Artificial wetland process design system and method based on neural network multi-boundary condition

Abstract

The invention discloses an artificial wetland process design system and method based on a neural network multi-boundary condition. The system comprises a data preprocessing module, a neural network model construction module, a multi-boundary constraint processing module and an optimal solution output module. Wherein the multi-boundary constraint processing module uses a Lagrange multiplier method and a penalty function method to manage constraint conditions of the constructed wetland involved in the optimization process; the optimal solution output module is used for randomly generating an initial constructed wetland process combination, inputting the initial constructed wetland process combination into the trained deep feed-forward neural network, and calculating a fitness value of the initial constructed wetland process combination after key performance indexes of operation of the constructed wetland are obtained; and on the basis of the fitness values and constraint conditions of the constructed wetland, correcting the process combination which does not meet the constraint conditions by adopting an evolutionary algorithm, and outputting the process combination with the highest fitness as an optimal solution after the optimization process is ended. The result obtained by the method is high in accuracy.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence technology and ecological environment engineering, and particularly relates to an artificial wetland process design system and method based on neural networks under multiple boundary conditions. Background Art

[0002] As an ecological sewage treatment technology, artificial wetlands are widely used in the treatment of domestic sewage, rural sewage, and some industrial wastewater due to their low cost, low energy consumption, simple maintenance, and significant landscape benefits. However, in the design and operation of artificial wetlands, they are often restricted by multi-dimensional and multi-constraint conditions, and these boundary conditions have an important impact on the treatment efficiency and operating cost of the wetland system. Regarding the design optimization problem of multiple boundary conditions in artificial wetland design, there are still many deficiencies in the existing technologies.

[0003] I. The multiple boundary conditions in the artificial wetland process design have the following characteristics

[0004] 1. Fluctuations in influent water quality and quantity

[0005] Artificial wetlands need to treat sewage from different sources, and its influent water quality (water quantity, COD concentration, ammonia nitrogen concentration, etc.) may change significantly with seasonal variations or fluctuations in pollution sources. This dynamic nature poses severe challenges to the microbial metabolism in the wetland, the plant absorption capacity, and the overall water quality purification effect. Traditional designs usually adopt fixed parameters, making it difficult to adapt to the actual influent water quality and quantity fluctuations in artificial wetlands, resulting in a decline in operating efficiency.

[0006] 2. Diversity of environmental conditions

[0007] The operating efficiency of artificial wetlands is significantly affected by environmental conditions (such as air temperature, humidity, precipitation). Especially in low-temperature environments, the metabolic rate of microorganisms decreases, and the treatment efficiency of the wetland drops significantly. How to operate stably under different climate conditions is a key issue to be solved in wetland design.

[0008] 3. Complex coupling among multiple parameters

[0009] The wetland process design involves multiple parameters, such as wetland area, types of bed fillers, plant types, flow velocity, etc. There is a highly non-linear coupling relationship among these parameters. For example, the flow guiding performance of the filler and the root distribution of the plant jointly affect the water flow distribution, and the water flow velocity in turn affects the degradation efficiency of pollutants. This complex coupling relationship poses challenges to the optimization design of wetlands.

[0010] 4. Diversification of external boundary conditions

[0011] The operation of constructed wetlands is not only restricted by internal parameters but also needs to meet external conditions (such as the effluent standards required by regulations, the spatial limitations of wetland sites, and economic cost constraints). The limitations of multi-dimensional boundary conditions increase the complexity of wetland optimization design.

[0012] II. Multifaceted Requirements for Process Optimization

[0013] 1. Treatment Efficiency and Stability

[0014] The primary goal of constructed wetland optimization is to improve the removal efficiency of pollutants (such as COD, ammonia nitrogen, total phosphorus, etc.) and maintain the operational stability of the system under dynamic conditions. For example, under high water volume or low temperature conditions, how to maintain stable effluent quality by adjusting wetland structure or operating parameters is an important issue.

[0015] 2. Balance between Energy Consumption and Economic Costs

[0016] As a low-energy consumption treatment technology, the optimization of constructed wetlands not only needs to focus on treatment efficiency but also needs to minimize operating energy consumption and construction costs as much as possible. In particular, the use of measures such as intelligent control and auxiliary heating requires a balance between energy efficiency and cost.

