Remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion

Abstract

The invention discloses an environmental protection equipment remote intelligent fault diagnosis system based on multi-source data fusion, and relates to the technical field of environmental protection equipment fault diagnosis, and the system comprises the steps: installing a sensor network at a key position of a plastic pyrolysis reactor, and storing a collected signal into a historical reaction database; performing signal decomposition on all data in the historical reaction database to obtain an environmental disturbance component and a catalyst memory component, and inputting the environmental disturbance component and the catalyst memory component into the constructed component prediction model to obtain a catalyst memory influence coefficient and an environmental disturbance influence coefficient; obtaining a reactor risk index according to the catalyst memory residual error and the environment disturbance residual error in combination with the temperature deviation; judging whether to operate a catalyst difference data module or not through a cleaning judgment strategy; and calculating a catalyst-temperature influence coefficient after cleaning, and transmitting the catalyst-temperature influence coefficient to the component prediction model. According to the invention, the problem of misjudgment caused by process fluctuation and a catalyst'memory 'effect is solved, and the operation safety, the monitoring precision and the operation and maintenance efficiency of the plastic pyrolysis reactor are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental protection equipment fault diagnosis, and particularly to a remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion. Background Art

[0002] With the rapid development of the new energy industry, the recycling of waste batteries has become a key link in circular economy and environmental protection; among them, the plastic diaphragms and packaging materials of waste lithium-ion batteries are recycled through pyrolysis technology in a plastic pyrolysis reactor, which can not only remove harmful binders and electrolyte residues, but also convert plastics into fuel oil and combustible gas, realizing resource utilization; therefore, high-precision fault monitoring of plastic pyrolysis reactors is of great significance for improving the operation reliability and intelligent level of pyrolysis devices and promoting the sustainable development of the waste plastic resource utilization industry; For example, the existing Chinese patent with the publication number CN110208451B discloses a bipolar micro fixed-bed reactor combined with a photoionization mass spectrometry on-line detection system and method, including a bipolar micro fixed-bed reactor, an off-line cooling device and an on-line detection device. The bipolar micro fixed-bed reactor is respectively connected to the off-line cooling device and the on-line detection device, and the bipolar micro fixed-bed reactor is connected to the off-line cooling device through a transmission line; the bipolar micro fixed-bed reactor performs catalytic pyrolysis on solid fuel samples; in the present invention, the pyrolysis reaction and the catalytic reaction do not interfere with each other, the reactor and the heating unit can be quickly separated and cooled down, and two different types of and primary state cokes can be accurately analyzed; a high-temperature valve is used to separate the two reactors, so that intermittent continuous sampling of solid fuel and real-time monitoring of catalyst deactivation can be realized. The device and the reactor are small in volume and the residence time of the reactants is short; However, for traditional plastic pyrolysis reactors, due to the high-temperature, oxygen-free environment and the adsorption-desorption behavior of pyrolysis products on the catalyst surface, the acidic products adsorbed in the equipment shift the acid-base balance on the catalyst surface to the acid side, and the activation energy required for pyrolysis changes accordingly, resulting in the drift of the rate constant at the same temperature, leading to the "history dependence" effect and transient disturbances inside the reactor, which will cause false alarms and pose major challenges to equipment operation monitoring and fault diagnosis.

[0003] Therefore, the present invention provides a remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion. Summary of the Invention

[0004] The purpose of the present invention is to provide a remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion to solve the existing problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion, including: A signal acquisition module, which is used to install a sensor network at key positions of a plastic pyrolysis reactor and store the acquired signals in a historical reaction database; A signal decomposition module, which is used to decompose all data in the historical reaction database to obtain an environmental disturbance component and a catalyst memory component, input them into a constructed component prediction model, and obtain a catalyst memory influence coefficient and an environmental disturbance influence coefficient; A safety threat prediction module, which obtains a reactor risk index according to the catalyst memory residual and the environmental disturbance residual combined with the temperature deviation; judges whether to run the catalyst difference data module through a cleaning judgment strategy; A catalyst difference data module, which is used to calculate the catalyst-temperature influence coefficient after cleaning and transport it to the component prediction model.

