Distributed wastewater treatment plant water quality soft-sensing method and system

Abstract

The application relates to a distributed sewage treatment facility water quality soft measurement method and system, which comprises the following steps: obtaining an operation electrical signal of a target power equipment in a distributed sewage treatment facility, and analyzing and generating an electrical energy fingerprint data sequence; based on a coupling mechanism of fluid rheology and oxygen mass transfer dynamics, performing decoupling operation on the electrical energy fingerprint data sequence to generate a first physical mechanism feature vector representing fluid physical rheological properties and a second physical mechanism feature vector representing gas-liquid contact surface mass transfer resistance, and then inputting into a preset cross-domain adaptive mapping model for calculation, outputting a target water quality monitoring index, and comparing the target water quality monitoring index with a preset safety threshold to generate an operation and maintenance control instruction for the target power equipment. The application has the advantages that rheology and oxygen mass transfer dynamics characteristics can be separated from the electrical signal, and individual differences of equipment and environmental noise interference can be overcome.

Description

Technical Field

[0001] This application relates to the field of water quality monitoring, and in particular to a soft measurement method and system for water quality in a distributed wastewater treatment facility. Background Technology

[0002] Currently, conventional water quality monitoring in the industry mainly relies on online chemical analysis instruments or optical sensors. However, in distributed sites, especially in unattended scenarios such as remote rural areas, these traditional hardware monitoring methods face numerous engineering constraints. Chemical analysis instruments require regular addition of reagents and cleaning of waste discharge pipelines, and the high-frequency mechanical movements can easily cause clogging of sampling pumps. Optical sensors, after prolonged immersion in activated sludge mixtures, are prone to biofilm buildup on their probe surfaces, leading to decreased light transmittance and measurement data drift. These factors increase the operation and maintenance costs of distributed facilities, and the continuity of water quality monitoring data is difficult to guarantee.

[0003] In existing soft sensing technologies for water quality, some schemes attempt to directly collect voltage or current data from wastewater treatment equipment and input it into conventional machine learning models to infer water quality. However, the operating electrical signals of power equipment such as submersible sewage pumps and aeration blowers are the result of the combined effects of multiple factors, including fluid dynamics load, equipment mechanical friction, and power grid fluctuations. Changes in the viscosity of the fluid medium and the mass transfer resistance at the gas-liquid interface are only one of the factors causing changes in the electrical load of the equipment.

[0004] When the power output of equipment at different wastewater treatment plants varies, or when there are transient fluctuations in the regional power grid, the original electrical signals will experience significant baseline shifts. This can cause models to misinterpret increased mechanical resistance due to equipment aging or decreased grid voltage as deterioration of water quality indicators. Furthermore, prediction models trained using data from specific benchmark sites show a significant decrease in generalization ability when transferred to ordinary rural sites with different equipment hardware attributes. This is due to the lack of mechanisms to suppress individual equipment differences, resulting in substantial errors in the output water quality indicators. Summary of the Invention

[0005] In order to separate rheological and oxygen mass transfer kinetic characteristics from electrical signals and overcome individual equipment differences and environmental noise interference, this application provides a soft measurement method and system for water quality in distributed wastewater treatment facilities.

[0006] Firstly, this application provides a soft measurement method for water quality in a distributed wastewater treatment facility, which adopts the following technical solution:

[0007] A method for soft water quality measurement in a distributed wastewater treatment facility includes the following steps:

[0008] S1. Acquire the operating electrical signals of the target power equipment in the distributed wastewater treatment facility, and parse and generate an electrical fingerprint data sequence; the target power equipment includes a submersible sewage pump and an aeration blower;

[0009] S2. Based on the coupling mechanism of fluid rheology and oxygen mass transfer dynamics, the electrical energy fingerprint data sequence is decoupled to generate a first physical mechanism feature vector characterizing the physical rheological properties of the fluid, and a second physical mechanism feature vector characterizing the mass transfer resistance at the gas-liquid interface.

[0010] S3. Input the first physical mechanism feature vector and the second physical mechanism feature vector into a preset cross-domain adaptive mapping model for calculation, and output the target water quality monitoring index; wherein, the cross-domain adaptive mapping model is constructed based on transfer learning training of source domain data and target domain data;

[0011] S4. Based on the comparison results between the target water quality monitoring index and the preset safety threshold, generate operation and maintenance control instructions for the target power equipment.

[0012] Optionally, step S1 includes the following sub-steps:

[0013] S11. Collect the operating electrical signals of the target power equipment, wherein the operating electrical signals include at least the active power time series, the current time series, and the voltage time series;

[0014] S12. The operating electrical signal is sampled according to a preset feature capture frequency, and the active power noise variance feature and current harmonic distortion rate feature are calculated within a preset time window to generate the electrical energy fingerprint data sequence; wherein, the feature capture frequency is set to be greater than or equal to the pressure pulsation frequency generated by the impeller rotation of the submersible sewage pump, and greater than or equal to the current noise frequency generated by the bubble breakage corresponding to the aeration blower.

[0015] Optionally, step S2 includes the following sub-steps:

[0016] S21. Extract the data components associated with the submersible sewage pump from the electrical fingerprint data sequence, calculate and generate the rheological equivalent viscosity index based on the mapping relationship between the non-Newtonian fluid viscosity characteristics and the pumping resistance, and use the rheological equivalent viscosity index as the first physical mechanism feature vector.

[0017] S22. Extract the data components associated with the aeration blower from the electrical fingerprint data sequence, calculate and generate the oxygen mass transfer resistance coefficient based on the inhibitory effect of sewage impurities on oxygen transfer efficiency, and use the oxygen mass transfer resistance coefficient as the second physical mechanism feature vector.

[0018] Optionally, step S21 includes the following sub-steps:

[0019] S211. Obtain the active power characteristics of the submersible sewage pump, and subtract a preset mechanical friction loss value and a preset gravitational potential energy work value from the active power characteristics to obtain the net fluid work power.

[0020] S212. Obtain the flow monitoring data of the submersible sewage pump, and divide the power of the clean fluid by the product of the flow monitoring data and the motor efficiency to generate the rheological equivalent viscosity index; use the feature set containing the rheological equivalent viscosity index as the first physical mechanism feature vector.

