment matrix, a first-level warning judgment is performed when the pump efficiency is higher than the set threshold and the fault is identified as normal, and a green normal operation warning signal is obtained;

[0091] The situation in which the pump efficiency in the water inflow-pump efficiency coupling judgment matrix is ​​in the medium range or a minor fault feature is detected is input into the secondary warning judgment for processing, and a yellow performance degradation warning signal is obtained;

[0092] When the pump efficiency in the water inflow-pump efficiency coupling judgment matrix is ​​in a low range or a medium fault feature is detected, a three-level warning judgment process is performed to obtain an orange fault risk warning signal;

[0093] Based on the water inflow-pump efficiency coupling judgment matrix, a four-level warning judgment process is performed when the pump efficiency is lower than the safety threshold or a serious fault feature is detected, and a red serious fault warning signal is obtained as the four-level warning output signal.

[0094] Specifically, a coupled water inflow-pump efficiency judgment matrix was constructed based on a comprehensive judgment framework using multidimensional data fusion and decision theory. The matrix is ​​designed with four rows and four columns, corresponding to different combinations of water inflow levels and pump efficiency ranges. Fault identification and classification results provide a qualitative assessment of the equipment's health status, encompassing four levels: normal, minor, moderate, and severe. Each level corresponds to a different fault probability distribution and maintenance requirements. Real-time pump efficiency data is calculated using simultaneous measurements from a flow meter, pressure transmitter, and power meter. The calculation utilizes the instantaneous power method to comprehensively analyze motor input power, transmission efficiency, and hydraulic power, yielding an accurate value reflecting the pump's current operating efficiency. The coupled judgment process combines the probability values ​​of the fault identification results with the pump efficiency values ​​through a weighted fusion. The fusion weights are determined based on the reliability and importance of the two types of information. The fault identification weight is set to 0.6, and the pump efficiency weight is set to 0.4. Matrix element values ​​are determined using a combination of fuzzy logic and expert systems, with each element representing the overall risk level for the corresponding water inflow and pump efficiency combination. The matrix construction process also considers the impact of the changing trend of water inflow on the judgment results. When the water inflow increases rapidly, the risk level needs to be increased even if the current pump efficiency is normal. When the water inflow is stable or decreasing, the risk assessment can be appropriately lowered.

[0095] Level 1 early warning verification confirms normal operation when the equipment is operating well. This verification requires that the pump efficiency exceeds a set threshold and the fault is identified as normal. The pump efficiency threshold is determined based on pump design parameters and operating experience, typically between 85 and 90 percent of the design efficiency. When the measured pump efficiency exceeds this threshold, it indicates that the pump is operating well within the high-efficiency range. Fault identification of a normal state requires that the neural network output has a normal state probability greater than 0.8 and all other fault state probabilities less than 0.1. This probability distribution indicates that the network has a high confidence level in its normal state judgment. Level 1 early warning verification also requires verification that the water inflow is within the design range and its changing trend is stable. If the water inflow exceeds the design range or experiences a sharp change, the system cannot be identified as normal, even if all other conditions are met. A green normal operation warning signal is displayed on the early warning system's display interface as a green indicator light or green text. The monitoring log also records the timestamp and key parameter values ​​of the normal operation status. The signal output also includes auxiliary information such as equipment operating efficiency, estimated maintenance time, and operational recommendations to provide operators with reference for equipment management.

[0096] Level 2 early warning provides early warning when equipment performance begins to decline but hasn't yet posed a serious threat. Criteria include pump efficiency being within the medium range or the detection of minor fault characteristics. Medium pump efficiency is defined as between 70% and 85% of design efficiency, indicating the pump is still functioning properly but with a decrease in efficiency that warrants attention. Minor fault characteristics are identified by a neural network output showing a probability of minor fault exceeding 0.5. The sum of the probabilities of medium and severe faults must be below 0.3 to ensure the severity of the fault is truly minor. Level 2 early warning processing utilizes a time window analysis method, requiring pump efficiency decline or minor fault characteristics to be detected within three consecutive sampling periods to avoid false alarms due to occasional fluctuations. The judgment process also considers the relationship between water inflow and pump efficiency. A natural decrease in pump efficiency when increased water inflow causes the pump to deviate from its optimal operating point is normal, and corrections should be made based on the pump's characteristic curve. A yellow performance degradation warning signal alerts operators that equipment performance has degraded but is still operational, and recommends a detailed inspection during the next scheduled maintenance. The signal output includes specific parameters of the performance degradation and an analysis of the possible causes.

