Electric field self-adaption sediment fluid adjusting method and system based on neurodynamics

Abstract

The embodiment of the invention discloses an electric field self-adaption sediment fluid adjusting method and device based on neurodynamics, the method and device are used for sediment separation in a hydrocyclone, and the method comprises the steps that sediment characteristic data in sewage are collected; generating simulation data according to a discrete element method of computational fluid dynamics, and training a neurodynamics model by using the simulation data; generating a particle aggregation probability and a settlement path by using a neurodynamic model; predicting particle distribution and sedimentation trend in a future preset time period according to the agglomeration probability, the sedimentation path and the sediment characteristic data; generating a continuous electric field spectrum by adopting a deep reinforcement learning model according to a prediction result; and dynamically adjusting the electric field intensity and the gradient rate parameter value of the hydrocyclone according to the continuous electric field spectrum. And the silt separation efficiency and the adaptability to different working conditions are improved.

Description

Technical Field

[0001] The present application relates to the field of solid-liquid separation and neural network technology, and in particular to an electric field adaptive sediment fluid regulation method and system based on neural dynamics. Background Art

[0002] If the large amount of sediment-laden wastewater generated during tunnel construction is not effectively treated, it will cause serious pollution to surrounding water bodies and the ecological environment. Hydrocyclones, as highly efficient solid-liquid separation equipment, are widely used for sediment separation due to their simple structure, small footprint, and low operating costs. However, the sediment characteristics (such as particle size and concentration) in tunnel sewage fluctuate frequently. Existing electric field-assisted technologies, which mostly use a fixed electric field mode, cannot adapt to changes in sediment characteristics in real time, resulting in high energy consumption (e.g., over 100W) or unstable separation efficiency.

[0003] In recent years, artificial intelligence (AI) technology has demonstrated potential in engineering optimization. However, existing AI methods are mostly based on static datasets and lack dynamic prediction and interdisciplinary modeling capabilities. Neurodynamics, a field at the intersection of neuroscience and physics, can simulate complex dynamic systems, but has yet to be applied to sediment separation. Therefore, an efficient, intelligent, and adaptive control method is urgently needed to address these challenges, improve sediment separation efficiency, and meet China's stringent environmental protection requirements. Summary of the Invention

[0004] The present application provides an electric field adaptive sediment fluid regulation method based on neurodynamics, which improves the sediment separation efficiency and the adaptability to different working conditions.

[0005] This application provides the following solutions: According to a first aspect, a neurodynamics-based electric field adaptive sediment flow regulation method for sediment separation in a hydrocyclone is provided, the method comprising: real-time collection of sediment characteristic data from sewage, the sediment characteristic data including sediment particle size, concentration, and flow velocity; generating simulation data for operating scenario parameters of the hydrocyclone based on a computational fluid dynamics discrete element method, and training a neurodynamics model using the simulation data; using the neurodynamics model to model sediment particle motion as a neurodynamics system based on the sediment characteristic data, and generating particle agglomeration probability and sedimentation path; predicting particle distribution and sedimentation trend within a predetermined future time period based on the particle agglomeration probability and sedimentation path, as well as the sediment characteristic data; generating a continuous electric field spectrum using a deep reinforcement learning model based on the predicted particle distribution and sedimentation trend; and dynamically adjusting the electric field strength and gradient rate parameter values ​​of the hydrocyclone based on the continuous electric field spectrum.

[0006] According to an achievable method in an embodiment of the present application, the use of the simulation data to train the neural dynamics model includes: the neural dynamics model adopts a pulse neural network model, and inputs particle trajectory data simulated based on the discrete element method of computational fluid dynamics, wherein the particle trajectory data includes particle position, velocity, collision force and electric field force; by optimizing the network parameters of the neural dynamics model, the particle agglomeration probability and sedimentation path are generated; wherein the optimization aims to minimize the prediction error of the agglomeration probability and the sedimentation path.

[0007] According to an achievable method in an embodiment of the present application, the method further includes: the pulse neural network model includes a leakage integration and firing neuron model, wherein the excitatory interaction triggers neuron firing based on the particle collision frequency, and the inhibitory interaction reduces the neuron potential based on the particle agglomeration event; according to the real-time collected sediment characteristic data, the weights of the excitatory interaction and the inhibitory interaction are dynamically adjusted, and the weights are associated with changes in sediment particle size and concentration.

[0008] According to an achievable method in an embodiment of the present application, the prediction of particle distribution and sedimentation trend within a predetermined time period in the future includes: based on the 5-second historical data of the sediment characteristic data and the agglomeration probability of the neural dynamics model, using a long short-term memory network to predict the particle concentration field and sedimentation velocity in the next 5-10 seconds, and generating a prediction result of particle distribution and sedimentation trend.

[0009] According to an achievable method in an embodiment of the present application, the training process of the deep reinforcement learning model includes: using the simulation data and the sediment characteristic data to construct a first reward function, wherein the first reward function comprehensively considers separation efficiency, energy consumption and electrode wear; through iterative optimization, generating a continuous electric field spectrum that adapts to changes in sediment particle size from 1 micron to 100 microns and concentration from 0.1 g / L to 10 g / L.

[0010] According to an achievable method in an embodiment of the present application, the use of a deep reinforcement learning model to generate a continuous electric field spectrum includes: the deep reinforcement learning model uses a proximal strategy optimization algorithm with a weighted sum of separation efficiency and energy consumption as a second reward function; based on the predicted results of the particle distribution and sedimentation trend and real-time sediment characteristic data, a continuous electric field spectrum is generated, and the continuous electric field spectrum includes any combination of electric field strength in the range of 0.1 to 10 kV / m and gradient rate in the range of 0 to 100 kV / m².

[0011] According to an achievable method in an embodiment of the present application, the method further includes: detecting abnormal changes in sediment characteristic data based on the predicted results of the particle distribution and sedimentation trend, the abnormal changes including sediment concentration exceeding 10 g / L or particle size suddenly changing to less than 5 microns; in response to the abnormal changes, triggering a protection mode to reduce the electric field strength to a predetermined safety range.

