Method and System for Predicting Clogging Faults in Wastewater Source Heat Pump Systems Based on Integrated Operational Data

Abstract

This application discloses a method and system for predicting fouling faults in a wastewater source heat pump system by integrating operational data. It relates to the field of equipment operation and maintenance data processing technology, and performs time alignment and reliability-weighted fusion of multi-source operational data to overcome the problems of ineffective fusion of asynchronous multi-source data and the susceptibility of single indicators to fluctuations in operating conditions. By constructing thermo-hydraulic characteristics and quantifying fouling features, and establishing a clean dynamic baseline based on operating condition classification, it abandons the fixed threshold judgment mode and adapts to the complex and ever-changing operating conditions of the system. A comprehensive fouling index is obtained through reliability suppression and nonlinear fusion, and risk prediction is achieved by combining trend, topology propagation, and cumulative operating data, accurately outputting the fault probability and remaining warning time. This effectively improves the accuracy of fouling fault identification, eliminates false alarms and missed alarms, ensures the stability of the system's thermo-hydraulic performance, reduces operation and maintenance costs, and enables online updates after operation and maintenance, adapting to the energy-saving and safe operation and maintenance requirements for long-term stable operation.

Description

Technical Field

[0001] This application relates to the field of equipment operation and maintenance data processing technology, specifically to a method and system for predicting dirt blockage faults in a sewage source heat pump system that integrates operation data. Background Technology

[0002] In recent years, wastewater source heat pump systems have been widely used in heating, cooling, and heat recovery in industrial parks and large buildings. These systems utilize the heat energy from wastewater for heating or cooling, significantly reducing energy consumption and improving operational efficiency. However, suspended solids and organic deposits in wastewater tend to accumulate in critical components such as heat exchangers, pipes, and pump sets. Over time, this can lead to scaling or blockages, affecting the system's thermal and hydraulic performance, reducing heating and cooling efficiency, and increasing maintenance costs.

[0003] Traditional methods for detecting blockages often rely on manual inspections or single-threshold alarms, such as judging faults based solely on pressure difference or flow rate changes. However, due to the complex operating conditions of wastewater source heat pump systems, single indicators are easily affected by flow fluctuations or load changes, leading to inaccurate blockage judgments, frequent false alarms, or serious missed alarms, making it difficult to meet the requirements for safe operation and energy conservation.

[0004] In the prior art, Chinese patent CN110779745B discloses a method for early fault diagnosis of heat exchangers based on BP neural networks. This method collects key parameters of the heat exchanger under normal operating conditions, constructs a BP neural network model, and judges early faults based on whether the predicted output error continuously exceeds the normal error range. This scheme can establish a fault diagnosis model using normal samples, but it mainly relies on the neural network prediction error for anomaly judgment. Chinese patent CN113108842A discloses a multi-parameter correlation monitoring and early warning method and system for heat exchangers. It acquires primary monitoring parameters such as temperature, pressure, and flow rate, and combines structural parameters and fluid property parameters to construct secondary monitoring parameters related to heat transfer performance, resistance performance, scaling faults, and leakage faults. Then, it uses a multi-parameter early warning model to judge state deviations and provide fault warnings. However, its object is mainly the heat exchanger body and does not integrate multi-source asynchronous data from wastewater source heat pump systems.

[0005] Therefore, although existing technologies can achieve early fault diagnosis and warning of heat exchangers to a certain extent, they still have the problem of inaccurate judgment for dirt blockage faults in sewage source heat pump systems under complex operating conditions. Summary of the Invention

[0006] In order to solve the above-mentioned technical problems, this application proposes the following technical solution: In a first aspect, embodiments of this application provide a method for predicting dirt blockage faults in a wastewater source heat pump system by integrating operational data, including: Multi-source operation data from the wastewater source heat pump system are collected, mapped onto a time grid, and weighted and fused at a unified time to generate credible fused data. Based on the aforementioned credibility fusion data, thermal and hydraulic features are constructed, and the dirt clogging features are quantified to obtain the dirt clogging characteristics. Based on the aforementioned thermal and hydraulic characteristics, a working condition vector is constructed and the working conditions are classified. Clean samples are screened within the same working condition and a clean dynamic baseline is established. The evidence set is formed by combining the aforementioned dirt and clogging characteristics with the generated clean dynamic baseline. After performing credibility suppression, nonlinear fusion is performed to obtain the comprehensive dirt and clogging index, and the trend and topological propagation are extracted. The system utilizes the comprehensive dirt and congestion index, trend, topology propagation volume, and cumulative operation volume to predict risks, output failure probability and remaining warning time, and performs online updates after maintenance events.

[0007] In one possible implementation, the multi-source operating data from the wastewater source heat pump system is uniformly mapped onto a time grid, and weighted fusion based on reliability is performed within a unified time frame to generate reliable fused data, including: Collect multi-source operational data from the monitoring platform of the wastewater source heat pump system; For data with different sampling frequencies or missing data, each variable is mapped to a unified time grid, and neighboring sampled values ​​are weighted using a kernel function. The sample confidence weights are calculated by combining the normalized residuals, the degree of normalized mutations, and the missing markers, and the resulting fused running data sequence is output at a unified time.

[0008] In one possible implementation, the step of combining normalized residuals, normalized mutation degree, and missing markers to calculate sample confidence weights and outputting a unified time-sequence fused running data sequence includes: Normalized residuals Normalized mutation degree and missing or imputation indicators Mapped to credibility weights: in: This represents the minimum confidence level. This represents the confidence bias parameter. , , These represent the penalty coefficients for residuals, abrupt changes, and missing or imputed terms, respectively. For any variable At the unified moment Constructing Credibility Fusion Data : in: Indicates the first Class variables at sampling time The sampled values, Indicates the first The set of sampling times for class variables. Indicates the sample confidence weight. Represents the time kernel function. Indicates the kernel width. Indicates a uniform time step. This indicates a smoothing term to prevent the denominator from being zero.

[0009] In one possible implementation, the step of constructing thermal and hydraulic features based on the credibility fusion data and quantifying the fouling features to obtain fouling characteristics includes: Based on the fused data, interpretable features of the thermal and hydraulic links are constructed respectively, transforming dirt and clogging from indirect indicators into accumulative thermal resistance and clogging intensity, thus eliminating the impact of operating condition changes on the indicators. It generates two explicit fouling quantities: thermal resistance and hydraulic pressure difference, and outputs heat exchange, logarithmic mean temperature difference, and comprehensive heat transfer capacity.

[0010] In one possible implementation, the construction of interpretable features of the thermal and hydraulic links based on fused data transforms fouling from an indirect indicator into accumulative thermal resistance and fouling intensity, eliminating the impact of operating condition changes on the indicators, including: Based on the fused data, the heat exchange on the wastewater side was calculated respectively. Heat exchange with load side ; The heat exchange balance residual is calculated based on the heat exchange on the wastewater side and the heat exchange on the load side. , and in conjunction with the above Obtain observation of comprehensive heat transfer capabilities ; Based on the above Perform thermal resistance series decomposition and inversion of fouling thermal resistance : in: Indicates the convective heat transfer coefficient on the wastewater side. Indicates the convective heat transfer coefficient on the load side. Indicates the heat exchange area. Indicates the thermal resistance of the heat exchange wall. Indicates the thermal resistance of dirt; And based on the observed pressure difference Pressure difference with clean baseline Obtaining hydraulic plugging strength : in: Represents a smooth nonnegative mapping function. Indicates the observed pressure difference. This represents the clean baseline pressure difference.

[0011] In one possible implementation, the step of constructing a working condition vector based on the thermodynamic and hydraulic characteristics and classifying the working conditions, screening clean samples within the same working condition, and establishing a clean dynamic baseline includes: Operating condition vectors are constructed based on flow rate, temperature, pump frequency, valve opening degree, and load rate. These operating conditions are then classified to form a sequence of operating condition categories, including: based on operating condition vectors. Calculate the current sample belonging to the first... The posterior probability of each working condition category is calculated, and the current working condition category is determined. : in: Indicates the first Prior weights for each work condition category Indicates the first The center vector of each working condition category Indicates the first Covariance matrix of each working condition category Represents the multivariate Gaussian density function; Establishing a dynamic baseline by screening clean samples under the same operating conditions makes the deviation of dirt and clogging indicators comparable under different operating conditions, providing a reference for constructing evidence of dirt and clogging, including: In operating condition category A clean dynamic baseline is generated using similarity kernel regression: in: Indicates the clean dynamic baseline. Indicates the type of working condition The clean sample library below The baseline output vector represents the clean sample. This represents the similarity weight between the current sample and the clean sample.