[0017] 3. Synergistic Optimization among Multiple Parameters

[0018] Multiple parameters in wetland process design do not exist independently but interact with each other. For example, increasing the wetland area can improve treatment capacity but may also lead to excessive construction costs; selecting high-thermal conductivity fillers can improve treatment efficiency under low temperature conditions, but the filler price may increase significantly. Therefore, the optimization design needs to find the optimal solution that meets multiple constraint conditions under the synergistic action of multiple parameters.

[0019] 4. Adaptability to Dynamic Environments

[0020] During the operation of wetlands, environmental conditions and influent water quality are constantly changing, and traditional design and optimization methods based on fixed parameters are difficult to adjust in real time. How to use dynamic data and real-time monitoring results to adaptively adjust wetland process parameters has become an important direction for improving wetland operation efficiency.

[0021] 5. Global Optimization Ability

[0022] Current wetland optimization methods mostly adopt experience-based linear models or traditional mathematical methods, which are unable to effectively handle the non-linear coupling of multiple parameters and global search problems. For example, design based on empirical formulas can only provide local optimization solutions and is difficult to meet the global optimization requirements under complex boundary conditions.

[0023] III. Deficiencies in Current Constructed Wetland Design Optimization

[0024] 1. Limitations of the optimization method: Traditional optimization methods all adopt general design specifications for constructed wetlands to design process parameters, lacking the ability to handle complex non-linear problems. Especially when facing high-dimensional multi-constraint problems, it is difficult to find the global optimal solution.

[0025] 2. Lack of dynamic adaptability: Most wetland designs are based on static models, lacking the real-time response ability to dynamic operating environments and being difficult to adapt to seasonal changes or sudden sewage shocks.

[0026] 3. Insufficient intelligence: Existing optimization methods lack the intelligent processing and learning ability for real-time data during the operation of wetlands, and it is difficult to rapidly iterate and optimize under multi-dimensional constraints. Summary of the Invention

[0027] In view of the deficiencies of the prior art, the present invention proposes a constructed wetland process design system and method under multi-boundary conditions based on neural networks. By combining the non-linear modeling ability of neural networks, the adaptive optimization characteristics of machine learning with the complex process design requirements of constructed wetlands, it can quickly solve the global optimal solution under multi-boundary constraint conditions, providing an intelligent and dynamic design and operation plan for constructed wetlands.

[0028] The object of the present invention is achieved by the following technical solutions:

[0029] A constructed wetland process design system under multi-boundary conditions based on neural networks, comprising the following modules:

[0030] A data preprocessing module, which uses the principal component analysis method to extract features from multi-dimensional data related to the design and operation of constructed wetlands, removes redundant information, and reduces data complexity; and normalizes the data with different dimensions after feature extraction;

[0031] A neural network model construction module, which is used to construct a deep feedforward neural network including an input layer, a hidden layer, and an output layer. The input of the deep feedforward neural network is the data normalized by the data preprocessing module, and the output is the key performance indicators of the operation of the constructed wetland;

[0032] A multi-boundary constraint processing module, which uses the Lagrange multiplier method and the penalty function method to manage the constraint conditions of the constructed wetland involved in the optimization process, ensuring that the solution result not only meets the constraint conditions but also can achieve the global optimal solution;

[0033] The constraint conditions include influent water quality constraints, environmental condition constraints, operation economy constraints, effluent standard constraints, and space constraints;

[0034] The optimized solution output module is used to randomly generate an initial constructed wetland process combination and input it into the trained deep feedforward neural network. Based on the key performance indicators of the operation of the constructed wetland output by the trained deep feedforward neural network, calculate the fitness value of the randomly generated initial constructed wetland process combination; based on these fitness values and the constraint conditions of the constructed wetland, use an evolutionary algorithm to correct the process combinations that do not meet the constraint conditions. When the fitness change rate of the process combination reaches the set threshold or the maximum number of iterations is reached, the optimization process terminates, and the process combination with the highest fitness is output as the optimal solution.

[0035] Further, the Lagrangian function in the Lagrange multiplier method is as follows:

[0036]

[0037] Among them, f(x) is the objective function that fuses several performance indicators, ɡ i (x) represents the constraint condition, λ i is the Lagrange multiplier, and m is the number of boundary conditions;

[0038] The penalty function in the penalty function method is as follows:

[0039]

[0040] Among them, F(x) represents the objective function corrected by the penalty function, P is the penalty factor, and the more times the constraint is violated or the greater the deviation from the range, the higher the penalty value.

