Incinerator intelligent diagnosis and maintenance method, device, equipment and storage medium

Abstract

The present invention relates to the field of intelligent diagnosis technology, and discloses an intelligent diagnosis and maintenance method, device, equipment and storage medium for an incinerator, wherein the method comprises: collecting original data sets of multiple target parts of the incinerator and preprocessing them to obtain a preprocessed data set; performing multi-scale feature extraction to obtain initial multi-dimensional features; performing local maximum average difference calculation to obtain aligned multi-dimensional features; inputting the aligned multi-dimensional features into a double Q network, learning the optimal incineration parameter combination, and obtaining a target operating parameter combination; using a two-dimensional coupled complex chaotic mapping to perform iterative prediction to obtain a dynamic behavior prediction result of the incinerator; performing remaining service life prediction and global dynamic adaptive maintenance analysis on the multiple target parts to obtain a target maintenance plan, wherein the method realizes flexible scheduling of maintenance activities, minimizes the impact of maintenance on production, and improves the overall operating efficiency of the incinerator.

Description

Technical Field

[0001] The present invention relates to the field of intelligent diagnosis technology, and in particular to an incinerator intelligent diagnosis and maintenance method, device, equipment and storage medium. Background Art

[0002] Long-term operation of incinerators faces problems such as equipment aging, performance degradation and failure risks, which not only affects the incineration efficiency, but also may cause environmental pollution and safety hazards. Traditional incinerator maintenance methods mainly rely on regular inspections and empirical diagnosis, which makes it difficult to detect potential failures in time and cannot achieve accurate predictive maintenance.

[0003] With the development of the Internet of Things and artificial intelligence technologies, intelligent diagnosis and maintenance methods have provided new ideas for the operation and management of incinerators. However, existing intelligent diagnosis systems often only focus on a single parameter or local features, making it difficult to fully capture the complex dynamic behavior of incinerators. In addition, due to the variable operating conditions of incinerators, there are significant differences in the data distribution under different operating conditions, which poses challenges to fault diagnosis and life prediction across operating conditions. On the other hand, maintenance decisions for incinerators involve multiple objectives, such as equipment reliability, maintenance costs, and production efficiency, and there are often conflicts between these objectives. How to minimize maintenance costs and maintain high production efficiency while ensuring equipment reliability is a key issue facing intelligent maintenance of incinerators. Summary of the invention

[0004] The present invention provides an incinerator intelligent diagnosis and maintenance method, device, equipment and storage medium, which are used to realize flexible scheduling of maintenance activities, minimize the impact of maintenance on production, and improve the overall operation efficiency of the incinerator.

[0005] In a first aspect, the present invention provides an incinerator intelligent diagnosis and maintenance method, the incinerator intelligent diagnosis and maintenance method comprising:

[0006] Collecting original data sets of multiple target parts of the incinerator through sensors, and preprocessing the original data sets to obtain preprocessed data sets;

[0007] Inputting the preprocessed data set into a preset residual network to perform multi-scale feature extraction to obtain initial multi-dimensional features;

[0008] Performing local maximum average difference calculation on the initial multi-dimensional features to align the feature distributions of the source domain and the target domain, thereby obtaining aligned multi-dimensional features;

[0009] Inputting the aligned multi-dimensional features into a dual Q network to learn an optimal incineration parameter combination to obtain a target operating parameter combination;

[0010] Based on the target operating parameter combination, two-dimensional coupled complex chaotic mapping is used to perform iterative prediction to obtain the dynamic behavior prediction result of the incinerator;

[0011] According to the dynamic behavior prediction results, remaining service life prediction and global dynamic adaptive maintenance analysis are performed on the multiple target parts to obtain a target maintenance plan.

[0012] In a second aspect, the present invention provides an incinerator intelligent diagnosis and maintenance device, the incinerator intelligent diagnosis and maintenance device comprising:

[0013] A collection module, used for collecting raw data sets of multiple target parts of the incinerator through sensors, and preprocessing the raw data sets to obtain preprocessed data sets;

[0014] An extraction module, used for inputting the preprocessed data set into a preset residual network to perform multi-scale feature extraction to obtain initial multi-dimensional features;

[0015] A calculation module, used to perform local maximum average difference calculation on the initial multi-dimensional features, align the feature distributions of the source domain and the target domain, and obtain aligned multi-dimensional features;

[0016] A learning module, used for inputting the aligned multi-dimensional features into a dual Q network, learning an optimal incineration parameter combination, and obtaining a target operating parameter combination;

[0017] A prediction module, used for iteratively predicting based on the target operating parameter combination using a two-dimensional coupled complex chaotic map to obtain a dynamic behavior prediction result of the incinerator;

[0018] The analysis module is used to perform remaining service life prediction and global dynamic adaptive maintenance analysis on the multiple target parts according to the dynamic behavior prediction results to obtain a target maintenance plan.

[0019] The third aspect of the present invention provides an incinerator intelligent diagnosis and maintenance device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the incinerator intelligent diagnosis and maintenance device executes the above-mentioned incinerator intelligent diagnosis and maintenance method.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned incinerator intelligent diagnosis and maintenance method.

[0021] In the technical solution provided by the present invention, the original data of multiple target parts of the incinerator are collected through a multi-source heterogeneous sensor network and preprocessed, thereby improving the comprehensiveness and quality of the data. The improved residual network is used for multi-scale feature extraction, which can effectively capture the feature information of different scales of the incinerator and improve the richness and accuracy of the feature representation. The local maximum mean difference calculation is introduced to achieve the alignment of the feature distribution of the source domain and the target domain, solve the problem of data distribution offset in cross-operating condition diagnosis, and improve the generalization ability of the model. The dual Q network is used to learn the optimal incineration parameter combination, realize the intelligent optimization of the incinerator operating parameters, and improve the incineration efficiency and environmental protection performance. The two-dimensional coupled complex chaotic mapping is used for iterative prediction, which can accurately simulate the complex dynamic behavior of the incinerator. Combined with the long short-term memory network and the multi-objective optimization algorithm, the remaining service life prediction of the key components of the incinerator and the global dynamic adaptive maintenance analysis are realized. By coordinating and optimizing the maintenance plan and the production plan, the flexible scheduling of maintenance activities is realized, the impact of maintenance on production is minimized, and the overall operation efficiency of the incinerator is improved.

[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of an embodiment of an incinerator intelligent diagnosis and maintenance method in an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of an embodiment of an incinerator intelligent diagnosis and maintenance device in an embodiment of the present invention;

[0026] Figure 3 It is a schematic diagram of an embodiment of the incinerator intelligent diagnosis and maintenance equipment in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0029] To facilitate understanding of this embodiment, firstly, a method for intelligent diagnosis and maintenance of an incinerator disclosed in an embodiment of the present invention is described in detail. Figure 1 As shown, the method comprises the following steps:

[0030] 101. Collect original data sets of multiple target parts of the incinerator through sensors, and preprocess the original data sets to obtain preprocessed data sets;

[0031] It is understandable that the execution subject of the present invention may be an intelligent diagnosis and maintenance device for an incinerator, or a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0032] Specifically, various types of sensors are installed at multiple target locations of the incinerator, including temperature sensors, pressure sensors, gas composition analyzers, flow meters, and vibration sensors. These sensors constitute a multi-source heterogeneous sensor network, which can collect various key data in real time during the operation of the incinerator, including temperature distribution in the furnace, flue gas pressure, flue gas composition, fuel supply, air supply, and equipment vibration data, etc., and form a comprehensive original data set through the collection of multi-source data. The original data set is preprocessed. In order to eliminate noise interference in the data, the original data set is denoised by wavelet transform. Wavelet transform effectively separates noise and useful information in the signal to obtain denoised multidimensional time series data. The denoised multidimensional time series data is normalized. Data of different dimensions and different ranges are mapped to the same scale, and the data is scaled to the interval of [0, 1] or [-1, 1]. Through normalization, the dimension differences between different data are eliminated, so that the data collected by different sensors can be compared and analyzed on the same scale, improving the efficiency and effect of data processing. Outlier detection is performed on the normalized multidimensional time series data to identify abnormal data points in the data set that may be caused by sensor failure, external interference or system abnormality. The outliers in the data are marked using methods based on statistical features, clustering algorithms or machine learning models to obtain outlier marking results. According to the marking results, outliers are removed from the normalized multidimensional time series data to prevent the abnormal data from misleading and affecting subsequent analysis, and the cleaned multidimensional time series data is obtained. The cleaned multidimensional time series data is segmented by time window. The data is divided according to a predetermined fixed time length, such as seconds, minutes or hours, and the entire time series data is divided into several fixed-length time segments. Fixed-length data segments capture the operating status and change trends of the incinerator in different time periods, which helps to discover regularities and abnormal patterns. The fixed-length data segments are stored in the time series database to form the final preprocessed data set. The time series database is used to store and manage time series data, can efficiently process a large number of time series data segments, and support fast query and complex analysis operations.