[0006] A further improvement of the present invention lies in that the signal decomposition module includes a signal separation unit, a component prediction model construction unit and an adaptive threshold generation unit; The signal separation unit is equipped with an adsorption-desorption critical frequency calculation strategy. First, the adsorption-desorption time constant is determined. In the catalytic cracking reaction, the adsorption-desorption of product molecule A on the catalyst surface S is expressed as , where represents the adsorption rate constant, represents the desorption rate constant; then its characteristic time constant is expressed as , where represents the instantaneous concentration of the adsorbate in the reactor gas phase, and the corresponding adsorption-desorption critical frequency is expressed as .

[0007] A further improvement of the present invention lies in that the signal separation unit further includes applying complete ensemble empirical mode decomposition to the original signal to obtain N intrinsic mode function components , performing Hilbert transform on each to obtain the instantaneous frequency , obtaining the average frequency of each mode. If is greater than k, the corresponding component is extracted and classified as the environmental disturbance component . If is less than or equal to k, the corresponding component is extracted and classified as the catalytic memory component .

[0008] A further improvement of the present invention lies in that the component prediction model construction unit includes aligning the environmental disturbance component and the catalyst memory component output by the signal separation unit according to the time stamp to form a parallel time series; for each moment t, intercepting the forward input with a window length w and performing normalization processing on the data within the window; Subsequently, an environmental disturbance channel and a catalytic memory channel are constructed. The environmental disturbance channel performs causal and dilated convolutions on the environmental disturbance terms of the parallel time series through the first convolutional kernel of the temporal convolutional network, and outputs an environmental disturbance feature map. The second convolutional kernel then performs convolution on the output of the first layer, and so on until the Mth convolutional kernel. After that, the M feature maps are compressed into a vector of a fixed dimension. , each layer of the decoder performs a deconvolution operation, gradually restoring the sequence length and channels, and generating a sequence of the same dimension as the input data at the last layer. , realizing the reconstruction of the environmental disturbance component. Taking the root mean square error of and as the reconstruction loss, calculating the difference between and , and outputting the environmental disturbance residual as the environmental disturbance influence coefficient.

[0009] A further improvement of the present invention lies in that the catalytic memory channel is implemented based on the LSTM model, specifically including: Calculating the sliding mean of the catalytic memory terms of the parallel time series and the autocorrelation coefficient . Concatenating , with the original catalytic memory component in terms of feature dimensions to form a three-channel catalytic memory input tensor Cmi. The LSTM model simultaneously accepts these three sequences at each step and then performs a normalization process, and maps the last layer to a scalar prediction through a fully connected layer to obtain the predicted value of the catalytic memory component at time t + 1 , and further obtaining . Calculating the difference between and , and outputting the catalytic memory residual as the catalytic memory influence coefficient.

[0010] A further improvement of the present invention lies in that the adaptive threshold generation unit includes extracting the adsorption-desorption critical frequency k, performing wavelet packet transform on the environmental disturbance influence coefficient and the catalytic memory influence coefficient to obtain the energy of each subband, and calculating the ratio of the sum of the energies corresponding to the environmental disturbance term and the catalytic memory term to obtain the catalytic memory energy ratio . Passing the adsorption-desorption critical frequency k and the catalytic memory energy ratio through the tanh activation function to obtain the fuzzy control input quantities and , obtaining the environmental interference threshold and the catalytic memory threshold , where is the comprehensive scaling coefficient output by fuzzy inference. represents the environmental disturbance influence coefficient within the sliding window [t−w,t], represents the catalytic memory influence coefficient within the sliding window [t−w,t].

[0011] A further improvement of the present invention lies in that the security threat prediction module includes a temperature deviation calculation unit and a device risk acquisition unit; the temperature deviation calculation unit obtains the temperature deviation by subtracting the digital twin predicted temperature from the measured temperature ; the device risk acquisition unit aggregates the catalyst memory influence coefficient, the environmental disturbance influence coefficient, and the temperature deviation after standardization using the Euclidean norm to obtain the reactor threat fusion value, and maps the reactor threat fusion value to [0,1] through the Logistic function to obtain the reactor risk index . .

