Copper recovery treatment system and method for flameless combustion of copper-containing sludge

Abstract

The invention discloses a copper recovery processing system and method for flameless combustion of copper-containing sludge, and relates to the technical field of intelligent control. A distributed sensor network is deployed, multi-source combustion data in a combustion furnace are acquired in real time, a hybrid convolution-long and short-term memory network model is constructed, cross-modal feature fusion is performed on the multi-source combustion data, and the copper recovery processing system and method for flameless combustion of the copper-containing sludge are obtained. The method comprises the following steps: extracting a multi-dimensional combustion feature vector, correcting system parameters of a differential-algebraic equation on line based on the multi-dimensional combustion feature vector, solving a real-time system state variable after the system parameters are corrected, and constructing a constraint condition by using the system state variable. A multi-agent reinforcement learning model is constructed based on the multi-source combustion data and the multi-dimensional combustion feature vectors, dynamic combustion parameters are optimized in parallel by adopting a multi-agent depth deterministic strategy gradient algorithm, and the dynamic combustion parameters are sent to a combustion execution mechanism for combustion parameter updating; the stability and the environmental protection performance in the combustion process are obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control, and specifically to a copper recovery treatment system and method for flameless combustion of copper-containing sludge. Background Art

[0002] As one of the industrial wastes, copper-containing sludge contains a certain amount of copper resources and also contains harmful components. During the flameless combustion process, through high-temperature treatment, not only can the sludge be stably disposed, but also the copper element can be promoted to be converted into an easily recyclable state, realizing the coordinated treatment of tail gas purification and copper recovery. The combustion furnace undertakes the important task of efficiently burning copper-containing sludge in a flameless environment during this process. However, the combustion process involves complex heat conduction, chemical reactions, and fluid dynamics behaviors, and its internal state distribution and local reaction conditions directly affect the combustion efficiency, pollutant emissions, and copper recovery effect.

[0003] Traditional combustion control systems mainly adopt traditional PID control or models based on preset fixed parameters. Such fixed combustion parameters cannot cope with dynamic changes such as changes in the composition of copper-containing sludge, equipment wear, and environmental disturbances, resulting in unstable combustion efficiency and difficult-to-meet environmental protection standards for tail gas emissions.

[0004] Therefore, there is an urgent need for intelligent dynamic combustion parameter control. By obtaining multi-source data in real time and online solving state parameters that are difficult to directly measure, the precise regulation of the combustion process can be achieved.

[0005] For this reason, the present invention proposes a copper recovery treatment system and method for flameless combustion of copper-containing sludge. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art. For this reason, the present invention proposes a copper recovery treatment system and method for flameless combustion of copper-containing sludge, which significantly improves the stability and environmental protection performance of the combustion process.

[0007] To achieve the above object, a copper recovery treatment method for flameless combustion of copper-containing sludge is proposed, including the following steps:

[0008] Step 1: Deploy a distributed sensor network to obtain multi-source combustion data in the combustion furnace in real time;

[0009] Step 2: Construct a hybrid convolutional-long short-term memory network model to perform cross-modal feature fusion on the multi-source combustion data and extract multi-dimensional combustion feature vectors;

[0010] Step 3: Based on the multi-dimensional combustion feature vectors, online correct the system parameters of the differential-algebraic equations, and after the system parameters are corrected, solve the real-time system state variables;

[0011] Step 4: Construct constraint conditions using the system state variables, build a multi-agent reinforcement learning model based on multi-source combustion data and multi-dimensional combustion feature vectors, and use the multi-agent deep deterministic policy gradient algorithm to parallelly optimize the dynamic combustion parameters;

[0012] Step 5: Send the dynamic combustion parameters to the combustion actuator for combustion parameter update.

[0013] The steps of deploying a distributed sensor network to obtain multi-source combustion data in the combustion furnace in real time are as follows:

[0014] Step 101: Select and deploy physical sensors for collecting physical parameter features according to the structure and process requirements of the combustion furnace to form a sensor network;

[0015] Temperature, gas composition, air pressure, and copper ion concentration are all used as physical parameter features;

[0016] Step 102: Collect the eigenvalues of the physical parameter features collected by each physical sensor in real time, and respectively construct corresponding spatio-temporal feature data based on the characteristics of the physical parameter features. The spatio-temporal feature data of each physical parameter feature constitutes multi-source combustion data.

[0017] The steps of constructing a hybrid convolutional-long short-term memory network model to perform cross-modal feature fusion on the multi-source combustion data and extract multi-dimensional combustion feature vectors are as follows:

[0018] Step 201: Align the time stamps of the spatio-temporal feature data of each physical parameter feature to construct synchronized multi-source combustion data composed of the synchronized spatio-temporal feature data of each item;

[0019] Step 202: For the spatio-temporal characteristics of each physical parameter feature, construct a network branch for each physical parameter feature, and input the spatio-temporal feature data of each physical parameter feature into each network branch. Each network branch extracts the spatio-temporal features of the corresponding spatio-temporal feature data; each of the network branches includes a convolutional branch and an LSTM branch;

[0020] Step 203: Each network branch inputs the extracted spatio-temporal features into the feature fusion layer, and the feature fusion layer fuses the spatio-temporal features of each item to obtain a time series splicing vector;

[0021] Step 204: Input the spliced time series splicing vector into one or more fully connected layers, and perform regularization to reduce the time series splicing vector to a multi-dimensional combustion feature vector with a preset dimension.