[0012] According to a second aspect, a neurodynamic-based electric field adaptive sediment flow regulation system is provided for sediment separation in a hydrocyclone. The system comprises: a real-time data acquisition unit configured to acquire sediment characteristic data from sewage in real time, the sediment characteristic data including sediment particle size, concentration, and flow velocity; a dynamic model training unit configured to generate simulation data for the working scenario parameters of the hydrocyclone based on a computational fluid dynamics discrete element method, and train a neurodynamic model using the simulation data; a dynamic model inference unit configured to use the neurodynamic model to model the movement of sediment particles as a neurodynamic system based on the sediment characteristic data, and generate a particle agglomeration probability and sedimentation path; a prediction unit configured to predict particle distribution and sedimentation trend within a predetermined future time period based on the agglomeration probability and sedimentation path and the sediment characteristic data; an electric field spectrum generation unit configured to generate a continuous electric field spectrum using a deep reinforcement learning model based on the predicted results of the particle distribution and sedimentation trend; and an electric field regulation unit configured to dynamically adjust the electric field strength and gradient rate parameter values ​​of the hydrocyclone based on the continuous electric field spectrum.

[0013] According to a third aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the steps of any one of the methods according to the first aspect are implemented.

[0014] According to a fourth aspect, an electronic device is provided, comprising: One or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, which, when read and executed by the one or more processors, execute the steps of the method described in any one of the first aspects above.

[0015] According to the specific embodiments provided in this application, the technical effects of the electric field adaptive sediment flow regulation method and system based on neural dynamics in this application are as follows: This application introduces neural dynamics modeling and deep reinforcement learning to achieve intelligent and adaptive optimization of the electric field control of hydrocyclones under complex sediment flow conditions. This method utilizes real-time sediment characteristic data, combined with the Computational Fluid Dynamics–Discrete Element Method (CFD-DEM) simulation to train a neural dynamics model, accurately simulating particle agglomeration and settling behavior and predicting particle distribution trends. Based on the predicted results, a continuous electric field spectrum is dynamically generated through deep reinforcement learning to achieve optimal regulation of the electric field strength and gradient rate, thereby improving sediment separation efficiency, reducing energy consumption, and enhancing the system's adaptability to different operating conditions.

[0016] Of course, any product implementing the present application does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 This is a flow chart of the electric field adaptive sediment flow regulation method based on neural dynamics provided in an embodiment of the present application; Figure 2 A structural block diagram of the electric field adaptive sediment flow regulation system based on neural dynamics provided in an embodiment of the present application; Figure 3 A schematic block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0020] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "an", "the" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0021] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0022] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0023] Several sediment separation methods currently exist. These use centrifugal force to separate sediment particles from wastewater onto the outer walls of a cyclone, where they settle to the underflow outlet. These methods utilize the high-speed rotation of the cyclone to create vortices, which drive larger sediment particles to settle, with clean water discharged through an overflow outlet. However, these methods require manual adjustment of cyclone parameters and are unable to adapt to changes in sediment characteristics in real time, resulting in fluctuating efficiency or excessive energy consumption. Furthermore, these traditional methods lack predictive and intelligent control of particle behavior, making them difficult to handle under complex operating conditions, leading to fine particle retention and increased equipment wear.

[0024] In view of this, this application provides a new idea. Figure 1 This is a flow chart of the neural dynamics-based electric field adaptive sediment flow regulation method provided in an embodiment of this application. This method is used for sediment separation in a hydrocyclone, a highly efficient solid-liquid separation device widely used for sediment separation in tunnel sewage treatment. Its operating principle is to use a high-speed rotating fluid to form a vortex within the cyclone, generating centrifugal force that flings sediment particles toward the outer wall and settles them to the underflow outlet, while clear water is discharged through the overflow outlet.

[0025] like Figure 1 As shown in , the method may include the following steps: Step S101: collecting sediment characteristic data in sewage in real time, wherein the sediment characteristic data includes sediment particle size, concentration and flow rate.

[0026] Step S102: Generate simulation data for the working scene parameters of the hydrocyclone according to the computational fluid dynamics discrete element method, and use the simulation data to train the neural dynamics model.

[0027] Step S103: Using the neural dynamics model, the movement of sediment particles is modeled as a neural dynamics system according to the sediment characteristic data, and the particle agglomeration probability and sedimentation path are generated.

[0028] Step S104: predicting the particle distribution and sedimentation trend within a predetermined time period in the future based on the particle agglomeration probability and sedimentation path and the sediment characteristic data.

[0029] Step S105: Based on the prediction results of the particle distribution and sedimentation trend, deep reinforcement learning is used to generate a continuous electric field spectrum, and the electric field intensity and gradient rate are dynamically adjusted.

[0030] Step S106: adjusting the parameter values ​​of the hydrocyclone according to the continuous electric field spectrum.

[0031] As can be seen from the above process, this application realizes intelligent and adaptive optimization of the electric field control of the hydrocyclone under complex sediment fluid conditions by introducing neural dynamics modeling and deep reinforcement learning. This method uses real-time sediment characteristic data, combined with CFD-DEM simulation to train the neural dynamics model, accurately simulates particle agglomeration and sedimentation behavior, and predicts particle distribution trends. Based on the prediction results, a continuous electric field spectrum is dynamically generated through deep reinforcement learning to achieve optimal regulation of the electric field strength and gradient rate, thereby improving sediment separation efficiency, reducing energy consumption, and enhancing the system's adaptability to different working conditions.

[0032] The following describes in detail the steps in the above process and the effects that can be further produced in conjunction with the embodiments.

[0033] First, the above step 101, namely, "real-time collection of sediment characteristic data in sewage, wherein the sediment characteristic data includes sediment particle size, concentration and flow velocity", is described in detail with reference to an embodiment.

[0034] This application uses sensor equipment to monitor the sediment characteristics of sewage in real time, including particle size, concentration, and flow rate. This data reflects the immediate state of sewage and can capture rapid changes in sediment characteristics during tunnel construction.

[0035] Sediment particle size is a parameter that describes the size of sediment particles, usually in the range of 1 to 100 microns. Particle size directly affects the sedimentation behavior of particles in the cyclone. For example, smaller particles require a stronger electric field force to promote agglomeration. Particle size distribution can be quickly measured using equipment such as laser particle size analyzers. Concentration indicates the mass content of sediment in sewage, usually between 0.1 and 10 grams per liter, reflecting the density of sediment. High concentration may increase the difficulty of separation, and real-time monitoring of concentration helps the system dynamically adjust the electric field parameters. Flow rate refers to the speed at which sewage enters the cyclone, usually between 0.5 and 5 meters per second, which affects the centrifugal force of the cyclone and the trajectory of particle motion. The flow rate can be obtained through real-time measurement of the flow meter.