[0021] Optionally, step S22 includes the following sub-steps:

[0022] S221. Obtain the active power characteristics of the aeration blower, and calculate the active power noise variance characteristics within a preset stable operation window;

[0023] S222. Perform spectral analysis on the operating electrical signal of the aeration blower to identify the power spectral density characteristics of specific frequency components caused by the generation and collapse of bubbles at the aeration head of the aeration blower;

[0024] S223. Calculate the ratio between the power spectral density feature and the power noise variance feature to generate the oxygen transfer resistance coefficient, which characterizes the degree of inhibition of the oxygen transfer process at the gas-liquid interface by organic impurities in wastewater, and use the feature set containing the oxygen transfer resistance coefficient as the second physical mechanism feature vector.

[0025] Optionally, the target water quality monitoring indicators include virtual sludge concentration and virtual chemical oxygen demand; the cross-domain adaptive mapping model is constructed based on an adversarial domain adaptation network, which includes a domain discriminator, a feature extractor, and a cascaded regressor; step S3 includes the following sub-steps:

[0026] S31. Using the feature extractor, the input first physical mechanism feature vector and the second physical mechanism feature vector are mapped to extract common electrical fingerprint features that are sensitive to water quality changes but not sensitive to individual differences of the target power equipment.

[0027] S32. Using the first calculation unit of the cascaded regressor, the virtual sludge concentration is calculated and output based on the common electrical fingerprint characteristics;

[0028] S33. Using the second calculation unit of the cascaded regressor, the common electrical fingerprint features and the virtual sludge concentration are recombined into a joint input vector, and the virtual chemical oxygen demand is calculated and output based on the joint input vector.

[0029] Optionally, the adversarial domain adaptation network is pre-built through a training process that includes the following sub-steps:

[0030] S301. Obtain the source domain original dataset of the benchmark site and the target domain original dataset of the ordinary site; using the decoupling operation described in step S2, convert the source domain original dataset into a source domain feature dataset containing a first physical mechanism feature vector and a second physical mechanism feature vector, and convert the target domain original dataset into a target domain feature dataset containing a first physical mechanism feature vector and a second physical mechanism feature vector; wherein, both the source domain original dataset and the target domain original dataset are composed of multiple energy fingerprint data sequences;

[0031] S302. Input the source domain feature dataset and the target domain feature dataset into the feature extractor to generate corresponding intermediate feature representations; input the intermediate feature representations into the domain discriminator for domain discrimination; and adjust the feature extractor based on the discrimination result output by the domain discriminator to suppress information about device differences between the source domain feature dataset and the target domain feature dataset in the intermediate feature representation, thereby generating a common power fingerprint feature with unified dimensions;

[0032] S303. Input the common electrical energy fingerprint features into the cascaded regressor and perform step-by-step training according to the following logic: Use the real sludge concentration labels in the source domain feature dataset that are time-aligned with the electrical energy fingerprint data sequence to perform supervised training on the first computing unit of the cascaded regressor, and determine the regression parameters by which the first computing unit maps the common electrical energy fingerprint features to virtual sludge concentration; combine the common electrical energy fingerprint features and the virtual sludge concentration into a joint input vector, and use the real chemical oxygen demand (COD) labels in the source domain feature dataset to perform supervised training on the second computing unit of the cascaded regressor, and determine the regression parameters by which the second computing unit maps the joint input vector to virtual COD;

[0033] S304. In response to the detection of newly added source domain data or target domain data, the basic weights that have been fixed in the cross-domain adaptive mapping model are maintained, the loss gradient is calculated using only the newly added source domain data or target domain data, and the local network parameters of the feature extractor, the cascaded regressor and the domain discriminator are fine-tuned online through the gradient inversion layer.

[0034] Optionally, it also includes step S5, which includes the following sub-steps:

[0035] S51. In response to the received maintenance work order status data, when the maintenance work order status data indicates that the maintenance is completed or the cleaning is completed, it is determined that the target power equipment has entered the clean water operation state; wherein, the maintenance work order status data is digital status information generated based on the equipment maintenance records fed back by the maintenance management terminal;

[0036] S52. Obtain the electrical energy fingerprint data sequence when the target power equipment is in the clean water operation state, and execute steps S2 to S3 to obtain the virtual sludge concentration benchmark value and / or virtual chemical oxygen demand benchmark value output by the cross-domain adaptive mapping model in the clean water operation state.

[0037] S53. Compare the virtual sludge concentration benchmark value and / or virtual chemical oxygen demand benchmark value with the corresponding preset clean water standard value, and calculate the corresponding baseline drift amount;

[0038] S54. The baseline drift is used to perform linear compensation correction on the corresponding virtual sludge concentration and / or virtual chemical oxygen demand to obtain the corrected target water quality monitoring index.

[0039] Optionally, it also includes step S6, which includes the following sub-steps:

[0040] S61. Acquire image data collected by on-site video monitoring equipment and analyze it to generate water clarity index;

[0041] S62. When the target water quality monitoring index shows that the water quality exceeds the standard, and the effluent clarity index is higher than the preset clarity threshold, a conflict alarm signal is triggered;

[0042] S63. Suspend the operation and maintenance control command and generate a manual review work order.

[0043] Secondly, the distributed wastewater treatment facility water quality soft measurement system provided in this application adopts the following technical solution:

[0044] A distributed wastewater treatment facility water quality soft measurement system, comprising:

[0045] The data acquisition module, deployed on the field intelligent gateway, is used to collect the operating electrical signals of the target power equipment and generate an electrical energy fingerprint data sequence;

[0046] The video surveillance module is used to collect image data from the scene.

[0047] A cloud computing platform, which is communicatively connected to the data acquisition module and the video monitoring module, is used to execute the above-mentioned soft measurement method for water quality in distributed wastewater treatment facilities.

[0048] The operation and maintenance management terminal is used to receive target water quality monitoring indicators and send operation and maintenance work order status data to the cloud computing platform.

[0049] In summary, this application includes at least one of the following beneficial technical effects:

[0050] 1. The cloud computing platform constructs an adversarial game between the domain discriminator and the feature extractor through a gradient inversion layer, eliminating the equipment background noise caused by differences in motor iron loss, pipeline head loss, and grid voltage fluctuations between the source and target domains. This decoupling process protects the cross-domain adaptive mapping model from electrical interference introduced by hardware aging or replacement when correcting water quality prediction biases. This allows for the lossless transfer of network parameters trained on benchmark sites to ordinary sites equipped only with power acquisition modules. By leveraging power fingerprinting to offset the lack of chemical analysis instruments, high-precision full-domain coverage can be achieved at low cost.

