A real-time monitoring method and system for the sterilization effect of an aldehyde disinfectant

Abstract

The present application is a kind of aldehyde disinfectant sterilization effect real-time monitoring method and system, relating to process monitoring technical field, including collecting microorganism electrochemical response signal, obtaining time domain impedance modulus, phase angle and environment parameter time sequence.Wavelet packet decomposition is carried out on impedance modulus time sequence, cell membrane permeability characteristic frequency band signal is extracted, and frequency domain envelope spectrum time sequence is generated.Four types of time sequence are aligned to form three-dimensional feature tensor, and sterilization efficiency feature vector is generated through gated attention fusion.Frequency domain relationship diagram is constructed, edge weight is calculated based on frequency spectrum similarity, and frequency domain enhancement matrix is aggregated through double-layer graph convolution.The time domain features are used as queries, and the frequency domain enhancement features are used as key values, and the biological constraint matrix is injected to modify the attention weight, and the time-frequency features are fused.The sterilization efficiency, inactivation logarithm and microorganism load prediction value are output in real time through bidirectional LSTM coding.The method combines electrochemical analysis and deep learning technology to realize fast and accurate evaluation of disinfection effect, and is suitable for medical disinfection monitoring scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of process monitoring, in particular to a real-time monitoring method and system for the sterilization effect of aldehyde disinfectant. BACKGROUND

[0002] Aldehyde disinfectants are widely used in the fields of medical instrument disinfection, epidemic source treatment, laboratory disinfection, etc. due to their broad-spectrum and high-efficiency sterilization ability. However, the sterilization effect of aldehyde disinfectants is significantly affected by many factors such as concentration, temperature, action time, and organic matter interference.

[0003] Currently, the evaluation of the sterilization effect of aldehyde disinfectants mainly relies on: offline chemical detection, which can only detect the residual concentration of disinfectants and cannot directly reflect the actual sterilization efficiency. Regular biological indicator culture takes a long time (usually 24-72 hours) and the results are lagging, which cannot realize process monitoring. End-point sampling culture method takes a long time and is a destructive detection, which cannot be continuously monitored. The above evaluation methods all have serious lag, which leads to the inability to judge whether the disinfection process is effective in real time, and there is a risk of disinfection failure or over-disinfection. It is impossible to dynamically adjust disinfection parameters such as concentration and time according to the actual microbial load, resulting in waste of resources or incomplete disinfection. In critical application scenarios such as surgical instrument sterilization and biosafety laboratories, it is impossible to provide immediate safety feedback. Therefore, there is an urgent need for a monitoring technology that can reflect the sterilization effect of aldehyde disinfectants in real time, online, and directly. SUMMARY

[0004] In order to overcome the shortcomings of the prior art that the monitoring data processing is lagging, the accuracy is greatly affected by environmental noise, and it is difficult to meet the high real-time requirement, the main purpose of the present application is to provide a real-time monitoring system and method for the sterilization effect of aldehyde disinfectants, which realizes continuous, online, and rapid monitoring of the activity state of microorganisms in the disinfection process, thereby realizing real-time evaluation and early warning of the sterilization effect.

[0005] To achieve the above purpose, the present application adopts the following technical scheme, a real-time monitoring method for the sterilization effect of aldehyde disinfectants, comprising:

[0006] Real-time acquisition of the electrochemical response signal of microorganisms in the disinfectant solution by a multi-frequency excitation three-electrode system, synchronous acquisition of the time-domain impedance modulus time sequence, phase angle time sequence, and environmental parameter time sequence;

[0007] Wavelet packet decomposition of the time-domain impedance modulus time sequence guided by the response characteristics of microorganisms, extraction of the pre-set cell membrane permeability change characteristic frequency band signal, and generation of the frequency domain envelope spectrum time sequence by fast Fourier transform;

[0008] The time-domain impedance modulus time sequence, the phase angle time sequence, the environmental parameter time sequence and the frequency-domain envelope spectrum time sequence are aligned along a time axis to form a three-dimensional feature tensor, and a gating attention mechanism is used to dynamically allocate feature weights and generate a sterilization efficiency feature vector by fusion;

[0009] A graph convolution-time-frequency joint attention LSTM network is constructed, the frequency band of the frequency-domain envelope spectrum time sequence is taken as a node, the node feature is the frequency-domain envelope spectrum time sequence, a frequency-domain relationship graph is constructed, the edge weight is calculated based on the spectral similarity between nodes, the neighborhood features are aggregated through double-layer graph convolution to obtain a frequency-domain enhanced feature matrix, the time-domain features are taken as query vectors, the frequency-domain enhanced features are taken as key-value vectors, a predefined biological constraint matrix is used to modify the attention weights, and the time-frequency fusion features with biological prior knowledge are obtained through weighted aggregation, after bidirectional encoding of the time-frequency fusion features by bidirectional LSTM, high-level features are extracted by a fully connected layer and decoupled, and real-time output of sterilization efficiency prediction value, inactivation logarithm prediction value and microbial load prediction value is taken as the real-time monitoring result of the sterilization effect.

[0010] Further comprising:

[0011] An inactivation logarithm target threshold and a microbial load safety threshold are set;

[0012] A disinfectant concentration adjustment instruction is generated based on the deviation of the inactivation logarithm prediction value from the target threshold;

[0013] A disinfection time adjustment instruction is generated based on the deviation of the microbial load prediction value from the safety threshold, and the disinfectant concentration adjustment instruction and the disinfection time adjustment instruction generate a multi-dimensional disinfection parameter adjustment instruction.