[0036] The above step 102, namely "generating simulation data for the working scenario parameters of the hydrocyclone according to the computational fluid dynamics discrete element method, and training the neural dynamics model using the simulation data" is described in detail below with reference to an embodiment.

[0037] This application uses the computational fluid dynamics discrete element method (CFD) to generate simulation data for the operating parameters of a hydrocyclone and uses this data to train a neural dynamics model. The CFD discrete element method (DEM) is an advanced simulation technique that accurately describes the interaction between sewage flow fields and sediment particles. By simulating the actual operating conditions within a hydrocyclone, such as varying flow rates and sediment characteristics, the generated data provides a physical basis for subsequent model training, ensuring that the model accurately reflects the aggregation and sedimentation patterns of particles.

[0038] The process of generating simulation data first requires setting the parameters of the hydrocyclone's operating scenario. These parameters can include the geometry of the cyclone, such as diameter and cone angle, and fluid properties such as density and viscosity. In addition, the characteristics of the sediment need to be defined, such as particle size ranging from 1 to 100 microns, concentration from 0.1 to 10 grams per liter, as well as the electric field strength and gradient rate. These parameters are based on typical operating conditions of tunnel sewage treatment and are determined through experimental measurements or engineering design. The simulation process uses specialized software, such as fluid dynamics analysis tools and discrete element method tools, to calculate the fluid's velocity distribution, pressure changes, and particle collisions, agglomeration, and sedimentation trajectories. These data describe in detail the dynamic behavior of sediment particles in the cyclone, providing rich training samples for the neural dynamics model.

[0039] The neurodynamic model simulates the motion of sediment particles by emulating the interactions between neurons. For example, particle collisions trigger excitation-like signals, while agglomeration produces an inhibitory effect. During training, simulated data, including parameters such as particle position, velocity, collision force, and agglomeration state, is used as input. The model learns the relationship between this data and agglomeration probability and sedimentation path. Once trained, the model can rapidly predict particle behavior based on real-time sediment characteristic data, eliminating the need for repeated, time-consuming simulations.

[0040] As an implementable method, using the simulation data to train the neural dynamics model includes: the neural dynamics model adopts a pulse neural network, inputs particle trajectory data simulated based on the discrete element method of computational fluid dynamics, and the particle trajectory data includes particle position, velocity, collision force and electric field force; by optimizing the network parameters of the neural dynamics model, the particle agglomeration probability and sedimentation path are generated; wherein, the optimization aims to minimize the prediction error of the agglomeration probability and sedimentation path.

[0041] A spiking neural network is a computational model that mimics the behavior of biological neurons and is particularly well-suited for processing dynamic, time-series data. By learning the patterns of particle motion in simulated data, it can quickly predict the agglomeration and settling behavior of sediment particles within a hydrocyclone, providing a basis for subsequent electric field adjustments.

[0042] The input data for the training process is particle trajectory data simulated using the discrete element method of computational fluid dynamics. This data describes in detail the motion characteristics of sediment particles within the cyclone, including particle position, velocity, collision force, and electric field force. Particle position and velocity reflect the particle's trajectory in the flow field, for example, whether the particle settles close to the outer wall. Collision force describes the interaction between particles or with the cyclone wall, affecting the possibility of agglomeration. Electric field force, derived from the applied electric field, drives particle agglomeration or accelerates sedimentation. By simulating different operating conditions, such as particle sizes ranging from 1 to 100 microns and concentrations from 0.1 to 10 grams per liter, these data provide a rich set of training samples, ensuring that the model can adapt to a variety of real-world scenarios.

[0043] The core of the training process is to generate the probability of particle agglomeration and sedimentation path by optimizing the network parameters of the pulse neural network. The probability of agglomeration indicates the probability of particles agglomerating into larger clusters. For example, the probability of fine particles agglomerating under the action of an electric field may reach 85%. The sedimentation path describes the movement trajectory of particles within the cyclone, for example, 80% of the particles settle to the bottom flow outlet. The optimization process aims to minimize the prediction error, ensuring that the agglomeration probability and sedimentation path output by the model are as close as possible to the simulation data. For example, by adjusting the connection weights of the network, the model can control the prediction error within 5%. This optimization usually uses a specialized algorithm and iterative calculations on a large amount of simulation data until the model achieves high accuracy.

[0044] Furthermore, the spiking neural network includes a leaky integration and firing neuron model, in which excitatory interactions trigger neuronal firing based on the frequency of particle collisions, and inhibitory interactions reduce neuronal potentials based on particle agglomeration events; the weights of excitatory and inhibitory interactions are dynamically adjusted based on real-time collected sediment characteristic data, and the weights are associated with changes in sediment particle size and concentration.

[0045] The spiking neural network configuration features a leaky integration and release neuron model, simulating the behavior of sediment particles through excitatory and inhibitory interactions. This model, through specific neuronal mechanisms, transforms the physical motion characteristics of particles into dynamic neural network calculations to efficiently generate agglomeration probabilities and sedimentation paths. The leaky integration and release neuron model is a computational framework that mimics the behavior of biological neurons, simulating how neurons accumulate potentials after receiving signals and release signals when a threshold is reached. In the present invention, this model maps particle collision and agglomeration events into interactions between neurons, thereby capturing the complex dynamic behavior of sediment particles within a hydrocyclone.

[0046] Excitation interactions trigger neuronal firing based on the frequency of particle collisions and are one of the core mechanisms of the model. The particle collision frequency reflects the frequency with which sediment particles come into contact with other particles or walls within the cyclone. For example, at high flow rates, the collision frequency may reach more than 10 times per second. When the collision frequency is high, the neurons in the model are simulated as receiving strong excitation signals, which trigger firing, similar to the response of biological neurons to stimuli. This process corresponds to an active state in particle motion and may cause small particles to begin agglomerating. By linking the collision frequency with excitation interactions, the model can accurately capture the dynamic changes in particle behavior and provide key information for predicting the probability of agglomeration.