[0051] 2. The mechanism feature decoupling unit decomposes the electrical fingerprint data sequence into the equivalent viscosity index, which characterizes the shear resistance of non-Newtonian fluids, and the oxygen mass transfer resistance coefficient, which characterizes the interfacial tension between the gas and liquid phases. This decoupling step cuts off the energy transfer coupling path between the submersible sewage pump impeller turbulence impedance and the energy consumption of bubble collapse in the aeration blower, preventing transient fluctuations in sludge concentration from causing a cross-shadowing effect on chemical oxygen demand (COD) prediction.

[0052] 3. The operation and maintenance management terminal captures the work order status data issued upon completion of equipment cleaning or maintenance as a physical anchor point calibration trigger signal. The cloud computing platform responds to the above signal and locks the electrical fingerprint data sequence of the target power equipment when it is in clean water operation, calculating the baseline drift between the current model output value and the clean water standard constant. This baseline drift is injected back into the generation link of the target water quality monitoring index as a linear compensation component, physically offsetting the long-term increase in mechanical resistance caused by impeller wear or microporous aeration head blockage, achieving model adaptive zeroing without human intervention and unaffected by the equipment life cycle. Attached Figure Description

[0053] Figure 1 A flowchart illustrating a method for soft water quality measurement in a distributed wastewater treatment facility according to an embodiment of the present invention is shown. Detailed Implementation

[0054] The present application will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the scope of the application.

[0055] This application discloses a soft measurement method for water quality in a distributed wastewater treatment facility. This method achieves water quality monitoring without chemical probes by exploring the implicit physical mapping relationship between the operating load of power equipment and water quality parameters. (Refer to...) Figure 1The method mainly includes the following steps S1 to S4.

[0056] S1. Acquire the operating electrical signals of the target power equipment in the distributed sewage treatment facility, and parse them to generate an electrical fingerprint data sequence; the target power equipment includes a submersible sewage pump and an aeration blower.

[0057] The on-site intelligent gateway is equipped with open-type current transformers and voltage transformers to directly collect operating electrical signals from the three-phase power supply lines of the submersible sewage pumps and aeration blowers. This hardware architecture maintains the integrity of the original electrical wiring and process piping, achieving non-intrusive data acquisition. These operating electrical signals are defined as electrical fingerprints. Electrical fingerprints differ from traditional billing electricity data; they characterize the electrical fluctuations modulated on the fundamental power supply frequency by the reverse transmission of fluid reaction force via the mechanical shaft to the motor stator side when the microscopic resistance characteristics of the fluid medium change.

[0058] Fluid motion possesses dynamic and continuous properties, and instantaneous electrical values ​​at a single point are easily affected by grid voltage dips or electromagnetic interference. The intelligent gateway introduces a time-domain dimension, aligning high-frequency sampled discrete electrical data by timestamps and parsing them to generate an electrical fingerprint data sequence. This electrical fingerprint data sequence reflects the dynamic evolution trajectory of equipment load as the fluid state fluctuates, preserving the frequency and amplitude information of load fluctuations, thus providing a data foundation for subsequent mechanism decoupling.

[0059] Distributed, unmanned rural wastewater treatment plants are limited by operational and maintenance conditions; online chemical oxygen demand (COD) analyzers are prone to failure due to reagent depletion or probe scaling. A smart gateway utilizes continuously operating sewage pumps and fans as sensing terminals, converting equipment energy consumption sequences into water quality monitoring data. This solution replaces chemical analysis instruments, reducing the construction cost and daily maintenance requirements of the water quality monitoring system.

[0060] Optionally, step S1 includes the following sub-steps S11-S12.

[0061] S11. Collect the operating electrical signals of the target power equipment, wherein the operating electrical signals include at least the active power time series, the current time series and the voltage time series.

[0062] S12. The operating electrical signal is sampled according to a preset feature capture frequency, and the active power noise variance feature and current harmonic distortion rate feature are calculated within a preset time window to generate the electrical energy fingerprint data sequence; wherein, the feature capture frequency is set to be greater than or equal to the pressure pulsation frequency generated by the impeller rotation of the submersible sewage pump, and greater than or equal to the current noise frequency generated by the bubble breakage corresponding to the aeration blower.

[0063] The intelligent gateway synchronously collects multi-dimensional electrical signals. The voltage time series is used to eliminate power grid fluctuations, the current time series reflects the instantaneous load of the motor-driven load, and the active power time series, composed of multiple active power signals in chronological order, characterizes the effective mechanical work done by the equipment to overcome fluid resistance. The non-uniformity and turbulent perturbations generated when high-concentration sludge flows through the pump chamber are mapped to the active power noise variance. The nonlinear acoustic pulsations generated by bubble bursts at the micropores of the aeration blower are transmitted in reverse to the motor side, modulating the current harmonic distortion rate.

[0064] The calculation module truncates electrical signals within a preset time window to filter out transient spikes in the power grid, ensuring that the generated statistical features characterize the steady state of fluid physics.

[0065] According to the Nyquist sampling theorem, the feature capture frequency must exceed the physical lower limit of the impeller pressure pulsation frequency and the bubble breakage frequency to prevent high-frequency hydraulic or aerodynamic features from being submerged in the low-frequency fundamental wave, thereby eliminating feature aliasing distortion.

[0066] During the wastewater biological treatment cycle, increased sludge concentration leads to increased fluid viscosity. The submersible sewage pump impeller performs high-frequency shearing on the viscous fluid; this shearing torque is transmitted through the rotor shaft system, inducing specific high-frequency current harmonic components in a frequency band higher than the impeller's mechanical rotation frequency. Feature capture frequency identifies and captures these high-frequency components, converting them into part of an electrical fingerprint data sequence.

[0067] S2. Based on the coupling mechanism of fluid rheology and oxygen mass transfer dynamics, the electrical energy fingerprint data sequence is decoupled to generate a first physical mechanism feature vector characterizing the physical rheological properties of the fluid, and a second physical mechanism feature vector characterizing the mass transfer resistance at the gas-liquid interface.