[0022] The steps of online correcting the system parameters of the differential-algebraic equation based on the multi-dimensional combustion feature vector and solving the real-time system state variables after the system parameters are corrected are as follows:

[0023] Step 301: Construct differential-algebraic equations;

[0024] Step 302: Construct a mapping function to map the states and parameters of the differential-algebraic equations into a multi-dimensional combustion feature space directly observed by sensors;

[0025] Step 303: Set an error function between the mapping function and the multi-dimensional combustion feature vectors, and update the system parameters to be corrected in the differential-algebraic equations by using the gradient descent algorithm;

[0026] Step 304: After the system parameters are updated, obtain the real-time values of each system state variable in the differential-algebraic equations by solving the differential-algebraic equations.

[0027] The method of using the gradient descent algorithm to update the system parameters to be corrected in the differential-algebraic equations is as follows:

[0028] Whenever a real-time multi-dimensional combustion feature vector is obtained, obtain the corresponding predicted feature through the current system parameters to be corrected and the current system state variables;

[0029] Calculate the error between the multi-dimensional combustion feature vector and the predicted feature, and calculate the gradient of the error;

[0030] Based on the calculated error and gradient, use the gradient descent algorithm to update the system parameters to be corrected in real time;

[0031] After the system parameters to be corrected are updated, feedback the updated system parameters to the differential-algebraic equations for the system parameter update in the next time step, forming a continuous online closed-loop feedback system.

[0032] The method of constructing constraint conditions with the system state variables, constructing a multi-agent reinforcement learning model based on multi-source combustion data and multi-dimensional combustion feature vectors, and using the multi-agent deep deterministic policy gradient algorithm to parallelly optimize the dynamic combustion parameters includes the following steps:

[0033] Step 401: Set upper and lower limit constraint conditions for each system state variable to ensure that during the optimization process, the dynamic regulation decisions meet the pre-set constraint conditions of the system state variables;

[0034] Step 402: According to each dynamic combustion parameter, divide the regulation task of the entire combustion system into multiple sub-tasks, and each sub-task is responsible for by an agent;

[0035] Step 403: Set the global state s to include all sensor data and the system state represented by the multi-dimensional combustion feature vectors; and design a deterministic policy network and a corresponding value network for each agent;

[0036] Step 404: Set the corresponding state space, action space, and reward function for each agent; the reward function includes at least the penalty for violating the constraint condition.

[0037] The state space of each agent is set to the global state s.

[0038] The action space represents the adjustment increment of each agent to the dynamic combustion parameters.

[0039] The reward function is set for the regulation task of the dynamic combustion parameters corresponding to each agent.

[0040] Step 405: Adopt the deep deterministic policy gradient algorithm. In each training cycle, each agent uses the data in the experience replay buffer to update the gradient of the value network, and through the policy value evaluation output by the value network, smoothly update the policy generated by the deterministic policy network.

[0041] The method of sending the dynamic combustion parameters to the combustion actuator for combustion parameter update is as follows:

[0042] The combustion actuator of the combustion furnace receives the adjustment values of the corresponding dynamic combustion parameters generated by the deterministic policy network of each agent, and adjusts the dynamic combustion parameters according to the adjustment values.

[0043] The copper recovery treatment system for flameless combustion of copper-containing sludge includes a multi-source data collection module, a feature extraction module, a state variable extraction module, and a combustion parameter adjustment module; among them, each module is connected electrically.

[0044] The multi-source data collection module deploys a distributed sensor network to obtain multi-source combustion data in the combustion furnace in real time, and sends the multi-source combustion data to the feature extraction module and the combustion parameter adjustment module.

[0045] The feature extraction module constructs a hybrid convolutional-long short-term memory network model to perform cross-modal feature fusion on the multi-source combustion data, extracts a multi-dimensional combustion feature vector, and sends the multi-dimensional combustion feature vector to the state variable extraction module and the combustion parameter adjustment module.

[0046] The state variable extraction module, based on the multi-dimensional combustion feature vector, online corrects the system parameters of the differential-algebraic equation, and after the system parameters are corrected, solves the real-time system state variables, and sends the system state variables to the combustion parameter adjustment module.

[0047] The combustion parameter adjustment module constructs constraint conditions based on the system state variables, builds a multi-agent reinforcement learning model based on multi-source combustion data and multi-dimensional combustion feature vectors, and uses the multi-agent deep deterministic policy gradient algorithm to parallelly optimize the dynamic combustion parameters, and sends the dynamic combustion parameters to the combustion actuator for combustion parameter update.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] The present invention establishes a differential-algebraic equation model of the combustion process, simulates and fits the multi-dimensional combustion feature vectors in the offline experiment, corrects the system parameters in the differential-algebraic equation model, then uses the multi-source data fusion technology to online solve the state parameters that are difficult to directly measure in the differential-algebraic equation model, and then constructs the constraint conditions of the multi-agent reinforcement learning model according to the state parameters that are difficult to directly measure, and finally parallelly optimizes the dynamic combustion parameters by using the multi-agent deep deterministic policy gradient algorithm; realizes the formation of a real-time online feedback closed loop through online monitoring, model solving and policy update, ensures that the combustion process always operates in an efficient, safe and environmentally friendly state, and significantly improves the stability and environmental performance of the combustion process. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flowchart of the copper recovery treatment method for flameless combustion of copper-containing sludge in Embodiment 1 of the present invention;

[0051] Figure 2 It is an example diagram of the configuration of various physical sensors in the combustion furnace in Embodiment 1 of the present invention;

[0052] Figure 3 It is an example diagram of the time curve of the pressure drop waveform in Embodiment 1 of the present invention;

[0053] Figure 4 It is a module connection relationship diagram of the copper recovery treatment system for flameless combustion of copper-containing sludge in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0055] In actual industrial applications, combustion equipment is often affected by factors such as fluctuations in raw material composition, environmental interference and equipment aging, resulting in difficult stable operation of the combustion process. At the same time, some key state parameters are difficult to obtain in real time due to harsh measurement conditions and high costs, which directly affect the judgment and control effect of the control system on the combustion state.