[0012] In one possible implementation, the evidence set is formed by combining the fouling features with the generated clean dynamic baseline, and after performing credibility suppression, a nonlinear fusion is performed to obtain a comprehensive fouling index. Trend and topology propagation are then extracted, including: The explicit amount of dirt and clogging is combined with the clean dynamic baseline to form multidimensional evidence of dirt and clogging, and the impact of abnormal noise is reduced by credibility suppression. Then, nonlinear fuzzy fusion is used to obtain the comprehensive dirt and blockage index, while extracting the dirt and blockage trend and constructing the system topology coupling quantity to reflect the coupling propagation and gradual degradation characteristics between devices.

[0013] In one possible implementation, the step of using nonlinear fuzzy fusion to obtain a comprehensive congestion index, while simultaneously extracting congestion trends and constructing system topology coupling quantities to reflect the coupling propagation and gradual degradation characteristics between devices includes: Evidence with suppressed credibility Sort by size as follows: remember: and order The comprehensive index of dirt and congestion is obtained by using Choquet fuzzy integral. : in: Indicates the amount of evidence. Represents a set The corresponding fuzzy measure; Based on the comprehensive dirt and congestion index Extracting trends: in, Indicates a trend of dirt and congestion. Indicates the span of trend calculation; Calculate the system coupling congestion amount based on system topology propagation. : in: Represents the node evidence vector. Represents the normalized topological diffusion matrix. This represents node evidence after topology propagation. Represents the node weight vector. This represents a vector consisting entirely of 1s.

[0014] In one possible implementation, the step of using the comprehensive congestion index, trend, topology propagation volume, and cumulative operating volume to perform risk prediction, output failure probability and remaining early warning time, and perform online updates after an operational event includes: A risk feature vector is constructed using the comprehensive dirt and blockage index, trend, topology propagation volume, and cumulative operation volume. This vector is then input into a time-varying hazard rate model to calculate the probability of dirt and blockage failures occurring in the future and the remaining warning time. When cleaning, maintenance, or manual confirmation events occur, the model parameters are updated online using event feedback to achieve adaptive fouling prediction and risk calibration across seasons and water quality conditions.

[0015] In one possible implementation, the step of constructing a risk feature vector using a comprehensive congestion index, trend, topology propagation volume, and cumulative operating volume, inputting it into a time-varying hazard rate model, and calculating the probability of a congestion failure occurring in the future time period and the remaining warning time includes: Based on risk feature vector Constructing a time-varying hazard rate model : in, Indicates the basic hazard rate. This represents the risk model weight vector; Calculating the Future The probability of a dirt blockage failure occurring within a given time period: in: Indicates the time when the dirt blockage fault occurred. Indicates the forecast time window, Indicates a uniform time step; At risk threshold Next, determine the remaining warning time. : in, This represents the probability threshold for triggering a risk warning.

[0016] Secondly, embodiments of this application provide a sewage source heat pump system with integrated operational data for predicting blockage faults, including: The multi-source data credibility fusion module is used to collect multi-source operating data from the sewage source heat pump system, map them uniformly onto a time grid, and perform credibility weighted fusion within a unified time to generate credibility fusion data; The thermal and hydraulic fouling feature quantification module is used to construct thermal and hydraulic features based on the credibility fusion data, and to quantify the fouling features to obtain fouling features. The working condition classification and clean baseline construction module is used to construct a working condition vector and classify the working conditions based on the thermal and hydraulic characteristics, screen clean samples within the same working condition and establish a clean dynamic baseline. The dirt and clogging index and degradation feature extraction module is used to combine the dirt and clogging features with the generated clean dynamic baseline to form an evidence set, perform credibility suppression and nonlinear fusion to obtain the comprehensive dirt and clogging index, and extract the trend and topological propagation amount. The fault risk prediction and adaptive update module is used to predict risks, output fault probability and remaining warning time using the comprehensive dirt and blockage index, trend, topology propagation volume and cumulative operation volume, and perform online updates after operation and maintenance events.

[0017] In this embodiment, time alignment and credibility-weighted fusion of multi-source operational data are performed, solving the problems of ineffective fusion of asynchronous multi-source data and susceptibility of single indicators to fluctuations in operating conditions. By constructing thermo-hydraulic features and quantifying fouling characteristics, a clean dynamic baseline is established in conjunction with operating condition classification, abandoning the fixed threshold judgment mode and adapting to the complex and ever-changing operating conditions of the system. Simultaneously, a comprehensive fouling index is obtained through credibility suppression and nonlinear fusion, and risk prediction is achieved by combining trend, topology propagation, and cumulative operating data, accurately outputting the probability of failure and the remaining warning time. Ultimately, this effectively improves the accuracy of fouling fault identification, eliminates false alarms and missed alarms, ensures the stability of the system's thermo-hydraulic performance, reduces operation and maintenance costs, and enables online updates after operation and maintenance, adapting to the energy-saving and safe operation and maintenance requirements for long-term stable operation. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the data acquisition and computing deployment architecture provided in the embodiments of this application; Figure 2 A flowchart illustrating a method for predicting dirt blockage faults in a wastewater source heat pump system that integrates operational data, provided in an embodiment of this application; Figure 3 A simplified topology diagram of the wastewater-side filter, heat exchanger, pipeline, and circulating pump unit provided in the embodiments of this application; Figure 4 This is a schematic diagram of the topology propagation calculation process provided in an embodiment of this application; Figure 5 Key dirt and clogging index comparison histograms provided for embodiments of this application; Figure 6 The comprehensive evolution and risk prediction results of the sewage source heat pump system provided in this application embodiment are as follows: (a) is the trend of the comprehensive dirt blockage index with the operating time; (b) is the relationship between the trend index and the operating time; (c) is the trend of the failure probability with the operating time; and (d) is the relationship between the remaining warning time and the operating time. Figure 7 The evolution curve of normalized key indicators provided for embodiments of this application; Figure 8 This is a schematic diagram of a sewage source heat pump system with integrated operating data for predicting dirt blockage faults, provided in an embodiment of this application. Detailed Implementation

[0019] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.

[0020] This embodiment can be deployed on a server, edge gateway, host computer, or local controller in a heat pump station. The system includes at least a data acquisition module, a time alignment and fusion module, a thermo-hydraulic calculation module, an operating condition identification and clean baseline module, an evidence fusion module, a risk prediction module, an online self-calibration module, and an alarm output module. Each module can be implemented using software programs, or it can be implemented in an industrial control computer, a PLC host system, or a cloud-edge collaborative platform.

[0021] The collected multi-source operational data includes at least the following categories: wastewater-side data, load-side data, main unit and pump set data, water quality data, and operation and maintenance event data. Wastewater-side data includes wastewater flow rate, differential pressure at key locations, wastewater inlet temperature, and wastewater outlet temperature; load-side data includes load flow rate, load inlet temperature, and load outlet temperature. Main unit and pump set data includes compressor power, circulating pump frequency, circulating pump power, and valve opening. Water quality data includes suspended solids, turbidity, chemical oxygen demand, and conductivity, and optionally includes oil content, ammonia nitrogen, etc. Operation and maintenance event data includes cleaning events, maintenance events, manually confirmed blockage events, sensor calibration events, and their timestamps.

[0022] This embodiment Figure 1 The term "in" does not merely refer to hardware connections, but rather illustrates the complete closed loop of this embodiment in an engineering setting, from data generation and acquisition to model calculation and application output. Specifically, Figure 1 The field equipment and data sources on the left correspond to the original input terminals of this embodiment, including wastewater flow rate, pressure differential at key locations, wastewater inlet temperature, wastewater outlet temperature, load-side flow rate, load-side inlet temperature, load-side outlet temperature, compressor power, circulating pump frequency or circulating pump power, water quality indicators, and operation and maintenance event records. These data reflect the flow state, heat exchange state, energy consumption state, water sedimentation tendency, and human maintenance facts, and serve as the basic data source for subsequent steps.