[0012] A further improvement of the present invention lies in that the device risk acquisition unit is also equipped with a cleaning judgment strategy, which is realized by setting a cleaning risk threshold Tcl and an alarm risk threshold Tal; When the reactor risk index is less than the cleaning risk threshold Tcl, the alarm level is normal, and at this time, continuous monitoring is carried out without any operation; When the reactor risk index is greater than or equal to the cleaning risk threshold Tcl and less than the alarm risk threshold Tal, the alarm level is a warning, and the flow rate is automatically adjusted according to the formula , and online ultrasonic or gas blowing cleaning is arranged, represents the initial flow rate; at the same time, the catalyst difference data module is run; When the reactor risk index is greater than the alarm risk threshold Tal, the alarm level is critical, catalyst oxidative regeneration or safe shutdown is initiated, a maintenance work order is triggered, and an alarm notification is sent.

[0013] A further improvement of the present invention lies in that the catalyst difference data module includes, at the cleaning trigger moment, recording the full-field temperature deviation of the digital twin before cleaning , after cleaning is completed and operation is resumed, recording the new deviation , obtaining the deviation recovery amount , pairing with the catalytic memory component within the corresponding time window to form an incremental learning sample set, adding the new sample to the LSTM training queue, and training using a mini-batch model with the catalytic memory component within the sliding window [t−w,t] to obtain the catalyst-temperature influence coefficient , after the feature dimension concatenation, update the model weights.

[0014] A further improvement of the present invention lies in that the key positions of the plastic pyrolysis reactor include arranging a thermocouple array at equal intervals along the height direction inside and outside the reactor cylinder to measure the local temperature; installing an in-situ spectral probe above the catalyst bed or at the sampling hole to detect the surface coverage of the catalyst.

[0015] Compared with the prior art, the beneficial effects of the present invention are: The present invention first realizes the accurate distinction between environmental disturbances and catalytic memory effects by comparing the adsorption-desorption critical frequency with the average instantaneous frequency of each signal component, reducing the misjudgment of "slow drift" caused by memory accumulation; Through the training of the "normal" environmental disturbances by the signal decomposition module, it can accurately reconstruct common patterns, and only generate significant residuals for truly abnormal situations, including sudden disturbances or equipment failures, thereby reducing the reconstruction false alarm rate; Introducing the high / low frequency energy ratio in the adaptive threshold generation can dynamically adjust the sensitivity according to the energy distribution, making the system robust to slow-changing memory noise and highly vigilant to transient disturbances; Dynamically generating the threshold based on the statistical features of the residual window can track the working condition changes in real time and automatically adjust the threshold bandwidth, greatly reducing the false alarms caused by process drift; integrating multi-source information such as adsorption driving factors, energy ratio, and temperature deviation through Mamdani inference effectively balances the adjustment of the threshold in the "memory accumulation" and "sudden disturbance" scenarios, ensuring the optimal balance between sensitivity and stability. Description of the Drawings

[0016] Figure 1 is the framework diagram of the remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion of the present invention; Figure 2 is the flowchart of the security threat prediction module of the present invention. Detailed Embodiments

[0017] The technical solutions of the present invention will be described in detail below with reference to the drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solutions of the present invention, rather than limitations on the technical solutions of the present invention. Without conflict, the technical features in the embodiments of the present invention and the embodiments can be combined with each other.

[0018] The term "and / or" only describes the associated relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " generally represents an "or" relationship between the associated objects before and after.

[0019] Example 1 Figure 1 The framework diagram of the remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion disclosed in this example is shown, including: A signal acquisition module, which is used to install a sensor network at key positions of the plastic pyrolysis reactor and store the collected signals in the historical reaction database; A signal decomposition module, which is used to decompose all the data in the historical reaction database to obtain an environmental disturbance component and a catalyst memory component, input them into the constructed component prediction model, and obtain a catalyst memory influence coefficient and an environmental disturbance influence coefficient; A safety threat prediction module, which obtains the reactor risk index according to the catalyst memory residual and the environmental disturbance residual combined with the temperature deviation; judges whether to run the catalyst difference data module through a cleaning judgment strategy; A catalyst difference data module, which is used to calculate the catalyst-temperature influence coefficient after cleaning and output it to the component prediction model.

[0020] The key positions of the plastic pyrolysis reactor include measuring the feed rate by a mass flowmeter at the feed inlet; upstream of the plastic particle / powder feed pipeline, between the inlet and the storage bin, the thermocouples and mass flowmeters at each key position of the reactor collect and store in the historical database at a frequency of 1 Hz through the SCADA system; and a thermocouple array (inlet, middle, outlet) is arranged equidistantly along the height direction inside and outside the reactor cylinder, and typically 5-7 measuring points are used to measure the local temperature; the full-field temperature distribution is calculated once per minute by CFD-DEM digital twin and synchronized with the measured temperature; an in-situ spectroscopic probe is installed above the catalyst bed or at the sampling hole to detect the surface coverage of the bed.