[0056] Example 1

[0057] As Figure 1 shown, a copper recovery treatment method for flameless combustion of copper-containing sludge includes the following steps:

[0058] Step 1: Deploy a distributed sensor network to obtain multi-source combustion data in the combustion furnace in real time;

[0059] Step 2: Construct a hybrid convolutional-long short-term memory network model to perform cross-modal feature fusion on the multi-source combustion data and extract multi-dimensional combustion feature vectors;

[0060] Step 3: Based on the multi-dimensional combustion feature vectors, online correct the system parameters of the differential-algebraic equations, and after the system parameters are corrected, solve for the real-time system state variables;

[0061] Step 4: Use the system state variables to construct constraint conditions, build a multi-agent reinforcement learning model based on multi-source combustion data and multi-dimensional combustion feature vectors, and use the multi-agent deep deterministic policy gradient algorithm to parallelly optimize the dynamic combustion parameters;

[0062] Step 5: Send the dynamic combustion parameters to the combustion actuator for combustion parameter update.

[0063] In the embodiment of the present invention, the deployment of the distributed sensor network to obtain multi-source combustion data in the combustion furnace in real time includes the following steps:

[0064] Step 101: Select and deploy physical sensors for collecting physical parameter features according to the structure and process requirements of the combustion furnace to form a sensor network;

[0065] Specifically, the deployment process of the sensor network includes:

[0066] Since the combustion process of copper-containing sludge mainly occurs in the furnace body of the combustion furnace, and the combustion temperature is the key factor controlling the combustion efficiency and coal / sludge conversion, too high or too low temperature will affect the combustion efficiency and pollutant emissions of the tail gas. Therefore, it is necessary to arrange thermocouple sensors at preset key positions along the furnace chamber to ensure coverage of all hot zones in the furnace chamber, and each thermocouple sensor is electrically connected to the data acquisition unit, so as to collect the temperature data at each position in the combustion furnace;

[0067] Since a large amount of tail gas is released during the combustion process of copper-containing sludge, the change in the tail gas composition reflects the combustion reaction efficiency and the generation of pollutants. Therefore, it is necessary to deploy a spectral sensor at the tail gas outlet position of the combustion furnace, and electrically connect the spectral sensor to the data acquisition unit through an optical fiber to collect the spectral data of the tail gas, so as to detect the gas components in the tail gas;

[0068] The adsorption tower is used to purify harmful components in the combustion exhaust gas. An increase in the pressure difference between the inlet and outlet of the adsorption tower indicates that the gas flow is blocked, which may mean that the adsorbent is saturated or blocked and needs to be processed in time to maintain the treatment efficiency of the adsorption tower. Therefore, differential pressure sensors are installed at the inlet and outlet positions inside the adsorption tower of the combustion furnace to collect the air pressure of the gas at the inlet and outlet of the adsorption tower;

[0069] During the flameless combustion process of copper-containing sludge, after the copper-containing sludge is heated to a high temperature, it will be partially converted into soluble copper compounds through the oxidation of the high-temperature atmosphere. Therefore, it is necessary to transport the leachate or reaction liquid containing dissolved metals from the leaching area to the subsequent treatment unit through the leachate channel of the soluble copper compound; Therefore, it is necessary to arrange ion-selective electrode sensors in the leachate channel of the combustion furnace to obtain the copper ion concentration at each point in the combustion furnace in real time;

[0070] An example diagram of the configuration of various physical sensors inside the combustion furnace is as Figure 2 shown. The structure of the combustion furnace is shown in this example diagram, and the positions of the furnace chamber, leachate channel, tail gas outlet, and adsorption tower are marked, and the installation positions of various physical sensors are correspondingly marked according to each position;

[0071] Temperature, gas composition, air pressure, and copper ion concentration are all used as physical parameter characteristics.

[0072] It should be noted that the above temperature, gas composition, air pressure, and copper ion concentration are only several exemplary physical parameter characteristics listed in this embodiment. During the actual flameless combustion process of copper-containing sludge, there may also be other characterization characteristics related to tail gas purification and copper recovery amount, and any other related characterization characteristics also fall within the protection scope of the present invention.

[0073] Step 102: Collect the characteristic values of the physical parameter characteristics collected by various physical sensors in real time, and respectively construct corresponding spatio-temporal characteristic data based on the characteristics of the physical parameter characteristics. The spatio-temporal characteristic data of each physical parameter characteristic constitute multi-source combustion data;

[0074] Specifically, the process of respectively constructing the corresponding spatio-temporal characteristic data includes:

[0075] For the thermocouple sensor, set the sampling frequency corresponding to the thermocouple sensor. Each time a sample is taken, each thermocouple sensor records the temperature value at the corresponding position;

[0076] According to the geometric position and shape of the furnace, the number of horizontal positions for deploying thermocouple sensors inside the furnace is marked as X. The number of vertical positions for deploying thermocouple sensors inside the furnace is marked as Y. The horizontal position numbers are marked as x, and the vertical position numbers are marked as y, that is, a total of X×Y thermocouple sensors are deployed;