[0023] Figure 1 The data acquisition layer corresponds to the data sensing and data governance section located on the field side or edge side in this embodiment. This layer includes at least sensors, This may include a controller, monitoring platform, historical database, and timestamp synchronization module. Sensors are used to collect temperature, flow rate, differential pressure, power, and water quality signals. The controller is used to complete on-site signal aggregation, preliminary range conversion, and equipment operating status reading; the monitoring platform is used to cache, display, and upload operating data; the historical database is used to store continuous operating data, cleaning records, fault records, and manual confirmation records; the timestamp synchronization module is used to unify the time base of different data sources, so that subsequent steps can be aligned and calculated according to a unified time.

[0024] Figure 1 The computation and analysis layer in the model corresponds one-to-one with the method execution chain in this embodiment. Figure 1 The application output layer on the right corresponds to the operation and maintenance interaction terminal in this embodiment, including a real-time monitoring interface, fault warnings, alarm push notifications, maintenance suggestions, and report archiving. The real-time monitoring interface displays indicators, fault warnings and alarm push notifications are used to notify operation and maintenance personnel when the risk reaches a threshold, maintenance suggestions are used to provide recommendations for cleaning, checking filters, checking heat exchangers, verifying pump sets, or calibrating sensors, etc., and report archiving is used to save risk change curves, alarm triggering reasons, changes in indicators before and after cleaning, and model update records.

[0025] Figure 1 The deployment location at the bottom indicates that this embodiment can be deployed in an edge gateway, host computer, server, or cloud-edge collaborative platform. For small heat pump stations, all calculations can be completed in the host computer or edge gateway. For large or multi-site systems, data preprocessing and anomaly filtering can be completed at the edge, and then baseline modeling, evidence fusion, and risk prediction can be completed by the server or cloud.

[0026] based on Figure 1 Deployment architecture, participate Figure 2 The wastewater source heat pump system clogging fault prediction method provided in this embodiment, which integrates operational data, includes: S101 collects multi-source operating data from the wastewater source heat pump system, maps it uniformly onto a time grid, and performs credibility-weighted fusion at a unified time to generate credibility fusion data.

[0027] The field data of the wastewater source heat pump system suffers from issues such as inconsistent sampling frequencies, sensor clock drift, partial data gaps, short-term abnormal spikes, and intermittent interpolation. Directly splicing data from different timestamps for pressure differential, heat exchange, or energy efficiency calculations will result in variables at the same moment not being truly synchronized, leading to misjudgments of subsequent fouling indicators. Therefore, this embodiment first maps all original variables to a unified time grid. Then, a weighted fusion is performed based on the credibility of each sample.

[0028] In this embodiment, the symbol The first occurrence indicates a unified calculation step size, which can be in seconds or minutes, preferably determined based on the monitoring platform's refresh cycle. If the sampling cycles for temperature, pressure difference, flow rate, and water quality indicators differ significantly, then... The sampling period can be taken as an integer multiple of the main operating data sampling period to balance real-time performance and computational stability. (Symbol) Represents a discrete index on a unified time grid. Indicates the first A unified calculation time.

[0029] To perform temporal neighborhood weighting, this embodiment employs a temporal kernel function. Preferably, Using a Gaussian kernel: in, This represents the normalized time distance between the original sampling time and the unified calculation time. (Time kernel function) Its function is to assign higher weights to original samples that are closer to the time of unification. Kernel width Used to control the coverage of the time neighborhood, which can be selected ,in: For positive numbers, usually take to When data sampling is dense and noise is low, A smaller value can be chosen to improve real-time performance. This is especially important when data sampling is sparse or short-term data gaps exist. A larger value can be selected to improve continuity.

[0030] For any variable At the unified moment Construction fusion value : In this formula, Representing variables The original set of sampling times, Indicates the first Each sampling time, Indicates the first The credibility weight of each sample For the above time neighborhood kernel function, For the above-mentioned kernel width, This is the smoothing term. The meaning of this formula is: to simultaneously weight the original data using both temporal proximity and sample reliability. When a variable has multiple neighboring sampling points, the closer the points are... Furthermore, more reliable samples contribute more; when there are few samples near that moment, the denominator... To ensure that the calculation remains stable.

[0031] To obtain the credibility weight This embodiment comprehensively considers three types of factors: short-term prediction residuals, the degree of adjacent mutations, and missing imputation markers. First, a short-term prediction model is used to obtain... When training conditions for complex models are lacking, exponential smoothing can be used for prediction. in: As the smoothing coefficient, when When the value is large, the predicted value depends more on the most recent sample. When the value is smaller, the predicted values ​​are smoother. The exponential smoothing model described above can also be replaced by an autoregressive model, a Kalman filter model, or a physically constrained prediction model, as long as the same variable can be output in different values. Short-term forecast value at time That's all.

[0032] Before calculating the residual scale, this embodiment first defines the moving median and the median absolute deviation. This represents the median within the sliding window. Indicates length is The median absolute deviation. For any sequence It can be defined as: in, The length of the sliding window can be set by... to The settings can also be adaptively determined based on the amount of stable operating data. Coefficients Used to make under approximate Gaussian noise Comparable to standard deviation. It is defined in this embodiment because it is directly used for and The normalization scale is determined.

[0033] Then calculate the normalized prediction residuals. : in: For variables The robustness scale within the sliding window can be taken as follows: . The larger the value, the more significantly the current sample deviates from the short-term operating trend, which may be due to sensor spikes, communication anomalies, or sudden changes in operating conditions.

[0034] Further calculation of normalized difference abrupt change : in: Represents the difference sequence of variables Robustness scale, can be adopted .when A large value indicates that the variable has abruptly changed between adjacent samples, and its impact on synchronous fusion should be reduced.

[0035] Logical functions In this embodiment, residual penalty, mutation penalty, and imputation penalty are mapped to... to The confidence factor within the interval is defined as follows: Indicator Function In this embodiment, the condition is used to describe whether a sample is a missing or imputed sample. When the condition in parentheses is met, ,otherwise .

[0036] Missing or imputation markers Defined as: When the sample is the true sampled value When the samples are obtained by linear interpolation, preserving previous values, model completion, or manual input, To avoid data breakage caused by completely discarding interpolated values, this embodiment does not directly delete the interpolated values, but instead penalizes them in the credibility weight.

[0037] Combining the above three types of penalties, the credibility weight is defined as follows: in: As the minimum confidence level, As a confidence benchmark bias, , , These are the penalty coefficients for prediction residuals, mutation severity, and missing imputation, respectively, all of which are non-negative. By setting... Even if some variables are temporarily missing, a minimum level of information input can be retained, preventing subsequent models from failing to compute due to univariate breakpoints. The determination method for the aforementioned confidence parameters is also completed in this embodiment. , , , Initialize based on experience first, to ensure normal and stable samples. near to This makes the abnormal peak samples Below The error can then be corrected using sensor calibration records, manual removal records, or maintenance confirmation records. Generally not exceeding Its purpose is to prevent the influence of a variable on subsequent calculations from being completely cut off due to a short-term anomaly of a single sensor.

[0038] This step outputs a unified time-based reliability fusion running data sequence. and the original sampling point confidence weight sequence In subsequent steps, all thermal, hydraulic, energy efficiency, water quality, and risk indicators are based on... Calculations can be performed. When evidence comes from low-confidence data, it can also be utilized... The statistical value further reduces the weight of this evidence in the fusion.

[0039] S102, construct thermal and hydraulic features based on the credibility fusion data, and quantify the dirt clogging features to obtain dirt clogging features.

[0040] The purpose of this step is to transform fouling, which is indirectly caused by increased pressure differential or decreased energy efficiency, into explicit state quantities with physical meaning and comparable cumulative characteristics. Specifically, this includes the fouling thermal resistance in the thermal link. and the intensity of blockage in the hydraulic link .in: This reflects the additional thermal resistance caused by deposits on the heat exchange walls and heat exchange channels. It reflects the relative deviation of the current pressure differential from the clean pressure differential under the same operating conditions.