[0014] The environmental parameter time sequence includes temperature, pH, oxidation-reduction potential and organic interferent concentration;

[0015] The excitation frequency of the microbial electrochemical response signal covers the frequency band of 50Hz to 200kHz, and the frequency scanning density is not less than 10 points per decade;

[0016] The wavelet packet decomposition guided by the microbial response features includes the following steps:

[0017] For the mechanism of aldehyde disinfectant, the microbial cell membrane rupture feature frequency band is preset as a low frequency band of 100Hz-5kHz and a medium frequency band of 5kHz-50kHz;

[0018] A db6 wavelet basis function is used for 5-layer wavelet packet decomposition, and the feature frequency band signals corresponding to the 3rd-5th layers of nodes are extracted.

[0019] The gating attention mechanism is used to dynamically allocate feature weights and generate a sterilization efficiency feature vector by fusion, including:

[0020] The environment parameter time sequence is dynamically interacted with the bioelectrochemical characteristics through cross attention, and an environment-modulated biological characteristic is obtained;

[0021] Multi-scale features are extracted from low-frequency bands, medium-frequency bands and phase angle differentials, a membrane rupture risk score is obtained, and a warning is adaptively triggered;

[0022] The environment-modulated biological characteristic and the membrane rupture risk score are used to generate a channel-time two-dimensional weight matrix through a gated residual structure;

[0023] The channel-time two-dimensional weight matrix is multiplied element by element with the time-domain impedance modulus time sequence, the phase angle time sequence, the environment parameter time sequence and the frequency domain envelope spectrum time sequence to obtain a weighted feature;

[0024] The time-domain impedance modulus time sequence, the phase angle time sequence, the environment parameter time sequence and the frequency domain envelope spectrum time sequence are connected through a residual connection with the weighted feature to obtain a sterilization efficiency feature vector.

[0025] The pre-defined biological constraint matrix includes:

[0026] According to prior knowledge of microbial electrochemical response, the time domain frequency band is divided into a cell membrane rupture characteristic frequency band, an ion channel disturbance frequency band and other frequency bands, and constraint weights are respectively assigned.

[0027] A system for real-time monitoring of the sterilization effect of a disinfectant includes:

[0028] An electrochemical signal acquisition module is configured with a multi-frequency excitation three-electrode sensing unit and a multi-channel environment sensor, for real-time acquisition of microbial electrochemical response signals in a disinfectant solution through a multi-frequency excitation three-electrode system, and for synchronous acquisition of time-domain impedance modulus time sequence, phase angle time sequence and environment parameter time sequence;

[0029] A biological characteristic extraction module is built-in with a wavelet packet-FFT coprocessor, configured to perform microbial response feature-oriented frequency domain envelope spectrum analysis, for wavelet packet decomposition of the time-domain impedance modulus time sequence in a microbial response feature-oriented manner, to extract a pre-set cell membrane permeability change characteristic frequency band signal, and to generate a frequency domain envelope spectrum time sequence through fast Fourier transform;

[0030] A dynamic fusion decision module includes a gated attention weight generation unit and a three-dimensional tensor processor, for aligning the time-domain impedance modulus time sequence, the phase angle time sequence, the environment parameter time sequence and the frequency domain envelope spectrum time sequence along the time axis to form a three-dimensional feature tensor, dynamically assigning feature weights through a gated attention mechanism and fusing to generate a sterilization efficiency feature vector;

[0031] The monitoring result acquisition module integrates a graph convolution-time-frequency joint attention LSTM network, which is used to take a frequency band of a frequency domain envelope spectrum time sequence as a node, take a node feature as the frequency domain envelope spectrum time sequence, construct a frequency domain relationship graph, calculate an edge weight based on a frequency spectrum similarity between nodes, aggregate neighborhood features through double-layer graph convolution, obtain a frequency domain enhanced feature matrix, take a time domain feature as a query vector, take a frequency domain enhanced feature as a key-value vector, inject a predefined biological constraint matrix to modify attention weights, and weighted aggregation is performed to obtain a time-frequency fusion feature of biological prior knowledge, after bidirectional encoding of the time-frequency fusion feature by using a bidirectional LSTM, high-level features are extracted by using a full connection layer and decoupled, and real-time output of a sterilization efficiency prediction value, an inactivation logarithm prediction value and a microorganism load prediction value are taken as real-time monitoring results of the sterilization effect.

[0032] Compared with the prior art, the beneficial effects of the present application are that the present application synchronously collects time domain impedance modulus, phase angle, environmental parameters and frequency domain envelope spectrum through a three-electrode system, forms a three-dimensional feature tensor, dynamically allocates weights in combination with a gated attention mechanism, solves the problem that a traditional single electrochemical signal is easily disturbed by the environment, significantly improves feature representation capability, and uses multi-modal data fusion to improve monitoring accuracy, and further uses wavelet packet decomposition to directionally extract a cell membrane permeability change feature band, generates a frequency domain envelope spectrum in combination with FFT, effectively captures the dynamic response of microorganisms under the action of aldehyde disinfectants, overcomes the defect of insufficient resolution in the low-frequency band of the traditional impedance spectrum, and uses time-frequency joint analysis to enhance feature extraction.

[0033] Further, a graph convolution-time-frequency joint attention LSTM network is constructed, a frequency band is taken as a node to construct a frequency domain relationship graph, neighborhood features are aggregated through double-layer graph convolution, cross-band harmonic correlation is explicitly captured, and the problem that the traditional method ignores the synergistic effect between frequency bands is solved, a predefined biological constraint matrix is introduced to modify attention weights, the prediction result is forced to comply with the physical law of microorganism inactivation, and the physical rationality of the prediction value is ensured, a time domain feature is taken as a query vector to actively retrieve a frequency domain feature, bidirectional encoding of time-frequency features is realized by using a bidirectional LSTM, the limitation of traditional single-direction feature extraction is overcome, sterilization efficiency, inactivation logarithm and microorganism load prediction values are output in real time, concentration / time adjustment instructions are generated based on threshold deviation, closed-loop control is realized, and the frequency band covered by the excitation signal is combined with environmental parameter time sequence correction, so that the monitoring result still maintains a preset error precision when organic interferents exist, and the anti-interference capability is optimized.