[0097] Level 3 early warning judgments address situations where the risk of equipment failure has significantly increased, necessitating preventive measures. This judgment is based on the presence of low pump efficiency or the detection of moderate fault characteristics. Low pump efficiency is defined as between 50% and 70% of the design efficiency. This level indicates a significant pump failure, and continued operation could exacerbate equipment damage. Moderate fault characteristics are identified by a medium failure probability output by the neural network exceeding 0.4, indicating that the equipment is experiencing significant fault indicators requiring attention. Level 3 early warning judgments also analyze fault development trends, comparing current failure probability with historical data to determine whether the fault is worsening. A continued increase in failure probability requires a higher warning level. The judgment process considers the coupled effects of multiple failure modes. When bearing failure and impeller wear occur simultaneously, the combined risk is higher than the simple summation of individual faults. An orange fault risk warning signal requires operators to closely monitor equipment status and prepare contingency plans. It recommends scheduling equipment maintenance at an appropriate time to prevent further escalation of the fault. The signal output includes fault type identification, risk level assessment, and recommended maintenance measures, providing technical support for maintenance decisions.

[0098] Level 4 early warning assessments address emergencies requiring immediate action when equipment is threatened with a critical failure. This assessment occurs when pump efficiency falls below a safety threshold or when critical fault characteristics are detected. The safety threshold is set at 50% of the design efficiency. Below this threshold, the pump essentially loses its ability to function normally, and continued operation poses a risk to equipment damage and personnel safety. Critical fault characteristics are identified by a neural network output showing a critical failure probability exceeding 0.3. While not the highest probability, this indicates the possibility of a critical failure. Level 4 early warning assessments utilize a multi-verification mechanism to prevent false alarms and unnecessary downtime. This requires at least two independent detection channels to simultaneously confirm the fault status, using methods including neural network diagnosis, threshold comparison, and trend analysis. The assessment process also considers the urgency of the water inflow. Even if pump efficiency is normal, a Level 4 early warning is triggered to prevent flooding. A red critical fault warning signal requires immediate equipment shutdown and the initiation of emergency response procedures. Signal output utilizes a combination of audible and visual alarms, SMS notifications, and automated control to ensure timely communication. The Level 4 early warning output also includes a fault diagnosis report, emergency response recommendations, and instructions for activating backup equipment.

[0099] In a specific embodiment, the process of executing the step of constructing a coupling judgment matrix based on the fault identification and classification results and the real-time pump efficiency data may specifically include the following steps:

[0100] The current water inflow condition is classified according to the real-time water inflow sensor data, and four water inflow grade identifications are obtained: low water inflow, medium-low water inflow, medium-high water inflow and high water inflow;

[0101] Continuously monitor the current water pump efficiency based on the flow meter, pressure transmitter and power meter data to obtain the real-time pump efficiency percentage value;

[0102] Cross-mapping the normal state, minor fault, moderate fault and severe fault classification in the fault identification classification results with the water inflow level identification to obtain the fault-water inflow correlation matrix;

[0103] The real-time pump efficiency percentage value and the fault-water inflow correlation matrix are coupled with weight calculation to obtain the pump efficiency-fault coupling coefficient.

[0104] The matrix elements are assigned based on the pump efficiency-fault coupling coefficient and the water inflow variation trend to obtain the water inflow-pump efficiency coupling judgment matrix.

[0105] Specifically, the magnitude classification of real-time water inflow sensor data is based on a grading standard established based on downhole hydrogeological conditions and pump design parameters. The classification process utilizes a combination of threshold judgment and sliding average to ensure the stability and accuracy of the classification results. Water inflow sensors are deployed at major water inflow points and catchment areas underground. Ultrasonic level meters and electromagnetic flowmeters monitor water flow in real time. Sensor data undergoes digital filtering to eliminate the effects of short-term fluctuations. The low water inflow level corresponds to normal downhole seepage, with a water inflow range of zero to 50 cubic meters per hour. At this level, the pump operates at partial load, resulting in low equipment pressure and relatively low efficiency. The medium-low water inflow level corresponds to mild water inflow, with a water inflow range of 50 to 100 cubic meters per hour. The pump enters a more ideal operating range, gradually improving efficiency. The medium-high water inflow level corresponds to moderate water inflow, with a water inflow range of 100 to 200 cubic meters per hour. The pump operates near its design operating conditions, achieving high efficiency but significantly increasing equipment load. The high water inflow level corresponds to large or sudden water inflow, with the water inflow exceeding 200 cubic meters per hour. The water pump operates in an overloaded state, its efficiency decreases, and there is a greater risk of failure.