[0041] In this embodiment, the subscript Indicates the sewage side. Indicates the load side, superscript or subscript This represents the observations calculated from the observed data. Indicates the clean dynamic baseline quantity. These symbols represent reference quantities used for normalization or comparison. The explanation of these symbols at their first appearance in the thermal and hydraulic formulas is to enable this embodiment to independently express the calculation meaning of thermal and hydraulic links.

[0042] The heat exchange area involved in this embodiment Wall thermal resistance and hydraulic diameter This can be determined from the heat exchanger's structural parameters, the manufacturer's technical manual, or on-site calibration data. Fluid density. Specific heat capacity at constant pressure Thermal conductivity and dynamic viscosity These parameters can be determined using water temperature gauges, water quality correction tables, online water quality estimation, or empirical constants during the commissioning period. If the above parameters cannot be measured in real time, approximate constants can be taken within the same operating season and verified using stable operating data after cleaning or maintenance.

[0043] First, the heat exchange on the wastewater side is calculated based on the fused data from S101. Heat exchange with load side : in: and The fluid densities on the wastewater side and the load side are respectively. and These are the isobaric specific heat capacities on the wastewater side and the load side, respectively. and These represent the volumetric flow rates on both sides. If real-time measurement is not possible on-site... and This information can be obtained from water temperature, water quality, and empirical tables. In scenarios where water quality changes are minimal, the calibration constant from the commissioning period can also be used.

[0044] To facilitate energy conservation verification, the heat transfer balance residual is defined. : when If the reading remains high, it indicates that there may be an abnormality in the temperature, flow rate, or power measurement, or that the system is in a non-steady-state process such as startup, shutdown, or commutation.

[0045] Then the logarithmic mean temperature difference was calculated. To be applicable to both heating and cooling conditions, this embodiment preferably uses a positive value for the temperature difference at the heat exchange ends: exist and When they are not equal, the logarithmic mean temperature difference is: when and The difference is less than the preset temperature difference threshold. To avoid numerical instability, we can approximate it using the limit form: The overall heat transfer capacity is calculated based on the load-side heat exchange and the logarithmic mean temperature difference. : in: The smaller the value, the lower the heat exchange capacity. However, It is also affected by operating conditions such as flow rate, inlet temperature, and load rate. Therefore, a fixed threshold cannot be used to directly determine dirt blockage. It is necessary to combine it with the subsequent clean dynamic baseline for comparison with the same operating conditions.

[0046] To ensure that the blockage has a clear physical meaning in the thermal link, this embodiment will... Write it in series thermal resistance form: This leads to the inversion of the thermal resistance of the dirt. : in: For heat exchange area, The thermal resistance of the heat exchange wall can be determined from the equipment design parameters, manufacturer's manual, or commissioning data. and These are the convective heat transfer coefficients on the wastewater side and the load side, respectively. If accurate calculation is not possible on-site... and During the commissioning period, adjustments can also be made based on cleanroom operation data. Perform overall calibration, but the calibration method used should be specified in the instruction manual.

[0047] The convective heat transfer coefficient can be calculated using the following general Nusselt number correlation formula. Let... If we indicate the wastewater side or the load side, then: in: Thermal conductivity, The hydraulic diameter, For Nusselt numbers, For Reynolds numbers, For Prandtl numbers, The average flow velocity, For fluid dynamic viscosity, This represents the dynamic viscosity at wall temperature. (Coefficient) , , , The determination can be made based on the heat exchanger type, flow conditions, manufacturer-recommended correlation formulas, or fitting of commissioning data.

[0048] To convert differential pressure deviation into non-negative and continuous evidence of hydraulic fouling, this embodiment employs a smoothed non-negative function. Its definition is: In software implementation, to avoid A large exponential overflow can be caused by an excessively large exponential structure, which can be addressed by using an equivalent stable form: This function is Significantly smaller than When the output is close to ,exist It grows approximately linearly in positive time, thus it is suitable for converting excess differential pressure relative to the clean baseline into hydraulic clogging intensity.

[0049] In a hydraulic linkage, to avoid incomparability of absolute pressure difference values ​​under different flow rates and pump frequencies, this embodiment will observe the pressure difference. Pressure difference with clean baseline under the same operating conditions The relative deviation is mapped to the blockage intensity. : in: This refers to the measured differential pressure of a filter, heat exchanger, or pipe section. Output from the clean dynamic baseline model in S103. Due to the... It depends on S103 In actual execution, calculations can be performed first in S102. , , The original features are then generated by S103. Post-backfill calculation If the equipment has just been put into operation and there is no clean baseline model yet, the manufacturer's differential pressure curve or initial clean data from the start of operation can be used as a baseline. The initial value.

[0050] This embodiment outputs a thermally displayed amount of dirt / clogging. Hydraulic explicit dirt blockage volume Observe the overall heat transfer capacity Heat exchange and thermal balance residual .

[0051] S103, construct the operating condition vector based on the thermal and hydraulic characteristics and classify the operating conditions, screen clean samples within the same operating condition and establish a clean dynamic baseline.

[0052] Under high flow rates and high pump frequencies, the pressure differential in the clean state is also relatively high. When the wastewater inlet temperature varies significantly, the pressure differential in the clean state... and The conditions also change with operating conditions. Without operating condition identification and cleanliness baseline modeling, normal operating condition changes can easily be misjudged as dirt and clogging. Therefore, this embodiment first identifies the operating condition category and then establishes a cleanliness dynamic baseline within the same operating condition.

[0053] In this embodiment, the number of working condition categories Prior probabilities in Gaussian mixture models can be obtained through clustering of historical operating data or set according to typical operating modes of heat pump systems, such as high heating load, low heating load, high cooling load, low cooling load, and transitional conditions. Mean vector Covariance Matrix It can be estimated from historical operating data or commissioning period data. If historical data is insufficient, operating conditions can be manually divided according to flow rate range, inlet water temperature range, and load rate range, and then gradually updated during operation. The above parameter descriptions are included in this embodiment because... , , It directly determines the working condition attribution of the current sample and the comparability of subsequent clean baselines.

[0054] First, construct the operating condition vector. : in: The load factor can be calculated as the ratio of the current load-side heat exchange capacity to the rated heat exchange capacity. This refers to the rated or designed heat exchange capacity of the unit. If the valve opening... If not collected, it can be obtained from... Delete this component. If the system also collects outdoor temperature, sewage tank level, or bypass valve status, these can also be added. .

[0055] Based on the Gaussian mixture model, calculate the current sample's category (i). The posterior probability of each work condition category: The operating condition category is determined as follows: in: Indicates time Operating condition categories, Indicates the first Prior weights for different operating conditions Indicates the first Similar working condition center, Represents the covariance matrix. This represents a multivariate Gaussian density function. The number of operating condition categories can be determined based on historical operating data through BIC, AIC, or manual maintenance partitioning; when historical samples are insufficient, regular operating condition categories can also be formed by combining flow ranges, pump frequency ranges, and load rate ranges.

[0056] Then in the same working condition category Internal screening clean sample library Clean samples should ideally meet the conditions of low differential pressure, high energy efficiency, good thermal balance, and no dirt clogging. This embodiment can adopt the following screening rules: in: Indicates the type of working condition Internal pressure difference Quantiles Indicates the type of working condition Inside of Quantiles The threshold value for thermal equilibrium residuals. As a marker for operational events, when the sample Take during the period when dirt blockage is confirmed, abnormality before cleaning, or sensor failure. Otherwise take . and Desirable to However, adjustments should be made based on the amount of equipment data and the proportion of clean samples.

[0057] In this embodiment, the quantile threshold and residual threshold involved in clean sample screening are determined along with the formula. Low quantile threshold for pressure differential. Desirable to High percentile threshold for energy efficiency Desirable to Thermal equilibrium residual threshold The similarity diffusion parameter can be determined based on stable operation data after cleaning. This can be determined by minimizing the baseline prediction error for the cleanroom period. Minimum sample size for the cleanroom sample library. The sample library can be determined based on covering at least one complete stable operating period for each operating condition. If there are insufficient clean samples for a certain operating condition, clean samples from adjacent operating conditions, manufacturer performance curves, or stable samples from the most recent cleaning can be temporarily used to supplement the sample library, and the sample library will be updated after subsequent cleaning events.