[0047] Inhibitory interactions, based on particle agglomeration events, reduce neuronal potentials, simulating the "stabilizing" effect of agglomeration on particle motion. When particles aggregate into larger clusters, such as when fine particles form larger clusters under the action of an electric field, the particles' motion activity decreases, and the frequency of collisions decreases accordingly. In the model, this agglomeration event is mapped as an inhibitory signal, causing the neuron's potential to drop, reducing the likelihood of firing. This inhibitory mechanism reflects the physical process by which particles tend to settle after agglomeration. For example, after agglomeration, particles are more likely to settle to the bottom flow outlet of a cyclone. Through the synergistic effect of excitation and inhibition interactions, the spiking neural network can accurately simulate the entire process of particles from active collisions to stable sedimentation.

[0048] The model dynamically adjusts the weights of excitatory and inhibitory interactions based on real-time sediment property data. This data includes particle size, concentration, and flow rate. These parameters fluctuate with changes in tunnel sewage treatment conditions, such as a sudden change in particle size from 10 microns to 5 microns, or an increase in concentration from 2 grams per liter to 8 grams per liter. The model adjusts the weights of the neural network based on this real-time data. For example, when the proportion of fine particles increases, the weight of excitatory interactions is increased to more sensitively capture collision events; at high concentrations, the weight of inhibitory interactions may be strengthened to reflect accelerated agglomeration. This dynamic adjustment enables the model to adaptively respond to changes in operating conditions, ensuring that the predicted agglomeration probability and sedimentation path remain close to reality.

[0049] Specifically, based on real-time sediment characteristic data, the characteristic values ​​of particle collision frequency and agglomeration events can be calculated, such as a collision frequency of 10 times per second or an agglomeration rate of 80%. These characteristic values ​​are input into the weight adjustment module of the spiking neural network, and the gradient is calculated using a predefined loss function. Using the gradient descent algorithm, the excitatory interaction weights (e.g., increasing the weight by 0.1 when the collision frequency is high) and the inhibitory interaction weights (e.g., decreasing the weight by 0.05 when the agglomeration rate is high) are dynamically updated. The adjustment step size is adaptively set based on the magnitude of changes in sediment particle size and concentration. For example, the step size increases to 0.01 when the particle size decreases from 10 microns to 5 microns, ensuring that the weights quickly adapt to changing operating conditions.

[0050] The above step 103, i.e., "using a neural dynamics model to model the movement of sediment particles into a neural dynamics system based on the sediment characteristic data, and generating particle agglomeration probability and sedimentation path," is described in detail below in conjunction with an embodiment.

[0051] Using a neurodynamic model, sediment particle motion is modeled as a neurodynamic system based on sediment property data. This process, based on real-time sediment property data, transforms the physical motion of particles into dynamic calculations within a neural network. By simulating the excitatory and inhibitory interactions between neurons, the neurodynamic system generates particle aggregation probabilities and settling paths, providing key information for subsequent particle behavior prediction and electric field regulation.

[0052] In a hydrocyclone, sediment particles are subjected to centrifugal forces, electric forces, and fluid drag, exhibiting complex behaviors such as collisions, agglomeration, and sedimentation. The neurodynamic model maps these physical processes to interactions between neurons. Particle collisions are simulated as excitatory interactions, similar to neurons firing after receiving stimulation. For example, frequent particle collisions at high flow rates trigger strong excitatory signals. Agglomeration events are simulated as inhibitory interactions, similar to a decrease in neuronal potential. For example, when particles aggregate into larger clusters, their motor activity decreases.

[0053] Particle aggregation probability and settling paths are direct outputs of the model. The aggregation probability indicates the likelihood that particles will aggregate into larger clusters. For example, fine particles may achieve an 85% aggregation rate under the influence of an electric field. This probability reflects the particle's tendency to settle, affecting separation efficiency. The settling path describes the particle's trajectory within the cyclone, e.g., 80% of particles settle to the underflow outlet or discharge overflow. By simulating the dynamic interactions between neurons, the model can rapidly calculate these outputs based on real-time sediment property data.

[0054] The above step 104, namely "predicting the particle distribution and sedimentation trend within a predetermined future time period based on the agglomeration probability and sedimentation path and the sediment characteristic data" is described in detail below with reference to an embodiment.

[0055] This step leverages the output of the neurodynamic model, combined with real-time data, to predict the future behavior of sediment particles within the hydrocyclone using a predictive model. Particle distribution refers to the spatial variation in particle concentration within the hydrocyclone; for example, the concentration at the outer wall may increase to 6 grams per liter. Sedimentation trends describe changes in particle settling velocity and proportion, such as a predicted settling velocity of 0.1 meters per second. This prediction can proactively identify changes in operating conditions, such as an increase in the proportion of fine particles that could lead to a decrease in settling efficiency, thereby guiding adjustments to the electric field strength.

[0056] This prediction process can be achieved in a variety of ways. For example, by using a neural network based on an attention mechanism to assign different attention weights to the input data, the features that have a greater impact on particle distribution and sedimentation trends can be highlighted; by using a probabilistic model based on Bayesian inference to infer the probability distribution of future particle behavior, etc.

[0057] Preferably, the present application uses a long short-term memory network for prediction, inputs 5 seconds of historical data of sediment characteristics data and the agglomeration probability of the neural dynamics model; predicts the particle concentration field and sedimentation velocity in the next 5-10 seconds, and generates prediction results of particle distribution and sedimentation trend.

[0058] A long short-term memory (LSTM) network is a neural network specifically designed to process time series data. It can capture long-term dependencies in data, such as concentration trends. By analyzing sediment characteristics and agglomeration probability over the past five seconds, the model can predict particle distribution and settling trends within the cyclone, ensuring the system adapts promptly to the dynamic conditions of tunnel sewage.

[0059] The LSTM network analyzes input data to predict the particle concentration field and settling velocity for the next 5 to 10 seconds. The particle concentration field describes the spatial distribution of particles within the cyclone. For example, the concentration at the outer wall may increase to 6 grams per liter, indicating a tendency for particles to settle toward the underflow outlet. The settling velocity, which measures the rate at which particles sink, is, for example, 0.1 meters per second, reflecting the combined effects of the electric field and flow field on particle motion. The 5- to 10-second prediction period matches the physical timescale of particle settling within the cyclone. For example, fine particles typically take several seconds to travel from the inlet to the outer wall.