[0106] When 50±5 / h and 100±8 When there is a level overlap phenomenon in the boundary area of ​​45-55, that is, when the water inflow is in these overlapping intervals, the system does not switch the level immediately, but makes a comprehensive judgment based on the water inflow change trend, duration and current operating status of the equipment. / h range and the rate of change is less than 2 / h / min, the system maintains the current level unchanged to avoid frequent switching; when the water inflow shows a clear upward trend in the overlapping interval and lasts for more than 20 minutes, the system enters a higher level in advance to ensure the timeliness of the warning.

[0107] In case of sudden water inrush, even if the water inrush exceeds 200 / h, the system first verifies the validity of the sensor data, eliminates sensor failures or external interference factors, and immediately triggers the emergency response mode after confirming a real water inflow event, rather than the conventional high-inflow level processing process. During the equipment startup and shutdown process, due to the hydraulic transient effect, the water inflow readings will fluctuate abnormally. The system suspends the level switching judgment within the first 10 minutes after detecting the water pump start and stop signal, and uses the stable operating condition data before the start and shutdown for early warning judgment. During sensor maintenance, when the main water inflow sensor goes offline, the system automatically switches to the backup sensor. At the same time, the accuracy requirements for the classification judgment are relaxed, and the original strict threshold is expanded to a fuzzy interval. The uncertainty during this period is marked in the system log.

[0108] The classification process uses a fifteen-minute sliding time window to calculate the average water inflow to avoid frequent switching of levels due to instantaneous fluctuations. The classification results are output in the form of water inflow level identification for easy use in subsequent processing.

[0109] Continuous monitoring of pump efficiency utilizes multi-sensor data fusion and real-time computing technology to achieve high-precision efficiency measurement. The monitoring process comprehensively considers the relationship between hydraulic power, mechanical power, and electrical power. The flowmeter utilizes an electromagnetic measurement principle, measuring the volumetric flow rate of conductive liquids based on Faraday's law of electromagnetic induction. The flowmeter is installed in a straight section of the pump outlet pipe. The upstream and downstream straight sections meet measurement accuracy requirements, achieving a measurement accuracy of 0.5% and a response time of less than one second. Pressure transmitters are installed at the pump inlet and outlet, respectively, to measure suction and discharge pressures. The pressure difference represents the actual pump head. The pressure transmitters utilize diffused silicon technology for high accuracy and long-term stability. A power meter measures the input power of the pump motor, acquiring three-phase current and voltage signals through current transformers and voltage transformers. Power calculations consider the influence of power factor and harmonic content to ensure measurement accuracy. Pump efficiency is calculated using the formula: hydraulic power divided by shaft power. Hydraulic power is calculated from flow rate, head, and liquid density, while shaft power is determined from motor power and transmission efficiency.

[0110] When the calculated pump efficiency value falls within the critical range of 75%-80%, the system doesn't simply classify the fault level according to a fixed threshold. Instead, it conducts a comprehensive assessment based on efficiency trends, operating condition deviations, and historical efficiency baselines. If the pump efficiency remains within this range but shows a stable or increasing trend, and the operating time does not exceed 120% of the design operating conditions, the system tends to maintain a lower fault risk rating. Conversely, if the pump efficiency is within an acceptable range but shows a rapid decline (a rate of decline exceeding 1% per hour), the system raises the fault risk level and increases the monitoring frequency.