[0058] Observe energy efficiency It can be calculated based on either the system port size or the host port size. The preferred system port size is: If only the heat exchange efficiency of the main unit is evaluated, the denominator can also be set to However, the same criteria should be maintained during system implementation to avoid inconsistencies in calculation criteria between the training and prediction periods.

[0059] In clean sample library Based on this, a multi-output clean dynamic baseline is established using similarity kernel regression. The baseline output vector is defined as follows: The clean baseline vector at the current moment is: Among them, weight Defined as: Similar distance It can be represented by the weighted Euclidean distance between the current operating condition vector and the clean sample operating condition vector: in: Operating condition category The characteristic scale matrix below can be taken as a diagonal matrix. . For the working condition vector, the first The scale of each component in a clean sample library This is the similarity diffusion parameter. The larger the value, the more clean samples are involved in the baseline estimation; The smaller the value, the more the baseline relies on the sample closest to the current operating conditions.

[0060] Depend on Take the clean baseline pressure difference separately Clean baseline overall heat transfer capacity Clean baseline energy efficiency and clean baseline dirt thermal resistance When there are insufficient clean samples during the initial commissioning phase, initialization can be performed in the following order: First, use stable operating samples after initial cleaning of the equipment during commissioning; second, use the pressure-flow curve and heat exchange performance curve provided by the manufacturer; and third, use historical clean data from the same model of unit. Once the system has accumulated sufficient local samples, gradually replace them with the system's own clean sample library.

[0061] This step outputs a sequence of operating condition categories. Clean dynamic baseline sequence and will Backfill to S102 and calculate hydraulic blockage intensity .

[0062] S104, combining the aforementioned dirt and clogging characteristics with the generated clean dynamic baseline to form an evidence set, performing credibility suppression and then performing nonlinear fusion to obtain a comprehensive dirt and clogging index, and extracting the trend and topological propagation amount.

[0063] This step transforms thermal, hydraulic, energy efficiency, water quality, and topological propagation information into unified fouling evidence, and obtains a stable comprehensive fouling index through credibility suppression and nonlinear fusion. Compared with simple linear weighting, nonlinear fusion can express the situation where the fouling risk is significantly enhanced when multiple pieces of evidence increase simultaneously, and can also suppress false alarms caused by single noisy evidence.

[0064] First, evidence of baseline deviation is constructed to avoid... When the denominator is too small near zero, a reference value for fouling thermal resistance is defined. : in: The normalized lower limit of fouling thermal resistance can be determined by the accuracy of heat transfer calculations or the cleanroom period. The upper bound of the fluctuation is determined.

[0065] Evidence of thermal fouling and deviation from thermal resistance for: Evidence of deviation in hydraulic pressure difference for: Evidence of overall heat transfer capacity degradation for: Evidence of energy efficiency degradation for: in: Emphasizing the increase in thermal resistance, Emphasizing the increase in pressure differential, Emphasizing the decline in heat exchange capacity, The study emphasizes the decline in energy efficiency. All four pieces of evidence are constructed relative to a clean baseline under similar operating conditions, thus reducing errors caused by variations in flow rate, inlet temperature, and load rate.

[0066] To demonstrate the driving force of wastewater quality on sedimentation, this embodiment constructs evidence of the driving force of wastewater sedimentation. .make Indicates the first Water quality indicators, such as , , , or First, robust standardization of water quality indicators is performed: Since water quality indicators below baseline should generally not increase the risk of sedimentation, a nonnegative standard value is defined as follows: Evidence for water sedimentation driving forces is defined as follows: in: For a single water quality indicator, the driving coefficient, This represents the interaction coefficient when two water quality indicators co-occur; a non-negative number is preferred. Interaction term. This is used to express situations where, for example, the simultaneous occurrence of high suspended solids and high turbidity poses a greater risk of sedimentation than a single increase. If certain water quality indicators are not collected online, the corresponding... and Set as Alternatively, low-frequency manual test values ​​can be maintained within the validity period.

[0067] In suppressing the credibility of evidence, this embodiment uses sliding variance. Measuring the short-term volatility of evidence. For evidence sequences. Its definition is: in, This is a window for evaluating the credibility of evidence, which can be set according to the data noise level and the rate of congestion evolution. If... Too short a timeframe can easily lead to misinterpreting normal fluctuations as instability. If the time is too long, it will reduce the speed of response to sudden anomalies.

[0068] The credibility of each piece of evidence was then suppressed. Credibility It can be determined by its short-term volatility or uncertainty: in: For the first The scale of fluctuation allowed by the evidence. This represents the moving variance. If a piece of evidence fluctuates drastically in the short term, then... The decrease indicates that the evidence may be affected by sensor noise, transient conditions, or calculation errors. The evidence after credibility suppression is as follows: In practice, the mean confidence level of the variables involved in the calculation of this evidence in S101 can also be used. Introduce a suppression term, for example, let However, this should be clearly stated in the implementation documents. The calculation range.

[0069] In obtaining Post-inhibition evidence Then, nonlinear fusion is performed using Choquet fuzzy integrals. In this embodiment... , respectively corresponding to .Will Sort by largest to smallest as follows: remember and order The comprehensive index of dirt and congestion for: in: For fuzzy measures, satisfying , , and when From time to time This constraint ensures that the larger the evidence set, the lower the fusion contribution.

[0070] To facilitate engineering parameterization, the following can be adopted: Fuzzy measure. For any disjoint set of evidence. and ,definition: The weight of a single element is denoted as ,satisfy .parameter It can be determined by the following normalized equation: when At that time, Choquet fusion degenerates into additive weighting, when When multiple pieces of evidence appear together, they can create a reinforcing effect. At that time, there was some redundancy among the evidence. and Corrections can be made using cleaning events, such as before cleaning. Higher, after cleaning The goal is to significantly reduce [the cost] as an optimization objective.

[0071] Fuzzy measure The determination of parameters is also part of this embodiment. The weight of a single element can be initialized according to the importance of thermal evidence, hydraulic evidence, energy efficiency evidence, and water quality evidence. For example, when the engineering site is more concerned with differential pressure risk, the weight can be increased. The single-element weight. When more attention is paid to the decline in heat transfer efficiency, the weight can be increased. or The single-element weights. Subsequently, the pre-cleaning events can be used. High, after cleaning Low objective function pairs Make corrections. Evidence credibility parameters. The corresponding evidence can be obtained at the standard deviation of the clean stability period. Doubled Times. Water quality driving parameters and It can be determined based on the correlation between water quality indicators and the rate of increase in pressure differential, cleaning cycle, or manually confirmed dirt blockage events.

[0072] To reflect the gradual degradation trend of dirt and grime, this embodiment extracts trend indicators from the comprehensive dirt and grime index. If the trend calculation span is... If you take a walk, then: in: Desirable to The specific value is the same as And related to system inertia. This indicates an increase in the degree of dirt and blockage. This indicates that the degree of blockage has stabilized or decreased. To reduce short-term fluctuations, it is also possible to first... Perform exponential smoothing before calculation .

[0073] Trend span In this embodiment, the determination is based on the rate of change of dirt and blockage and the noise level, preferably using... cover to The runtime. If the system's runtime fluctuations are small, A shorter length can be chosen to improve response speed; however, if the pressure difference and water quality data have significant noise, A longer value can be chosen to reduce misjudgment of trends. This parameter directly affects... Its smoothness and responsiveness are therefore explained in the context of the trend indicator formula.

[0074] Furthermore, to illustrate the coupling and propagation relationships among filters, heat exchangers, piping, and pump units, the system is constructed as a topology graph. Let the node evidence vector be... : in: This indicates evidence of filter node clogging. This indicates evidence of dirt and blockage at the heat exchanger nodes. This indicates evidence of dirt or blockage at pipe nodes. This indicates evidence of abnormal pump load. If the system only uses overall differential pressure and overall heat exchange parameters, this can also be used. It is simplified to an overall node evidence vector.

[0075] Let the adjacency matrix be The self-connection matrix is degree matrix Defined as: The normalized topological diffusion matrix is: The node evidence after system topology propagation is as follows: System coupling dirt blockage for: in: This is the node importance weight vector. For the whole vector. The weighting can be determined based on the impact of nodes on the system's heat exchange capacity, energy consumption, or failure consequences. For example, heat exchanger nodes and filter nodes can be assigned higher weights, while bypass pipe nodes can be assigned lower weights. Through topology propagation, when the filter differential pressure increases and causes abnormal pump power, It can reflect the systemic impact of this anomaly.