[0068] Directly inputting raw high-dimensional electrical signals into a black-box model leads to dimensionality mismatch and loss of physical meaning. The cloud computing platform utilizes fluid dynamics formulas to perform prior dimensionality reduction and mapping on the electrical signals, transforming abstract electrical signals into concrete fluid physical quantities. Decoupling operations physically separate the rheological properties of the submersible sewage pump from the mass transfer resistance of the corresponding aeration blower. The mechanical work of the submersible sewage pump primarily overcomes fluid viscosity, while the work fluctuations of the aeration blower mainly stem from changes in surface tension at the gas-liquid interface. This forced decoupling mechanism cuts off the energy coupling path between power devices, preventing cross-shadowing interference from sludge viscosity fluctuations on oxygen mass transfer resistance characteristics.

[0069] Optionally, step S2 includes the following sub-steps S21-S22.

[0070] S21. Extract the data components associated with the submersible sewage pump from the electrical fingerprint data sequence, calculate and generate the rheological equivalent viscosity index based on the mapping relationship between the non-Newtonian fluid viscosity characteristics and the pumping resistance, and use the rheological equivalent viscosity index as the first physical mechanism feature vector.

[0071] Activated sludge mixtures exhibit pseudoplastic fluid characteristics with shear thinning at different shear rates. When the impeller of a submersible sewage pump cuts high-concentration sludge, the fluid shear stress on the impeller surface increases dramatically. This shear stress is directly fed back to the stator side of the motor through the rotor shaft system, thereby causing a specific shift in the fundamental values ​​of stator current and active power. The rheological equivalent viscosity index eliminates the influence of hydraulic friction along the pipeline and purely characterizes the internal micro-friction coefficient of the fluid.

[0072] Optionally, step S21 includes the following sub-steps S211-S212.

[0073] S211. Obtain the active power characteristics of the submersible sewage pump, and subtract a preset mechanical friction loss value and a preset gravitational potential energy work value from the active power characteristics to obtain the net fluid work power.

[0074] S212. Obtain the flow monitoring data of the submersible sewage pump, and divide the power of the clean fluid by the product of the flow monitoring data and the motor efficiency to generate the rheological equivalent viscosity index; use the feature set containing the rheological equivalent viscosity index as the first physical mechanism feature vector.

[0075] The total active power of a submersible sewage pump consists of mechanical bearing friction losses, the work done by gravitational potential energy to raise the fluid height, and the work done to overcome the internal shear forces of the fluid. The cloud computing platform, by subtracting constants such as mechanical friction losses and gravitational potential energy work, and by removing background mechanical noise of non-fluid properties, obtains the net fluid work power contributed purely by rheological characteristics.

[0076] Based on dimensional analysis, net fluid power represents the amount of work done by the fluid per unit time, while flow monitoring data represents the volume of fluid moving per unit time. The cloud computing platform divides the net fluid power by the product of the flow monitoring data and the motor efficiency, which is physically equivalent to the specific energy consumed by the fluid to overcome internal viscous resistance, thus rigorously deriving the rheological equivalent viscosity index.

[0077] This division operation mathematically normalizes the flow fluctuations. When grid voltage fluctuations cause a simultaneous decrease in pump speed and flow, the above normalization mechanism maintains the calculation benchmark of the viscosity index unchanged, preventing the model from misinterpreting a decrease in speed as a decrease in sludge concentration, and improving the anti-interference ability of the feature vector under complex operating conditions.

[0078] S22. Extract the data components associated with the aeration blower from the electrical fingerprint data sequence, calculate and generate the oxygen mass transfer resistance coefficient based on the inhibitory effect of sewage impurities on oxygen transfer efficiency, and use the oxygen mass transfer resistance coefficient as the second physical mechanism feature vector.

[0079] In the wastewater biological treatment tank, surfactants and dissolved organic matter accumulate on the surface of air bubbles, forming a liquid film resistance layer and reducing the gas-liquid interfacial tension. This reduction in interfacial tension alters the bubble formation size and detachment frequency at the microporous aerator head. When these microbubbles burst in the fluid, they release high-frequency pressure pulses, which propagate in the reverse direction as acoustic waveforms to the aerator blower impeller. The acoustic fretting experienced by the impeller then causes a perturbation in the air gap magnetic field of the blower motor, ultimately modulating specific high-frequency components in the current and power signals.

[0080] Optionally, step S22 includes the following sub-steps S221-S223.

[0081] S221. Obtain the active power characteristics of the aeration blower, and calculate the active power noise variance characteristics within a preset stable operation window.

[0082] S222. Perform spectral analysis on the operating electrical signal of the aeration blower to identify the power spectral density characteristics of specific frequency components caused by the generation and collapse of bubbles at the aeration head of the aeration blower.

[0083] S223. Calculate the ratio between the power spectral density feature and the power noise variance feature to generate the oxygen transfer resistance coefficient, which characterizes the degree of inhibition of the oxygen transfer process at the gas-liquid interface by organic impurities in wastewater, and use the feature set containing the oxygen transfer resistance coefficient as the second physical mechanism feature vector.

[0084] The cloud computing platform uses Fourier transform or wavelet decomposition to perform spectral deconstruction on electrical signals, extracting power spectral density features in specific high-frequency bands corresponding to the characteristic frequencies of the bubble diameter. This deconstruction algorithm accurately separates bubble detachment and bursting events from background mechanical noise.

[0085] Simply using the absolute value of power spectral density cannot accurately characterize impurity content, as fluctuations in fan speed synchronously alter the signal energy across the entire frequency band. The cloud computing platform normalizes this by dividing the power spectral density at a specific frequency by the global active power noise variance. This ratio calculation eliminates the floor noise from equipment mechanical vibrations, ensuring that the output value reflects only the gas-liquid dynamic state.

[0086] When the concentration of chemical oxygen demand (COD) in wastewater increases, the bubble size decreases and the detachment frequency increases, leading to an increase in the proportion of power spectral density in a specific frequency band. This increase is physically equivalent to a decrease in the oxygen mass transfer resistance coefficient and an increase in the degree of impurity suppression, thereby establishing a deterministic mapping from electrical noise to the degree of chemical pollution and realizing the physical transformation from the fan load state to water quality pollution indicators.

[0087] S3. Input the first physical mechanism feature vector and the second physical mechanism feature vector into a preset cross-domain adaptive mapping model for calculation, and output the target water quality monitoring index; wherein, the cross-domain adaptive mapping model is constructed based on transfer learning training of source domain data and target domain data.