[0111] The real-time calculation cycle is set to five seconds. Each calculation result is processed through a Kalman filter to reduce the impact of measurement noise. Pump efficiency percentage values ​​are rounded to two decimal places and displayed as a percentage. The fault identification and classification results are cross-mapped with the water inflow level identifier to establish a multidimensional correlation analysis framework. This mapping process considers the probability and impact of failure modes under different water inflow conditions. The fault classification results include a probability distribution for four states, each corresponding to a different equipment health level and maintenance requirement. The probability value reflects the neural network's confidence in each state. The water inflow level identifier provides information about the current operating conditions. Different water inflow levels correspond to different equipment operating conditions and failure risk levels. The cross-mapping process uses a four-by-four matrix structure, with rows corresponding to four fault states and columns corresponding to four water inflow levels. Each matrix element represents the risk weight for the corresponding fault state and water inflow level combination. The risk weights are determined based on historical fault statistics and expert experience. Normal states have the lowest weight under low water inflow conditions, indicating the lowest risk, while severe faults have the highest weight under high water inflow conditions, indicating the highest risk.

[0112] When the system detects the abnormal combination of "normal status + high water inflow," it doesn't directly conduct a risk assessment based on the pre-set matrix. Instead, it initiates a secondary verification process, focusing on checking sensor calibration status, the confidence interval of the fault identification algorithm, and historical records of similar operating conditions. If the system confirms that the equipment is operating normally under extreme operating conditions, it marks this condition as "borderline normal" and appropriately increases monitoring frequency without triggering a fault warning. If it detects the combination of "serious fault + low water inflow," the system prioritizes the possibility of internal equipment failure, such as bearing damage or impeller wear, which are unrelated to external operating conditions. In this case, low water inflow cannot mask serious equipment problems.

[0113] The mapping calculation performs matrix operations on the fault probability vector and the water inflow level identifier to obtain a modified fault probability distribution that takes into account the influence of operating conditions. The element values ​​of the fault-water inflow correlation matrix are calculated using a weighted average method, with the weight coefficient determined based on the historical frequency and severity of each combination. The pump efficiency percentage value is coupled with the fault-water inflow correlation matrix to establish a quantitative risk assessment model. The calculation process comprehensively considers equipment performance indicators and fault status information. The pump efficiency percentage value provides a direct measurement of the current equipment performance. Lower values ​​indicate worse equipment performance and higher failure risk. The fault-water inflow correlation matrix provides a fault risk assessment that takes into account the influence of operating conditions. Larger matrix element values ​​indicate a higher risk level for the corresponding combination. The coupling weight calculation uses a combination of fuzzy logic and linear weighting. First, the pump efficiency percentage value is converted into a fuzzy set using a membership function. The membership function uses a trapezoidal distribution to describe the relationship between pump efficiency and risk level.

[0114] When the pump efficiency value is within the fuzzy boundary of 60%-65%, the system does not use a deterministic membership function value. Instead, it introduces an uncertainty interval. The membership function exhibits a gradual change characteristic within this range, while also considering the impact of the pump efficiency measurement confidence interval (±2%) on the final weight calculation. If the pump efficiency change rate is abnormal (the absolute value exceeds 5% / hour), the system automatically expands the fuzzy boundary range from the original ±2.5% to ±5% to accommodate the rapid changes in equipment status.

[0115] The weight calculation formula multiplies the fuzzy value of pump efficiency by the fault-inflow correlation matrix to produce a coupling coefficient that comprehensively considers both performance and fault factors. The calculation also considers the rate of change of pump efficiency. Rapidly declining pump efficiency requires an increase in risk weight, even if the current value is acceptable. The rate of change is determined by calculating the derivative of the pump efficiency time series. The pump efficiency-fault coupling coefficient ranges from zero to ten, with zero representing no risk and ten representing extremely high risk. The coefficient is mapped to the warning level using piecewise linear interpolation.

[0116] Matrix element assignment is based on a comprehensive assessment of the pump efficiency-fault coupling coefficient and the water inflow trend. The assignment process utilizes a combination of dynamic weight adjustment and trend analysis. The water inflow trend is determined through linear regression analysis of the water inflow time series. The regression slope reflects the increase or decrease trend and rate of change in water inflow. Trend analysis uses a 30-minute time window to calculate the rate of change in water inflow. A positive value indicates an increase in water inflow, while a negative value indicates a decrease. The absolute value of the rate of change reflects the severity of the change. The matrix element assignment rules are determined based on the base value of the pump efficiency-fault coupling coefficient and the modified value of the water inflow trend. The base value provides a steady-state risk assessment, while the modified value accounts for the impact of dynamic changes.