[0033] 102. Input the preprocessed data set into a preset residual network to extract multi-scale features to obtain initial multi-dimensional features;

[0034] Specifically, the preprocessed data set is input into the first-layer multi-scale convolution structure of the preset residual network. The multi-scale convolution structure contains three convolution kernels of different sizes, namely 1x1, 3x3 and 5x5 convolution kernels, which extract feature information within different receptive fields. The preprocessed data set is convolved by these convolution kernels to obtain three sets of feature maps of different scales, which respectively capture the information of the incinerator data at different spatial scales, including local detail features and global features. The three sets of feature maps of different scales are batch normalized to eliminate the internal covariate offset generated by the feature maps during the training process, stabilize the training process and accelerate convergence. The rectified linear unit activation function is applied to the normalized feature maps, which can introduce nonlinear features and improve the expression ability and generalization performance of the network. After batch normalization and ReLU activation function processing, three sets of normalized activated feature maps are obtained. The three sets of normalized activated feature maps are input into the feature fusion module. The self-attention mechanism is introduced to determine the weight coefficient by calculating the importance of each normalized activated feature map in the overall feature. The self-attention mechanism can capture the correlation between features of different scales, highlight key features, and suppress irrelevant or redundant information. The weighted sum operation is performed on the three sets of normalized activated feature maps according to the calculated weight coefficients to obtain the multi-scale fused feature map. The multi-scale fused feature map is transferred to the deep layer of the network through residual connection and added to the deep features at the element level. The residual connection method can effectively alleviate the problems of gradient vanishing and gradient explosion in deep networks, making it easier for deep networks to learn complex feature representations and obtain fused deep feature maps. The fused deep feature maps are processed by stacking residual blocks. Each residual block consists of two 3x3 convolutional layers, two batch normalization layers, two rectified linear unit activation function layers, and a shortcut connection. Through this structure, the residual block can perform multiple residual learning, continuously optimize the feature representation, and maintain the transmission of the original feature information. The stacking of multiple residual blocks gradually extracts more abstract and deep features to obtain the residual processed feature map. The residual processed feature map is input into the global average pooling layer. The global average pooling layer performs an average operation on the feature map of each channel in the spatial dimension, converting the feature map into a fixed-dimensional channel-level vector representation, effectively reducing the dimension of the feature map while retaining the global feature information, reducing the number of parameters, and reducing the risk of overfitting. The fixed-dimensional channel-level vector representation is input into the fully connected layer, and the deep features are extracted through linear transformation and nonlinear activation to obtain the initial multi-dimensional features.

[0035] 103. Calculate the local maximum average difference of the initial multi-dimensional features to align the feature distributions of the source domain and the target domain, and obtain the aligned multi-dimensional features;

[0036] Specifically, the initial multi-dimensional features are divided into source domain features and target domain features according to the source of the data. The source domain features come from historical data or labeled data during model training, while the target domain features come from the real-time operation data of the incinerator. The two types of features are divided into intra-class and inter-class samples, that is, according to the category to which the data belongs, the source domain and target domain features are divided into intra-class sample sets of different categories and inter-class sample sets containing the relationship between different categories. The feature vectors are calculated for the intra-class sample sets of the source domain and the target domain, respectively, to obtain the average feature vector within the source domain class and the average feature vector within the target domain class, which representatively captures the feature distribution of the source domain and the target domain within each category. Similarly, the inter-class average feature vector is calculated for the inter-class sample sets of the source domain and the target domain, respectively, to obtain the source domain inter-class average feature vector and the target domain inter-class average feature vector. The intra-class average feature vector reflects the feature concentration trend within the same category, while the inter-class average feature vector reveals the distribution differences between different categories. The Euclidean distance between the average feature vector within the source domain and the average feature vector within the target domain is calculated to obtain the intra-class inter-domain distance, which measures the distribution difference between the source domain and the target domain within the same category, reflecting the consistency of the source domain and the target domain in the same sample features. Similarly, the Euclidean distance between the average feature vector between the source domain and the target domain is calculated to obtain the inter-class inter-domain distance, which measures the distribution difference between the source domain and the target domain in different categories, reflecting the consistency of the inter-class relationship between different domains. Through the measurement of these two distances, the feature distribution difference between the source domain and the target domain at the fine-grained level is captured. The intra-class inter-domain distance and the inter-class inter-domain distance are added to obtain the local maximum average difference value. The local maximum average difference value is a measure of the overall difference in the feature distribution between the source domain and the target domain within and between classes. In order to effectively minimize this difference, a loss function is constructed based on the local maximum average difference value. The goal of the loss function is to minimize the distribution difference between the source domain and the target domain within and between classes by optimizing the network parameters to achieve feature alignment. The initial multi-dimensional features are optimized by gradient descent based on the constructed loss function. Through the back-propagation algorithm and gradient descent method, the parameters of the network are continuously adjusted so that the value of the loss function gradually decreases, and the difference in feature distribution between the source domain and the target domain is gradually reduced. The optimization process can improve the adaptability of the model to the target domain data and retain the effective information contained in the source domain data. The optimized multi-dimensional features are nonlinearly transformed. Nonlinear transformation maps features to a new feature space through nonlinear activation functions such as ReLU, Sigmoid, or Tanh, etc., captures more complex relationships and patterns in the original features, and obtains aligned multi-dimensional features.

[0037] 104. Input the aligned multi-dimensional features into the double Q network, learn the optimal incineration parameter combination, and obtain the target operation parameter combination;

[0038] Specifically, the aligned multi-dimensional features are input as states into the dual Q network. The dual Q network includes two neural networks with the same structure, namely, the online network and the target network. The aligned multi-dimensional features are respectively input into the input layers of the online network and the target network, and linear transformation is performed through the fully connected layer to map the input features to a higher-dimensional feature space so that the subsequent network layers can better learn the complex relationships in the data. Batch normalization is performed in the linear transformation to eliminate the scale differences of the input data in different dimensions and improve the stability and convergence speed of the training. The normalized data is nonlinearly transformed using the rectified linear unit activation function (ReLU) to obtain the initial state representation. The initial state representation is input into the hidden layer of the multilayer perceptron of the online network for nonlinear transformation. The hidden layer of the multilayer perceptron includes multiple layers of fully connected layers, each of which is followed by a batch normalization layer and a rectified linear unit activation function layer. Through the stacking of these layers and the action of the activation function, the potential complex nonlinear relationships in the input features are effectively captured, and a richer intermediate feature representation is generated. The intermediate feature representation of the online network is respectively input into the state value function estimator and the advantage function estimator. The state value function estimator is used to estimate the overall value of the current state. It consists of a fully connected layer and outputs a scalar state value estimate. The advantage function estimator is used to estimate the advantage of taking different actions in the current state. It consists of another fully connected layer and outputs an advantage function estimate in vector form. The state value estimate and the advantage function estimate are added, and the mean of the advantage function estimate is subtracted from the result to remove unnecessary bias to obtain the Q value estimate of the online network. The Q value estimate represents the expected total reward of taking each possible action in the current state. Based on the Q value estimate, the action with the largest Q value is selected as the candidate action. In order to improve the exploratory nature of the strategy, the candidate action is added with Gaussian noise with a mean of zero and a variance that decays exponentially with the training process to obtain the actual execution action. This strategy increases exploration in the early stage of training to avoid the model falling into a local optimum. The actual execution action is applied to the simulation environment of the incinerator, and the operating parameters of the incinerator, such as furnace temperature, fuel supply rate, and air supply, are updated. A time step is simulated to obtain the next state and immediate reward. The immediate reward is used to measure the impact of the action taken in the current state on the operating performance of the incinerator. The higher the reward value, the better the current action. The new state is input into the target network, and the Q value estimate of the target network is calculated through the neural network layer with the same structure as the online network. Based on the immediate reward and the Q value estimate of the target network, the target Q value is calculated using the Bellman equation. The target Q value represents the expected cumulative return in all future time steps after taking a certain action in the current state. The target Q value is compared with the Q value estimate of the online network, and the mean square error between them is calculated to obtain the loss function value. The parameters of the online network are optimized using the gradient descent method to minimize the loss function value.In order to avoid the impact of drastic parameter updates on model stability, the learning rate will decay exponentially as the training progresses. Every fixed number of steps, the parameters of the online network are soft-updated to the target network. The soft update strategy can gradually pass the latest learned information to the target network while maintaining the stability of the target network. After this series of training and updating processes, the optimal burning parameter combination is finally obtained, that is, the target operating parameter combination.