[0088] This application defines a benchmark site equipped with high-precision chemical analysis instruments as the source domain and a distributed rural site equipped only with power monitoring equipment as the target domain. Differences in the aging of submersible sewage pumps, head loss in process pipelines, and voltage drop in the regional power grid between different sites lead to significant statistical distribution biases in the physical mechanism feature vectors. Conventional mapping algorithms applied directly across sites will fail due to these distribution biases; therefore, the cloud computing platform must introduce a transfer learning mechanism to align the data distribution between the source and target domains.

[0089] Optionally, the target water quality monitoring indicators include virtual sludge concentration and virtual chemical oxygen demand; the cross-domain adaptive mapping model is constructed based on an adversarial domain adaptation network, which includes a domain discriminator, a feature extractor, and a cascaded regressor; step S3 includes the following sub-steps S31-S33.

[0090] S31. Using the feature extractor, the input first physical mechanism feature vector and the second physical mechanism feature vector are mapped to extract common electrical fingerprint features that are sensitive to water quality changes but not sensitive to individual differences of the target power equipment.

[0091] S32. Using the first calculation unit of the cascaded regressor, the virtual sludge concentration is calculated and output based on the common electrical fingerprint features.

[0092] S33. Using the second calculation unit of the cascaded regressor, the common electrical fingerprint features and the virtual sludge concentration are recombined into a joint input vector, and the virtual chemical oxygen demand is calculated and output based on the joint input vector.

[0093] The feature extractor constructs a nonlinear transformation matrix in a high-dimensional space to perform hardware filtering on the input mechanistic features. This matrix filters out residual device-specific hardware deviations in the first and second physical mechanism feature vectors, retaining only the feature projection components that are sensitive to the absolute values ​​of fluid viscosity and gas-liquid surface tension, thereby generating universally applicable common electrical fingerprint features.

[0094] The step-by-step computational architecture of the cascaded regressor strictly follows the logic of biochemistry and physics. Biomass, composed of activated sludge, is the physical carrier for degrading chemical oxygen demand (COD), and the electrical signal responds most directly and dramatically to the viscosity of the solid-liquid two-phase flow caused by the increase in sludge concentration. Therefore, the model forces the first computational unit to prioritize determining the total amount of biomass.

[0095] The cloud computing platform feeds forward the virtual sludge concentration output from the first computing unit to the second computing unit, essentially establishing a physical boundary for the maximum reaction rate in predicting chemical oxygen demand (COD). This joint input feedforward mechanism prevents the model from outputting non-physicochemical oxygen demand (NOD) values ​​that exceed the current upper limit of sludge degradation capacity when encountering input characteristic fluctuations induced by transient power grid changes, thus establishing the limit constraints on the model's output data.

[0096] The adversarial domain adaptation network used in the aforementioned online inference phase is pre-constructed through an offline and incremental training process including the following sub-steps. The cloud computing platform first executes sub-step S301 to obtain the original source domain dataset for benchmark sites and the original target domain dataset for ordinary sites; using the decoupling operation described in step S2, the original source domain dataset is converted into a source domain feature dataset containing a first physical mechanism feature vector and a second physical mechanism feature vector, and the original target domain dataset is converted into a target domain feature dataset containing both the first and second physical mechanism feature vectors; wherein, both the original source domain dataset and the original target domain dataset consist of multiple electrical energy fingerprint data sequences.

[0097] The benchmark site is equipped with high-power power equipment and high-precision water quality instruments, while ordinary rural sites are equipped with low-power power equipment and no chemical instruments. If the cloud computing platform directly inputs the original current and voltage sequences into the neural network, the electrical parameters of the high-power motor and the low-power motor will differ greatly in value, resulting in the feature space not overlapping and causing domain offset faults. The cloud computing platform performs mechanism preprocessing through the division operation in step S2 above to eliminate the influence of absolute power on feature amplitude. Taking the 10 kW sewage pump of the benchmark site and the 5 kW sewage pump of the ordinary site as examples, the mechanism preprocessing step normalizes the active power using motor efficiency and flow data, converting absolute electrical energy into dimensionless equivalent viscosity index and mass transfer resistance coefficient, thus achieving initial alignment of the underlying hydrodynamic characteristics of motors with different power.

[0098] The cloud computing platform then executes sub-step S302, inputting the source domain feature dataset and the target domain feature dataset into the feature extractor to generate corresponding intermediate feature representations; inputting the intermediate feature representations into the domain discriminator for domain discrimination; and adjusting the feature extractor based on the discrimination result output by the domain discriminator to suppress information about device differences between the source domain feature dataset and the target domain feature dataset in the intermediate feature representation, thereby generating a common power fingerprint feature with unified dimensions.

[0099] The domain discriminator receives intermediate feature representations, calculates the classification probability of whether they originate from benchmark site equipment or ordinary site equipment, and outputs a domain classification loss. The cloud computing platform inserts a gradient inversion layer between the feature extractor and the domain discriminator. During the backpropagation phase, the gradient inversion layer inverts the sign of the gradient output by the domain discriminator, generating a negative gradient that is backpropagated to the feature extractor. The feature extractor is forced to adjust the weight matrix of its internal convolutional kernels in the opposite direction to the domain discriminator's target. For example, when the domain discriminator identifies high-frequency noise in a specific frequency band as originating from the bearing structure of a specific brand of wind turbine unique to the benchmark site, the negative gradient drives the feature extractor to erase the weight coefficients for that specific frequency band. As iterations deepen, the feature extractor gradually forgets hardware features such as the stator and rotor air gaps and mechanical bearing types of the equipment. The final output common electrical fingerprint features only contain physical quantities determined by the fluid medium load, achieving forced overlap of feature spaces across equipment.

[0100] After feature alignment is completed, the cloud computing platform executes sub-step S303, inputting the common electrical fingerprint features into the cascaded regressor and performing step-by-step training according to the following logic: using the real sludge concentration labels in the source domain feature dataset that are time-aligned with the electrical fingerprint data sequence, the first computing unit of the cascaded regressor is supervised and trained to determine the regression parameters by which the first computing unit maps the common electrical fingerprint features to virtual sludge concentration; the common electrical fingerprint features and the virtual sludge concentration are combined into a joint input vector, and the second computing unit of the cascaded regressor is supervised and trained using the real chemical oxygen demand (COD) labels in the source domain feature dataset to determine the regression parameters by which the second computing unit maps the joint input vector to virtual COD.