[0117] When a conflict arises between the pump efficiency-fault coupling coefficient and the water inflow trend, indicating different risk levels, the system uses a weighted voting mechanism, combining equipment operating history, maintenance records, and current environmental factors for a comprehensive decision. If the pump efficiency indicator indicates low risk but the water inflow increases sharply, the system prioritizes the urgency of the change in water inflow, but will mark this inconsistency in the warning message and recommend manual verification by the operator. Under extreme operating conditions (such as a water inflow change rate exceeding 50 m³ / h / hour), the system suspends the normal matrix assignment process and directly enters emergency warning mode. At this time, all matrix elements are assigned high risk weights, ensuring that the system maintains a conservative safety strategy even in the event of data anomalies or sensor failures.

[0118] When water inflow is increasing, the matrix element values ​​need to be adjusted toward higher risk. The faster the rate of increase, the greater the adjustment. When water inflow is decreasing, the risk level can be appropriately lowered, but the reduction is limited. The assignment process also considers the statistical patterns of historical data, combining prior probabilities with current observations through Bayesian inference to generate a posterior risk assessment. The final element values ​​of the water inflow-pump efficiency coupling judgment matrix are determined through the comprehensive calculation of all correction factors. The matrix has a four-row, four-column structure, with each element corresponding to the warning level for a specific combination of water inflow level and pump efficiency range.

[0119] The above describes the pump room automatic monitoring and early warning method in the embodiment of the present application. The following describes the pump room automatic monitoring and early warning system in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the pump room automatic monitoring and early warning system includes:

[0120] The acquisition module is used to collect and process multi-dimensional data of the underground drainage pump room through hydraulic vibration sensors, impeller cavitation acoustic sensors and bearing temperature rise sensors to obtain hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data;

[0121] an optimization module for performing feature optimization processing on the hydraulic pulsation amplitude, cavitation acoustic emission intensity, and bearing temperature rise data according to an adaptive variable selection algorithm for downhole environment to obtain a pump efficiency attenuation feature vector;

[0122] a modeling module, configured to perform fault mode modeling processing on the pump efficiency attenuation characteristic vector through a hydraulic attenuation model to obtain pump health status prediction data;

[0123] A training module is used to perform deep learning training on the water pump health status prediction data according to the downhole water pump fault propagation network to obtain a fault identification and classification result;

[0124] The early warning module is used to perform graded early warning processing on the pump room operation status based on the fault identification and classification results through water inflow-pump efficiency coupling judgment, and obtain a four-level early warning output signal.

[0125] above Figure 2 The automatic monitoring and early warning system for the pump room in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The automatic monitoring and early warning equipment for the pump room in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0126] Reference Figure 3 In the embodiment of the present invention, there is also provided a pump room automatic monitoring and early warning device, which can be a server, and its internal structure can be as follows Figure 3As shown. The pump room automation monitoring and early warning device includes a processor, a memory, a display screen, an input device, a network interface and a database connected via a system bus. Among them, the computer-designed processor is used to provide computing and control capabilities. The memory of the pump room automation monitoring and early warning device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the pump room automation monitoring and early warning device is used to store the corresponding data in this embodiment. The network interface of the pump room automation monitoring and early warning device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0127] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the pump room automation monitoring and early warning equipment to which the solution of the present invention is applied.

[0128] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions. When the instructions are run on a computer, the computer executes the steps of the pump room automation monitoring and early warning method.