[0039] 105. Based on the target operating parameter combination, two-dimensional coupled complex chaos mapping is used for iterative prediction to obtain the dynamic behavior prediction results of the incinerator;

[0040] Specifically, the furnace temperature distribution and velocity field parameters in the target operating parameter combination are mapped to the complex plane. The complex physical parameters are converted into complex form for subsequent chaotic mapping processing. The furnace temperature distribution and velocity field parameters represent the thermodynamic state and fluid dynamic state inside the incinerator, respectively, which are converted into an initial state complex vector z through a mapping function, in which the real part and the imaginary part represent different physical quantities, respectively. More complex nonlinear transformations and mapping operations are performed on the complex plane in the form of complex vectors. The two-dimensional coupled complex chaotic mapping function is applied to the initial state complex vector z for an iterative calculation. The form of the chaotic mapping function is f(z)=az+bconj(z)+c*z^2+d, where a, b, c, d are preset complex parameters, and conj(z) represents the complex conjugate of the initial state complex vector z. The setting of the complex parameters determines the evolution behavior of the chaotic mapping, which can produce diverse and complex dynamic characteristics. By substituting the initial state complex vector z into the chaotic mapping function, the state complex vector at the next moment is obtained, capturing the nonlinear evolution characteristics of the internal state of the incinerator over time. The Euclidean distance between the state complex vector at the next moment and the initial state complex vector is calculated to measure the change amplitude of the state within the current iteration step and obtain the state change amount. The state change amount is compared with the preset threshold. If the state change amount is less than the threshold, it means that the system has stabilized and meets the convergence condition; if the state change amount is greater than the threshold, it means that the system is still changing significantly, and the iterative calculation continues. If the convergence condition is not met, the state complex vector at the next moment is used as the new initial state, and the iterative calculation process of the chaotic mapping function is repeated. The iterative process will continue until the convergence condition is met or the maximum number of iterations N is reached, and the target state complex vector of the system is obtained. The target state complex vector is reversely mapped from the complex plane back to the physical space, and the complex form of the target state is converted back to the physical quantity, that is, it is restored to the furnace temperature distribution and velocity field of the incinerator, which intuitively reflects the temperature change trend and airflow characteristics of the incinerator in the future time period. In order to evaluate the stability and safety of the system, the corresponding Lyapunov exponent is calculated based on the predicted furnace temperature distribution and velocity field. The Lyapunov index is an indicator used to measure the stability of a dynamic system, which describes the sensitivity of the system to the initial conditions. If the Lyapunov index is positive, it indicates that the system is highly sensitive to disturbances in the initial state, the predicted trajectory will diverge exponentially, and the system is in a chaotic state; if the Lyapunov index is negative, the trajectory of the system will converge and exhibit stable behavior. Based on information such as the predicted furnace temperature distribution, velocity field, and Lyapunov index, the dynamic behavior prediction results of the incinerator are generated, including temperature fluctuation trends and combustion stability assessments, as well as potential risk warnings.The temperature fluctuation trend can reflect the amplitude and period of the temperature change in the furnace, which helps to identify possible situations where the temperature is too high or too low; the combustion stability assessment is based on the coordinated analysis of the velocity field and temperature distribution to determine whether the incineration process is uniform and smooth; the potential risk warning is based on the Lyapunov index and the characteristics of unstable states in historical data to issue warnings for abnormal situations that may occur in the future.

[0041] 106. Based on the dynamic behavior prediction results, the remaining service life prediction and global dynamic adaptive maintenance analysis are performed on multiple target parts to obtain the target maintenance plan.

[0042] Specifically, the key feature information of each target part is extracted from the dynamic behavior prediction results, including temperature fluctuation trend, combustion stability assessment and potential risk warning. The temperature fluctuation trend can reflect the changing law and fluctuation amplitude of the temperature inside the incinerator. The combustion stability assessment is used to judge the stability and consistency of the incineration process, while the potential risk warning is based on the prediction model to give early warning of possible future failures or abnormal conditions. The feature information is integrated to construct a feature matrix describing the current state of the target part to obtain the target part state feature matrix. The target part state feature matrix is ​​fused with the historical operation data of the incinerator. The historical operation data includes information such as the working status, operating parameters and maintenance records of the equipment in the past period of time. The fusion uses the sliding window method to construct a time series sample, and arranges the historical data and the current prediction features into continuous time segments in chronological order. The sliding window method can retain the changing trend of the target part in different time periods and obtain the fused time series feature sequence. The fused time series feature sequence is input into the long short-term memory network (LSTM) for processing. LSTM is a deep learning model for processing time series data, which can effectively capture long-term dependencies and complex nonlinear characteristics in time series. By learning and predicting the time series features of each target part, LSTM can output the predicted value of the remaining service life of each target part. The predicted value of the remaining service life indicates the length of time that the equipment can continue to work normally in the current state. Based on the predicted values ​​of the remaining service life of each target part, a multi-objective optimization problem is constructed. The objective function of the optimization problem includes three factors: equipment reliability, maintenance cost, and production efficiency. Equipment reliability refers to the probability that the equipment will not fail within the predicted remaining service life. The maintenance cost includes spare parts, replacement costs, and labor costs. Production efficiency refers to the degree of impact of the maintenance plan on normal production operations. By comprehensively considering the objective functions, an initial optimization model is constructed. The initial optimization model is solved by a genetic algorithm. A genetic algorithm is a search algorithm based on the principles of natural selection and genetics that can effectively handle complex optimization problems. The genetic algorithm continuously generates new solution sets through iterative evolution of the population, and finally obtains the Pareto optimal solution set. Each solution in the Pareto optimal solution set reaches a balance state on all objectives, that is, it is impossible to optimize one objective without losing the performance of another objective. A compromise solution is selected from the Pareto optimal solution set as the preliminary maintenance plan. The choice of compromise solution takes into account the comprehensive balance of equipment reliability, maintenance cost and production efficiency, ensuring that the maintenance plan can effectively reduce the risk of failure without excessively increasing maintenance costs or causing unnecessary interruptions to production. The preliminary maintenance plan is coordinated and optimized with the production plan of the incinerator. The production plan includes the production task arrangement, production cycle and capacity requirements of the equipment. The maintenance time window of the preliminary maintenance plan is adjusted to ensure that the maintenance work can be carried out without affecting normal production.The coordination and optimization process comprehensively considers factors such as equipment downtime, spare parts preparation, and operator scheduling, and ultimately determines an optimal maintenance time window. The target maintenance plan obtained includes a specific maintenance schedule, and clearly defines the required spare parts and detailed operating steps.

[0043] In the embodiment of the present invention, the original data of multiple target parts of the incinerator are collected through a multi-source heterogeneous sensor network and preprocessed, thereby improving the comprehensiveness and quality of the data. The improved residual network is used for multi-scale feature extraction, which can effectively capture the feature information of different scales of the incinerator and improve the richness and accuracy of the feature representation. The local maximum mean difference calculation is introduced to achieve the alignment of the feature distribution of the source domain and the target domain, solve the problem of data distribution offset in cross-operating condition diagnosis, and improve the generalization ability of the model. The dual Q network is used to learn the optimal incineration parameter combination, realize the intelligent optimization of the incinerator operating parameters, and improve the incineration efficiency and environmental protection performance. The two-dimensional coupled complex chaotic mapping is used for iterative prediction, which can accurately simulate the complex dynamic behavior of the incinerator. Combined with the long short-term memory network and the multi-objective optimization algorithm, the remaining service life prediction of the key components of the incinerator and the global dynamic adaptive maintenance analysis are realized. By coordinating and optimizing the maintenance plan and the production plan, the flexible scheduling of maintenance activities is realized, the impact of maintenance on production is minimized, and the overall operation efficiency of the incinerator is improved.

[0044] In a specific embodiment, the process of executing step 101 may specifically include the following steps:

[0045] Temperature sensors, pressure sensors, gas composition analyzers, flow meters, and vibration sensors are installed on multiple target locations to obtain a multi-source heterogeneous sensor network;

[0046] Based on a multi-source heterogeneous sensor network, the temperature distribution in the incinerator, flue gas pressure, flue gas composition, fuel supply, air supply and equipment vibration data are collected during the operation of the incinerator to obtain the original data set;

[0047] Performing wavelet transform denoising on the original data set to obtain denoised multidimensional time series data, and performing normalization on the denoised multidimensional time series data to obtain normalized multidimensional time series data;

[0048] Perform outlier detection on the normalized multidimensional time series data to obtain an outlier marking result, and remove outliers from the normalized multidimensional time series data according to the outlier marking result to obtain cleaned multidimensional time series data;

[0049] The cleaned multidimensional time series data is segmented into time windows to obtain data segments of fixed length, and the fixed-length data segments are stored in a time series database to obtain a preprocessed data set.

[0050] Specifically, a scientific and reasonable sensor layout is carried out at different key positions of the incinerator, including temperature sensors, pressure sensors, gas composition analyzers, flow meters and vibration sensors. The temperature sensor is mainly used to collect the temperature distribution inside the furnace and related parts to reflect the temperature changes during the incineration process. The pressure sensor is used to monitor parameters such as flue gas pressure and fuel delivery pressure. These parameters can directly reflect the pressure fluctuations during the incineration process. The gas composition analyzer detects the concentration of components in the flue gas, such as oxygen (O 2 ), carbon dioxide (CO 2 ), carbon monoxide (CO), and nitrogen oxides (NOx) and other gas concentrations. These data can reflect the combustion efficiency and pollutant emission levels of the incinerator. Flow meters are used to monitor the supply of fuel and air, including fuel delivery flow and air supply flow, which are important factors affecting the combustion process of the incinerator. Vibration sensors are used to monitor the vibration of the incinerator and effectively capture the working status and potential abnormalities of the mechanical structure. Based on a multi-source heterogeneous sensor network, the temperature distribution in the furnace, flue gas pressure, flue gas composition, fuel supply, air supply and equipment vibration data are collected during the operation of the incinerator. These data are summarized through the data acquisition system to form a complete original data set. The original data set is a time series data matrix containing multiple types and multiple dimensions, denoted as ,in Indicates at time The moment is The data collected by the sensors contain detailed operating status of the incinerator at different time points. The original data set is denoised by wavelet transform. Wavelet transform is a time-frequency analysis tool that can effectively separate noise and useful signals in the data. Perform wavelet transform to obtain the wavelet coefficient matrix ,in Represented on the wavelet scale and location The high-frequency wavelet coefficients are soft-thresholded to eliminate the influence of noise, and finally the inverse wavelet transform is performed to reconstruct the denoised data. ,in is the denoised time series data. The denoised multidimensional time series data can more accurately reflect the actual state of the incinerator. The denoised multidimensional time series data is normalized to eliminate the dimensional differences between different sensor data, so that the data of each dimension are on the same scale, and the normalized multidimensional time series data is obtained. Outlier detection is performed on the normalized multidimensional time series data. An outlier refers to an observation value that deviates significantly from the normal state, which is caused by sensor failure, external environmental changes, or system abnormalities. Outlier detection uses statistical methods or machine learning algorithms, such as the 3σ principle based on mean and standard deviation:

[0051] ;

[0052] in, and Respectively represent The mean and standard deviation of the sensor data. If the data point With the mean The deviation exceeds 3 times the standard deviation , it is determined as an outlier and marked as 1, otherwise it is 0. Through outlier detection, the outlier marking result matrix is ​​obtained Anomaly According to the outlier marking results, the normalized multi-dimensional time series data is subjected to outlier removal. For data points marked as abnormal, the data of the previous and next time points are used for interpolation filling, or the abnormal points are directly deleted. After outlier removal, the cleaned multi-dimensional time series data is obtained. ,in is the value after cleaning. The cleaned multi-dimensional time series data is segmented into time windows, and the time series data is cut into multiple time segments according to a fixed time length. Set the time window length to , then each time segment is represented as:

[0053] ;

[0054] in, Indicates from time The starting length is By converting the entire time series data The data is divided into multiple fixed-length data segments to obtain a set of data segments. These fixed-length data segments are stored in a time series database, which can effectively manage a large amount of time series data and provide efficient data access and processing capabilities for subsequent feature extraction, pattern recognition and model training. All preprocessed data segments constitute a preprocessed data set, which contains the status information of the incinerator in different time windows.

[0055] In a specific embodiment, the process of executing step 102 may specifically include the following steps:

[0056] The preprocessed data set is input into the first-layer multi-scale convolution structure of the preset residual network. The multi-scale convolution structure contains convolution kernels of three scales: 1x1, 3x3, and 5x5. Convolution operation is performed on the preprocessed data set to obtain three sets of feature maps of different scales.

[0057] The three groups of feature maps of different scales are subjected to batch normalization and rectified linear unit activation function processing respectively to obtain three groups of normalized and activated feature maps;

[0058] The three sets of normalized activated feature maps are input into the feature fusion module, and the weight coefficient of each normalized activated feature map is calculated using the self-attention mechanism. The weighted sum is performed according to the weight coefficient to obtain the multi-scale fused feature map.

[0059] The multi-scale fused feature map is transferred to the deep layer of the network through residual connection, and the element-level addition operation is performed with the deep features to obtain the fused deep feature map;

[0060] The fused deep feature map is processed with stacked residual blocks. Each residual block contains two 3x3 convolutional layers, two batch normalization layers, two rectified linear unit activation function layers and a shortcut connection. Multiple residual learning is performed to obtain the feature map after residual processing.

[0061] The feature map after residual processing is input into the global average pooling layer, and the feature map of each channel is averaged in the spatial dimension to obtain a channel-level vector representation of a fixed dimension. The channel-level vector representation of the fixed dimension is input into the fully connected layer for linear transformation and nonlinear activation to obtain the initial multi-dimensional features.

[0062] Specifically, the preprocessed data set is a high-dimensional time series data matrix, denoted as ,in represents the number of time steps, The dimension of the feature, such as data collected by multiple sensors. The preprocessed data set is input into the first layer of the multi-scale convolution structure of the residual network to capture the multi-level features in the data at different receptive fields. The first layer contains three convolution kernels of different sizes, namely , and These convolution kernels can extract feature information of different granularities from different scales. For a given input data matrix During the convolution operation, each convolution kernel traverses the entire input data and calculates the convolution sum at each position. Assume that the size of the convolution kernel is , No. The convolution kernel parameters of channels are expressed as , the convolution operation is expressed as:

[0063] ;

[0064] in, Indicates that after Channel The convolution kernel is at position The convolution output result is: Indicates that the input data is at an offset The value of The radius of the convolution kernel size is rounded. This operation matches the convolution kernel with the local area of ​​the input data by weighted summation to obtain a feature map. For a 1x1 convolution kernel, its receptive field is extremely small, and it only performs linear combination at a local position, which is mainly used for dimensionality reduction or dimensionality increase operations; while 3x3 and 5x5 convolution kernels capture larger local spatial features and extract richer feature representations. After convolution operations with three convolution kernels of different sizes, three sets of feature maps of different scales are obtained, which are recorded as and Each set of feature maps represents the characteristic performance of the input data at different scales. These three sets of feature maps of different scales are batch normalized separately. Batch normalization is a regularization technique that eliminates the scale differences between different feature channels by normalizing the data in each small batch. The formula is:

[0065] ;

[0066] in, Indicates The feature map after channel normalization, and are the mean and variance of the small batch data, is a small constant used to prevent the denominator from being zero. The normalized feature map is activated by the rectified linear unit (ReLU), namely:

[0067]

[0068] in, Indicates The feature map of the channel after ReLU activation. The ReLU activation function can introduce nonlinear features and enhance the network's expressiveness. For the three sets of feature maps of different scales, batch normalization and ReLU activation are performed respectively to obtain three sets of normalized and activated feature maps. and . The three sets of normalized activated feature maps are input into the feature fusion module to effectively integrate features of different scales so that the subsequent network can comprehensively utilize multi-scale information. In the feature fusion module, the self-attention mechanism is used to calculate the weight coefficient of each normalized activated feature map. The self-attention mechanism determines the importance of features by calculating the similarity matrix between features. For each set of feature maps, their weight coefficients are calculated:

[0069] ;

[0070] in, Indicates The weight coefficient of the feature map, is the weight parameter of the self-attention mechanism, which is obtained through training. Use this weight coefficient to perform weighted summation on the three sets of normalized activated feature maps to obtain the multi-scale fused feature map:

[0071] ;

[0072] It represents a comprehensive feature map that integrates information of different scales and contains feature information of more levels. The multi-scale fused feature map is transmitted to the deep layer of the network through residual connection and is added to the deep features at the element level. Assume that the deep feature map is represented as , then the fusion of the two can be written as:

[0073] ;

[0074] in, It is a residual feature map that fuses multi-scale feature maps and deep feature maps. Residual connections can effectively alleviate the gradient vanishing and gradient exploding problems in deep networks, making it easier for the network to learn complex feature representations. The fused deep feature map is processed by stacking residual blocks. Each residual block contains two 3x3 convolutional layers, two batch normalization layers, two ReLU activation function layers and a shortcut connection. Stacking multiple residual blocks gradually optimizes the feature representation while maintaining the transmission of the original feature information. The structure of the residual block is represented as:

[0075] ;

[0076] in, Indicates The output feature map of the residual block, It is The combined result of the convolution, normalization and activation function operations of the residual blocks, For the The residual block parameters. The feature map after residual processing is input into the global average pooling layer. The global average pooling layer performs an average operation on the feature map of each channel in the spatial dimension and compresses the spatial information of each feature map into a scalar. Let the feature map of each channel be represented as , the global average pooling operation is expressed as:

[0077] ;

[0078] in, Indicates The scalar value after global average pooling of channels, and are the height and width of the feature map, respectively. After global average pooling, a fixed-dimensional channel-level vector representation is obtained. The fixed-dimensional channel-level vector representation is input into the fully connected layer, processed by linear transformation and nonlinear activation function, and the final initial multi-dimensional features are obtained.

[0079] In a specific embodiment, the process of executing step 103 may specifically include the following steps:

[0080] The initial multi-dimensional features are divided into source domain features and target domain features, and the source domain features and target domain features are divided into intra-class and inter-class samples, respectively, to obtain the intra-class sample sets and inter-class sample sets of the source domain and target domain;

[0081] The intra-class average feature vectors are calculated for the intra-class sample sets of the source domain and the target domain respectively, and the intra-class average feature vectors of the source domain and the intra-class average feature vectors of the target domain are obtained. The inter-class average feature vectors are calculated for the inter-class sample sets of the source domain and the target domain respectively, and the inter-class average feature vectors of the source domain and the inter-class average feature vectors of the target domain are obtained.

[0082] The Euclidean distance between the average feature vector within the source domain class and the average feature vector within the target domain class is calculated to obtain the intra-class inter-domain distance, and the Euclidean distance between the average feature vector between the source domain classes and the average feature vector between the target domain classes is calculated to obtain the inter-class inter-domain distance;

[0083] Add the intra-class inter-domain distance and the inter-class inter-domain distance to obtain the local maximum average difference value, and construct the loss function based on the local maximum average difference value;

[0084] Based on the loss function, the initial multi-dimensional features are optimized by gradient descent to obtain the optimized multi-dimensional features, and the optimized multi-dimensional features are nonlinearly transformed to obtain the aligned multi-dimensional features.