[0034] In summary, the present application realizes high-precision, fully-automated real-time monitoring and dynamic optimization of the sterilization effect of aldehyde disinfectants through innovative combination of multi-frequency electrochemical signals and a deep learning model, especially graph convolution frequency domain modeling and a biological constraint attention mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0035] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. The detailed description of the application refers to the accompanying drawings.

[0036] Figure 1 is a schematic diagram of the monitoring process of the present application.

[0037] Figure 2 is a schematic diagram of the signal acquisition parameters of the present application.

[0038] Figure 3 is a schematic diagram of the wavelet packet decomposition process of the present application.

[0039] Figure 4 is a schematic diagram of the gating attention mechanism process of the present application.

[0040] Figure 5 is a schematic diagram of the process of generating the bactericidal efficacy feature vector of the present application.

[0041] Figure 6 is a schematic diagram of the biological constraint matrix of the present application.

[0042] Figure 7 is a schematic diagram of the electrical signal acquisition module of the present application. DETAILED DESCRIPTION

[0043] At present, the evaluation of the bactericidal effect of aldehyde disinfectants mainly relies on the following methods: offline chemical detection: only the residual concentration of disinfectants can be detected, and the actual bactericidal efficacy cannot be directly reflected. Regular biological indicator culture: time-consuming, usually 24-72 hours, the result is seriously lagging behind, and process monitoring and end-point sampling culture method cannot be realized.

[0044] These methods all have serious lag, which leads to the inability to judge whether the disinfection process is effective in real time, and there is a risk of disinfection failure or over-disinfection; it is impossible to dynamically adjust disinfection parameters such as concentration and time according to the actual microbial load, resulting in waste of resources or incomplete disinfection; in critical application scenarios such as surgical instrument sterilization and biosafety laboratories, it is impossible to provide immediate safety feedback. Therefore, there is an urgent need for a monitoring technology that can reflect the bactericidal effect of aldehyde disinfectants in real time, online and directly.

[0045] In order to overcome the shortcomings of the prior art that the monitoring data processing is lagging, the accuracy is greatly affected by environmental noise, and it is difficult to meet the high real-time requirement, the main purpose of the present application is to provide a real-time monitoring system and method for the bactericidal effect of aldehyde disinfectants, which realizes continuous, online and rapid monitoring of the activity state of microorganisms in the disinfection process, so as to evaluate the bactericidal effect in real time and warn the failure risk. Referring to Figures 1 to 7 The following technical solutions are adopted:

[0046] The microorganism electrochemical response signal in the disinfectant solution is collected in real time by a multi-frequency excitation three-electrode system, and time-domain impedance modulus time series, phase angle time series and environmental parameter time series are synchronously obtained; the environmental parameter time series includes temperature, pH, oxidation-reduction potential and organic interferent concentration; and the excitation frequency of the microorganism electrochemical response signal covers a frequency band of 50 Hz to 200 kHz, and the frequency scanning density is not less than 10 points per decade;

[0047] The time-domain impedance modulus time series is subjected to wavelet packet decomposition in the direction of microorganism response characteristics, and a preset cell membrane permeability change characteristic frequency band signal is extracted; that is, for the action mechanism of aldehyde disinfectants, the preset microorganism cell membrane rupture characteristic frequency band is a low frequency band of 100 Hz-5 kHz and a medium frequency band of 5 kHz-50 kHz; db6 wavelet basis function is used for 5-layer wavelet packet decomposition, and the characteristic frequency band signals corresponding to the 3rd-5th layers of nodes are extracted. Further, the frequency domain envelope spectrum time series is generated by fast Fourier transform;

[0048] The time-domain impedance modulus time series, the phase angle time series, the environmental parameter time series and the frequency domain envelope spectrum time series are aligned along the time axis to form a three-dimensional feature tensor, and a disinfection efficiency feature vector is generated by dynamically allocating feature weights and fusion through a gated attention mechanism; wherein the disinfection efficiency feature vector is generated by dynamically allocating feature weights and fusion through a gated attention mechanism, which includes dynamically interacting the environmental parameter time series and the bioelectrochemical characteristics through cross-attention to obtain environment-modulated biological characteristics; multi-scale features are extracted from the low frequency band, the medium frequency band spectrum and the phase angle differential to obtain a membrane rupture risk score and adaptively trigger an early warning; the environment-modulated biological characteristics and the membrane rupture risk score are used to generate a channel-time two-dimensional weight matrix through a gated residual structure; the channel-time two-dimensional weight matrix is multiplied element by element with the time-domain impedance modulus time series, the phase angle time series, the environmental parameter time series and the frequency domain envelope spectrum time series to obtain weighted features; the time-domain impedance modulus time series, the phase angle time series, the environmental parameter time series and the frequency domain envelope spectrum time series are connected through a residual connection with the weighted features to obtain a disinfection efficiency feature vector.

[0049] A graph convolution-time-frequency joint attention LSTM network is constructed, the frequency band of the frequency domain envelope spectrum time series is taken as a node, the node feature is the frequency domain envelope spectrum time series, a frequency domain relationship graph is constructed, the edge weight is calculated based on the spectral similarity between nodes, the neighborhood features are aggregated through double-layer graph convolution to obtain a frequency domain enhanced feature matrix, the time domain features are taken as query vectors, the frequency domain enhanced features are taken as key-value vectors, a predefined biological constraint matrix is used to modify the attention weight, and the time-frequency fusion features of biological prior knowledge are obtained by weighted aggregation, after the time-frequency fusion features are bidirectionally encoded by bidirectional LSTM, high-level features are extracted by a fully connected layer and decoupled, and real-time output of the sterilization efficiency prediction value, the inactivation logarithm prediction value and the microorganism load prediction value are taken as real-time monitoring results of the sterilization effect.