[0101] This step-by-step supervised training path cuts off the disordered spread of prediction errors among multi-layer networks. The actual sludge concentration data measured in the benchmark site laboratory serves as the first anchor point, guiding the weight matrix of the first computational unit to converge towards the rheological mapping direction. The regression parameters locked by the first computational unit then cease updating, and its output virtual sludge concentration serves as the known spatial mass boundary, fed forward and concatenated to the original common features to form a joint input vector. The second computational unit performs supervised learning of the actual chemical oxygen demand (COD) label under the constraint of this joint input vector. Taking the benchmark site's sludge concentration of 3000 mg / L and COD of 150 mg / L as an example, the model first solidifies the parameter mapping the equivalent viscosity to 3000 mg / L of total biomass. Then, based on this total biomass, it solidifies the parameter mapping the oxygen mass transfer resistance coefficient to 150 mg / L of COD, establishing a rigorous causal chain of biochemical degradation.

[0102] To adapt to environmental evolution throughout the entire lifecycle, the cloud computing platform executes sub-step S304. In response to the detection of newly added source domain data or target domain data, it maintains the fixed basic weights in the cross-domain adaptive mapping model, calculates the loss gradient using only the newly added source domain data or target domain data, and fine-tunes the local network parameters of the feature extractor, the cascaded regressor, and the domain discriminator online through the gradient inversion layer.

[0103] The fixed foundational weights embody the universal mapping logic of fluid dynamics and electromagnetic induction. With the arrival of winter, the drop in influent temperature at the wastewater treatment plant triggers microbial community succession, altering the apparent viscosity and oxygen uptake of the activated sludge. The cloud computing platform freezes the aforementioned foundational physical weights, calculating the loss gradient only for local parameters in the shallow layers of the network. The newly added fine-tuning gradients generated from the temperature drop and community succession data essentially offset the static mapping error of the original model under this specific seasonal condition.

[0104] As a specific implementation of the aforementioned adversarial domain adaptation network construction process, the cloud computing platform has designed a specific topology for the network's input layer, intermediate layer, and output layer. The input layer receives a two-dimensional matrix whose dimension is the number of mechanistic features multiplied by the time step. The feature extractor of the intermediate layer consists of alternating stacked one-dimensional convolutional layers and long short-term memory (LSM) network layers. The one-dimensional convolutional layers are responsible for extracting local abrupt peaks of the rheological equivalent viscosity index via a sliding window on the time axis, while the LSM network layers are responsible for capturing the long-range temporal dependence of the oxygen mass transfer resistance coefficient as it evolves with the biochemical cycle. The domain discriminator of the intermediate layer consists of a gradient inversion layer, a fully connected layer, and a normalized exponential function layer connected in series, outputting a two-dimensional probability vector, corresponding to the probability that the input data comes from a benchmark site or a regular site. The cascaded regressor of the output layer is composed of stacked multilayer perceptron networks, with linear neurons at their ends without activation functions, directly outputting continuous concentration values.

[0105] During the iterative update of network parameters, the cloud computing platform defines a composite loss function to guide gradient descent. For the domain discriminator in the intermediate layer, the cloud computing platform uses a binary cross-entropy function to calculate the domain classification loss between its output probability and the real site label. For the cascaded regressor in the output layer, the cloud computing platform uses a mean squared error function to calculate the first regression loss between virtual sludge concentration and real test data, and the second regression loss between virtual chemical oxygen demand and real test data.

[0106] In the specific execution path of adversarial feedback adjustment, the overall objective function is set to minimize the first regression loss and the second regression loss, while maximizing the domain classification loss. The cloud computing platform introduces an adaptive adversarial weight hyperparameter during gradient backpropagation. When the gradient flows through the gradient reversal layer, this layer multiplies the binary cross-entropy loss gradient from the domain discriminator by a negative adaptive adversarial weight hyperparameter before backpropagating it to the feature extractor.

[0107] S4. Based on the comparison results between the target water quality monitoring index and the preset safety threshold, generate operation and maintenance control instructions for the target power equipment.

[0108] In summary, due to the time lag of several hours in the microbial metabolism and chemical oxygen demand (COD) degradation process within the biological treatment tank, traditional feedback control based on dissolved oxygen probes is prone to overshoot. The cloud computing platform utilizes frequently updated target water quality monitoring indicators as feedforward control signals to adjust the aeration fan speed in advance, before the peak COD reaches the biological reaction tank but before causing oxygen deficiency in the water. This pre-compensation logic effectively offsets potential future dissolved oxygen deficits, ensuring that the rural unmanned sampling station can maintain effluent standards even during sudden changes in influent load. Furthermore, the generation mechanism of operation and maintenance control commands is not simply a one-way increase in aeration volume or fluid pumping volume. Because there is a strong nonlinear relationship between virtual sludge concentration and fluid transport energy consumption, the cloud computing platform searches for the Pareto optimality of energy consumption and oxygen transfer efficiency within the control domain.

[0109] Optionally, step S5 may also be included, which includes the following sub-steps S51-S54.

[0110] S51. In response to the received maintenance work order status data, when the maintenance work order status data indicates that the maintenance is completed or the cleaning is completed, the target power equipment is determined to enter the clean water operation state; wherein, the maintenance work order status data is digital status information generated based on the equipment maintenance records fed back by the maintenance management terminal.

[0111] S52. Obtain the electrical energy fingerprint data sequence when the target power equipment is in the clean water operation state, and execute steps S2 to S3 to obtain the virtual sludge concentration benchmark value and / or virtual chemical oxygen demand benchmark value output by the cross-domain adaptive mapping model in the clean water operation state.

[0112] S53. Compare the virtual sludge concentration benchmark value and / or virtual chemical oxygen demand benchmark value with the corresponding preset clean water standard value, and calculate the corresponding baseline drift amount.

[0113] S54. The baseline drift is used to perform linear compensation correction on the corresponding virtual sludge concentration and / or virtual chemical oxygen demand to obtain the corrected target water quality monitoring index.

[0114] The "repair completed" or "cleaning completed" status in maintenance work orders strictly corresponds to two extreme boundary conditions in physics: first, the deposited sludge in the sewage pipes has been removed, and the fluid medium has changed to pure water; second, the deposits on the surface of the sewage pump impeller and the blockages in the fan filter have been removed, and the mechanical friction coefficient of the equipment has been reduced to its initial value. This strong binding relationship between the physical environment and management data makes the work order status data a deterministic signal that triggers baseline calibration.