[0085] Specifically, source domain features refer to the data features used in model training, which come from historical data or labeled training data. Target domain features refer to the feature data obtained in real time in the actual application scenario of the model, which come from real-time collection or different distribution data sets. Source Domain Features and target domain features ,in Indicates The multi-dimensional feature vector of samples, represents the number of source domain samples, is the total number of samples, is the dimension of the feature. The samples of the source domain and the target domain are divided into intra-class and inter-class. The intra-class sample set represents samples of the same category, while the inter-class sample set represents a combination of samples of different categories. Assume that there are categories, for source domain features , and classify them into Divide into each category The sample set within the class is expressed as The inter-class sample set is obtained by random sampling between different categories, that is, Similarly, for the target domain features In the same way, the intra-class and inter-class samples are divided to obtain the intra-class sample set of the target domain. and between-class sample sets . Calculate the intra-class and inter-class average feature vectors of the source domain and the target domain respectively. The intra-class average feature vector represents the average feature within the same category and is calculated by the following formula:

[0086] ;

[0087] in, Indicates the source domain The within-class average eigenvector of the class, Indicates the source domain Similarly, the number of samples in the target domain The average feature vector within a class is expressed as:

[0088] ;

[0089] in, Indicates the target domain The within-class average eigenvector of the class, Indicates the target domain The number of class samples. The inter-class average feature vector represents the average features between different classes, which is obtained by calculating the mean of the inter-class sample set. For the inter-class average feature vector of the source domain, it is expressed as:

[0090] ;

[0091] in, Indicates the source domain Class and The average feature vector between classes. Similarly, Class and The inter-class average feature vector between classes is expressed as:

[0092] ;

[0093] in, Indicates the target domain Class and The inter-class average feature vector between classes. The Euclidean distance of the intra-class average feature vector between the source domain and the target domain is calculated. The intra-class inter-domain distance is expressed as:

[0094] ;

[0095] in, Indicates the source domain and the target domain The distance between domains within a class, represents the Euclidean distance. The intra-class inter-domain distance reflects the difference in feature distribution between the source domain and the target domain within the same category. Similarly, the inter-class inter-domain distance is expressed as:

[0096] ;

[0097] in, Indicates the source domain and the target domain Class and Inter-class and inter-domain distances between classes. Inter-class and inter-domain distances reflect the difference in feature distribution between source and target domains in different categories. The intra-class and inter-domain distances and inter-class and inter-domain distances of all categories are accumulated to obtain the global intra-class and inter-domain distances:

[0098] ;

[0099] Add the intra-class inter-domain distance and the inter-class inter-domain distance to get the local maximum average difference value:

[0100] ;

[0101] The local maximum average difference measures the distribution difference between the source domain and the target domain within the same category and between different categories. In order to minimize the distribution difference, a loss function is constructed based on the local maximum average difference. The loss function is expressed as:

[0102] ;

[0103] in, represents the task loss function, such as the cross entropy loss in the classification task, Is a balancing coefficient used to adjust the weight of the task loss and the local maximum average difference loss. By minimizing this loss function, the difference in feature distribution between the source domain and the target domain is minimized while optimizing the task objective. Based on this loss function, the initial multi-dimensional features are optimized by gradient descent. The gradient of the loss function relative to the network parameters is calculated using the back propagation algorithm. , and continuously update the network parameters through the gradient descent method , so that the loss function Minimize. The iterative update of the optimization process is expressed as:

[0104] ;

[0105] in, Indicates The network parameters for the iteration, is the learning rate. After multiple iterations of optimization, the optimized multi-dimensional features are obtained. In order to improve the nonlinear expression ability of the features, the optimized multi-dimensional features are nonlinearly transformed. Common nonlinear transformations include activation functions such as ReLU, Sigmoid or Tanh. Assume that the optimized features are , the nonlinear transformation is expressed as:

[0106] ;

[0107] in, Represents a nonlinear activation function, such as the ReLU function: After nonlinear transformation, the aligned multi-dimensional features are obtained. It can better capture the complex feature distribution relationship between the source domain and the target domain, thereby improving the generalization performance and adaptability of the model in practical applications.

[0108] In a specific embodiment, the process of executing step 104 may specifically include the following steps:

[0109] The aligned multi-dimensional features are used as state inputs to the input layers of the online network and the target network in the dual Q network, and are linearly transformed and batch normalized through the fully connected layer, and then activated by the rectified linear unit function to obtain the initial state representation.

[0110] The initial state representation is nonlinearly transformed through the hidden layer of the multi-layer perceptron of the online network. The hidden layer of the multi-layer perceptron includes a fully connected layer, a batch normalization layer, and a rectified linear unit activation function to obtain the intermediate feature representation of the online network;

[0111] The intermediate feature representation of the online network is input into the state value function estimator and the advantage function estimator respectively. Both estimators are composed of fully connected layers to obtain scalar state value estimation and vector advantage function estimation.

[0112] Add the scalar state value estimate and the vector advantage function estimate, and subtract the mean of the advantage function estimate to obtain the Q value estimate of the online network, and select the action corresponding to the maximum Q value based on the Q value estimate of the online network to obtain the candidate action;

[0113] The candidate action is added with Gaussian noise with zero mean and variance exponentially decaying with the training process to obtain the actual execution action, and the actual execution action is applied to the incinerator simulation environment to update the furnace temperature, fuel supply rate, and air supply parameters, simulate a time step, and obtain the next state and immediate reward;

[0114] The next state is input into the target network, and the Q value estimation of the target network is obtained through the neural network layer with the same structure as the online network. The target Q value is calculated using the Bellman equation based on the immediate reward and the Q value estimation of the target network, and the mean square error is calculated with the Q value estimation of the online network to obtain the loss function value;

[0115] The online network parameters are optimized by gradient descent, and the learning rate decays exponentially with the training process to minimize the loss function value. The online network parameters are soft-updated to the target network every fixed number of steps to obtain the target operating parameter combination.

[0116] Specifically, the aligned multi-dimensional features are input into the input layer of the online network and the target network of the dual Q network. The dual Q network is composed of two independent neural networks: the online network is used to generate action strategies and the target network is used to help update and optimize the strategies of the online network. The multi-dimensional features input into the network are recorded as ,in is the dimension of the feature. After entering the network, the input features are linearly transformed through the fully connected layer. The fully connected layer linearly combines the input features and maps them to a higher-dimensional feature space. The formula is as follows:

[0117] ;

[0118] in, represents the initial state after linear transformation, is the weight matrix of the fully connected layer, is the bias vector, Indicates the size of the hidden layer. In order to prevent the different scales of features in different dimensions from affecting the training process, batch normalization is performed to normalize the features. The formula is as follows:

[0119] ;

[0120] in, is the normalized initial state representation, and are the mean and variance of the current batch features, is a very small constant used to prevent the denominator from being zero. After normalization, it is processed by the rectified linear unit activation function (ReLU) to introduce nonlinear features and make the network more expressive. The formula of ReLU is:

[0121] ;

[0122] Get the initial state representation The initial state represents a nonlinear transformation through the hidden layer of the multilayer perceptron of the online network. The hidden layer of the multilayer perceptron contains multiple fully connected layers, batch normalization layers, and ReLU activation functions. Assuming that two hidden layers are used, then The calculation of the hidden layer is expressed as:

[0123] ;

[0124] ;

[0125] ;

[0126] After the hidden layer transformation of the multi-layer perceptron, the intermediate feature representation of the online network is obtained ,in Represents the index of the last hidden layer. The intermediate feature representation of the online network is input into the state value function estimator and the advantage function estimator respectively. The state value function estimator is used to estimate the overall value of the current state, and the advantage function estimator is used to estimate the relative advantages of taking different actions in this state. The state value function is estimated through a fully connected layer, and the formula is:

[0127] ;

[0128] in, is the state value function, is the weight matrix of the state value function estimator, is the bias. The advantage function is estimated through a fully connected layer and is expressed as:

[0129] ;

[0130] in, is the advantage function, and are the weight and bias of the advantage function estimator. and advantage function Perform the addition operation and subtract the mean of the advantage function to get the final Q value estimate:

[0131] ;

[0132] in, is the Q-value estimate, indicating that in state Next select action The expected return that can be obtained. After calculating the Q value estimate of each action, the action with the largest Q value is selected as the candidate action:

[0133] ;

[0134] In order to maintain the exploratory nature, the candidate actions Gaussian noise with zero mean and variance that gradually decays as training progresses Add them together to get the actual execution action:

[0135] ;

[0136] As the training process progresses, the variance of the noise will gradually decrease, allowing the network to make more use of its learned strategy. Applied in the simulation environment of the incinerator, update the operating parameters of the incinerator (such as furnace temperature, fuel supply rate and air supply), and simulate a time step . Through the simulation environment, get the next state and instant rewards . Instant Rewards It is a measure of the impact of the current action on the efficiency and stability of the incinerator. Enter the target network. The target network has the same structure as the online network and is used to calculate the target Q value. The calculation of the target Q value is based on the Bellman equation:

[0137] ;

[0138] in, Is the discount factor, which indicates the discount rate of future returns. Compare the target Q value with the Q value of the online network and calculate the loss function:

[0139] L ;

[0140] in, is the parameter of the online network. Update the parameters of the online network by gradient descent:

[0141] ;

[0142] in, is the learning rate, which indicates the step size of parameter update. In order to maintain the consistency between the target network and the online network, the parameters of the online network are soft-updated to the target network every fixed number of steps:

[0143] ;

[0144] in, is a soft update coefficient, which usually takes a small value (such as 0.01) to smoothly update the parameters of the target network. Through the above process, the dual Q network can learn the optimal operating parameter combination of the incinerator, such as furnace temperature, fuel supply rate and air supply. This strategy can not only optimize the operating efficiency of the incinerator, but also ensure the stability and safety of the combustion process. For example, in actual applications, the dual Q network continuously adjusts the ratio of fuel supply and air supply to ensure that the incinerator can maintain the optimal combustion state under different loads.