[0050] setting an inactivation logarithm target threshold value and a microbial load safety threshold value;

[0051] generating a disinfectant concentration adjustment instruction based on the deviation of the inactivation logarithm prediction value from the target threshold value;

[0052] generating a disinfection time adjustment instruction based on the deviation of the microbial load prediction value from the safety threshold value, the disinfectant concentration adjustment instruction and the disinfection time adjustment instruction generating a multi-dimensional disinfection parameter adjustment instruction.

[0053] and a system for real-time monitoring of the bactericidal effect of an aldehyde-based disinfectant, comprising:

[0054] an electrochemical signal acquisition module configured with a multi-frequency excitation three-electrode sensing unit and a multi-channel environmental sensor, for real-time acquisition of microbial electrochemical response signals in a disinfectant solution through a multi-frequency excitation three-electrode system, and for synchronous acquisition of time-domain impedance modulus time series, phase angle time series and environmental parameter time series;

[0055] a biological feature extraction module with a wavelet packet-FFT coprocessor configured to perform microbial response feature-oriented frequency domain envelope spectrum analysis, for wavelet packet decomposition of the time-domain impedance modulus time series with microbial response feature orientation, extraction of pre-set cell membrane permeability change feature frequency band signals, and generation of frequency domain envelope spectrum time series through fast Fourier transform;

[0056] a dynamic fusion decision module including a gating attention weight generation unit and a three-dimensional tensor processor, for aligning the time-domain impedance modulus time series, the phase angle time series, the environmental parameter time series and the frequency domain envelope spectrum time series along the time axis to form a three-dimensional feature tensor, dynamically allocating feature weights and fusing to generate a bactericidal efficiency feature vector through a gating attention mechanism;

[0057] a monitoring result acquisition module integrating a graph convolution-time-frequency joint attention LSTM network, for constructing a frequency domain relationship graph with the frequency band of the frequency domain envelope spectrum time series as the node and the node feature as the frequency domain envelope spectrum time series, calculating the edge weight based on the spectral similarity between nodes, obtaining a frequency domain enhanced feature matrix through double-layer graph convolution to aggregate neighborhood features, taking the time domain feature as the query vector and the frequency domain enhanced feature as the key-value vector, injecting a predefined biological constraint matrix to modify the attention weight, and weighting and aggregating to obtain a time-frequency fusion feature with biological prior knowledge, and after bidirectional encoding of the time-frequency fusion feature using a bidirectional LSTM, extracting high-level features through a fully connected layer and decoupling, and real-time outputting a bactericidal efficiency prediction value, an inactivation logarithm prediction value and a microbial load prediction value as real-time monitoring results of the bactericidal effect.

[0058] The aldehyde disinfectant sterilization effect is monitored and dynamically optimized in high precision and full automation in real time by combining multi-frequency electrochemical signals and a deep learning model.

[0059] The application will be further described below in combination with the drawings and embodiments.

[0060] Embodiment 1: This embodiment aims to solve the problem that the traditional method relies on fixed time or periodic sampling culture, resulting in low efficiency, poor accuracy, and inability to respond to environmental changes and microbial load fluctuations when medical institutions such as operating rooms and endoscopy centers use 2% glutaraldehyde disinfectant to soak medical devices such as laparoscopes and arthroscopes. This embodiment provides a real-time monitoring method, and can intelligently adjust disinfection parameters according to the monitoring results to ensure disinfection effect.

[0061] It is applied to a hospital endoscope disinfection workstation. When a used gastroscope is cleaned, it is placed in a disinfection tank containing fresh 2% glutaraldehyde disinfectant. The monitoring system described in this embodiment is started immediately to monitor the disinfectant in real time, and the disinfectant concentration is adjusted or the disinfection time is extended / shortened according to the monitoring results until the preset disinfection effect standard is reached.

[0062] In this embodiment, a PARSTAT4000 electrochemical workstation produced by Gamry Instruments is used as the core excitation and acquisition device. The working electrode is a 3mm diameter platinum black electrode for enhancing electrochemical activity, the reference electrode is a saturated calomel electrode SCE with stable potential, and the auxiliary electrode is a platinum wire electrode. The three-electrode system is completely immersed in the disinfectant, the initial volume of the disinfectant is 5L, the initial temperature is 25℃, and the surface of the gastroscope to be disinfected is adjacent.

[0063] The excitation is an alternating signal with a frequency range of 50Hz to 200kHz. To meet the requirement of "frequency scanning density not less than 10 points per decade", the specific frequency points are set as:

[0064] 50Hz, 70Hz, 100Hz, 140Hz, 200Hz, 280Hz, 400Hz, 560Hz, 800Hz, 1.1kHz, 1.6kHz, 2.2kHz, 3.2kHz, 4.5kHz, 6.3kHz, 9kHz, 12.6kHz, 18kHz, 25kHz, 36kHz, 50kHz, 70kHz, 100kHz, 140kHz, 200kHz, a total of 24 frequency points. An alternating voltage of 10mV is applied to each frequency point, and the sampling time is 5 seconds per frequency point.

[0065] In this embodiment, the environmental parameter sensors are temperature sensors, pH sensors, and oxidation-reduction potential (ORP) sensors.