[0077] The thermocouple sensor in the x-th row and y-th column is marked as xy;

[0078] The time of each sampling is marked as t, and the temperature sampled by the thermocouple sensor xy at the t-th moment is marked as T xy (t), and the temperature matrix sampled by all thermocouple sensors at the t-th moment is marked as T(t);

[0079] Then the expression of T(t) is:

[0080]

[0081] Among them, the value in the x-th row and y-th column represents the temperature value collected by the thermocouple sensor xy at the t-th moment;

[0082] All temperature matrices are saved in chronological order as temperature field time-series data;

[0083] For the spectral sensor, set the corresponding acquisition period of the spectral sensor. Each time a sample is taken, the spectral sensor measures the light absorption intensity at different wavelengths to generate a spectral diagram P(λ, t) of the exhaust gas; where λ is the wavelength and t is the sampling time;

[0084] The output of the spectral sensor includes multiple bands in the visible light region such as 200 nm to 800 nm, and the absorption intensity of each band represents the concentration of a specific component;

[0085] Arrange the spectral data sampled by the spectral sensor at the t-th moment in chronological order to obtain a spectral data matrix:

[0086] P(t) = [P(λ1, t), P(λ2, t), …, P(λ e , t)];

[0087] Among them, e is the number of measured spectral bands. Each row in the matrix represents the spectral data at a certain time point, and each column represents the intensity at a certain wavelength;

[0088] The spectral data matrices at all times are sorted in chronological order to form a spectral time-series data matrix; and the spectral data matrix at each time point can be used to analyze the component changes of pollutants in the exhaust gas;

[0089] For the differential pressure sensor, set the corresponding acquisition period of the differential pressure sensor. Each time the differential pressure is collected, the gas pressure difference at the inlet and outlet of the adsorption tower is collected in real time;

[0090] The differential pressure sensor measures the pressure difference ΔPr(t) at both ends of the adsorption tower in real time;

[0091] The calculation formula for the pressure difference ΔPr(t) is:

[0092] ΔPr(t) = Pr in (t) - Pr out (t);

[0093] where Pr in (t) and Pr out (t) are the gas pressures at the inlet and outlet of the adsorption tower respectively;

[0094] Through continuous sampling, the time curve of the adsorption tower formed by the pressure difference changing with time is used as the pressure drop waveform; An example diagram of the time curve of a typical pressure drop waveform is as Figure 3 shown. In this pressure drop waveform, three different stages of pressure difference characteristics are presented: In the initial stage, the adsorption tower operates stably and the pressure drop remains at a low level. As time goes by, the adsorbent begins to adsorb pollutants in the tail gas and the gas flow resistance gradually increases; In the gradually rising stage, as the adsorbent gradually saturates, the pressure drop gradually rises; In the sharply rising stage, the adsorbent is saturated or blocked, and the pressure drop increases rapidly, indicating that the treatment effect of the adsorption tower decreases and needs to be treated in time;

[0095] For the ion selective electrode sensor, set the acquisition period corresponding to the ion selective electrode sensor. Each ion selective electrode sensor will calculate the concentration C(x',t) of copper ions in the leaching solution according to the measured potential change, where x' is the spatial position of the liquid and t is the time; For example, set multiple sampling points in the leaching solution channel and arrange ion selective electrode sensors along the leaching solution flow channel to obtain concentration data at different positions;

[0096] According to the positions of the ion selective electrode sensors, record the concentration data of copper ions at different positions and construct the spatial distribution of copper ion concentration. The concentration data at each moment can be expressed in the form of a concentration position sequence C(t), and C(t) can be expressed as:

[0097] C(t) = [C(x1,t), C(x2,t), C(x3,t), …, C(xh,t)];

[0098] where h is the number of all ion selective electrode sensors, and x1, x2, ……, xh are the positions of the measurement points of each ion selective electrode sensor respectively;

[0099] The concentration position sequences at all moments form a concentration gradient data set containing time and position changes;

[0100] The time-series data of the temperature field, the spectral time-series data matrix, the pressure-drop waveform, and the concentration-gradient data set are used as the spatio-temporal feature data of temperature, gas composition, air pressure, and copper-ion concentration, respectively.

[0101] Further, the steps of constructing a hybrid convolutional-long short-term memory network model to perform cross-modal feature fusion on the multi-source combustion data and extract multi-dimensional combustion feature vectors are as follows:

[0102] Step 201: Align the time stamps of the spatio-temporal feature data of each physical parameter feature to construct synchronous multi-source combustion data composed of the spatio-temporal feature data of each item after time synchronization.

[0103] Specifically, a common time grid is preset as a common acquisition period. For the spatio-temporal feature data with inconsistent sampling frequencies, such as the time-series data of the temperature field, the spectral time-series data matrix, the pressure-drop waveform, and the concentration-gradient data set, an interpolation algorithm is used for interpolation filling so that the spatio-temporal feature data of each item is aligned with the acquisition period of the common time grid.

[0104] For example, if the spectral sensor samples once per second and the time-stamp interval is much larger than the 0.1 second of the common time grid, the linear interpolation method is used to extend each sampling value to all 0.1-second time nodes within its corresponding 1 second. For example, for the spectral data at time t = 1 second, it can be set to this value at all common time points between 1 second and 1.1 seconds, thus achieving alignment with the common time grid in terms of time;

[0105] Similarly, the linear interpolation method can be used for interpolation filling of the time-series data of the temperature field, the pressure-drop waveform, and the concentration-gradient data set to ensure time alignment of the spatio-temporal feature data of each item in terms of the acquisition period.