[0145] In a specific embodiment, the process of executing step 105 may specifically include the following steps:

[0146] S1: Map the furnace temperature distribution and velocity field parameters in the target operating parameter combination to the complex plane to obtain the initial state complex vector z;

[0147] S2: Apply the two-dimensional coupled complex chaos mapping function f(z)=a*z+b*conj(z)+c*z to the initial state complex vector z 2 +d performs an iterative calculation, where a, b, c, d are preset complex parameters, conj(z) is the complex conjugate of the initial state complex vector z, and the state complex vector at the next moment is obtained;

[0148] S3: Calculate the Euclidean distance between the next state complex vector and the initial state complex vector to obtain the state change, and compare the state change with a preset threshold to determine whether the convergence condition is met;

[0149] S4: If the convergence condition is not met, the next state complex vector is used as the new initial state, and steps S2 to S4 are repeated until the convergence condition is met or the maximum number of iterations N is reached, and the target state complex vector is obtained;

[0150] S5: reverse mapping the target state complex vector, mapping the state on the complex plane back to the physical space, and obtaining the predicted furnace temperature distribution and velocity field;

[0151] S6: Calculate the Lyapunov exponent corresponding to the predicted furnace temperature distribution and velocity field to obtain the system stability index, and generate the dynamic behavior prediction results of the incinerator based on the predicted furnace temperature distribution, velocity field and system stability index. The dynamic behavior prediction results include temperature fluctuation trend, combustion stability assessment and potential risk warning.

[0152] Specifically, the physical parameters are converted into complex vector form. The target operating parameter combination of the incinerator includes the furnace temperature distribution and the velocity field parameters, which are recorded as and ,in Indicates The temperature of the measuring point, Indicates To map these parameters to the complex plane, temperature and velocity are combined into a complex vector , where each complex number represents the imaginary unit. The complex vector of the initial state is obtained Apply a two-dimensional coupled complex chaos mapping function to the initial state complex vector z for an iterative calculation. The chaos mapping function is expressed as:

[0153] ;

[0154] in, , , , is a preset complex parameter that determines the dynamic characteristics of the chaotic mapping; Indicates plural The complex conjugate of , which is used to introduce nonlinear coupling effects. For the initial state complex vector , each complex component After mapping function The transformation of the state complex vector at the next moment is obtained ,in The calculation method is:

[0155] ;

[0156] The iterative process reflects the dynamic evolution of the furnace temperature and velocity field on the complex plane. , , , The complex property of the mapping shows complex chaotic behavior, which can capture the nonlinear dynamic characteristics of the temperature and airflow changes in the furnace. Calculate the complex vector of the state at the next moment With the initial state complex vector The Euclidean distance between is used to measure the change in state. The calculation formula of Euclidean distance is as follows:

[0157] ;

[0158] in, Indicates The modulus of the difference of the complex components. The calculated state change Indicates the magnitude of the overall displacement of the complex vector on the complex plane. The state change is compared with the preset threshold Compare and determine whether the convergence condition is met. , the system is considered to have stabilized or entered a periodic orbit and meets the convergence condition; otherwise, if the state change is greater than the threshold, it is necessary to continue iterating. If the convergence condition is not met, the next state complex vector As the new initial state complex vector, the iterative process of the chaotic mapping function is repeated, that is, returning to step S2. The iterative process will continue until the state change meets the convergence condition or reaches the preset maximum number of iterations. This process captures the long-term behavioral characteristics of the system and obtains the target state complex vector . The target state complex vector Perform the reverse mapping and convert back to physical space. Each complex component Contains the predicted furnace temperature distribution and velocity field information, namely:

[0159] ;

[0160] in, Indicates plural The real part of corresponds to the predicted furnace temperature distribution , Indicates plural The imaginary part of . Through the reverse mapping process, the predicted values ​​of the furnace temperature distribution and velocity field in the physical space corresponding to the target state complex vector on the complex plane are obtained. Based on the predicted furnace temperature distribution and velocity field, the corresponding Lyapunov index is calculated. The Lyapunov index is an indicator used to measure the sensitivity of a dynamic system to initial conditions, which can reflect the chaos and stability of the system. For the predicted state, the calculation formula of the Lyapunov index is:

[0161] ;

[0162] in, represents the Lyapunov exponent, is the number of iterations, Indicates that the mapping function is at each iteration point The derivative at . A positive Lyapunov exponent indicates that the system is very sensitive to the initial conditions, the predicted trajectory will diverge exponentially, and the system is in a chaotic state; a negative Lyapunov exponent indicates that the system tends to be stable. By analyzing the Lyapunov exponent, the stability index of the system is obtained, thereby judging the operating state of the incinerator. Combining the above steps, the dynamic behavior prediction results of the incinerator are generated according to the predicted furnace temperature distribution, velocity field and system stability index. The dynamic behavior prediction results include temperature fluctuation trend, combustion stability assessment and potential risk warning. The temperature fluctuation trend reflects the characteristics of the change of temperature in the furnace over time, which helps to judge the stability of the incineration process; the combustion stability assessment combines the velocity field information to assess whether the combustion in the furnace is uniform and whether there is a flow deviation phenomenon; the potential risk warning is based on the Lyapunov exponent and historical data to provide early warning of possible combustion abnormalities, excessively high or low temperatures in the future.

[0163] In a specific embodiment, the process of executing step 106 may specifically include the following steps:

[0164] Extract the temperature fluctuation trend, combustion stability assessment and potential risk warning information of each target part from the dynamic behavior prediction results to obtain the state feature matrix of the target part;

[0165] The target part state feature matrix is ​​fused with the historical operation data, and the time series samples are constructed using the sliding window method to obtain the fused time series feature sequence;

[0166] The fused time series feature sequence is input into the long short-term memory network to predict the remaining service life of each target part, and the predicted value of the remaining service life of each target part is obtained;

[0167] Based on the predicted value of the remaining service life of each target part, a multi-objective optimization problem is constructed. The objective function includes equipment reliability, maintenance cost and production efficiency, and the initial optimization model is obtained.

[0168] The initial optimization model is solved by genetic algorithm to obtain the Pareto optimal solution set, and a compromise solution is selected from the Pareto optimal solution set to obtain a preliminary maintenance plan;

[0169] The preliminary maintenance plan is coordinated and optimized with the production plan of the incinerator, and the maintenance time window is adjusted to obtain the target maintenance plan, which includes maintenance time, required spare parts and operating steps.

[0170] Specifically, the operating status of each target part of the incinerator is analyzed. The dynamic behavior prediction results are based on the simulation and prediction of the complex dynamic behavior inside the incinerator based on the previous model. These results include the temperature fluctuation trend of different parts in the furnace, the stability of the combustion process, and potential risk factors. For each target part, the following three types of key feature information are extracted: temperature fluctuation trend , combustion stability evaluation , and potential hazard warning These feature information are represented as the state feature vector of the target part ,in Indicates The temperature fluctuation characteristics of the target parts Indicates the combustion stability evaluation results, It represents the potential risk warning information of the part. The state feature vectors of all target parts are combined together to obtain a target part state feature matrix ,in is the number of target parts. The state feature matrix is ​​integrated with the historical operation data. The historical operation data contains the operation status, operation parameters, and maintenance records of the incinerator in different time periods. With historical operation data The sliding window method is used to construct time series samples. The sliding window method is to slice the time series data into fixed time lengths to form multiple time window samples. Set the time window length to , the sliding step length is , then A time series sample is represented as:

[0171] ;

[0172] in, Indicates from time The initial fused time series feature sequence, and Respectively represent the state characteristics of the target part and the historical operation data in time In this way, a set of fused time series feature sequences is obtained , contains the state feature information of the current target part, and integrates the dynamic change rules of historical data. The fused time series feature sequence is input into the long short-term memory network (LSTM) to predict the remaining service life of each target part. LSTM is a recursive neural network that can handle long-term dependencies and is suitable for modeling and predicting time series data. Set the input of LSTM to The output is the predicted value of the remaining service life of each target part ,in Indicates The RUL prediction value of the target part. The calculation of LSTM is expressed as:

[0173] ;

[0174] ;

[0175] ;

[0176] ;

[0177] in, Indicates the hidden state at the current moment, is the state of the memory cell, and are the activation values ​​of the forget gate, input gate, and output gate, respectively. is the weight matrix of LSTM, , is the bias term, Represents the Sigmoid activation function, and © represents the element-wise product operation. Through the multi-layer recursive processing of the LSTM network, the complex dynamic relationship in the time series features is captured, so as to accurately predict the remaining service life of each target part. A multi-objective optimization problem is constructed based on the predicted value of the remaining service life of each target part. The objective function of the optimization problem includes equipment reliability, maintenance cost, and production efficiency. Equipment reliability is expressed as the weighted sum of the remaining service life of all parts:

[0178] ;