[0066] Temperature sensor is PT100 platinum resistance thermometer, accuracy ±0.1℃, measurement range 0-50℃, installed on the wall of the sterilization tank, real-time monitoring of the disinfectant temperature.

[0067] pH sensor is HORIBA B-715 portable pH meter, accuracy ±0.02pH, measurement range 0-14pH, inserted into the disinfectant, monitoring the pH change caused by glutaraldehyde degradation.

[0068] Oxidation-reduction potential ORP sensor is Mettler Toledo InPro 6800i, range-200mV to +1000mV, accuracy ±1mV, inserted into the disinfectant, used to reflect the oxidation-reduction capacity of the disinfectant.

[0069] Organic interferent concentration sensor is an online analyzer based on the principle of TOC total organic carbon, and in this embodiment, Shimadzu TOC-VCSH is specifically selected, with a detection limit of 0.02mg / L, sampling once every 5 minutes, used to monitor organic matter such as residual blood and protein in surgery.

[0070] All sensor data is collected synchronously through NI USB-6363 data acquisition card with a sampling rate of ≥1MHz, and the timestamp accuracy reaches 1ms. A complete EIS scan and environmental parameter collection is completed every 30 seconds to form the time-domain impedance modulus time sequence, phase angle time sequence, and environmental parameter time sequence.

[0071] The preset microbial cell membrane rupture characteristic frequency band is selected as the low frequency band 50Hz-5kHz and the medium frequency band 5kHz-50kHz. The basis is that these two frequency bands are most sensitive to the destruction of cell membrane structure by glutaraldehyde, corresponding to the changes in cell membrane capacitance and membrane resistance.

[0072] For each time-domain impedance modulus sequence obtained by collection, the length is 24 frequency points x 5 seconds / second = 120 data points, or the average value of multiple scans is processed, and db6 wavelet basis function is used for 5-layer wavelet packet decomposition. After decomposition, a total of 32 sub-bands 2^5 are obtained.

[0073] For feature band extraction, the 3rd layer cell membrane initial penetration response, node 3-1: 50Hz-625Hz, the 4th layer membrane protein denaturation feature, node 4-2: 3.1kHz-9.8kHz, and the 5th layer membrane rupture and leakage feature, node 5-2: 15.6kHz-49.2kHz, are extracted. These nodes cover the preset characteristic frequency bands of 100Hz-5kHz and 5kHz-50kHz. The extracted sub-band signals are reconstructed to obtain the characteristic frequency band signals.

[0074] Then the reconstructed characteristic band signal is subjected to fast Fourier transform (FFT) to obtain a frequency domain signal. Then, an envelope spectrum is calculated, and in this embodiment, a frequency domain envelope spectrum time series is obtained by Hilbert transform or wavelet packet energy spectrum method. The time series reflects the change of energy in the characteristic band over time.

[0075] For three-dimensional feature tensor construction and gated attention mechanism, first, the tensor needs to be aligned. The time-domain impedance modulus time series, phase angle time series, and environmental parameter time series collected within 30 seconds, each 1 point, and the energy values of the corresponding characteristic frequency bands calculated, assuming 3 frequency bands, 3 points, are aligned along the time axis. The 24-dimensional time-domain impedance modulus time series, 24-dimensional phase angle time series, 4-dimensional environmental parameter time series, and 3-dimensional frequency domain envelope spectrum time series are aligned along the time axis to form a three-dimensional feature tensor with a shape of [time step, 55, 1]. The specific frequency points are set according to logarithmic intervals: 50 Hz, 63 Hz, 79 Hz,... 200 kHz, a total of 40 frequency points. It should be noted that the number of time / frequency points can be adjusted according to the actual processing method. This is an example of the application of this embodiment.

[0076] Then, the feature weights are dynamically allocated and fused to generate a bactericidal efficiency feature vector through the gated attention mechanism: the environmental parameter time series and the bioelectrochemical features, i.e., impedance modulus, phase angle, and envelope spectrum, are dynamically interacted through cross-attention to obtain environment-modulated biological features; multi-scale features are extracted from the 50 Hz-5 kHz low-frequency band, 5 kHz-50 kHz mid-frequency band spectrum, and phase angle differential to obtain a membrane rupture risk score, with a low-frequency band amplitude increase of +0.3 points and a mid-frequency band amplitude decrease of +0.5 points; the environment-modulated biological features and the membrane rupture risk score are generated through a gated residual structure to generate a channel-time two-dimensional weight matrix; the channel-time two-dimensional weight matrix is multiplied element by element with the time-domain impedance modulus time series, phase angle time series, environmental parameter time series, and frequency domain envelope spectrum time series to obtain weighted features; the time-domain impedance modulus time series, phase angle time series, environmental parameter time series, and frequency domain envelope spectrum time series are connected through a residual connection with the weighted features to obtain a bactericidal efficiency feature vector.

[0077] A time-frequency joint attention LSTM network is constructed, which includes frequency bands of the frequency domain envelope spectrum time series as nodes, 3 nodes in this embodiment, and the node features are the frequency domain envelope spectrum time series; neighborhood features are aggregated through double-layer graph convolution to obtain a frequency domain enhanced feature matrix; edge weights are calculated based on the spectral similarity between nodes: neighborhood features are aggregated through double-layer graph convolution to obtain a frequency domain enhanced feature matrix; time-domain features are used as query vectors, and frequency domain enhanced features are used as key-value vectors.

[0078] The predefined biological constraint matrix is injected to modify the attention weight, and the edge weight is calculated based on the spectral similarity between nodes: the neighborhood features are aggregated by double-layer graph convolution to obtain a frequency domain enhanced feature matrix; and the time domain features are used as query vectors, and the frequency domain enhanced features are used as key value vectors.