[0106] Step 202: For the spatio-temporal characteristics of each physical parameter feature, construct network branches for each physical parameter feature, input the spatio-temporal feature data of each physical parameter feature into each network branch, and each network branch extracts the spatio-temporal features of the corresponding spatio-temporal feature data; each of the network branches includes a convolutional branch and an LSTM branch;

[0107] Specifically, the construction method of the network branch includes:

[0108] Since the temperature field data has obvious spatial distribution information, a two-dimensional convolutional network can capture local temperature gradients, hot spot regions, and temperature distribution patterns, thereby forming high-level spatial features; for the time-series data of the temperature field, a two-dimensional convolutional network is used to process the temperature field image. The first layer is set with a preset number of filters, such as 32 filters, the convolutional kernel size is 3×3×3, the stride is set to 1, the edge padding method uses "same" padding, the activation function uses the ReLU function, and then a 2×2 max pooling layer is connected for downsampling. After repeating the convolution-pooling stack 2-3 times, the local spatial features of the temperature at each time step are extracted, and the local spatial features of the temperature extracted at each time step are flattened into a one-dimensional vector to form temperature time-series features; the shape of the temperature time-series features is: B×T1×d temp , where B is the batch size of the preset model input, T1 is the preset time step, and d temp is the feature dimension output by the pooling layer;

[0109] Since there are local correlations between different bands in the spectral data, one-dimensional convolution can extract information such as specific absorption peaks and baseline drifts, which helps to identify the characteristics of each component in the exhaust gas; for the spectral time-series data matrix, one-dimensional convolutional layers are used for convolution. The first layer is set with a preset number of filters, the convolutional kernel length is 5, the activation function uses ReLU, and then a one-dimensional max pooling layer is connected to reduce the dimension of the local band information; the number of filters is, for example, 64; after one-dimensional convolution stacking, the spectral local features at each time step are extracted as spectral vectors;

[0110] Since the pressure drop waveform and copper ion concentration data mainly reflect the equipment state and fluid dynamic changes, LSTM can capture long-term dependencies and dynamic change laws, providing time evolution features for the model; therefore, for the pressure drop waveform and copper ion concentration data, the pressure drop waveform data and the concentration gradient dataset of copper ions in the leaching solution are extracted respectively;

[0111] For the pressure drop waveform data or the concentration gradient dataset, a one-dimensional convolutional layer such as filtering or local pattern recognition is first used for preliminary local feature extraction, and the output local feature dimensions are respectively denoted as and

[0112] The local feature sum is concatenated with the original signal to form a comprehensive vector at each time step

[0113] The comprehensive vector is input into a long short-term memory network to obtain the output time-series dynamic feature F dyn (t); the long short-term memory network is set with, for example, 128 hidden units, and the gating mechanism is used to capture the changing trends of the pressure drop and concentration over time.

[0114] Step 203: Each network branch inputs the extracted spatio-temporal features into the feature fusion layer. After the feature fusion layer fuses the spatio-temporal features, a temporal splicing vector is obtained.

[0115] Specifically, the method of fusing the spatio-temporal features by the feature fusion layer to obtain the temporal splicing vector is as follows:

[0116] For each time step t, the features from each network branch are spliced to generate the temporal splicing vector; that is, the temporal splicing vector is represented as [F temp (t), F spec (t), F dyn (t)]; where F temp (t), F spec (t) are the temperature temporal features extracted from the temperature field temporal data by the network branch and the spectral vectors extracted from the spectral temporal data matrix by the corresponding network branch, respectively.

[0117] It can be understood that the spliced temporal splicing vector contains the spatial features of the temperature field, the frequency domain information of the spectrum, and the temporal dynamic features of the pressure drop and concentration, and the dimensions of the feature data are unified.

[0118] Step 204: Input the spliced temporal splicing vector into one or more fully connected layers, and use Dropout and Batch Normalization for regularization to reduce the temporal splicing vector to a multi-dimensional combustion feature vector with a preset dimension.

[0119] It should be noted that the flameless combustion process of copper-containing sludge usually involves complex heat conduction, chemical reactions, and fluid mechanics behaviors physically. Its state evolution over time can be described by differential-algebraic equations (DAE). This model not only includes the dynamic differential equation part but also includes algebraic equations representing instantaneous constraints. For example, the mass conservation, energy conservation, and chemical equilibrium conditions can all be represented by algebraic equations.

[0120] Furthermore, the steps of online correcting the system parameters of the differential-algebraic equation based on the multi-dimensional combustion feature vector and solving the real-time system state variables after the system parameter correction include the following:

[0121] Step 301: Construct a differential-algebraic equation.

[0122] Specifically, the differential-algebraic equation is constructed according to the physical laws of combustion. For example, the form of the differential-algebraic equation can be:

[0123]

[0124] Among them, x(t) represents the system state variables that are difficult to directly measure. In this embodiment, it may include physical quantities such as the real-time reactant concentration, flow rate, dynamic heat transfer coefficient, and heat exchange efficiency in the combustion furnace. Since these physical quantities are difficult to directly obtain by sensors, some physical models are required for solution; u(t) represents the external control input. In this embodiment, it may include controllable parameters during the combustion process such as combustion temperature, air intake, nozzle opening, fuel supply rate, and auxiliary heating; θ is the system parameter to be corrected, such as dynamic combustion parameters such as reaction rate constant, heat transfer coefficient, and mixing efficiency. The function f(·) is used to describe the dynamic change of the system state. The function f(·) calculates the state change rate according to the current state of the combustion furnace. The function g(·) is used to describe the algebraic constraint conditions that must be satisfied, such as algebraic equations of energy conservation and mass conservation; represents the derivative of the system state variable x(t) with respect to time change, that is, the state change rate, which describes the state change speed of the system at a certain moment and reflects the dynamic response.