[0021] Subsequently, the original signal is first applied with band-pass filtering to remove electromagnetic interference above 50 Hz, and then linear detrending is performed; the temperature and flow signals are linearly mapped to the [0,1] interval to provide a unified dimension for subsequent decomposition and model training.

[0022] The signal decomposition module includes a signal separation unit, a component prediction model construction unit, and an adaptive threshold generation unit; The signal separation unit is equipped with an adsorption-desorption critical frequency calculation strategy. First, the adsorption-desorption time constant is determined. In the catalytic cracking reaction, the adsorption-desorption of product molecule A on the catalyst surface S is expressed as , where represents the adsorption rate constant, represents the desorption rate constant; it is set by those skilled in the art. Regarding this second-order adsorption-desorption process as a first-order system, its characteristic time constant is expressed as , where Represents the instantaneous concentration of the adsorbate (pyrolysis product) in the gas phase of the reactor, with the unit of , which can be obtained in real time by on-line gas chromatography or mass spectrometry; Adsorption rate Refers to the amount of gas-phase molecule A adsorbed on the free surface sites per unit time per unit surface area , where The unit is , Is the concentration of free active sites per unit area or unit volume, with the unit ; Desorption rate , The unit is , Represents the concentration of sites occupied by adsorbed molecules, with the same unit as ; For the surface coverage Conduct a mass balance, Represents the total site concentration, and the transient behavior of the adsorption-desorption process can be obtained , making the equilibrium state , there is , and we get , where , represents the adsorption equilibrium constant, reflecting the affinity of molecules on the surface; The adsorption / desorption rate constants satisfy the Arrhenius relationship, , , where and Represent the pre-exponential factor, and Represent the activation energy, RT represents the gas constant, and through temperature scanning experiments, their respective and A can be obtained; then correct T online to get the real-time and ; When is small or is large, the desorption is slow, the adsorption is fast, the surface coverage accumulates higher, and the influence on the subsequent reaction rate is more significant, resulting in an enhanced "history dependence". The characteristic frequency of the adsorption-desorption can be used as the critical value for signal separation. The components with frequencies lower than this value are classified as catalytic memory components, and those higher than this value are classified as environmental perturbation components. Then the corresponding adsorption-desorption critical frequency is expressed as , representing the main dynamic time scale of the "memory" effect on the catalyst surface.

[0023] The signal separation unit further includes applying complete ensemble empirical mode decomposition to the original signal to obtain N intrinsic mode function components , for each Perform the Hilbert transform to obtain the instantaneous frequency , and obtain the average frequency of each mode , , if is greater than k, then extract the corresponding component and classify it as the environmental disturbance component , if is less than or equal to k, then extract the corresponding component and classify it as the catalytic memory component ; The Hilbert–Huang transform ensures the accurate extraction of the local frequency of non-stationary and non-linear signals.

[0024] The component prediction model construction unit includes aligning the environmental disturbance component and the catalyst memory component output by the signal decomposition unit according to the time stamp to form a parallel time series; for each moment t, intercept the forward input with a window length w (for example, 100 steps), , , and perform standardization processing on the data within the window; Subsequently, construct an environmental disturbance channel and a catalytic memory channel. The environmental disturbance channel performs causal and dilated convolutions on the environmental disturbance terms of the parallel time series through the first convolutional kernel of the temporal convolutional network, and outputs an environmental disturbance feature map. The second convolutional kernel then performs convolution on the output of the first layer, and so on until the Mth convolutional kernel. After that, compress the M feature maps into a vector of a fixed dimension , each layer of the decoder performs an anti-convolution operation to gradually restore the sequence length and channels, so that the time dimension and the number of channels of the feature map are restored in turn, and a sequence of the same dimension as the input data is generated at the last layer , realizing the reconstruction of the environmental disturbance component. This vector is the "latent representation", which condenses all the time series information within the window. Take and as the root mean square error of reconstruction loss, , this loss not only measures the reconstruction quality, but also drives the model to learn the "normal dynamics" of the conversion perturbation. Calculate and to output the environmental disturbance residual as the environmental disturbance influence coefficient.