[0115] Increased mechanical work caused by long-term wear of the submersible sewage pump impeller or long-term scaling of the microporous aeration head may be misinterpreted by the cross-domain adaptive mapping model as an increase in fluid viscosity or organic matter concentration. In this case, the virtual benchmark value output in step S52 is not a true water quality parameter, but rather a spurious fluid load generated by equipment aging. The cloud computing platform calculates the difference between this benchmark value and the zero point of the pure water standard, accurately quantifying the long-term irreversible energy loss of the mechanical system and achieving separation between equipment degradation characteristics and the true fluid load.

[0116] The increase in resistance caused by microscopic wear or permanent biofilm accumulation is a constant static constant in the short term. Therefore, the cloud computing platform uses this baseline drift as a reverse compensation factor, directly injecting it into the end of the water quality indicator generation chain. This linear compensation mechanism offsets the computational bias introduced by mechanical aging in the numerical output, achieving accuracy conservation of the soft measurement system throughout the entire equipment lifecycle, without requiring any professionals to go to remote rural sites for probe calibration.

[0117] Optionally, step S6 may also be included, which includes the following sub-steps S61-S63.

[0118] S61. Acquire image data collected by on-site video monitoring equipment and analyze it to generate water clarity index.

[0119] S62. When the target water quality monitoring index shows that the water quality exceeds the standard, and the effluent clarity index is higher than the preset clarity threshold, a conflict alarm signal is triggered.

[0120] S63. Suspend the operation and maintenance control command and generate a manual review work order.

[0121] The cloud computing platform performs cross-modal physical inconsistency discrimination between the microscopic biochemical characteristics reflected by electrical fingerprints and the macroscopic transmittance characteristics reflected by optical images. According to the law of extinction in optics, the concentration of solids in a fluid medium is strictly positively correlated with the turbidity of the fluid. When the equivalent viscosity deconstructed from the electrical signal is extremely high, indicating that the target water quality monitoring index shows that the water quality is seriously exceeding the standard, while the effluent clarity index analyzed by the on-site video monitoring equipment is higher than the preset clarity threshold, this phenomenon violates the law of extinction in physical optics, indicating that the system's sensing data chain is disconnected from the actual fluid conditions.

[0122] Based on the aforementioned cross-modal physical inconsistencies, the closed-loop control center, through logical deduction, determined that the sudden change in electrical data did not originate from deterioration of the fluid water quality, but rather from a sudden hardware failure of the power equipment itself, such as sudden jamming of the motor bearings, fibrous foreign matter entanglement on the impeller of the submersible sewage pump, or a momentary voltage drop on the power supply bus. Upon triggering the conflict alarm signal, the control center immediately suspended the operation and maintenance control commands. This circuit breaker mechanism fundamentally prevents the control system from executing excessive aeration or forced pumping based on erroneous water quality signals, preventing the equipment from falling into energy waste and secondary damage caused by erroneous adjustments and mechanical overload.

[0123] When maintenance personnel at remote rural sites respond to the work order, go to the site to remove obstructions from the equipment, and report the completion status on the maintenance management terminal, this manual operation is directly converted into the maintenance work order status data required in step S51. This data flow mechanism automatically triggers the baseline drift calculation and linear compensation correction in step S5, completing a full physical closed loop from fault identification and manual intervention repair to automatic system recalibration of the zero point, thus establishing the system's fully automated self-healing maintenance logic.

[0124] Furthermore, this application also discloses a distributed wastewater treatment facility water quality soft measurement system, including:

[0125] The data acquisition module, deployed on the field intelligent gateway, is used to collect the operating electrical signals of the target power equipment and generate an electrical energy fingerprint data sequence;

[0126] The video surveillance module is used to collect image data from the scene.

[0127] A cloud computing platform, which is communicatively connected to the data acquisition module and the video monitoring module, is used to execute the above-mentioned soft measurement method for water quality in distributed wastewater treatment facilities.

[0128] The operation and maintenance management terminal is used to receive target water quality monitoring indicators and send operation and maintenance work order status data to the cloud computing platform.

[0129] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0130] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for soft water quality measurement in a distributed wastewater treatment facility, characterized in that, Includes the following steps: S1. Acquire the operating electrical signals of the target power equipment in the distributed wastewater treatment facility, and parse and generate an electrical fingerprint data sequence; the target power equipment includes a submersible sewage pump and an aeration blower; S2. Based on the coupling mechanism of fluid rheology and oxygen mass transfer dynamics, the electrical energy fingerprint data sequence is decoupled to generate a first physical mechanism feature vector characterizing the physical rheological properties of the fluid, and a second physical mechanism feature vector characterizing the mass transfer resistance at the gas-liquid interface. S3. Input the first physical mechanism feature vector and the second physical mechanism feature vector into a preset cross-domain adaptive mapping model for calculation, and output the target water quality monitoring index; wherein, the cross-domain adaptive mapping model is constructed based on transfer learning training of source domain data and target domain data; S4. Based on the comparison results between the target water quality monitoring index and the preset safety threshold, generate operation and maintenance control instructions for the target power equipment; Step S2 includes the following sub-steps: S21. Extract the data components associated with the submersible sewage pump from the electrical fingerprint data sequence, calculate and generate the rheological equivalent viscosity index based on the mapping relationship between the non-Newtonian fluid viscosity characteristics and the pumping resistance, and use the rheological equivalent viscosity index as the first physical mechanism feature vector. S22. Extract the data components associated with the aeration blower from the electrical fingerprint data sequence, calculate and generate the oxygen mass transfer resistance coefficient based on the inhibitory effect of sewage impurities on oxygen transfer efficiency, and use the oxygen mass transfer resistance coefficient as the second physical mechanism feature vector.

2. The method for soft water quality measurement in distributed wastewater treatment facilities according to claim 1, characterized in that, Step S1 includes the following sub-steps: S11. Collect the operating electrical signals of the target power equipment, wherein the operating electrical signals include at least the active power time series, the current time series, and the voltage time series; S12. The operating electrical signal is sampled according to a preset feature capture frequency, and the active power noise variance feature and current harmonic distortion rate feature are calculated within a preset time window to generate the electrical energy fingerprint data sequence; wherein, the feature capture frequency is set to be greater than or equal to the pressure pulsation frequency generated by the impeller rotation of the submersible sewage pump, and greater than or equal to the current noise frequency generated by the bubble breakage corresponding to the aeration blower.