[0125] An example of the function f is:

[0126]

[0127] Among them, Q1 and Q2 respectively represent the energy function of combustion and the energy function of heat transfer. k is the temperature-related reaction rate constant, C A is the reactant concentration; m’ represents the mass of the object, C p represents the specific heat capacity, and T is the temperature;

[0128] Each line in this formula represents the state change rate of a physical parameter characteristic with time. Each line of the formula is obtained by fitting according to actual experience or historical data. For example, the first line is the empirical fitting formula for the state change rate of temperature, and the second line is the empirical fitting formula for the state change rate of copper ion concentration.

[0129] An example of the function g is:

[0130]

[0131] Among them, T b represents the equilibrium temperature at time t, C A (t) and C B (t) respectively represent the reactant and product concentrations at time t, while C total represents the total concentration of substances;

[0132] Each line in this formula represents a constraint condition. The first line represents temperature conservation, and the second line represents mass conservation.

[0133] Step 302: Construct a mapping function to map the differential-algebraic equation states and parameters into a multi-dimensional combustion feature space directly observed by sensors;

[0134] Specifically, the mapping into the multi-dimensional combustion feature space directly observed by sensors means that the predicted features after mapping are made consistent with the multi-dimensional combustion feature vectors obtained by actual measurement;

[0135] Mark the mapping function as h(x(t),θ), and this mapping function can be constructed by combining physical formulas and data fitting. For example, map the temperature distribution data into feature vectors through statistical functions or non-linear functions based on physical meanings, or use a neural network such as a fully connected network as a non-linear mapping function to perform end-to-end mapping on the model state parameters, and use some pre-annotated data for supervised training during the learning process;

[0136] Step 303: Set an error function between the mapping function and the multi-dimensional combustion feature vectors, and update the system parameters to be corrected in the differential-algebraic equation by using the gradient descent algorithm;

[0137] Specifically, the error function E(θ) between the mapping function and the multi-dimensional combustion feature vectors can be set as:

[0138] E(θ)=||h(x(t),θ)-F final (t)|| 2 where, F final (t) is the multi-dimensional combustion feature vector, and ||·|| represents the Euclidean norm;

[0139] Step 304: After the system parameters are updated, obtain the real-time values of each system state variable in the differential-algebraic equation by solving the differential-algebraic equation.

[0140] Furthermore, the method of using the gradient descent algorithm to update the system parameters to be corrected in the differential-algebraic equation is as follows:

[0141] Whenever a real-time multi-dimensional combustion feature vector is obtained, obtain the corresponding predicted features through the current system parameters to be corrected and the current system state variables;

[0142] Calculate the error between the multi-dimensional combustion feature vector and the predicted features, and calculate the gradient of this error;

[0143] Based on the calculated error and gradient, use the gradient descent algorithm to update the system parameters to be corrected in real time;

[0144] After the system parameters to be corrected are updated, the updated system parameters are immediately fed back into the differential-algebraic equation model for the update of the system parameters at the next time step, forming a continuous online closed-loop feedback system.

[0145] It can be understood that the differential-algebraic equation model describes the parameter simulation process of the combustion process. By inputting the real-time state and the system parameters to be corrected that have been pre-fitted in the model, a predicted multi-dimensional combustion characteristic vector is obtained, realizing the mapping from the real-time state of combustion to the spatio-temporal characteristics of the physical parameter characteristics representing the combustion results, thereby achieving the purpose of predicting the combustion results on the premise of inputting the real-time combustion state.

[0146] Furthermore, constructing the constraint conditions with the system state variables, building a multi-agent reinforcement learning model based on multi-source combustion data and multi-dimensional combustion characteristic vectors, and using the multi-agent deep deterministic policy gradient algorithm to parallelly optimize the dynamic combustion parameters include the following steps:

[0147] Step 401: Set the upper and lower limit constraint conditions for each system state variable to ensure that during the optimization process, the dynamic regulation decision satisfies the pre-set constraint conditions of the system state variables;

[0148] For example, set the upper and lower limits of the change rate of the reactant concentration to ensure that the combustion reaction speed is neither too fast nor too slow; this constraint condition is only an example, and other constraint conditions can be further defined according to actual needs, and the present invention does not limit this.