[0025] The catalytic memory channel is implemented based on the LSTM model, specifically including Calculating the sliding mean and the autocorrelation coefficient of the catalytic memory terms of the parallel time series, and taking , Perform feature dimension concatenation with the original catalytic memory component to form a three-channel catalytic memory input tensor Cmi; the LSTM model simultaneously accepts these three sequences at each step and performs normalization processing, so that when updating the hidden state, it simultaneously refers to the original signal and auxiliary features, strengthening the expression ability of trends and dependencies, and mapping the last layer of fully connected to a scalar prediction to obtain the predicted value of the catalytic memory component at time t+1 , and then obtain , calculate and The difference of is output as the catalytic memory residual as the catalytic memory influence coefficient, reflecting the cumulative "memory" effect not captured by the model.

[0026] By continuously calculating , a much smoother curve than can be obtained to depict the slow-varying trend and steady-state drift of the system; at the same time, As the model input, it can provide a "local baseline" for the network to help it distinguish true structural changes (trend drift) from random jitter (residual), thereby reducing the over-sensitivity to short-term fluctuations; A high positive autocorrelation (close to 1) means Changes slowly at adjacent times and has strong lag memory; approaching 0 indicates a more random dynamic; calculate within a sliding window (e.g., window=50) to quantify the "smoothness" and "trend strength" of the window and provide a sequence dependence index for the model; The auxiliary features participate in weight update both in the forward feed and backpropagation of the network, making the model pay more attention to low-frequency information related to trends rather than being dominated by short-term jitter; the sliding mean filters high-frequency anomalies, and the autocorrelation value reflects the signal persistence. The two can complement each other to help the model distinguish "short-term perturbations" from "long-term trends" and reduce the false alarm rate.

[0027] The adaptive threshold generation unit includes extracting the adsorption-desorption critical frequency k. When the adsorption accumulation is significant (k is large), the threshold is amplified to tolerate "memory" drift; when k is small, the threshold is contracted to sensitively capture sudden perturbations, and the environmental perturbation influence coefficient and the catalytic memory influence coefficient Perform wavelet packet transform to obtain the energy of each subband, and calculate the ratio of the sum of the energies corresponding to the environmental perturbation term and the catalytic memory term to obtain the catalytic memory energy ratio ; A large value indicates prominent transient perturbations, and the threshold should be appropriately reduced to improve sensitivity; When it is small, the threshold can be increased to robustly resist slow-varying noise; the adsorption-desorption critical frequency k and the catalytic memory energy ratio Pass through the tanh activation function to obtain the fuzzy control input quantity and , the environmental interference threshold is obtained and the catalytic memory threshold , where is the comprehensive scaling coefficient of the fuzzy inference output, represents the environmental disturbance influence coefficient within the sliding window [t−w,t], represents the catalytic memory influence coefficient within the sliding window [t−w,t]. The fuzzy rule base is expressed as if is high and is low, the threshold increases. If is low and is high, the threshold decreases. The inference and defuzzification adopt Mamdani inference and defuzzify by the centroid method; represents the mean value within the window, represents the standard deviation within the window.

[0028] Example 2 Based on the same inventive concept as in Example 1, this example presents the specific implementation of the security threat prediction module in Example 1, Figure 2 showing the flowchart of the security threat prediction module in this example, specifically including a temperature deviation calculation unit and a device risk acquisition unit; The temperature deviation calculation unit obtains the temperature deviation by the difference between the measured temperature and the digital twin predicted temperature ; The device risk acquisition unit aggregates the catalyst memory influence coefficient, environmental disturbance influence coefficient, and temperature deviation after standardization using the Euclidean norm to obtain the reactor threat fusion value, and maps the reactor threat fusion value to [0,1] through the Logistic function to obtain the reactor risk index .