[0060] The above step 105, i.e., "generating a continuous electric field spectrum using a deep reinforcement learning model based on the prediction results of the particle distribution and sedimentation trend," is described in detail below in conjunction with an embodiment.

[0061] This process uses the predicted particle behavior to dynamically generate a continuously changing sequence of electric field parameters, called a continuous electric field spectrum, through a deep reinforcement learning algorithm. This sequence is used to adjust the electric field strength and gradient rate in the hydrocyclone. Deep reinforcement learning is an intelligent decision-making algorithm that can learn optimal strategies through trial and error in complex environments. In the present invention, the algorithm uses particle distribution and sedimentation trends as input, with the goal of generating a continuous electric field spectrum, such as a smooth change in electric field strength from 5 kilovolts per meter to 8 kilovolts per meter, and an adjustment of the gradient rate from 50 kilovolts per square meter to 80 kilovolts per square meter. The continuous electric field spectrum forms a smoothly changing control sequence by dynamically adjusting the electric field parameters, which can accurately match the predicted results of particle behavior. For example, when the prediction shows an increase in the proportion of fine particles, the electric field spectrum may briefly increase in intensity to 8 kilovolts per meter to promote agglomeration, and then drop to 4 kilovolts per meter to save energy. This dynamic adjustment significantly improves the efficiency of fine particle separation while optimizing energy consumption.

[0062] The algorithm evaluates the effects of different electric field parameters by defining a reward mechanism, such as rewarding high separation efficiency and low energy consumption. For example, an electric field spectrum that improves separation efficiency from 70% to 84% while reducing energy consumption to 85 watts would receive a high reward.

[0063] The deep reinforcement learning model of the present application is obtained through pre-training. As an implementable method, the training process of the present application includes: using the simulation data and the sediment characteristic data to construct a first reward function, which comprehensively considers separation efficiency, energy consumption and electrode wear; through iterative optimization, generating a continuous electric field spectrum that adapts to sediment particle size changes from 1μm to 100μm and concentration changes from 0.1 g / L to 10 g / L.

[0064] Specifically, the training process first constructs a reward function using simulation data and sediment property data. The simulation data comes from the computational fluid dynamics discrete element method and provides detailed information such as particle trajectories, collision forces, agglomeration rates, and sedimentation velocities. For example, the probability of agglomeration is 85% under conditions of a particle size of 10 microns and a concentration of 5 grams per liter. Sediment property data includes historical or experimentally measured particle size, concentration, and flow rate, such as a concentration of 2 grams per liter and a flow rate of 2 meters per second. Together, these data describe the particle behavior and separation effect within the cyclone. The first reward function comprehensively considers three optimization objectives: separation efficiency, such as achieving an 84% sediment removal rate; energy consumption, such as maintaining a low power of 85 watts; and electrode wear, such as an electrode mass loss rate of less than 5%. The reward function may be set as a weighted sum of efficiency, energy consumption, and wear, for example, 60% weighting efficiency, 30% weighting energy consumption, and 10% weighting wear, to balance performance and equipment life.

[0065] Furthermore, the first reward function in this application can be expressed as: R = w1*S + w2 * (1-E / Emax) + w3*(1 - W / Wmax) Where R is the reward value, S represents the separation efficiency, for example, 84% sediment removal rate; E is the actual energy consumption, for example, 85 watts, and Emax is the maximum energy consumption reference value, for example, 100 watts; W is the electrode wear rate, for example, 5% mass loss, and Wmax is the maximum wear rate reference value, for example, 10%; w1, w2, and w3 are weight coefficients, for example, w1=0.6, w2=0.3, and w3=0.1, and the sum is 1.

[0066] Through iterative optimization, the deep reinforcement learning model learns to generate a continuous electric field spectrum. The iterative optimization uses a proximal strategy optimization algorithm to adjust the model parameters through thousands of cycles to select the best electric field parameters under different working conditions. For example, under high-concentration conditions, the model may generate a high-intensity electric field such as 10 kilovolts per meter to improve efficiency, while reducing it to 0.1 kilovolts per meter at low concentrations to save energy. The training covers a range of particle sizes from 1 micron to 100 microns and concentrations from 0.1 grams per liter to 10 grams per liter, ensuring that the model can adapt to the dynamic changes in tunnel sewage, such as sudden influx of fine particles. The optimized model is able to generate a continuous electric field spectrum, such as smoothly adjusting the intensity and gradient rate within 10 seconds, keeping the separation efficiency stable at more than 84%, while reducing energy consumption to less than 85 watts.

[0067] As an implementable approach, a deep reinforcement learning model uses a proximal policy optimization algorithm in the model inference stage, with the weighted sum of separation efficiency and energy consumption as the reward function; based on the predicted results of the particle distribution and sedimentation trend and real-time sediment characteristic data, a continuous electric field spectrum is generated, which includes any combination of electric field strength in the range of 0.1-10 kV / m and gradient rate in the range of 0-100 kV / m².

[0068] The proximal policy optimization algorithm is an efficient reinforcement learning method that can stably learn control policies in complex environments. It iteratively optimizes the strategy, balancing the exploration of new parameters with the utilization of existing knowledge to ensure that the generated electric field spectrum is both effective and stable. The second reward function, the core of the algorithm, is defined as the weighted sum of separation efficiency and energy consumption. For example, 80% of the weight is allocated to separation efficiency and 20% to minimizing energy consumption. Separation efficiency and energy consumption are two key performance indicators used by deep reinforcement learning models to optimize continuous electric field spectra. They are used to measure the effectiveness and economic efficiency of hydrocyclones for sediment separation in tunnel sewage treatment. Separation efficiency refers to the ability of a hydrocyclone to remove sediment from sewage, typically expressed as the percentage of sediment mass collected at the underflow outlet to the total sediment mass. Energy consumption refers to the power consumed by the electric field generator and other equipment during cyclone operation, expressed in watts. This design incentivizes the model to minimize power consumption for electric field operation while improving sediment removal efficiency. For example, when predictions indicate an increase in fine particle concentration, the model may select a higher electric field intensity to improve efficiency while avoiding unnecessary energy waste.