[0129] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0130] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a pump room automation monitoring and early warning device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0131] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A pump room automatic monitoring and early warning method, characterized in that: The method comprises: Multi-dimensional data collection and processing of underground drainage pump room is carried out through hydraulic vibration sensors, impeller cavitation acoustic sensors and bearing temperature rise sensors to obtain hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data; According to the downhole environment adaptive variable selection algorithm, the hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data are subjected to feature optimization processing to obtain a pump efficiency attenuation feature vector, including: dividing the downhole water inflow condition into four levels of low flow, medium-low flow, medium-high flow and high flow based on the water inflow classification matrix to obtain a water inflow condition classification label; extracting key inflection point parameters of the pump head-flow characteristic curve according to the water inflow condition classification label to obtain the best efficiency point flow, shut-off head, maximum flow and efficiency decline inflection point parameters; inputting the hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data into the principal component analysis algorithm for correlation weight calculation processing to obtain a sensor parameter weight matrix; performing dynamic correction processing on the sensor parameter weight matrix for changes in downhole temperature, humidity and pH value to obtain an environmental adaptive weight coefficient; screening and optimizing the core characteristic variables based on the environmental adaptive weight coefficient to obtain a pump efficiency attenuation feature vector; The pump efficiency attenuation characteristic vector is subjected to fault mode modeling through a hydraulic attenuation model to obtain water pump health status prediction data, including: calculating the cavitation coefficient based on the cavitation acoustic emission intensity and impeller cavitation acoustic sensor data in the pump efficiency attenuation characteristic vector to obtain the cavitation loss coefficient; calculating the wear loss coefficient based on the water quality pH value and particle concentration parameters in the pump efficiency attenuation characteristic vector to obtain the impeller wear loss coefficient; inputting the cavitation loss coefficient and the impeller wear loss coefficient into the water pump similarity law to perform pump efficiency attenuation law modeling to obtain pump efficiency time-varying attenuation data; performing polynomial fitting calculation on the pump efficiency time-varying attenuation data to obtain water pump head-flow characteristic curve degradation parameters; performing comprehensive health status evaluation based on the water pump head-flow characteristic curve degradation parameters and bearing water lubrication theory to obtain water pump health status prediction data; Performing deep learning training on the water pump health status prediction data based on the downhole water pump fault propagation network to obtain a fault identification and classification result; The fault identification and classification results are used to perform graded early warning processing on the pump room operation status through water inflow-pump efficiency coupling judgment, and a four-level early warning output signal is obtained.

2. The pump room automatic monitoring and early warning method according to claim 1 is characterized in that: The multi-dimensional data acquisition and processing of the underground drainage pump room is performed through the hydraulic vibration sensor, the impeller cavitation acoustic sensor and the bearing temperature rise sensor to obtain the hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data, including: The hydraulic vibration sensor collects and processes signals from the piezoelectric vibration detection unit and the water pressure pulsation measurement unit at the water pump inlet to obtain hydraulic excitation spectrum data in the 0.1Hz-5kHz frequency band. Performing amplitude extraction processing on the hydraulic excitation spectrum data to obtain the hydraulic pulsation amplitude; performing broadband acoustic emission detection processing on the 20kHz-200kHz frequency band of the impeller housing using the impeller cavitation acoustic sensor to obtain the bubble burst acoustic characteristic signal; Performing acoustic emission intensity calculation processing on the bubble collapse acoustic characteristic signal to obtain cavitation acoustic emission intensity; The bearing seat temperature change is monitored and processed with a resolution of 0.1°C based on the bearing temperature rise sensor to obtain the bearing temperature rise data.

3. The pump room automatic monitoring and early warning method according to claim 1 is characterized in that: The deep learning training process is performed on the water pump health status prediction data according to the downhole water pump fault propagation network to obtain a fault identification and classification result, including: Inputting the water pump health status prediction data into the input layer of the multi-layer neural network for data reception and processing, and obtaining input data corresponding to the impeller cavitation acoustic emission intensity, bearing temperature rise rate, and hydraulic vibration main frequency; Performing ReLU activation function calculation processing on the hidden layer neuron nodes based on the input data to obtain a nonlinear feature mapping result; According to the nonlinear feature mapping results, the four nodes in the output layer are classified into fault states to obtain the classification probabilities of normal state, minor fault, moderate fault and major fault; Inputting the classification probability into the cross entropy loss function and the mean square error loss function to perform weighted combination loss calculation processing to obtain the network training loss value; The network parameters are back-propagated and optimized based on the network training loss value to obtain a fault identification and classification result.

4. The pump room automatic monitoring and early warning method according to claim 1 is characterized in that: The fault identification and classification results are subjected to graded early warning processing on the pump room operation status through water inflow-pump efficiency coupling judgment, and a four-level early warning output signal is obtained, including: A coupling judgment matrix is ​​constructed based on the fault identification and classification results and the real-time pump efficiency data to obtain a water inflow-pump efficiency coupling judgment matrix; According to the water inflow-pump efficiency coupling judgment matrix, a first-level warning judgment process is performed when the pump efficiency is higher than the set threshold and the fault is identified as a normal state, and a green normal operation warning signal is obtained; Inputting the pump efficiency in the medium range or detecting a minor fault feature in the water inflow-pump efficiency coupling judgment matrix into the secondary warning judgment for processing to obtain a yellow performance degradation warning signal; Performing a three-level warning judgment process on the situation where the pump efficiency in the water inflow-pump efficiency coupling judgment matrix is ​​in a low range or a medium fault feature is detected, and obtaining an orange fault risk warning signal; Based on the situation in the water inflow-pump efficiency coupling judgment matrix where the pump efficiency is lower than the safety threshold or a serious fault feature is detected, a four-level warning judgment process is performed to obtain a red serious fault warning signal as a four-level warning output signal.