[0029] The device risk acquisition unit also incorporates a cleaning judgment strategy, which is implemented by setting a cleaning risk threshold Tcl and an alarm risk threshold Tal; When the reactor risk index is less than the cleaning risk threshold Tcl, the alarm level is normal, and monitoring continues without any operation; When the reactor risk index is greater than or equal to the cleaning risk threshold Tcl and less than the alarm risk threshold Tal, the alarm level is a warning. According to the formula automatically adjusts the flow rate and arranges for on-line ultrasonic or air-blowing cleaning, represents the initial flow rate; meanwhile, the catalyst difference data module is run; represents the flow rate that increases as the reactor risk index increases, It also increases accordingly, enabling the flow rate to be adaptively adjusted according to the risk indicators of the reactor; When the reactor risk indicator is greater than the alarm risk threshold Tal, the alarm level is critical, catalyst oxidative regeneration or safe shutdown is initiated, a maintenance work order is triggered, and an alarm notification is sent.

[0030] This embodiment also proposes a specific implementation method of the catalyst difference data module in Embodiment 1, including recording the full-field temperature deviation of the digital twin before cleaning at the cleaning trigger moment , after the cleaning is completed and the operation is resumed, recording the new deviation , obtaining the deviation recovery amount , and pairing with the catalytic memory component within the corresponding time window to form an incremental learning sample set, adding the new sample to the LSTM training queue, and training with a small batch model and the catalytic memory component within the sliding window [t−w,t] , obtaining the function f, calculating the catalyst-temperature influence coefficient , after feature dimension splicing, updating the model weights. Since the perturbation residual discrimination is more reliable, the overall false alarm rate of the system is reduced; the present invention can automatically optimize the cleaning cycle according to the distribution, extend the cleaning interval when is small, saving maintenance costs; the online learning mechanism enables the model to adapt to different raw material batches and operating conditions, improving the system stability.

[0031] The set values such as the threshold and weight can be set according to the default settings of the present invention or can be set by those skilled in the art themselves.

[0032] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0033] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0034] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0035] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or means for implementing the functions specified in one block or multiple blocks.

[0036] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit and scope protected by the present invention and the claims. All of these are within the protection scope of the present invention.

Claims

1. A remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion, characterized in that: Including: A signal acquisition module, which is used to install a sensor network at key positions of a plastic pyrolysis reactor and store the acquired signals in a historical reaction database; A signal decomposition module, which is used to decompose all the data in the historical reaction database to obtain an environmental disturbance component and a catalyst memory component, input them into a constructed component prediction model, and obtain a catalyst memory influence coefficient and an environmental disturbance influence coefficient; A safety threat prediction module, which obtains a reactor risk index according to the catalyst memory residual and the environmental disturbance residual combined with the temperature deviation; judges whether to run a catalyst difference data module through a cleaning judgment strategy; A catalyst difference data module, which is used to calculate the catalyst-temperature influence coefficient after cleaning and deliver it to the component prediction model.

2. The remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion according to claim 1, wherein: The signal decomposition module includes a signal separation unit, a component prediction model construction unit and an adaptive threshold generation unit; The signal separation unit is equipped with an adsorption-desorption critical frequency calculation strategy. First, the adsorption-desorption time constant is determined. In the catalytic cracking reaction, the adsorption-desorption of product molecule A on the catalyst surface S is expressed as , where represents the adsorption rate constant, represents the desorption rate constant; then its characteristic time constant is expressed as , where represents the instantaneous concentration of the adsorbate in the gas phase of the reactor, and the corresponding adsorption-desorption critical frequency is expressed as .

3. The remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion according to claim 2, characterized in that: The signal separation unit further includes applying complete ensemble empirical mode decomposition to the original signal to obtain N intrinsic mode function components , performing Hilbert transform on each to obtain the instantaneous frequency , obtaining the average frequency of each mode , if is greater than k, then extracting the corresponding component and classifying it as an environmental disturbance component , if is less than or equal to k, then extracting the corresponding component and classifying it as a catalytic memory component .

4. The remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion according to claim 3, characterized in that: The component prediction model construction unit includes aligning the environmental disturbance component and the catalyst memory component output by the signal separation unit according to the time stamp to form a parallel time series; for each moment t, intercepting the forward input with a window length w and performing normalization processing on the data within the window; Subsequently, an environmental disturbance channel and a catalytic memory channel are constructed. The environmental disturbance channel performs causal and dilated convolutions on the environmental disturbance terms of the parallel time series through the first convolutional kernel of the temporal convolutional network, and outputs an environmental disturbance feature map. The second convolutional kernel then performs convolution on the output of the first layer, and so on until the Mth convolutional kernel. After that, the M feature maps are compressed into a vector of a fixed dimension. , each layer of the decoder performs an anti-convolution operation, gradually restoring the sequence length and channels, and generating a sequence of the same dimension as the input data at the last layer , realizing the reconstruction of the environmental disturbance component. Taking the root mean square error of and as the reconstruction loss, calculate and , and output the environmental disturbance residual as the environmental disturbance influence coefficient.