[0069] Inputs to generating a continuous electric field spectrum include predicted particle distribution and settling trends, as well as real-time sediment property data. Particle distribution predictions may indicate concentration changes in the outer wall region, suggesting the need to increase the electric field to promote agglomeration; settling trends may indicate settling velocity, suggesting adjustments to the gradient rate to accelerate settling. Real-time sediment property data provides dynamic information about the current operating conditions. The model integrates these inputs to generate an electric field spectrum, for example, gradually ramping the electric field strength from 5 kilovolts per meter to 8 kilovolts per meter over 10 seconds, forming a smooth control sequence. This continuity ensures that the electric field synchronizes with the dynamic changes in particle behavior, overcoming the hysteresis associated with traditional fixed electric field control.

[0070] The output range of the continuous electric field spectrum is clearly defined, with field strengths ranging from 0.1 to 10 kilovolts per meter and gradients from 0 to 100 kilovolts per square meter, covering the typical requirements of tunnel wastewater treatment. Low field strengths are suitable for low-concentration conditions, for example, 0.1 kilovolts per meter can treat wastewater with a concentration of 0.1 grams per liter; high field strengths address high concentrations or fine particles, such as 10 kilovolts per meter promoting agglomeration of particles as small as 5 microns. The range of gradients allows for spatial variations in the electric field within the cyclone, for example, a high gradient of 60 kilovolts per square meter enhances axial sedimentation.

[0071] This application can also perform anomaly detection based on the prediction of particle distribution and sedimentation trends. Based on the predicted results of the particle distribution and sedimentation trends, it detects abnormal changes in sediment characteristic data, such as sediment concentration exceeding 10 g / L or particle size suddenly changing to less than 5 μm. In response to such abnormal changes, a protection mode is triggered, reducing the electric field strength to a predetermined safe range.

[0072] Detection of abnormal changes focuses on two key indicators of sediment characteristics: concentration and particle size. A sediment concentration exceeding 10 grams per liter indicates an abnormally high sediment content in the wastewater, possibly due to tunnel construction blasting or silt release. This high concentration increases the complexity of particle collisions and agglomeration within the cyclone, potentially leading to electric field overload or reduced separation efficiency. A sudden change in particle size below 5 microns indicates a surge in the proportion of fine particles, such as the release of large amounts of fine sediment due to construction disturbances. Fine particles are difficult to settle using conventional centrifugal forces and require a stronger electric field to promote agglomeration. However, excessively high electric fields can cause electrode overheating or a surge in energy consumption. The detection process utilizes predicted results and real-time data to identify anomalies using preset thresholds, such as a concentration threshold of 10 grams per liter or a particle size threshold of 5 microns, ensuring timely detection of potential risks.

[0073] First, the particle distribution and sedimentation trends predicted by the long short-term memory network are used as a reference benchmark, for example, the outer wall concentration will increase to 6 grams per liter and the sedimentation rate will be 0.1 meters per second in the next 5 seconds. Real-time sediment characteristic data, including particle size, concentration, and flow rate, are collected by sensors at a frequency of 10 times per second. Using a sliding window, for example, 30 data points in the past 3 seconds, the mean and standard deviation of the concentration are calculated, for example, a mean of 5 grams per liter and a standard deviation of 0.5 grams per liter. If the current concentration exceeds the preset threshold of 10 grams per liter, or deviates from the mean by 3 times the standard deviation, for example, greater than 6.5 grams per liter, it is marked as a concentration anomaly. Particle size anomaly detection is similar, calculating the particle size distribution within the sliding window, for example, a mean of 10 microns. If the current particle size is less than 5 microns or the proportion suddenly increases by more than 30%, it is marked as a particle size anomaly.

[0074] When the concentration is detected to be over 10 grams per liter or the particle size is less than 5 microns, the system automatically switches to protection mode and reduces the electric field strength to a predetermined safety range, for example, from 8 kilovolts per meter to 2 kilovolts per meter. This safety range is preset according to the equipment design and operating requirements, aiming to reduce the load on the electric field generator and prevent increased electrode wear or excessive energy consumption. For example, reducing the electric field strength under high concentration conditions can avoid electrode overheating while maintaining basic separation efficiency. The protection mode may also suspend high gradient rates, for example, from 60 kilovolts per square meter to 20 kilovolts per square meter, to stabilize the operation of the cyclone. This response mechanism is implemented through edge computing devices, and the detection and adjustment time is usually within 50 milliseconds, meeting real-time control requirements.

[0075] The above step 106, namely "dynamically adjusting the electric field intensity and gradient rate parameter values ​​of the hydrocyclone according to the continuous electric field spectrum", is described in detail below with reference to an embodiment.

[0076] By applying the electric field spectrum to the electric field generator of the cyclone, this system can precisely adjust the electric field strength and gradient rate, enhancing fine particle agglomeration and sedimentation, and achieving efficient separation. The present invention achieves smooth regulation through a continuous electric field spectrum, and the regulation is controlled by the partitioned electrodes through edge computing devices.

[0077] The electric field strength indicates the magnitude of the electric field force and reflects the driving force of the electric field applied to the sediment particles. In the present invention, the electric field strength ranges from 0.1 to 10 kilovolts per meter and is applied through partitioned electrodes, such as 6 kilovolts per meter for the conical section and 4 kilovolts per meter for the cylindrical section. A higher electric field strength, such as 8 kilovolts per meter, can enhance the agglomeration of fine particles. For example, the probability of agglomeration of particles with a particle size of 5 microns increases from 70% to 85%, thereby promoting sedimentation to the bottom flow outlet. Lower strengths, such as 0.1 kilovolts per meter, are suitable for low-concentration conditions, such as 0.1 grams per liter, saving energy.

[0078] The gradient rate represents the rate of change of the electric field intensity in space, with the unit of kilovolts per square meter, reflecting the uneven distribution of the electric field in the cyclone. In the present invention, the gradient rate ranges from 0 to 100 kilovolts per square meter, which is achieved by adjusting the voltage difference between the electrodes. For example, the voltage of the conical section is 1500 volts and the cylindrical section is 800 volts, forming a gradient of 60 kilovolts per square meter. A high gradient rate enhances the axial and radial movement of particles. For example, it accelerates the sedimentation of particles with a particle size of 10 microns, and the sedimentation rate increases from 0.05 meters per second to 0.1 meters per second. The gradient rate is controlled by a continuous electric field spectrum, for example, it gradually changes from 40 kilovolts per square meter to 60 kilovolts per square meter within 10 seconds to adapt to high-concentration working conditions. A low gradient rate, such as 20 kilovolts per square meter, is used in protection mode to prevent electrode overload.