[0179] in, Indicates the overall reliability of the equipment. For the The maintenance cost is expressed as the total cost of maintenance activities, including replacement of spare parts, labor costs and downtime losses, etc., expressed as:

[0180] ;

[0181] in, represents the total maintenance cost, For the The maintenance cost weight of each part, is the actual cost of each maintenance. Production efficiency is related to the downtime of the equipment and is expressed as:

[0182] ;

[0183] in, Indicates production efficiency, is the total production time, is the downtime. The multi-objective optimization problem is expressed as:

[0184] ;

[0185] In the multi-objective optimization model, the goal is to minimize maintenance costs while maximizing equipment reliability and production efficiency. In order to solve the multi-objective optimization problem, a genetic algorithm is used. The genetic algorithm searches for the optimal solution in the solution space by simulating the process of natural selection and biological evolution. The individual representation of the genetic algorithm is the encoding of the maintenance strategy, and the fitness function is calculated based on the weighted sum of maintenance cost, equipment reliability and production efficiency. The basic steps of the genetic algorithm include selection, crossover and mutation. Through multiple generations of evolution, the Pareto optimal solution set is obtained. A compromise solution is selected from the Pareto optimal solution set as the preliminary maintenance plan. The choice of the compromise solution is based on the decision maker's preference weights for different objectives to find an optimal balance between maintenance cost, equipment reliability and production efficiency. The preliminary maintenance plan is coordinated and optimized with the production plan of the incinerator. The production plan usually includes constraints such as production cycle, equipment utilization, and order demand. The preliminary maintenance plan is comprehensively considered with the production plan, and the maintenance time window is adjusted so that the maintenance activities can not only ensure the normal progress of the production tasks, but also reduce the maintenance cost and improve the operation reliability of the equipment. The target maintenance plan includes a specific maintenance schedule, required spare parts, and detailed operation steps. The maintenance schedule refers to when the equipment will be shut down for inspection and maintenance, the required spare parts are a list of parts that may need to be replaced for each target part, and the operation steps are the specific process of actually performing the maintenance work.

[0186] The above describes the intelligent diagnosis and maintenance method of the incinerator in the embodiment of the present invention. The following describes the intelligent diagnosis and maintenance device of the incinerator in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an intelligent diagnosis and maintenance device for an incinerator includes:

[0187] The acquisition module 201 is used to acquire original data sets of multiple target parts of the incinerator through sensors, and preprocess the original data sets to obtain preprocessed data sets;

[0188] An extraction module 202 is used to input the preprocessed data set into a preset residual network to extract multi-scale features and obtain initial multi-dimensional features;

[0189] A calculation module 203 is used to perform local maximum average difference calculation on the initial multi-dimensional features, so as to align the feature distributions of the source domain and the target domain, and obtain the aligned multi-dimensional features;

[0190] A learning module 204 is used to input the aligned multi-dimensional features into a dual Q network, learn an optimal incineration parameter combination, and obtain a target operating parameter combination;

[0191] The prediction module 205 is used to perform iterative prediction based on the target operation parameter combination by using a two-dimensional coupled complex chaotic mapping to obtain a dynamic behavior prediction result of the incinerator;

[0192] The analysis module 206 is used to perform remaining service life prediction and global dynamic adaptive maintenance analysis on multiple target parts according to the dynamic behavior prediction results to obtain a target maintenance plan.

[0193] Through the collaboration of the above components, the original data of multiple target parts of the incinerator are collected through a multi-source heterogeneous sensor network and preprocessed, which improves the comprehensiveness and quality of the data. The improved residual network is used for multi-scale feature extraction, which can effectively capture the feature information of different scales of the incinerator and improve the richness and accuracy of feature representation. The local maximum mean difference calculation is introduced to achieve the alignment of feature distributions of the source domain and the target domain, solve the problem of data distribution offset in cross-condition diagnosis, and improve the generalization ability of the model. The dual Q network is used to learn the optimal incineration parameter combination, realize the intelligent optimization of the incinerator operating parameters, and improve the incineration efficiency and environmental performance. The two-dimensional coupled complex chaotic mapping is used for iterative prediction, which can accurately simulate the complex dynamic behavior of the incinerator. Combined with the long short-term memory network and the multi-objective optimization algorithm, the remaining service life prediction and global dynamic adaptive maintenance analysis of the key components of the incinerator are realized. By coordinating and optimizing the maintenance plan and the production plan, the flexible scheduling of maintenance activities is realized, which minimizes the impact of maintenance on production and improves the overall operation efficiency of the incinerator.

[0194] above Figure 2 The intelligent diagnosis and maintenance device for the incinerator in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The intelligent diagnosis and maintenance device for the incinerator in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0195] Figure 3It is a structural schematic diagram of an incinerator intelligent diagnosis and maintenance device provided by an embodiment of the present invention. The incinerator intelligent diagnosis and maintenance device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the incinerator intelligent diagnosis and maintenance device 300. Furthermore, the processor 310 can be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the incinerator intelligent diagnosis and maintenance device 300 to implement the steps of the above-mentioned incinerator intelligent diagnosis and maintenance method.

[0196] The incinerator intelligent diagnosis and maintenance device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It can be understood by those skilled in the art that Figure 3 The structure of the incinerator intelligent diagnosis and maintenance equipment shown does not constitute a limitation on the incinerator intelligent diagnosis and maintenance equipment provided by the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0197] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the incinerator intelligent diagnosis and maintenance method.

[0198] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0199] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0200] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An incinerator intelligent diagnosis and maintenance method, characterized in that: The method comprises: Collecting original data sets of multiple target parts of the incinerator through sensors, and preprocessing the original data sets to obtain preprocessed data sets; Inputting the preprocessed data set into a preset residual network to perform multi-scale feature extraction to obtain initial multi-dimensional features; Performing local maximum average difference calculation on the initial multi-dimensional features to align the feature distributions of the source domain and the target domain, thereby obtaining aligned multi-dimensional features; Inputting the aligned multi-dimensional features into a dual Q network to learn an optimal incineration parameter combination to obtain a target operating parameter combination; Based on the target operating parameter combination, a two-dimensional coupled complex chaotic mapping is used for iterative prediction to obtain the dynamic behavior prediction result of the incinerator; specifically comprising: S1: mapping the furnace temperature distribution and velocity field parameters in the target operating parameter combination to a complex plane to obtain an initial state complex vector z; S2: applying a two-dimensional coupled complex chaotic mapping function f(z)=a*z+b*conj(z)+c*z to the initial state complex vector z 2 +d for one iteration, where a, b, c, d are preset complex parameters, conj(z) is the complex conjugate of the initial state complex vector z, and the next state complex vector is obtained; S3: the Euclidean distance between the next state complex vector and the initial state complex vector is calculated to obtain the state change amount, and the state change amount is compared with a preset threshold to determine whether the convergence condition is met; S4: if the convergence condition is not met, the next state complex vector is used as the new initial state, and steps S2 to S4 are repeated until the convergence condition is met. or reaches the maximum number of iterations N, and obtains the target state complex vector; S5: reverse mapping the target state complex vector, mapping the state on the complex plane back to the physical space, and obtaining the predicted furnace temperature distribution and velocity field; S6: calculating the Lyapunov exponents corresponding to the predicted furnace temperature distribution and velocity field, obtaining the system stability index, and generating the dynamic behavior prediction results of the incinerator according to the predicted furnace temperature distribution, velocity field and system stability index, the dynamic behavior prediction results including temperature fluctuation trend, combustion stability assessment and potential risk warning; According to the dynamic behavior prediction results, the remaining service life prediction and global dynamic adaptive maintenance analysis are performed on the multiple target parts to obtain a target maintenance plan; specifically, the method comprises: extracting the temperature fluctuation trend, combustion stability evaluation and potential risk warning information of each target part from the dynamic behavior prediction results to obtain a target part state feature matrix; fusing the target part state feature matrix with historical operation data, constructing a time series sample using a sliding window method, and obtaining a fused time series feature sequence; inputting the fused time series feature sequence into a long short-term memory network, predicting the remaining service life of each target part, and obtaining a remaining service life prediction value of each target part; constructing a multi-objective optimization problem based on the remaining service life prediction value of each target part, wherein the objective function includes equipment reliability, maintenance cost and production efficiency, and obtaining an initial optimization model; solving the initial optimization model by a genetic algorithm to obtain a Pareto optimal solution set, and selecting a compromise solution from the Pareto optimal solution set to obtain a preliminary maintenance plan; coordinating and optimizing the preliminary maintenance plan with the production plan of the incinerator, and adjusting the maintenance time window to obtain a target maintenance plan, wherein the target maintenance plan includes maintenance time, required spare parts and operation steps.

2. The incinerator intelligent diagnosis and maintenance method according to claim 1 is characterized in that: The original data sets of multiple target parts of the incinerator are collected by sensors, and the original data sets are preprocessed to obtain preprocessed data sets, including: Installing temperature sensors, pressure sensors, gas composition analyzers, flow meters, and vibration sensors on the multiple target locations to obtain a multi-source heterogeneous sensor network; Based on the multi-source heterogeneous sensor network, the temperature distribution in the incinerator, flue gas pressure, flue gas composition, fuel supply, air supply and equipment vibration data are collected during the operation of the incinerator to obtain an original data set; Performing wavelet transform denoising processing on the original data set to obtain denoised multidimensional time series data, and performing normalization processing on the denoised multidimensional time series data to obtain normalized multidimensional time series data; Performing outlier detection on the normalized multidimensional time series data to obtain an outlier marking result, and removing outliers from the normalized multidimensional time series data according to the outlier marking result to obtain cleaned multidimensional time series data; The cleaned multidimensional time series data is segmented into time windows to obtain data segments of fixed length, and the data segments of fixed length are stored in a time series database to obtain a preprocessed data set.