[0079] The predefined biological constraint matrix is injected to modify the attention weight, and the predefined biological constraint matrix weight distribution is specifically: the weight of the 100 Hz-50 kHz cell membrane rupture characteristic frequency band is 0.7, the weight of the 1-10 MHz ion channel disturbance frequency band is 0.2, and the weight of other frequency bands is 0.1; then the features output by the joint attention layer are input into the final prediction layer, which includes three fully connected units corresponding to three outputs: a sterilization efficiency prediction value, a range of 0%-100%, an inactivation logarithm prediction value range of 0-6 log CFU / mL, and a microbial load prediction value, unit CFU / mL, on a logarithmic scale. The network output is the three real-time prediction values.

[0080] According to the hospital disinfection specification, such as WS 507-2016 “Disinfection Supply Center Part 2: Cleaning, Disinfection and Sterilization Technical Operation Specification”, the inactivation logarithm target threshold is set to be ≥3.0 log CFU, and the microbial load safety threshold is set to be ≤100 CFU / mL.

[0081] The inactivation logarithm target threshold is set to be 3, and the microbial load safety threshold is set to be 100. The system generates a regulation instruction based on the deviation of the prediction value from the threshold. When the deviation of the inactivation logarithm prediction value from the target threshold is greater than 0.2, a disinfectant concentration adjustment instruction is generated, and the adjustment range is 0.02. When the deviation of the microbial load prediction value from the safety threshold is greater than 25, a disinfection time adjustment instruction is generated, and the time is extended or shortened by 0.5 minutes. The generated concentration adjustment instruction and time adjustment instruction are packaged into a multi-dimensional disinfection parameter adjustment instruction, which is sent to the control unit of the disinfection workstation.

[0082] In this embodiment, the initial microbial load in a certain disinfection process is about 10^5 CFU / mL.

[0083] Stage 1 (0-5 minutes): The system monitors that the inactivation logarithm prediction value rapidly rises to 2.8 log CFU, and the microbial load prediction value drops to about 125 CFU / mL. Since the inactivation logarithm prediction value of 2.8 is close to the target threshold of 3.0, the deviation ΔL=0.2, which triggers the concentration fine adjustment instruction=0.02%, and the concentration is increased to 2.02%. The microbial load prediction value of 125 exceeds the safety threshold of 100, and the deviation=25, which triggers the time extension instruction=0.5 minutes.

[0084] Phase 2 (5-10 minutes): After the concentration is increased, the inactivation log predictive value quickly reaches 3.1 log CFU, and the microbial load predictive value drops to 80 CFU / mL. At this time, the inactivation log predictive value exceeds the target threshold, and the microbial load predictive value is below the safety threshold. The system using the method of the present application maintains the current concentration and time.

[0085] Phase 3 (10-15 minutes): If the TOC rises due to organic interference, the inactivation log predictive value drops to 2.9 log CFU, and the system will again determine whether further adjustment of the concentration is needed according to AL = 0.1.

[0086] Through real-time monitoring and intelligent control, the parameters can be dynamically adjusted according to the actual sterilization efficiency and microbial load of the disinfectant. Compared with the fixed time disinfection method, the average disinfection time can be shortened, while ensuring that the disinfection effect always meets the standards and reducing the overuse of disinfectants.

[0087] Example 2: This embodiment provides a specific application of a real-time monitoring method for the sterilization effect of aldehyde disinfectants. The implementation process and related technical details of the method are described in detail below.

[0088] First, real-time collection of microbial electrochemical response signals, configured with a multi-frequency excitation three-electrode sensing unit and a multi-channel environmental sensor, a multi-frequency excitation three-electrode system is used to electrochemically stimulate microorganisms in the disinfectant solution, and a multi-channel environmental sensor is used to synchronously collect environmental parameters. The excitation frequency used covers the frequency band of 50 Hz to 200 kHz, and the frequency scanning density is not less than 10 sampling points per decade. Through the excitation signal, the microbial electrochemical response signal in the disinfectant solution is collected in real time, and the following time series data is obtained:

[0089] Time-domain impedance modulus time series: by real-time monitoring of electrochemical impedance, the response of microorganisms to different frequency signals is recorded.

[0090] Phase angle time series: Obtain the phase angle change at each frequency, reflecting the change characteristics of the microbial cell membrane.

[0091] Environmental parameter time series: Synchronously collect environmental parameters such as temperature, pH, redox potential, and organic interferent concentration of the solution.

[0092] For the electrochemical response signal of microorganisms, a wavelet packet-FFT coprocessor is built in, which is configured to perform microbial response feature-oriented frequency domain envelope spectrum analysis. Wavelet packet decomposition is performed to extract the change characteristics of the microbial cell membrane permeability. db6 wavelet basis function is used for 5-layer wavelet packet decomposition, and the feature frequency band signal of the 3rd to 5th layer nodes is extracted, which is divided into two frequency bands, 50Hz-5kHz low frequency band and 5kHz-50kHz medium frequency band. The frequency domain envelope spectrum time series is generated by fast Fourier transform, which specifically includes:

[0093] Microbial response characteristics in low frequency band 50Hz-5kHz and middle frequency band 5kHz-50kHz.

[0094] Performing fast Fourier transform (FFT) on the extracted time-domain signal to generate a frequency-domain envelope spectrum time series, which represents the dynamic changes of the microbial cell membrane.

[0095] Aligning the four time series data along the time axis to form a three-dimensional feature tensor: time-domain impedance modulus time series, phase angle time series, environmental parameter time series, and frequency-domain envelope spectrum time series, which provides input data for subsequent deep learning models.