3. The method for soft water quality measurement in distributed wastewater treatment facilities according to claim 1, characterized in that, Step S21 includes the following sub-steps: S211. Obtain the active power characteristics of the submersible sewage pump, and subtract a preset mechanical friction loss value and a preset gravitational potential energy work value from the active power characteristics to obtain the net fluid work power. S212. Obtain the flow monitoring data of the submersible sewage pump, and divide the power of the clean fluid by the product of the flow monitoring data and the motor efficiency to generate the rheological equivalent viscosity index; use the feature set containing the rheological equivalent viscosity index as the first physical mechanism feature vector.

4. The method for soft water quality measurement in distributed wastewater treatment facilities according to claim 1, characterized in that, Step S22 includes the following sub-steps: S221. Obtain the active power characteristics of the aeration blower, and calculate the active power noise variance characteristics within a preset stable operation window; S222. Perform spectral analysis on the operating electrical signal of the aeration blower to identify the power spectral density characteristics of specific frequency components caused by the generation and collapse of bubbles at the aeration head of the aeration blower; S223. Calculate the ratio between the power spectral density feature and the power noise variance feature to generate the oxygen transfer resistance coefficient, which characterizes the degree of inhibition of the oxygen transfer process at the gas-liquid interface by organic impurities in wastewater, and use the feature set containing the oxygen transfer resistance coefficient as the second physical mechanism feature vector.

5. The method for soft water quality measurement in distributed wastewater treatment facilities according to claim 1, characterized in that, The target water quality monitoring indicators include virtual sludge concentration and virtual chemical oxygen demand; the cross-domain adaptive mapping model is constructed based on an adversarial domain adaptation network, which includes a domain discriminator, a feature extractor, and a cascaded regressor; step S3 includes the following sub-steps: S31. Using the feature extractor, the input first physical mechanism feature vector and the second physical mechanism feature vector are mapped to extract common electrical fingerprint features that are sensitive to water quality changes but not sensitive to individual differences of the target power equipment. S32. Using the first calculation unit of the cascaded regressor, the virtual sludge concentration is calculated and output based on the common electrical fingerprint characteristics; S33. Using the second calculation unit of the cascaded regressor, the common electrical fingerprint features and the virtual sludge concentration are recombined into a joint input vector, and the virtual chemical oxygen demand is calculated and output based on the joint input vector.

6. The method for soft water quality measurement in distributed wastewater treatment facilities according to claim 5, characterized in that, The adversarial domain adaptation network is pre-built through a training process that includes the following sub-steps: S301. Obtain the source domain original dataset of the benchmark site and the target domain original dataset of the ordinary site; using the decoupling operation described in step S2, convert the source domain original dataset into a source domain feature dataset containing a first physical mechanism feature vector and a second physical mechanism feature vector, and convert the target domain original dataset into a target domain feature dataset containing a first physical mechanism feature vector and a second physical mechanism feature vector; wherein, both the source domain original dataset and the target domain original dataset are composed of multiple energy fingerprint data sequences; S302. Input the source domain feature dataset and the target domain feature dataset into the feature extractor to generate corresponding intermediate feature representations; input the intermediate feature representations into the domain discriminator for domain discrimination; and adjust the feature extractor based on the discrimination result output by the domain discriminator to suppress information about device differences between the source domain feature dataset and the target domain feature dataset in the intermediate feature representation, thereby generating a common power fingerprint feature with unified dimensions; S303. Input the common electrical energy fingerprint features into the cascaded regressor and perform step-by-step training according to the following logic: Use the real sludge concentration labels in the source domain feature dataset that are time-aligned with the electrical energy fingerprint data sequence to perform supervised training on the first computing unit of the cascaded regressor, and determine the regression parameters by which the first computing unit maps the common electrical energy fingerprint features to virtual sludge concentration; combine the common electrical energy fingerprint features and the virtual sludge concentration into a joint input vector, and use the real chemical oxygen demand (COD) labels in the source domain feature dataset to perform supervised training on the second computing unit of the cascaded regressor, and determine the regression parameters by which the second computing unit maps the joint input vector to virtual COD; S304. In response to the detection of newly added source domain data or target domain data, the basic weights that have been fixed in the cross-domain adaptive mapping model are maintained, the loss gradient is calculated using only the newly added source domain data or target domain data, and the local network parameters of the feature extractor, the cascaded regressor and the domain discriminator are fine-tuned online through the gradient inversion layer.

7. The method for soft water quality measurement in distributed wastewater treatment facilities according to claim 1, characterized in that, It also includes step S5, which comprises the following sub-steps: S51. In response to the received maintenance work order status data, when the maintenance work order status data indicates that the maintenance is completed or the cleaning is completed, it is determined that the target power equipment has entered the clean water operation state; wherein, the maintenance work order status data is digital status information generated based on the equipment maintenance records fed back by the maintenance management terminal; S52. Obtain the electrical energy fingerprint data sequence when the target power equipment is in the clean water operation state, and execute steps S2 to S3 to obtain the virtual sludge concentration benchmark value and / or virtual chemical oxygen demand benchmark value output by the cross-domain adaptive mapping model in the clean water operation state. S53. Compare the virtual sludge concentration benchmark value and / or virtual chemical oxygen demand benchmark value with the corresponding preset clean water standard value, and calculate the corresponding baseline drift amount; S54. The baseline drift is used to perform linear compensation correction on the corresponding virtual sludge concentration and / or virtual chemical oxygen demand to obtain the corrected target water quality monitoring index.

8. The method for soft water quality measurement in distributed wastewater treatment facilities according to claim 1, characterized in that, The method further includes step S6, which comprises the following sub-steps: S61. Acquire image data collected by on-site video monitoring equipment and analyze it to generate water clarity index; S62. When the target water quality monitoring index shows that the water quality exceeds the standard, and the effluent clarity index is higher than the preset clarity threshold, a conflict alarm signal is triggered; S63. Suspend the operation and maintenance control command and generate a manual review work order.

9. A soft water quality measurement system for a distributed wastewater treatment facility, characterized in that, include: The data acquisition module, deployed on the field intelligent gateway, is used to collect the operating electrical signals of the target power equipment and generate an electrical energy fingerprint data sequence; The video surveillance module is used to collect image data from the scene. A cloud computing platform, communicatively connected to the data acquisition module and the video monitoring module, is used to execute the distributed wastewater treatment facility water quality soft measurement method as described in any one of claims 1 to 8; an operation and maintenance management terminal is used to receive target water quality monitoring indicators and send operation and maintenance work order status data to the cloud computing platform.

Images