[0076] The above topological coupling calculation and Figure 3 and Figure 4 Correspondingly. Figure 3 A simplified topology of the wastewater-side filter, heat exchanger, piping, and circulating pump unit is shown, where the nodes... and nodes These represent the sewage inlet and sewage outlet, respectively, and are boundary nodes. This indicates that evidence of localized clogging in the filter primarily stems from deviations in the pressure differential across the filter, denoted as... Or may be by Representation. Node The evidence of localized fouling in heat exchangers primarily stems from fouling thermal resistance. Overall heat transfer capacity attenuation Or the pressure differential of the heat exchanger deviates. Node and nodes These represent different pipe sections, and the evidence of localized blockage can be expressed as follows: and .node For circulating pump sets, evidence of local anomalies may include deviations in pump power, pump frequency, or energy consumption. Characterization. Figure 3 The solid arrows in the diagram indicate the direction of sewage flow, illustrating the sequential connection between filters, heat exchangers, pipelines, and pump sets. The dashed evidence boxes indicate that each node can form evidence of localized blockage or abnormality.

[0077] exist Figure 4In topological propagation computation, node evidence vectors Used to summarize the time of each node Local evidence. If the system uses a seven-node topology, it can be written as: in: and It can represent the inlet and outlet boundary status or be set to... . Evidence of filter differential pressure deviation can be obtained. Evidence of heat exchanger fouling thermal resistance can be obtained. and The appropriate blockage strength can be selected for the corresponding pipe section. Evidence of pump unit energy consumption deviation can be obtained. If only some sensors are installed on-site, then the unobserved nodes... It can be estimated by propagation from neighboring nodes, or it can be set to... And reduce its impact by improving node credibility.

[0078] Figure 4 Adjacency matrix in Used to describe system connectivity. When there is a direct pipe connection between two nodes. When there is no direct connection between two nodes, Add a self-connect matrix Later formed The purpose is to ensure that nodes retain their own evidence while also absorbing evidence from neighboring nodes during propagation. Degree matrix The normalized diffusion matrix is ​​used to record the connectivity of each node. This is used to suppress scale differences between nodes with different connectivity. It is obtained after one topology diffusion. Then, through the node importance weight vector Summarized as system coupling congestion quantity .therefore, Figure 3 and Figure 4 This further illustrates that the present invention does not isolate the judgment of a single differential pressure point or heat exchanger, but rather utilizes the topology between devices to transform local anomalies into a system-level characterization of congestion that considers coupling relationships.

[0079] The evidence indicators and comprehensive index formed in this embodiment and Figure 5 The key indicators of dirt and clogging are compared in the corresponding histograms. Figure 5 Used to illustrate pressure differential deviation rate and fouling thermal resistance under different degrees of fouling. , Attenuation rate and comprehensive index The normalized response relationship. Figure 5The horizontal axis represents the operating status, including clean operation, slight fouling, moderate fouling, and severe fouling. The vertical axis represents the normalized index value, used to eliminate dimensional differences between pressure difference, thermal resistance, energy efficiency, and the comprehensive index. Normalization can be performed using the following formula: in: Indicates the first One indicator to be normalized and These represent the lower and upper limits of the indicator in the calibration sample or historical sample, respectively. This represents the normalized indicator value. For indicators that decrease with increasing contamination, such as... or It can be first converted into a decay rate and then normalized, for example: Figure 5 The differential pressure deviation rate is used to reflect the degree of increase in flow resistance, and can be determined by... or Characterization. Fouling thermal resistance Used to reflect the degree of fouling in a heat exchanger, it can be determined by... or Characterization. The attenuation rate is used to reflect the degree of decline in system energy efficiency, and can be determined by... Characterization, comprehensive index This is used to reflect the overall level of fouling after the fusion of multiple evidences. As the operating condition progresses from clean to severe fouling, the four types of indicators generally show a continuous upward trend, indicating that the thermal evidence, hydraulic evidence, energy efficiency evidence, and fusion index constructed in this invention can all provide a consistent response to the severity of fouling.

[0080] In implementation, Figure 5 This can be used to explain the state classification logic of this invention. Under clean operating conditions, all normalized indicators should be in the lower range; in the case of mild fouling, the pressure differential deviation rate or fouling thermal resistance may first show a slight increase, indicating that early deposition has occurred in the system; in the case of moderate fouling, the pressure differential, thermal resistance, Attenuation and Simultaneous increases indicate degradation in the thermal, hydraulic, and energy efficiency pathways; in cases of severe clogging, the comprehensive index... A value in the high range indicates that multiple pieces of evidence point to worsening congestion. Therefore, Figure 5 The content was used to support the rationality of the evidence construction and nonlinear fusion in S104, and to provide an interpretable basis for the risk level classification in the subsequent S105.

[0081] This step outputs the comprehensive dirt and blockage index. The trend of dirt and blockage System coupling dirt blockage and water sedimentation driving force These quantities together constitute the risk characteristics required for subsequent risk prediction.

[0082] S105, using the comprehensive dirt and blockage index, trend, topology propagation volume and cumulative operation volume, performs risk prediction, outputs failure probability and remaining warning time, and performs online updates after maintenance events.

[0083] The purpose of this step is to convert the comprehensive dirt and clogging index, trend, topology propagation volume, and cumulative operating volume obtained from S104 into an executable failure probability and remaining warning time. Compared with fixed threshold alarms, the time-varying risk model can output the probability of dirt and clogging failures occurring within the future maintenance window, and continuously adjust the model parameters based on cleaning, maintenance, or manual confirmation events.

[0084] The risk output process in this embodiment and Figure 6 Correspondingly. Figure 6 The time axis is used as the horizontal axis to display the comprehensive index of dirt and congestion. Trend indicators ,future Failure probability and remaining warning time The changing relationship. Figure 6 (a) This indicates the current overall level of congestion. The higher the value, the higher the level of congestion after the fusion of multi-source evidence. The warning threshold is used to indicate the boundary for judging whether the system enters the warning state from the normal state. Figure 6 (b) express The rate of change, if Persistently greater than This indicates that dirt and blockage are accumulating. If After cleaning, it dropped to near or less This indicates that maintenance measures have an inhibitory or restorative effect on the development of dirt and blockage. Figure 6 (c) Indicates the current moment Predicting the future The probability of internal blockage failure, threshold This indicates the risk assessment threshold. Figure 6 (d) or This indicates the remaining time to reach the preset risk threshold; as the risk increases, this value gradually decreases.

[0085] Figure 6 The warning trigger time corresponds to , or The time when preset alarm conditions are met. For example, the following alarm conditions can be set: in: The threshold for the comprehensive index. As a probability threshold, This represents the minimum lead time for maintenance. The determination of the aforementioned threshold is completed in this embodiment: Historical cleaning before The lower quantile or the empirical lower limit for manually confirmed contaminated samples. Can be selected based on operational risk preference to . It can be determined based on spare parts preparation time, shift schedule, and allowable downtime window, for example, taking , or By embedding the threshold determination method into the risk output step, the early warning strategy can be matched with specific operational conditions. Figure 6 The gray shaded area in the diagram represents the risk escalation zone, indicating that the system has entered a stage requiring close monitoring. The presence of this area demonstrates that the output of this invention does not simply provide a binary result of whether a fault has occurred, but rather presents a continuous risk evolution process from normal, attention-based, warning to alarm. (The right-hand output box...) , , , Together with alarm levels, they constitute operational decision-making information, enabling operational personnel to simultaneously grasp the current status, rate of change, probability of recent failures, and remaining maintainable time.

[0086] First, define the risk feature vector. : in: This indicates the cumulative operating volume since the last cleaning or maintenance. You can use the cumulative runtime, cumulative sewage flow, or cumulative heat exchange integral, or a weighted combination thereof: in: This indicates the time since the most recent cleaning, maintenance, or confirmation of restoration to cleanliness. , , This is a cumulative weight. If only the cumulative runtime is used in the implementation, then it can be set as follows: , , .