[0096] Based on the gating attention weight generation unit and the three-dimensional tensor processor, the above features are weighted and fused using the gating attention mechanism, with the following steps: dynamically interacting the environmental parameter time series and the bioelectrochemical features through cross-attention to obtain the environment-modulated biological features; extracting multi-scale features from the low frequency band, middle frequency band spectrum, and phase angle differential to obtain the membrane rupture risk score and adaptively trigger the warning; generating a channel-time two-dimensional weight matrix through the gating residual structure for the environment-modulated biological features and the membrane rupture risk score; multiplying the channel-time two-dimensional weight matrix with the time-domain impedance modulus time series, phase angle time series, environmental parameter time series, and frequency-domain envelope spectrum time series element by element to obtain the weighted features; connecting the time-domain impedance modulus time series, phase angle time series, environmental parameter time series, and frequency-domain envelope spectrum time series with the weighted features through the residual connection to obtain the sterilization efficiency feature vector.

[0097] Then construct a graph convolution-time-frequency joint attention LSTM network, which is specifically: taking the frequency band of the frequency-domain envelope spectrum time series as the node, and the node feature as the frequency-domain envelope spectrum time series, constructing a frequency-domain relationship graph, calculating the edge weight based on the spectral similarity between nodes, aggregating the neighborhood features through double-layer graph convolution to obtain the frequency-domain enhanced feature matrix, taking the time-domain feature as the query vector, the frequency-domain enhanced feature as the key-value vector, injecting a predefined biological constraint matrix to modify the attention weight, and weighted aggregation to obtain the time-frequency fusion feature of biological prior knowledge, and after bidirectional encoding of the time-frequency fusion feature using bidirectional LSTM, extracting high-level features through the fully connected layer and decoupling, and real-time output including the following three key indicators:

[0098] Sterilization efficiency prediction value: real-time prediction of the overall effect of sterilization in the disinfection process.

[0099] Inactivation log prediction value: represents the inactivation of microorganisms in the disinfection process.

[0100] Microbial load prediction value: reflects the remaining amount of microorganisms in the disinfection process.

[0101] According to the real-time monitoring results, generate corresponding disinfection parameter adjustment instructions, including:

[0102] Inactivation logarithm target threshold, a target value of inactivation logarithm is set to generate disinfectant concentration adjustment instructions.

[0103] Microbial load safety threshold, a safety threshold of microbial load is set to generate disinfection time adjustment instructions.

[0104] After the adjustment instructions are generated, the disinfectant concentration and the disinfection time can be adjusted according to the deviation value to achieve more accurate sterilization effect.

[0105] In the experiment of the embodiment, the concentration of aldehyde disinfectant is 500 ppm, the environmental conditions are temperature 25℃, pH 6.5, and ORP 300 mV. A group of microorganisms are subjected to sterilization treatment. The electrochemical response signal is collected in real time by the system applying the method of the application, and the following time sequence data is obtained:

[0106] Time-domain impedance modulus time sequence: within the frequency range of 50 Hz to 200 kHz, the impedance modulus changes with time.

[0107] Phase angle time sequence: the phase angle reflecting the change of the cell membrane changes with time.

[0108] Environmental parameter time sequence: records the environmental temperature 25℃, the pH 6.5, and the ORP 300 mV, etc.

[0109] After wavelet packet decomposition and Fourier transform, the frequency domain envelope spectrum time sequence is generated, and a three-dimensional feature tensor is further formed. Through the gated attention mechanism and the LSTM network, the sterilization efficiency is predicted in real time to be 92%, the inactivation logarithm is 3.5, and the microbial load is 4.2 log.

[0110] Based on the deviation of 3.5 log between the predicted value of the microbial load and the safety threshold, the system issues a disinfection time adjustment instruction, recommending to extend the disinfection time by 15 minutes. Based on the deviation of 3.0 log between the inactivation logarithm and the target value, the system issues a disinfectant concentration adjustment instruction, recommending to increase the concentration to 600 ppm.

[0111] Finally, the adjusted parameters of the system can effectively improve the disinfection efficiency and realize real-time monitoring and optimal adjustment.

[0112] It is to be noted that the terms such as first and second, etc. are used herein merely to differentiate one entity or operation from another entity or operation, and do not necessarily require or imply there is any such actual relationship or order between these entities or operations. Also, the terms "include", "comprise" or any other variant thereof are intended to cover a non-exclusive inclusion, so that a process, method, article, or apparatus that includes a list of elements does not only include those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0113] The above examples are only illustrative of the present application and are not intended to limit the scope of protection of the present application. Any design identical or similar to the present application falls within the scope of protection of the present application.

Claims

1. A method for real-time monitoring of the bactericidal effect of aldehyde disinfectants, characterized in that, include: The electrochemical response signals of microorganisms in the disinfection solution are acquired in real time by a multi-frequency excitation three-electrode system, and the timing sequence of time domain impedance modulus, phase angle and environmental parameters are acquired simultaneously. The time-domain impedance modulus time series is subjected to wavelet packet decomposition guided by microbial response characteristics to extract the preset cell membrane permeability change characteristic frequency band signal, and then subjected to Fourier transform to generate frequency domain envelope spectrum time series. The time-domain impedance modulus time series, phase angle time series, environmental parameter time series and frequency domain envelope spectrum time series are aligned along the time axis to form a three-dimensional feature tensor. Feature weights are dynamically allocated and fused through a gating attention mechanism to generate a bactericidal efficacy feature vector. A graph convolutional-time-frequency joint attention LSTM network is constructed, using frequency bands of the frequency domain envelope spectrum time series as nodes and the node features as the frequency domain envelope spectrum time series. A frequency domain relationship graph is constructed, and edge weights are calculated based on the spectral similarity between nodes. Neighborhood features are aggregated through two-layer graph convolution to obtain a frequency domain enhanced feature matrix. The time domain features are used as query vectors, and the frequency domain enhanced features are used as key vectors. A predefined biological constraint matrix is ​​injected to correct the attention weights, and weighted aggregation is used to obtain time-frequency fusion features of biological prior knowledge. The time-frequency fusion features are bidirectionally encoded using a bidirectional LSTM, and then high-level features are extracted and decoupled through a fully connected layer. The predicted values ​​of sterilization efficiency, inactivation logarithm, and microbial load are output in real time as real-time monitoring results of sterilization effect.