[0079] The above-mentioned method provided in the embodiments of this application can be applied to a variety of application scenarios, including but not limited to: tunnel construction wastewater treatment, mine tailings treatment, river dredging projects, and industrial wastewater treatment. Through real-time data-driven, intelligent prediction, and dynamic electric field regulation, efficient sediment separation is achieved in tunnel construction, mine tailings, river dredging, and industrial wastewater treatment. This provides an intelligent solution for environmental protection projects, significantly improves economic benefits and environmental compliance, and has strong potential for industrial promotion.

[0080] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0081] According to an embodiment of another aspect, a neural dynamics-based electric field adaptive sediment flow regulation system is provided. Figure 2 FIG. 1 shows a schematic block diagram of the electric field adaptive sediment flow regulation system based on neural dynamics according to one embodiment. Figure 2 As shown, the device 200 includes: The real-time data collection unit 201 is configured to collect sediment characteristic data in sewage in real time, wherein the sediment characteristic data includes sediment particle size, concentration and flow velocity.

[0082] The dynamic model training unit 202 is configured to generate simulation data for the working scenario parameters of the hydrocyclone according to the computational fluid dynamics discrete element method, and use the simulation data to train the neural dynamic model.

[0083] The dynamic model inference unit 203 is configured to use the neural dynamic model to model the movement of sediment particles as a neural dynamic system according to the sediment characteristic data, and generate particle agglomeration probability and sedimentation path.

[0084] The prediction unit 204 is configured to predict the particle distribution and sedimentation trend within a predetermined time period in the future based on the agglomeration probability and sedimentation path and the sediment characteristic data.

[0085] The electric field spectrum generating unit 205 is configured to generate a continuous electric field spectrum using a deep reinforcement learning model according to the prediction results of the particle distribution and sedimentation trend.

[0086] The electric field regulating unit 206 is configured to dynamically regulate the electric field intensity and gradient rate parameter values ​​of the hydrocyclone according to the continuous electric field spectrum.

[0087] As an implementable manner, the dynamic model training unit 202 can be configured as the neural dynamic model adopts a pulse neural network model, inputs particle trajectory data simulated based on the computational fluid dynamics discrete element method, and the particle trajectory data includes particle position, velocity, collision force and electric field force; by optimizing the network parameters of the neural dynamic model, the particle agglomeration probability and sedimentation path are generated; wherein, the optimization aims to minimize the prediction error of the agglomeration probability and sedimentation path.

[0088] As an implementable manner, the dynamic model training unit 202 can be configured as the pulse neural network model including a leakage integration and firing neuron model, wherein the excitatory interaction triggers neuron firing based on the particle collision frequency, and the inhibitory interaction reduces the neuron potential based on the particle agglomeration event; according to the real-time collected sediment characteristic data, the weights of the excitatory interaction and the inhibitory interaction are dynamically adjusted, and the weights are associated with changes in sediment particle size and concentration.

[0089] As an implementable method, the prediction unit 204 predicts the particle distribution and sedimentation trend within a predetermined time period in the future, including: based on the 5-second historical data of the sediment characteristic data and the agglomeration probability of the neural dynamics model, using the long short-term memory network to predict the particle concentration field and sedimentation velocity in the next 5-10 seconds, and generating a prediction result of the particle distribution and sedimentation trend.

[0090] As an implementable approach, the training process of the deep reinforcement learning model includes: using the simulation data and the sediment characteristic data to construct a first reward function, which comprehensively considers separation efficiency, energy consumption and electrode wear; through iterative optimization, generating a continuous electric field spectrum that adapts to sediment particle size changes from 1 micron to 100 microns and concentration changes from 0.1 g / L to 10 g / L.

[0091] As an implementable manner, the dynamic model reasoning unit 203 can be configured as follows when generating a continuous electric field spectrum using a deep reinforcement learning model: the deep reinforcement learning model uses a proximal policy optimization algorithm with a weighted sum of separation efficiency and energy consumption as the second reward function; based on the predicted results of the particle distribution and sedimentation trend and the real-time sediment characteristic data, a continuous electric field spectrum is generated, and the continuous electric field spectrum includes any combination of electric field strength in the range of 0.1 to 10 kV / m and gradient rate in the range of 0 to 100 kV / m².

[0092] As an practicable manner, the device may further include an anomaly detection unit 207, which is configured to: detect abnormal changes in sediment characteristic data based on the predicted results of the particle distribution and sedimentation trend, the abnormal changes including sediment concentration exceeding 10 g / L or a sudden change in particle size to less than 5 microns; in response to the abnormal changes, trigger a protection mode and reduce the electric field strength to a predetermined safety range.

[0093] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For the relevant parts, refer to the partial description of the method embodiment. The system embodiment described above is only exemplary, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without making any creative efforts.

[0094] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0095] In addition, an embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of any one of the methods in the aforementioned method embodiments are implemented.

[0096] And an electronic device comprising: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, which, when read and executed by the one or more processors, execute the steps of the method described in any one of the aforementioned method embodiments.

[0097] The present application also provides a computer program product, comprising a computer program, which implements the steps of any one of the methods described in the aforementioned method embodiments when executed by a processor.

[0098] in, Figure 3 The electronic device architecture is shown as an example, and may include a processor 310, a video display adapter 311, a disk drive 312, an input / output interface 313, a network interface 314, and a memory 320. The processor 310, the video display adapter 311, the disk drive 312, the input / output interface 313, the network interface 314, and the memory 320 may be communicatively connected via a communication bus 330.

[0099] The processor 310 may be implemented as a general-purpose CPU, a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and may be used to execute relevant programs to implement the technical solutions provided in this application.

[0100] The memory 320 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 320 can store an operating system 321 for controlling the operation of the electronic device 300, and a basic input and output system (BIOS) 322 for controlling the low-level operations of the electronic device 300. In addition, a web browser 323, a data storage management system 324, and a neurodynamics-based electric field adaptive sediment fluid regulation system 325, etc. can also be stored. The above-mentioned neurodynamics-based electric field adaptive sediment fluid regulation system 325 can be an application program that specifically implements the operations of the aforementioned steps in the embodiment of the present application. In short, when the technical solution provided by the present application is implemented by software or firmware, the relevant program code is stored in the memory 320 and is called and executed by the processor 310.