[0087] Constructing time-varying hazard rates : in: Basic hazard rate, This is the risk feature weight vector. This represents the instantaneous risk intensity of a dirt blockage failure occurring per unit of time. Because it uses an exponential form, Always positive, and when , , or As the risk increases, so does the base hazard rate. and weight vector The initial value can be estimated from historical failure events, cleaning events, or manually confirmed events. If there is insufficient event data, it can be initially estimated based on... , and The high quantile values ​​are used to set an empirical threshold, which is then converted into initial risk weights. In subsequent operations, and Online updates are performed through cleaning, maintenance, or manual confirmation of events.

[0088] Set the maintenance and operation focus window as The corresponding distance from the walk number is: future The probability of a dirt blockage failure occurring within a given time period is: in: The time when the dirt blockage failure occurs. Indicates at time For the future The predicted hazard rate for each step. If there is no future operating condition plan, the prediction can be made by maintaining the current risk or extrapolating a linear trend: And Substitution calculate To avoid over-extrapolation, one can... Set upper and lower limits, for example, limit it to history. of Partial to Within the range of quantiles.

[0089] At confidence level Below, remaining warning time Defined as the minimum time it takes for the failure probability to first reach the threshold: in: The alarm confidence threshold can be set to [value]. , , Or it may be determined by the operation and maintenance strategy. When Less than the preset maintenance lead time When, the system outputs a warning; when Exceeding the probability threshold At that time, the system outputs a fault risk alarm.

[0090] When the system undergoes cleaning, maintenance, or manual verification events, an online calibration tag is constructed. . Can be defined by the event window: if the time If it is located within the window of confirmed dirt blockage or pre-cleaning anomalies, then If at any time If it is located within the stable clean window after cleaning, then If the sample state is uncertain, the sample will not participate in the update or will be given a lower update weight. To achieve lightweight online updates, this embodiment can use recursive least squares updates. : in: For the gain vector, Let covariance be the parameter. Forgetting factor, preferred selection to . The smaller the size, the faster the model adapts to seasonal changes and long-term drift of wastewater quality; The closer The more stable the model update, the better. The initial covariance matrix can be taken as... ,in: For larger positive numbers, This is an identity matrix. If historical event data is already available, it can also be obtained through maximum likelihood estimation or batch regression. and The initial value of .

[0091] Furthermore, the online self-calibration and maintenance effect evaluation process in this embodiment is similar to... Figure 7 Correspondingly. Figure 7 Plotting long-term operating time on the x-axis and normalized values ​​on the y-axis, it also shows... Deviation rate Attenuation rate, comprehensive index and failure probability The evolution curve. Figure 7 The risk accumulation stage in the system indicates that during continuous operation, suspended solids, organic matter, grease, biofilm, or inorganic sediments gradually adhere to filters, heat exchangers, or pipelines, leading to... Increased deviation rate Increased attenuation rate This increases and drives up the probability of failure. This stage corresponds to the continuous online execution and output of risk indicators from S101 to S105.

[0092] Figure 7 The cleaning events in the system correspond to filter backflushing, heat exchanger cleaning, pipeline flushing, pump unit maintenance, or manual confirmation of recovery events in actual operation and maintenance. After the event occurs, the system should record the cleaning or maintenance time as... And reinitialize the cumulative runtime, for example: Stable recovery window after cleaning Within the clean sample library, samples can be labeled as clean or low-risk samples for updating the clean sample library. : in: This represents the upper limit of the comprehensive cleanliness index after cleaning. For thermal equilibrium residuals, This is the thermal balance residual threshold. If after cleaning... Deviation rate Attenuation rate and If all indicators show a significant decrease, it indicates that the maintenance was effective in restoring the system to its normal state. If the above indicators do not decrease significantly after cleaning, it may indicate a sensor malfunction, irreversible scaling on the heat exchanger, unresolved partial blockage in the pipeline, or degradation of the pump unit's performance.

[0093] Figure 7 The cleaning and recovery phase in the model represents the system re-entering its operational cycle after maintenance, with key indicators slowly rising from lower levels. The curve shape during this phase can be used to evaluate the model's self-calibration effectiveness: if after cleaning... and The timely decline indicates that the cleanliness baseline, evidence fusion, and risk model are responding correctly to maintenance events. If the indicators subsequently rise slowly again, it suggests that the model is continuously tracking new dirt and grime accumulation. Figure 7 It is evident that this application can not only predict failure probability, but also update the clean baseline, reset the cumulative operating volume, correct model parameters, and evaluate maintenance effectiveness based on cleaning events.

[0094] During the execution of this embodiment, if , , , or If an invalid value is found, that moment is marked as a low-confidence sample, and model parameter updates are skipped. If water quality indicators are missing, the corresponding sedimentation driving force component can be set to... Alternatively, the most recent valid value can be retained, and its confidence level can be reduced using the missing value flag in S101. This boundary processing is directly embedded in the risk prediction step, preventing abnormal inputs from causing incorrect self-learning of the risk model.

[0095] This embodiment ultimately outputs the failure probability. Remaining warning time Risk level and updated model parameters The risk level can be set to four levels: Normal, Attention, Early Warning, and Alarm, according to project needs. For example, when... and This is normal; when In to Attention is always on our minds; when or This serves as an early warning; when or An alarm will be triggered at this time. The above thresholds are for illustrative purposes only and can be adjusted based on maintenance costs, downtime risks, and cleaning cycles.

[0096] Through the above steps, this application can unify operational data of different frequencies and confidence levels into a time series that can be used for joint calculation, and decompose the fouling state into multiple interpretable evidences such as thermal fouling resistance, hydraulic pressure difference deviation, water quality deposition drive, energy efficiency degradation, and topology propagation effects. Since all major evidences are constructed relative to a clean dynamic baseline under the same operating conditions, this application can reduce the interference of flow rate changes, inlet temperature changes, load changes, and pump frequency changes on the prediction results. By introducing confidence suppression and Choquet nonlinear fusion, this application can improve early warning sensitivity when multiple pieces of evidence co-occur, while suppressing false alarms caused by a single anomalous sensor. By introducing time-varying hazard rates and online self-calibration, this invention can continuously adapt to seasonal water quality drift based on cleaning, maintenance, and manual confirmation events, thereby achieving early prediction and continuous optimization of fouling faults in wastewater source heat pump systems.

[0097] Corresponding to the above embodiment's method for predicting blockage faults in a wastewater source heat pump system by integrating operational data, this application also provides an embodiment of a system for predicting blockage faults in a wastewater source heat pump system by integrating operational data.

[0098] See Figure 8 The wastewater source heat pump system clogging and fault prediction system 20, which integrates operational data, includes: The multi-source data credibility fusion module 201 is used to collect multi-source operating data from the sewage source heat pump system, map them uniformly onto a time grid, and perform credibility weighted fusion at a unified time to generate credibility fusion data.

[0099] The thermal and hydraulic fouling feature quantification module 202 is used to construct thermal and hydraulic features based on the credibility fusion data, and to quantify the fouling features to obtain fouling features.

[0100] The operating condition classification and clean baseline construction module 203 is used to construct an operating condition vector and classify the operating conditions based on the thermal and hydraulic characteristics, screen clean samples within the same operating condition, and establish a clean dynamic baseline.

[0101] The clogging index and degradation feature extraction module 204 is used to combine the clogging features with the generated clean dynamic baseline to form an evidence set, perform credibility suppression and nonlinear fusion to obtain the clogging comprehensive index, and extract the trend and topological propagation.

[0102] The fault risk prediction and adaptive update module 205 is used to predict risks, output fault probability and remaining warning time using the comprehensive dirt and blockage index, trend, topology propagation volume and cumulative operation volume, and perform online updates after maintenance events.

[0103] For the system implementation, since the methods for implementing the modules are the same as those in the above method implementation, the description is relatively simple. For relevant details, please refer to the description in the method implementation.