5. The remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion according to claim 4, characterized in that: The catalytic memory channel is implemented based on an LSTM model, specifically including: Calculate the sliding mean of the catalytic memory term of the parallel timing sequence and the autocorrelation coefficient , concatenate , with the original catalytic memory component in the feature dimension to form a three-channel catalytic memory input tensor Cmi; the LSTM model simultaneously accepts these three sequences at each step and then performs normalization processing, and maps the last fully connected layer to a scalar prediction to obtain the predicted value of the catalytic memory component at time t+1 , and then obtain , calculate and The difference outputs the catalytic memory residual as the catalytic memory influence coefficient.

6. The remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion according to claim 5, characterized in that: The adaptive threshold generation unit includes extracting the adsorption-desorption critical frequency k and the environmental disturbance influence coefficient and the catalytic memory influence coefficient performing wavelet packet transform to obtain the energy of each subband, and calculating the ratio of the sum of the energies corresponding to the environmental disturbance term and the catalytic memory term to obtain the catalytic memory energy ratio ; using the adsorption-desorption critical frequency k and the catalytic memory energy ratio to obtain the fuzzy control input variables and through the tanh activation function, and obtaining the environmental disturbance threshold and the catalytic memory threshold , where is the comprehensive scaling coefficient output by fuzzy inference, represents the environmental disturbance influence coefficient within the sliding window [t - w, t], and represents the catalytic memory influence coefficient within the sliding window [t - w, t].

7. The remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion according to claim 1, characterized in that: The security threat prediction module includes a temperature deviation calculation unit and a device risk acquisition unit; the temperature deviation calculation unit obtains the temperature deviation by subtracting the digital twin predicted temperature from the measured temperature from the digital twin predicted temperature to obtain the temperature deviation ; the device risk acquisition unit aggregates the catalyst memory influence coefficient, the environmental disturbance influence coefficient, and the standardized temperature deviation using the Euclidean norm to obtain the reactor threat fusion value, and maps the reactor threat fusion value to [0,1] through the Logistic function to obtain the reactor risk index .

8. The remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion according to claim 7, characterized in that: The equipment risk acquisition unit is also equipped with a cleaning judgment strategy, and the cleaning judgment strategy is implemented by setting a cleaning risk threshold Tcl and an alarm risk threshold Tal; When the reactor risk index is less than the cleaning risk threshold Tcl, the alarm level is normal. At this time, continue to monitor without any operation; When the reactor risk index is greater than or equal to the cleaning risk threshold Tcl and less than the warning risk threshold Tal, the alarm level is a warning. According to the formula automatically adjust the flow rate and arrange for on-line ultrasonic or air-blowing cleaning, represents the initial flow rate; at the same time, run the catalyst difference data module; When the reactor risk index is greater than the warning risk threshold Tal, the alarm level is critical, catalyst oxidative regeneration or safe shutdown is initiated, a maintenance work order is triggered, and an alarm notification is sent.

9. The remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion according to claim 1, characterized in that: The catalyst difference data module includes recording the full-field temperature deviation of the digital twin before cleaning at the cleaning trigger moment , recording the new deviation after cleaning is completed and operation resumes , obtaining the deviation recovery amount , and pairing with the catalytic memory component within the corresponding time window to form an incremental learning sample set, adding the new sample to the LSTM training queue, and training using a mini-batch model with the catalytic memory component within the sliding window [t−w,t] , obtaining the catalyst-temperature influence coefficient , and updating the model weights after feature dimension splicing.

10. The remote intelligent fault diagnosis system for environmental protection equipment based on multi-source data fusion according to claim 1, characterized in that: The key positions of the plastic pyrolysis reactor include arranging a thermocouple array at equal intervals along the height direction inside and outside the reactor cylinder to measure the local temperature; installing an in-situ spectroscopic probe above the catalyst bed or at the sampling hole to detect the catalyst surface coverage.

Images