[0101] The input / output interface 313 is used to connect to input / output modules to enable information input and output. The input / output modules can be configured as components within the device (not shown) or externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, and various sensors. Output devices may include a display, speaker, vibrator, indicator light, and the like.

[0102] The network interface 314 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.).

[0103] The bus 330 comprises a pathway for transmitting information between the various components of the device (eg, the processor 310 , the video display adapter 311 , the disk drive 312 , the input / output interface 313 , the network interface 314 , and the memory 320 ).

[0104] It should be noted that although the above device only shows the processor 310, video display adapter 311, disk drive 312, input / output interface 313, network interface 314, memory 320, bus 330, etc., in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may also include only the components necessary to implement the solution of the present application, and does not necessarily include all the components shown in the figure.

[0105] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer program product. The computer program product can be stored in a storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0106] The above is a detailed introduction to the technical solutions provided by this application. Specific examples are used herein to illustrate the principles and implementation methods of this application. The description of the above embodiments is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the contents of this specification should not be understood as limiting this application.

Claims

1. A method for regulating sediment flow in an electric field adaptively based on neural dynamics, used for sediment separation in a hydrocyclone, characterized in that: The method comprises: Real-time collection of sediment characteristic data in sewage, including sediment particle size, concentration, and flow rate; generating simulation data for the working scenario parameters of the hydrocyclone according to the computational fluid dynamics discrete element method, and training the neural dynamics model using the simulation data; Using the neural dynamics model, the movement of sediment particles is modeled as a neural dynamics system based on the sediment characteristic data to generate particle agglomeration probability and sedimentation path; Predicting particle distribution and sedimentation trends within a predetermined time period in the future based on the particle agglomeration probability, the sedimentation path, and the sediment characteristic data; generating a continuous electric field spectrum using a deep reinforcement learning model based on the predicted results of the particle distribution and sedimentation trend; The electric field intensity and gradient rate parameter values ​​of the hydrocyclone are dynamically adjusted according to the continuous electric field spectrum.

2. The method for electric field adaptive sediment flow regulation based on neural dynamics according to claim 1 is characterized in that: The training of the neural dynamics model using the simulated data comprises: The neural dynamics model adopts a pulse neural network model and inputs particle trajectory data simulated based on the discrete element method of computational fluid dynamics, wherein the particle trajectory data includes particle position, velocity, collision force and electric field force; By optimizing the network parameters of the neural dynamics model, the particle agglomeration probability and the sedimentation path are generated; wherein the optimization aims to minimize the prediction error of the particle agglomeration probability and the sedimentation path.

3. The method for electric field adaptive sediment flow regulation based on neural dynamics according to claim 2 is characterized in that: The spiking neural network model includes a leaky integrate-and-spark neuron model, wherein the excitatory interaction triggers neuronal firing based on particle collision frequency, and the inhibitory interaction reduces neuronal potential based on particle aggregation events; The weights of the excitatory interaction and the inhibitory interaction are dynamically adjusted according to the sediment characteristic data collected in real time, and the weights are associated with changes in sediment particle size and concentration.

4. The method for electric field adaptive sediment flow regulation based on neural dynamics according to claim 1 is characterized in that: The prediction of particle distribution and sedimentation trend within a predetermined time period in the future includes: Based on the 5-second historical data of the sediment characteristic data and the particle agglomeration probability of the neural dynamics model, the long short-term memory network is used to predict the particle concentration field and sedimentation velocity in the next 5-10 seconds, and the prediction results of particle distribution and sedimentation trend are generated.

5. The method for electric field adaptive sediment flow regulation based on neural dynamics according to claim 1 is characterized in that: The training process of the deep reinforcement learning model includes: constructing a first reward function using the simulation data and the sediment characteristic data, the first reward function comprehensively considering separation efficiency, energy consumption, and electrode wear; Through iterative optimization, a continuous electric field spectrum was generated that adapts to changes in sediment particle size from 1 μm to 100 μm and concentration from 0.1 g / L to 10 g / L.

6. The method for electric field adaptive sediment flow regulation based on neural dynamics according to claim 1 is characterized in that: The method of using a deep reinforcement learning model to generate a continuous electric field spectrum includes: The deep reinforcement learning model uses a proximal policy optimization algorithm with a weighted sum of separation efficiency and energy consumption as the second reward function; Based on the predicted results of the particle distribution and sedimentation trend and the real-time sediment characteristic data, a continuous electric field spectrum is generated, wherein the continuous electric field spectrum includes any combination of electric field strengths in the range of 0.1 to 10 kV / m and gradient rates in the range of 0 to 100 kV / m².

7. The method for electric field adaptive sediment flow regulation based on neural dynamics according to claim 1, characterized in that: The method further comprises: detecting abnormal changes in sediment characteristic data based on the predicted results of the particle distribution and sedimentation trend, wherein the abnormal changes include sediment concentration exceeding 10 g / L or particle size suddenly changing to less than 5 microns; In response to the abnormal change, a protection mode is triggered to reduce the electric field strength to a predetermined safety range.

8. A neurodynamics-based electric field adaptive sediment flow regulation system for sediment separation in a hydrocyclone, characterized in that: The system comprises: A real-time data acquisition unit is configured to acquire sediment characteristic data in the sewage in real time, wherein the sediment characteristic data includes sediment particle size, concentration and flow rate; a dynamic model training unit configured to generate simulation data for the working scenario parameters of the hydrocyclone according to a computational fluid dynamics discrete element method, and train a neural dynamic model using the simulation data; a dynamic model inference unit configured to utilize the neural dynamic model to model the movement of sediment particles as a neural dynamic system based on the sediment characteristic data, and generate particle agglomeration probability and sedimentation path; a prediction unit configured to predict particle distribution and sedimentation trends within a predetermined time period in the future based on the particle agglomeration probability, the sedimentation path, and the sediment characteristic data; an electric field spectrum generating unit configured to generate a continuous electric field spectrum using a deep reinforcement learning model based on the prediction results of the particle distribution and sedimentation trend; The electric field regulating unit is configured to dynamically regulate the electric field intensity and gradient rate parameter values ​​of the hydrocyclone according to the continuous electric field spectrum.

9. An electronic device, characterized in that: include: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, wherein the program instructions, when read and executed by the one or more processors, execute the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images