[0149] Step 402: According to each dynamic combustion parameter, divide the regulation task of the entire combustion system into multiple sub-tasks, and each sub-task is responsible for by an agent; for example, one agent focuses on regulating the combustion temperature, another agent regulates the air intake volume, and the third agent is responsible for the nozzle opening, etc.;

[0150] Step 403: Set the global state s to include all sensor data and the system state represented by the multi-dimensional combustion characteristic vector; and design a deterministic policy network and a corresponding value network for each agent;

[0151] Among them, the policy network continuously outputs the adjustment amount of the dynamic combustion parameters, and the value network is used to estimate the future cumulative reward under the current state and action;

[0152] Step 404: Set the corresponding state space, action space and reward function for each agent; the reward function at least includes the penalty for violating the constraint conditions;

[0153] Generally, the state space of each agent is set to the global state s;

[0154] The action space represents the adjustment increment of each agent to the dynamic combustion parameters. The action space is usually continuous and needs to be normalized when output to ensure that the foregoing constraint conditions are met;

[0155] The reward function is set for the regulation task of the dynamic combustion parameters corresponding to each agent. For example, for the agent that regulates the combustion temperature, a comprehensive reward function of combustion efficiency and violation of constraint conditions is set, that is, if the combustion efficiency is improved, a positive reward is given, and if the constraint conditions are violated, a greater penalty is given; it can be understood that during the dynamic adjustment of various dynamic combustion parameters, various system state variables will change accordingly. Therefore, the judgment of violating the constraint conditions is as follows: calculate various system state variables through the process of step three above, and then judge whether the system state variable meets its corresponding constraint conditions;

[0156] Step 405: Adopt the deep deterministic policy gradient algorithm. In each training cycle, each agent uses the data in the experience replay buffer to update the gradient of the value network, and through the policy value evaluation output by the value network, smoothly update the policy generated by the deterministic policy network.

[0157] Furthermore, the method of sending the dynamic combustion parameters to the combustion actuator for combustion parameter update is as follows:

[0158] The combustion actuator of the combustion furnace receives the adjustment values of the corresponding dynamic combustion parameters generated by the deterministic policy network of each agent, and adjusts the dynamic combustion parameters according to the adjustment values.

[0159] Embodiment 2

[0160] As Figure 4 shown, the copper recovery treatment system for flameless combustion of copper-containing sludge includes a multi-source data collection module, a feature extraction module, a state variable extraction module, and a combustion parameter adjustment module; among them, each module is connected electrically.

[0161] The multi-source data collection module deploys a distributed sensor network to obtain multi-source combustion data in the combustion furnace in real time, and sends the multi-source combustion data to the feature extraction module and the combustion parameter adjustment module;

[0162] The feature extraction module constructs a hybrid convolutional-long short-term memory network model to perform cross-modal feature fusion on the multi-source combustion data, extracts a multi-dimensional combustion feature vector, and sends the multi-dimensional combustion feature vector to the state variable extraction module and the combustion parameter adjustment module;

[0163] A state variable extraction module, based on the multi-dimensional combustion feature vector, online corrects the system parameters of the differential-algebraic equations, and after the system parameters are corrected, solves the real-time system state variables and sends the system state variables to the combustion parameter adjustment module;

[0164] A combustion parameter adjustment module constructs constraint conditions with the system state variables, constructs a multi-agent reinforcement learning model based on multi-source combustion data and multi-dimensional combustion feature vectors, and uses the multi-agent deep deterministic policy gradient algorithm to parallelly optimize the dynamic combustion parameters, and sends the dynamic combustion parameters to the combustion actuator for combustion parameter update.

[0165] In addition, parts of the above technical solutions provided in the embodiments of the present application that are consistent with the corresponding technical solutions in the prior art in terms of implementation principles are not described in detail to avoid excessive elaboration.

[0166] As described above in the specific embodiments, the purpose, technical solutions, and beneficial effects of the present invention are further described in detail. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

[0167] The above preset parameters or preset thresholds are all set by those skilled in the art according to the actual situation or obtained through a large amount of data simulation.

[0168] The above embodiments are only used to illustrate the technical methods of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical methods of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for copper recovery treatment by flameless combustion of copper-containing sludge, characterized in that, It includes the following steps: Step 1: Deploy a distributed sensor network to obtain multi-source combustion data in the combustion furnace in real time; Step 2: Construct a hybrid convolutional-long short-term memory network model to perform cross-modal feature fusion on the multi-source combustion data and extract multi-dimensional combustion feature vectors; Step 3: Based on the multi-dimensional combustion feature vectors, online correct the system parameters of the differential-algebraic equations, and after the system parameters are corrected, solve for the real-time system state variables; Step 4: Use the system state variables to construct constraint conditions, build a multi-agent reinforcement learning model based on the multi-source combustion data and multi-dimensional combustion feature vectors, and use the multi-agent deep deterministic policy gradient algorithm to parallelly optimize the dynamic combustion parameters; Step 5: Send the dynamic combustion parameters to the combustion actuator for combustion parameter update.

2. The copper recovery and treatment method for flameless combustion of copper-containing sludge according to claim 1, characterized in that The deployment of the distributed sensor network to obtain multi-source combustion data in the combustion furnace in real time includes the following steps: Step 101: According to the structure and process requirements of the combustion furnace, select and deploy physical sensors for collecting physical parameter features to form a sensor network; Temperature, gas composition, air pressure, and copper ion concentration are all used as physical parameter features; Step 102: Real-time collect the feature values of the physical parameter features collected by each physical sensor, and based on the characteristics of the physical parameter features, respectively construct corresponding spatio-temporal feature data, and the spatio-temporal feature data of each physical parameter feature form multi-source combustion data.

3. The copper recovery treatment method for flameless combustion of copper-containing sludge according to claim 2, characterized in that, The construction of the hybrid convolutional-long short-term memory network model to perform cross-modal feature fusion on the multi-source combustion data and extract multi-dimensional combustion feature vectors includes the following steps: Step 201: Align the time stamps of the spatio-temporal feature data of each physical parameter feature to construct synchronous multi-source combustion data composed of the time-synchronized spatio-temporal feature data of each item; Step 202: For the spatio-temporal characteristics of each physical parameter feature, construct a network branch for each physical parameter feature, and input the spatio-temporal feature data of various physical parameter features into each network branch, and each network branch extracts the spatio-temporal features of the corresponding spatio-temporal feature data; Step 203: Each network branch inputs the extracted spatio-temporal features into the feature fusion layer, and after the feature fusion layer fuses the spatio-temporal features of each item, a time-series splicing vector is obtained; Step 204: Input the spliced time-series splicing vector into one or more fully connected layers, and perform regularization to reduce the time-series splicing vector to a multi-dimensional combustion feature vector with a preset dimension.