[0104] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0105] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method for predicting dirt blockage faults in a wastewater source heat pump system by integrating operational data, characterized in that, include: Multi-source operation data from the wastewater source heat pump system are collected, mapped onto a time grid, and weighted and fused at a unified time to generate credible fused data. Based on the aforementioned credibility fusion data, thermal and hydraulic features are constructed, and the dirt clogging features are quantified to obtain the dirt clogging characteristics. Based on the aforementioned thermal and hydraulic characteristics, a working condition vector is constructed and the working conditions are classified. Clean samples are screened within the same working condition and a clean dynamic baseline is established. The evidence set is formed by combining the aforementioned dirt and clogging characteristics with the generated clean dynamic baseline. After performing credibility suppression, nonlinear fusion is performed to obtain the comprehensive dirt and clogging index, and the trend and topological propagation are extracted. The system utilizes the comprehensive dirt and congestion index, trend, topology propagation volume, and cumulative operation volume to predict risks, output failure probability and remaining warning time, and performs online updates after maintenance events.

2. The method for predicting blockage faults in a wastewater source heat pump system by integrating operational data as described in claim 1, characterized in that, The multi-source operational data collected from the wastewater source heat pump system are uniformly mapped onto a time grid, and weighted and fused at a unified time to generate reliable fused data, including: Collect multi-source operational data from the monitoring platform of the wastewater source heat pump system; For data with different sampling frequencies or missing data, each variable is mapped to a unified time grid, and neighboring sampled values ​​are weighted using a kernel function. The sample confidence weights are calculated by combining the normalized residuals, the degree of normalized mutations, and the missing markers, and the resulting fused running data sequence is output at a unified time.

3. The method for predicting blockage faults in a wastewater source heat pump system by integrating operational data as described in claim 2, characterized in that, The step of calculating sample confidence weights by combining normalized residuals, normalized mutation degrees, and missing markers to output a unified time-sequence fused running data sequence includes: Normalized residuals Normalized mutation degree and missing or imputation indicators Mapped to credibility weights: in: This represents the minimum confidence level. This represents the confidence bias parameter. , , These represent the penalty coefficients for residuals, abrupt changes, and missing or imputed terms, respectively. For any variable At the unified moment Construction of credibility fusion data : in: Indicates the first Class variables at sampling time The sampled values, Indicates the first The set of sampling times for class variables. Indicates the sample confidence weight. Represents the time kernel function. Indicates the kernel width. Indicates a uniform time step. This indicates a smoothing term to prevent the denominator from being zero.

4. The method for predicting blockage faults in a wastewater source heat pump system by integrating operational data as described in claim 1, characterized in that, The process of constructing thermal and hydraulic features based on the credibility fusion data and quantifying the clogging features to obtain clogging characteristics includes: Based on the fused data, interpretable features of the thermal and hydraulic links are constructed respectively, transforming dirt and clogging from indirect indicators into accumulative thermal resistance and clogging intensity, thus eliminating the impact of operating condition changes on the indicators. It generates two explicit fouling quantities: thermal resistance and hydraulic pressure difference, and outputs heat exchange, logarithmic mean temperature difference, and comprehensive heat transfer capacity.

5. The method for predicting blockage faults in a wastewater source heat pump system by integrating operational data as described in claim 4, characterized in that, The method of constructing interpretable features of thermal and hydraulic links based on fused data transforms fouling from an indirect indicator into accumulative thermal resistance and fouling intensity, eliminating the impact of operating condition changes on the indicators, including: Based on the fused data, the heat exchange on the wastewater side was calculated respectively. Heat exchange with load side ; The heat exchange balance residual is calculated based on the heat exchange on the wastewater side and the heat exchange on the load side. , and in conjunction with the above Obtain observation of comprehensive heat transfer capabilities ; Based on the above Perform thermal resistance series decomposition and inversion of fouling thermal resistance : in: Indicates the convective heat transfer coefficient on the wastewater side. Indicates the convective heat transfer coefficient on the load side. Indicates the heat exchange area. Indicates the thermal resistance of the heat exchange wall. Indicates the thermal resistance of dirt; And based on the observed pressure difference Pressure difference from clean baseline Obtaining hydraulic plugging strength : in: Represents a smooth nonnegative mapping function. Indicates the observed pressure difference. This represents the clean baseline pressure difference.

6. The method for predicting blockage faults in a wastewater source heat pump system by integrating operational data as described in claim 1, characterized in that, The step of constructing a working condition vector based on the thermal and hydraulic characteristics and classifying the working conditions, screening clean samples within the same working condition, and establishing a clean dynamic baseline includes: Operating condition vectors are constructed based on flow rate, temperature, pump frequency, valve opening degree, and load rate. These operating conditions are then classified to form a sequence of operating condition categories, including: based on operating condition vectors. Calculate which sample belongs to the first... The posterior probability of each working condition category is calculated, and the current working condition category is determined. : in: Indicates the first Prior weights for each work condition category, Indicates the first The center vector of each working condition category Indicates the first Covariance matrix of each working condition category Represents the multivariate Gaussian density function; Establishing a dynamic baseline by screening clean samples under the same operating conditions makes the deviation of dirt and clogging indicators comparable under different operating conditions, providing a reference for constructing evidence of dirt and clogging, including: In operating condition category A clean dynamic baseline is generated using similarity kernel regression: in: Indicates the clean dynamic baseline, Indicates the type of working condition The clean sample library below The baseline output vector represents the clean sample. This represents the similarity weight between the current sample and the clean sample.

7. The method for predicting blockage faults in a wastewater source heat pump system based on integrated operational data as described in claim 4, characterized in that, The evidence set is formed by combining the aforementioned fouling characteristics with the generated clean dynamic baseline. After performing credibility suppression, nonlinear fusion is performed to obtain the fouling comprehensive index, and the trend and topological propagation are extracted, including: The explicit amount of dirt and clogging is combined with the clean dynamic baseline to form multidimensional evidence of dirt and clogging, and the impact of abnormal noise is reduced by credibility suppression. Then, nonlinear fuzzy fusion is used to obtain the comprehensive dirt and blockage index, while extracting the dirt and blockage trend and constructing the system topology coupling quantity to reflect the coupling propagation and gradual degradation characteristics between devices.

8. The method for predicting blockage faults in a wastewater source heat pump system by integrating operational data as described in claim 7, characterized in that, The process then employs nonlinear fuzzy fusion to obtain a comprehensive congestion index, simultaneously extracting congestion trends and constructing system topology coupling quantities to reflect the coupling propagation and gradual degradation characteristics between devices, including: Evidence with suppressed credibility Sort by size as follows: remember: and order The comprehensive index of dirt and congestion is obtained by using Choquet fuzzy integral. : in: Indicates the amount of evidence. Represents a set The corresponding fuzzy measure; Based on the comprehensive dirt and congestion index Extracting trends: in, Indicates a trend of dirt and congestion. Indicates the span of trend calculation; Calculate the system coupling congestion amount based on system topology propagation. : in: Represents the node evidence vector. Represents the normalized topological diffusion matrix. This represents node evidence after topology propagation. Represents the node weight vector. This represents a vector consisting entirely of 1s.

9. The method for predicting blockage faults in a wastewater source heat pump system by integrating operational data as described in claim 1, characterized in that, The process of using the comprehensive congestion index, trend, topology propagation volume, and cumulative operational volume to predict risks, output failure probability and remaining warning time, and perform online updates after maintenance events includes: A risk feature vector is constructed using the comprehensive dirt and blockage index, trend, topology propagation volume, and cumulative operation volume. This vector is then input into a time-varying hazard rate model to calculate the probability of dirt and blockage failures occurring in the future and the remaining warning time. When cleaning, maintenance, or manual confirmation events occur, the model parameters are updated online using event feedback to achieve adaptive fouling prediction and risk calibration across seasons and water quality conditions.

10. The method for predicting blockage faults in a wastewater source heat pump system based on integrated operational data as described in claim 9, characterized in that, The method involves constructing a risk feature vector using a comprehensive dirt-clogging index, trend, topology propagation volume, and cumulative operational volume. This vector is then input into a time-varying hazard rate model to calculate the probability of dirt-clogging failures occurring in the future and the remaining warning time. This includes: Based on risk feature vector Constructing a time-varying hazard rate model : in, Indicates the basic hazard rate. This represents the risk model weight vector; Calculating the Future The probability of a dirt blockage failure occurring within a given time period: in: Indicates the time when the dirt blockage fault occurred. Indicates the forecast time window, Indicates a uniform time step; At risk threshold Next, determine the remaining warning time. : in, This represents the probability threshold for triggering a risk warning.

Images