3. The incinerator intelligent diagnosis and maintenance method according to claim 1 is characterized in that: The step of inputting the preprocessed data set into a preset residual network to extract multi-scale features to obtain initial multi-dimensional features includes: Inputting the preprocessed data set into the first-layer multi-scale convolution structure of a preset residual network, wherein the multi-scale convolution structure includes convolution kernels of three scales: 1x1, 3x3, and 5x5, performing convolution operation on the preprocessed data set to obtain three sets of feature maps of different scales; The three groups of feature maps of different scales are respectively subjected to batch normalization processing and rectified linear unit activation function processing to obtain three groups of normalized and activated feature maps; The three groups of normalized activated feature maps are input into the feature fusion module, the weight coefficient of each normalized activated feature map is calculated by the self-attention mechanism, and weighted summation is performed according to the weight coefficient to obtain a multi-scale fused feature map; The multi-scale fused feature map is transmitted to the deep layer of the network through a residual connection, and an element-level addition operation is performed with the deep features to obtain a fused deep feature map; Performing stacked residual block processing on the fused deep feature map, each residual block includes two 3x3 convolutional layers, two batch normalization layers, two rectified linear unit activation function layers and a shortcut connection, performing multiple residual learning, and obtaining a feature map after residual processing; The feature map after residual processing is input into the global average pooling layer, and the feature map of each channel is averaged in the spatial dimension to obtain a channel-level vector representation of a fixed dimension, and the channel-level vector representation of the fixed dimension is input into the fully connected layer for linear transformation and nonlinear activation to obtain the initial multi-dimensional features.

4. The incinerator intelligent diagnosis and maintenance method according to claim 1 is characterized in that: The performing local maximum average difference calculation on the initial multi-dimensional features to align the feature distributions of the source domain and the target domain to obtain the aligned multi-dimensional features includes: The initial multi-dimensional features are divided into source domain features and target domain features, and the source domain features and the target domain features are divided into intra-class and inter-class samples, respectively, to obtain intra-class sample sets and inter-class sample sets of the source domain and the target domain; For the intra-class sample sets of the source domain and the target domain, respectively, the intra-class average feature vectors are calculated to obtain the intra-class average feature vectors of the source domain and the intra-class average feature vectors of the target domain, and for the inter-class sample sets of the source domain and the target domain, respectively, the inter-class average feature vectors are calculated to obtain the inter-class average feature vectors of the source domain and the inter-class average feature vectors of the target domain; Calculating the Euclidean distance between the average feature vector within the source domain class and the average feature vector within the target domain class to obtain the intra-class inter-domain distance, and calculating the Euclidean distance between the average feature vector between the source domain classes and the average feature vector between the target domain classes to obtain the inter-class inter-domain distance; Adding the intra-class inter-domain distance and the inter-class inter-domain distance to obtain a local maximum average difference value, and constructing a loss function based on the local maximum average difference value; Based on the loss function, the initial multi-dimensional features are optimized by gradient descent to obtain optimized multi-dimensional features, and the optimized multi-dimensional features are nonlinearly transformed to obtain aligned multi-dimensional features.

5. The incinerator intelligent diagnosis and maintenance method according to claim 1 is characterized in that: The aligned multi-dimensional features are input into the dual Q network to learn the optimal incineration parameter combination to obtain the target operation parameter combination, including: Inputting the aligned multi-dimensional features as states into the input layers of the online network and the target network in the dual Q network, performing linear transformation and batch normalization processing through a fully connected layer, and passing through a rectified linear unit activation function to obtain an initial state representation; The initial state representation is subjected to nonlinear transformation through a hidden layer of a multilayer perceptron of an online network, wherein the hidden layer of the multilayer perceptron comprises a fully connected layer, a batch normalization layer, and a rectified linear unit activation function, to obtain an intermediate feature representation of the online network; Inputting the intermediate feature representation of the online network into a state value function estimator and an advantage function estimator respectively, both estimators are composed of fully connected layers, to obtain a scalar state value estimation and a vector advantage function estimation; Performing an addition operation on the scalar state value estimate and the vector advantage function estimate, and subtracting a mean value of the advantage function estimate to obtain a Q value estimate of the online network, and selecting an action corresponding to a maximum Q value based on the Q value estimate of the online network to obtain a candidate action; Add the candidate action to Gaussian noise with zero mean and variance exponentially decaying with the training process to obtain the actual execution action, and apply the actual execution action to the incinerator simulation environment, update the furnace temperature, fuel supply rate, and air supply parameters, simulate a time step, and obtain the next state and immediate reward; The next state is input into the target network, and the Q value estimation of the target network is obtained through the neural network layer with the same structure as the online network. The target Q value is calculated using the Bellman equation based on the instant reward and the Q value estimation of the target network, and the mean square error is calculated with the Q value estimation of the online network to obtain the loss function value; The online network parameters are optimized by gradient descent, and the learning rate decays exponentially with the training process to minimize the loss function value. The online network parameters are soft-updated to the target network every fixed number of steps to obtain the target operating parameter combination.

6. An intelligent diagnosis and maintenance device for an incinerator, characterized in that: The device is used to perform the incinerator intelligent diagnosis and maintenance method according to any one of claims 1 to 5, and comprises: A collection module, used for collecting raw data sets of multiple target parts of the incinerator through sensors, and preprocessing the raw data sets to obtain preprocessed data sets; An extraction module, used for inputting the preprocessed data set into a preset residual network to perform multi-scale feature extraction to obtain initial multi-dimensional features; A calculation module, used to perform local maximum average difference calculation on the initial multi-dimensional features, align the feature distributions of the source domain and the target domain, and obtain aligned multi-dimensional features; A learning module, used for inputting the aligned multi-dimensional features into a dual Q network, learning an optimal incineration parameter combination, and obtaining a target operating parameter combination; The prediction module is used to perform iterative prediction based on the target operating parameter combination using a two-dimensional coupled complex chaotic mapping to obtain a dynamic behavior prediction result of the incinerator; specifically comprising: S1: mapping the furnace temperature distribution and velocity field parameters in the target operating parameter combination to a complex plane to obtain an initial state complex vector z; S2: applying a two-dimensional coupled complex chaotic mapping function f(z)=a*z+b*conj(z)+c*z to the initial state complex vector z 2 +d for one iteration, where a, b, c, d are preset complex parameters, conj(z) is the complex conjugate of the initial state complex vector z, and the next state complex vector is obtained; S3: the Euclidean distance between the next state complex vector and the initial state complex vector is calculated to obtain the state change amount, and the state change amount is compared with a preset threshold to determine whether the convergence condition is met; S4: if the convergence condition is not met, the next state complex vector is used as the new initial state, and steps S2 to S4 are repeated until the convergence condition is met. or reaches the maximum number of iterations N, and obtains the target state complex vector; S5: reverse mapping the target state complex vector, mapping the state on the complex plane back to the physical space, and obtaining the predicted furnace temperature distribution and velocity field; S6: calculating the Lyapunov exponents corresponding to the predicted furnace temperature distribution and velocity field, obtaining the system stability index, and generating the dynamic behavior prediction results of the incinerator according to the predicted furnace temperature distribution, velocity field and system stability index, the dynamic behavior prediction results including temperature fluctuation trend, combustion stability assessment and potential risk warning; The analysis module is used to perform remaining service life prediction and global dynamic adaptive maintenance analysis on the multiple target parts according to the dynamic behavior prediction results, and obtain a target maintenance plan, which specifically includes: extracting the temperature fluctuation trend, combustion stability evaluation and potential risk warning information of each target part from the dynamic behavior prediction results to obtain a target part state feature matrix; fusing the target part state feature matrix with historical operation data, using the sliding window method to construct a time series sample, and obtaining a fused time series feature sequence; inputting the fused time series feature sequence into a long short-term memory network, predicting the remaining service life of each target part, and obtaining a remaining service life prediction value of each target part; constructing a multi-objective optimization problem based on the remaining service life prediction value of each target part, wherein the objective function includes equipment reliability, maintenance cost and production efficiency, and obtaining an initial optimization model; solving the initial optimization model through a genetic algorithm to obtain a Pareto optimal solution set, and selecting a compromise solution from the Pareto optimal solution set to obtain a preliminary maintenance plan; coordinating and optimizing the preliminary maintenance plan with the production plan of the incinerator, and adjusting the maintenance time window to obtain a target maintenance plan, wherein the target maintenance plan includes maintenance time, required spare parts and operation steps.

7. An intelligent diagnosis and maintenance device for an incinerator, characterized in that: The incinerator intelligent diagnosis and maintenance device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory so that the incinerator intelligent diagnosis and maintenance device executes the incinerator intelligent diagnosis and maintenance method as described in any one of claims 1-5.

8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the incinerator intelligent diagnosis and maintenance method according to any one of claims 1 to 5 is implemented.

Images