2. The method for real-time monitoring of the bactericidal effect of aldehyde disinfectants as described in claim 1, characterized in that, Also includes: Set the target threshold for inactivation logarithmic inactivation and the safe threshold for microbial load; Disinfectant concentration adjustment instructions are generated based on the deviation between the predicted logarithm of inactivation and the target threshold. Based on the deviation between the predicted microbial load and the safety threshold, a disinfection time adjustment instruction is generated. The disinfectant concentration adjustment instruction and the disinfection time adjustment instruction generate a multidimensional disinfection parameter adjustment instruction.

3. The method for real-time monitoring of the bactericidal effect of aldehyde disinfectants as described in claim 1, characterized in that, The environmental parameters time series include temperature, pH, redox potential and concentration of organic interfering substances; The excitation frequency of the microbial electrochemical response signal covers the frequency band from 50Hz to 200kHz, and the frequency scanning density is not less than 10 points / ten octaves.

4. The method for real-time monitoring of the bactericidal effect of aldehyde disinfectants as described in claim 1, characterized in that, The microbial response feature-guided wavelet packet decomposition includes the following steps: Based on the mechanism of action of aldehyde disinfectants, the preset characteristic frequency bands for microbial cell membrane rupture are the low frequency band (50Hz-5kHz) and the mid frequency band (5kHz-50kHz). The db6 wavelet basis function is used to perform 5-level wavelet packet decomposition to extract the characteristic frequency band signals corresponding to the nodes of the 3rd to 5th levels.

5. The method for real-time monitoring of the bactericidal effect of aldehyde disinfectants as described in claim 1, characterized in that, The process of dynamically allocating feature weights and fusing them to generate a bactericidal efficacy feature vector through a gating attention mechanism includes: By dynamically interacting environmental parameters with bioelectrochemical characteristics through cross-attention, environmental-modulated biological characteristics can be obtained. Multi-scale features are extracted from low-frequency and mid-frequency spectra and phase angle differentials to obtain membrane rupture risk scores and adaptively trigger early warnings. The environmental modulated biomarkers and membrane rupture risk scores are used to generate a channel-time dual-dimensional weight matrix through a gated residual structure; The channel-time dual-dimensional weight matrix is ​​multiplied element-wise with the time-domain impedance magnitude time series, phase angle time series, environmental parameter time series, and frequency domain envelope spectrum time series to obtain the weighted features. By connecting the time-domain impedance modulus time series, phase angle time series, environmental parameter time series, and frequency domain envelope spectrum time series with the weighted features via residual concatenation, a bactericidal efficacy feature vector is obtained.

6. The method for real-time monitoring of the bactericidal effect of aldehyde disinfectants as described in claim 1, characterized in that, The predefined biological constraint matrix includes: Based on prior knowledge of microbial electrochemical response, the time-domain frequency band is divided into the cell membrane rupture characteristic band, the ion channel perturbation band, and other bands, and each is assigned a constraint weight.

7. A system for real-time monitoring of the bactericidal effect of an aldehyde disinfectant according to any one of claims 1-6, characterized in that, include: The electrochemical signal acquisition module is equipped with a multi-frequency excitation three-electrode sensing unit and a multi-channel environmental sensor. It is used to acquire the electrochemical response signal of microorganisms in the disinfection solution in real time through the multi-frequency excitation three-electrode system, and simultaneously acquire the time domain impedance modulus timing, phase angle timing and environmental parameter timing. The biofeature extraction module, with a built-in wavelet packet-FFT coprocessor, is configured to perform microbial response feature-guided frequency domain envelope spectrum analysis. It is used to perform microbial response feature-guided wavelet packet decomposition on the time-domain impedance modulus time series, extract the preset cell membrane permeability change feature frequency band signal, and generate the frequency domain envelope spectrum time series through fast Fourier transform. The dynamic fusion decision module includes a gated attention weight generation unit and a three-dimensional tensor processor. It is used to align the time-domain impedance modulus time series, phase angle time series, environmental parameter time series and frequency domain envelope spectrum time series according to the time axis to form a three-dimensional feature tensor. It dynamically allocates feature weights and fuses them to generate a bactericidal efficacy feature vector through a gated attention mechanism. The monitoring result acquisition module integrates a graph convolutional-time-frequency joint attention LSTM network. It receives the sterilization efficacy feature vector and uses it to construct a frequency domain relationship graph with frequency bands of the frequency domain envelope spectrum time series as nodes and node features as the frequency domain envelope spectrum time series. It calculates edge weights based on the spectral similarity between nodes, aggregates neighborhood features through two-layer graph convolution to obtain a frequency domain enhanced feature matrix. Using the time domain features as the query vector and the frequency domain enhanced features as the key vector, it injects a predefined biological constraint matrix to correct the attention weights and weighted aggregates to obtain time-frequency fusion features of biological prior knowledge. After bidirectional encoding of the time-frequency fusion features using a bidirectional LSTM, it extracts high-level features and decouples them through a fully connected layer, and outputs the sterilization efficiency prediction value, inactivation log prediction value, and microbial load prediction value in real time as the real-time monitoring results of sterilization effect.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 6.

9. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of claims 1 to 6.

Images