5. The pump room automatic monitoring and early warning method according to claim 4 is characterized in that: The coupling judgment matrix is ​​constructed based on the fault identification and classification results and the real-time pump efficiency data to obtain a water inflow-pump efficiency coupling judgment matrix, including: The current water inflow condition is classified according to the real-time water inflow sensor data, and four water inflow grade identifications are obtained: low water inflow, medium-low water inflow, medium-high water inflow and high water inflow; Continuously monitor the current water pump efficiency based on the flow meter, pressure transmitter and power meter data to obtain the real-time pump efficiency percentage value; Cross-mapping the normal state, minor fault, moderate fault and severe fault classification in the fault identification classification result with the water inflow level identifier to obtain a fault-water inflow association matrix; Performing coupling weight calculation processing on the real-time pump efficiency percentage value and the fault-water inflow association matrix to obtain a pump efficiency-fault coupling coefficient; Matrix element assignment processing is performed based on the pump efficiency-fault coupling coefficient and the water inflow variation trend to obtain a water inflow pump efficiency coupling judgment matrix.

6. A pump room automatic monitoring and early warning system, characterized in that: For implementing the pump room automatic monitoring and early warning method according to any one of claims 1 to 5, the pump room automatic monitoring and early warning system comprises: The acquisition module is used to collect and process multi-dimensional data of the underground drainage pump room through hydraulic vibration sensors, impeller cavitation acoustic sensors and bearing temperature rise sensors to obtain hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data; An optimization module is used to perform feature optimization processing on the hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data according to a downhole environment adaptive variable selection algorithm to obtain a pump efficiency attenuation feature vector, including: classifying the downhole water inflow condition into four levels of low flow, medium-low flow, medium-high flow and high flow based on a water inflow classification matrix to obtain a water inflow condition classification label; extracting key inflection point parameters of the pump head-flow characteristic curve according to the water inflow condition classification label to obtain the best efficiency point flow, shut-off head, maximum flow and efficiency decline inflection point parameters; inputting the hydraulic pulsation amplitude, cavitation acoustic emission intensity and bearing temperature rise data into a principal component analysis algorithm for correlation weight calculation processing to obtain a sensor parameter weight matrix; performing dynamic correction processing on the sensor parameter weight matrix for changes in downhole temperature, humidity and pH value to obtain an environmental adaptive weight coefficient; and screening and optimizing the core characteristic variables based on the environmental adaptive weight coefficient to obtain a pump efficiency attenuation feature vector. A modeling module is used to perform fault mode modeling processing on the pump efficiency attenuation characteristic vector through a hydraulic attenuation model to obtain water pump health status prediction data, including: performing cavitation coefficient calculation processing based on the cavitation acoustic emission intensity and impeller cavitation acoustic sensor data in the pump efficiency attenuation characteristic vector to obtain a cavitation loss coefficient; performing wear loss coefficient calculation processing based on the water quality pH value and particle concentration parameters in the pump efficiency attenuation characteristic vector to obtain an impeller wear loss coefficient; inputting the cavitation loss coefficient and the impeller wear loss coefficient into the water pump similarity law to perform pump efficiency attenuation law modeling processing to obtain pump efficiency time-varying attenuation data; performing polynomial fitting calculation processing on the pump efficiency time-varying attenuation data to obtain water pump head-flow characteristic curve degradation parameters; performing comprehensive health status evaluation processing based on the water pump head-flow characteristic curve degradation parameters and bearing water lubrication theory to obtain water pump health status prediction data; A training module is used to perform deep learning training on the water pump health status prediction data according to the downhole water pump fault propagation network to obtain a fault identification and classification result; The early warning module is used to perform graded early warning processing on the pump room operation status based on the fault identification and classification results through water inflow-pump efficiency coupling judgment, and obtain a four-level early warning output signal.

7. A pump room automatic monitoring and early warning device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the pump room automatic monitoring and early warning method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is enabled to execute the pump room automatic monitoring and early warning method according to any one of claims 1 to 5.

Images