4. The copper recovery and treatment method for flameless combustion of copper-containing sludge according to claim 3, wherein Based on the multi-dimensional combustion feature vectors, online correct the system parameters of the differential-algebraic equations, and after the system parameters are corrected, solve for the real-time system state variables includes the following steps: Step 301: Construct a differential-algebraic equation; Step 302: Construct a mapping function to map the differential-algebraic equation state and parameters into the multi-dimensional combustion feature space directly observed by the sensor; Step 303: Set an error function between the mapping function and the multi-dimensional combustion feature vector, and update the system parameters to be corrected in the differential-algebraic equation by using the gradient descent algorithm; Step 304: After the system parameters are updated, obtain the real-time values of various system state variables in the differential-algebraic equations by solving the differential-algebraic equations.

5. The copper recovery and treatment method for flameless combustion of copper-containing sludge according to claim 4, wherein The method of using the gradient descent algorithm to update the system parameters to be corrected in the differential-algebraic equations is as follows: Whenever a real-time multi-dimensional combustion feature vector is obtained, obtain the corresponding predicted feature through the current system parameters to be corrected and the current system state variables. Calculate the error between the multi-dimensional combustion feature vector and the predicted feature, and calculate the gradient of the error. Based on the calculated error and gradient, use the gradient descent algorithm to update the system parameters to be corrected in real time. After the system parameters to be corrected are updated, feedback the updated system parameters into the differential-algebraic equations for the update of the system parameters in the next time step.

6. The copper recovery treatment method for flameless combustion of copper-containing sludge according to claim 5, characterized in that The steps of constructing constraint conditions with the system state variables, constructing a multi-agent reinforcement learning model based on multi-source combustion data and multi-dimensional combustion feature vectors, and using the multi-agent deep deterministic policy gradient algorithm to parallelly optimize the dynamic combustion parameters are as follows: Step 401: Set the upper and lower limit constraint conditions for each system state variable. Step 402: According to each dynamic combustion parameter, divide the regulation task of the entire combustion system into multiple sub-tasks, and each sub-task is responsible for by an agent. Step 403: Set the global state s to include all sensor data and the system state represented by the multi-dimensional combustion feature vector; and design a deterministic policy network and a corresponding value network for each agent. Step 404: Set the corresponding state space, action space and reward function for each agent. Step 405: Adopt the deep deterministic policy gradient algorithm. In each training cycle, each agent uses the data in the experience replay buffer to update the gradient of the value network, and through the policy value evaluation output by the value network, smoothly update the policy generated by the deterministic policy network.

7. The copper recovery and treatment method for flameless combustion of copper-containing sludge according to claim 6, characterized in that The reward function includes at least the penalty for violating the constraint conditions.

8. The copper recovery and treatment method for flameless combustion of copper-containing sludge according to claim 7, wherein The state space of each agent is set to the global state s. And the action space represents the regulation increment of each agent for the dynamic combustion parameters. And the reward function is set for the regulation task of the dynamic combustion parameters corresponding to each agent.

9. The copper recovery and treatment method for flameless combustion of copper-containing sludge according to claim 3, characterized in that, Each of the network branches includes a convolutional branch and an LSTM branch.

10. The copper recovery and treatment method for flameless combustion of copper-containing sludge according to claim 8, characterized in that, The method of sending the dynamic combustion parameters to the combustion actuator for updating the combustion parameters is as follows: The combustion actuator of the combustion furnace receives the adjustment values of the corresponding dynamic combustion parameters generated by the deterministic policy networks of each agent, and adjusts the dynamic combustion parameters according to the adjustment values.

11. A copper recovery treatment system for flameless combustion of copper-containing sludge, which is used to implement the copper recovery treatment method for flameless combustion of copper-containing sludge described in any one of claims 1-10, and is characterized in that, It includes a multi-source data collection module, a feature extraction module, a state variable extraction module and a combustion parameter adjustment module; among them, each module is connected electrically. The multi-source data collection module deploys a distributed sensor network, obtains multi-source combustion data in the combustion furnace in real time, and sends the multi-source combustion data to the feature extraction module and the combustion parameter adjustment module. The feature extraction module constructs a hybrid convolutional-long short-term memory network model to perform cross-modal feature fusion on the multi-source combustion data, extracts a multi-dimensional combustion feature vector, and sends the multi-dimensional combustion feature vector to the state variable extraction module and the combustion parameter adjustment module; The state variable extraction module online corrects the system parameters of the differential-algebraic equations based on the multi-dimensional combustion feature vector, and after the system parameters are corrected, solves the real-time system state variables and sends the system state variables to the combustion parameter adjustment module; The combustion parameter adjustment module constructs constraint conditions with the system state variables, constructs a multi-agent reinforcement learning model based on the multi-source combustion data and the multi-dimensional combustion feature vector, and uses the multi-agent deep deterministic policy gradient algorithm to parallelly optimize the dynamic combustion parameters, and sends the dynamic combustion parameters to the combustion actuator for combustion parameter update.

Images