[0041] Further, the calculation formula of the fitness is as follows:

[0042] Fitness = w1·(COD removal rate) + w2·(ammonia nitrogen removal rate) - w3·(operating cost)

[0043] Among them, w1, w2, and w3 are the weight coefficients of the COD removal rate, ammonia nitrogen removal rate, and operating cost, respectively.

[0044] Further, the deep feedforward neural network includes three hidden layers. The first hidden layer includes 128 nodes, the second hidden layer includes 64 nodes, and the third hidden layer includes 32 nodes.

[0045] Further, the key performance indicators of the operation of the constructed wetland output by the trained deep feedforward neural network include the COD removal rate, ammonia nitrogen removal rate, total phosphorus removal rate, and total operating cost.

[0046] Further, the influent water quality constraints include the maximum and minimum values of the pollutant concentrations including COD, ammonia nitrogen, and total phosphorus;

[0047] The environmental condition constraints include the temperature range tolerated by the wetland system;

[0048] The operating economy constraints include the upper limits of operating energy consumption and total construction cost;

[0049] The effluent standard constraints are the effluent quality indicators stipulated by national or regional regulations;

[0050] The space constraints are the upper limit of the wetland area.

[0051] A method for designing an artificial wetland process under multi-boundary conditions based on a neural network includes the following steps:

[0052] S1: Collect data related to the design and operation of the artificial wetland, including influent water quality parameters, hydraulic parameters, environmental conditions, and wetland structure characteristics; use the principal component analysis method to extract features from these data to extract key influencing parameters; finally, normalize the data with different dimensions of the key influencing parameters extracted by feature extraction.

[0053] S2: Construct a deep feedforward neural network including an input layer, a hidden layer, and an output layer. The input of the deep feedforward neural network is the data normalized in step S1, and the output is the key performance indicators of the operation of the artificial wetland; use the historical data normalized in S1 to train the deep feedforward neural network.

[0054] S3: Randomly generate an initial artificial wetland process combination, and each combination represents a set of wetland design and operation parameter combinations; after performing the same feature extraction and normalization processing on the parameters of the initial artificial wetland process combination as in step S1, input them into the deep feedforward neural network trained in step S2 to obtain the key performance indicators of the operation of the artificial wetland corresponding to each combination.

[0055] S4: Embed multi-boundary constraint conditions into the evolutionary algorithm, introduce the Lagrange multiplier method and the penalty function method to manage the multi-boundary constraint conditions, correct the process combinations that do not meet the constraint conditions to ensure that the optimization process is carried out in the feasible solution space; when the fitness change rate of the process combination is less than the set threshold or reaches the maximum number of iterations, the optimization process terminates; based on the key performance indicators of the operation of the artificial wetland corresponding to each process combination obtained in S3, calculate the fitness value of each process combination obtained at the end of the optimization, evaluate each combination until the termination condition is met, and output the combination with the highest fitness as the optimal design scheme.

[0056] Further, in step S4, heuristic rules and a dynamic constraint adjustment mechanism are introduced to improve the optimization ability of the evolutionary algorithm under multi-boundary constraint conditions: when the iteration of the evolutionary process terminates, if the output optimal solution still does not satisfy all constraint conditions, the penalty factor is adaptively adjusted, the fitness function is optimized, the iteration scale is extended, and the constraint conditions are dynamically adjusted in combination with heuristic strategies to ensure that the finally output process design scheme is optimal in terms of water quality compliance, economic feasibility, and space adaptation, meeting the requirements of engineering practical applications.

[0057] Further, the influent water quality parameters in S1 include COD concentration, ammonia nitrogen concentration, and total phosphorus concentration; the hydraulic parameters include the average daily sewage flow and peak flow; the environmental conditions include the environmental temperature range and temperature control scheme in the area where the wetland is located; the wetland structure characteristics include wetland area, filler thermal conductivity, plant configuration scheme, and water flow velocity in the wetland.

[0058] Further, the key influencing parameters in S1 include wetland area, filler thermal conductivity, plant species, water flow velocity in the wetland, and temperature control scheme.

[0059] The beneficial effects of the present invention are as follows:

[0060] 1. Effective handling of complex multi-boundary conditions

[0061] The present invention combines the Lagrange multiplier method, penalty function method, and dynamic constraint adjustment mechanism, and can efficiently handle complex multi-boundary conditions (such as influent water quality range, environmental temperature fluctuation, effluent standard, and upper limit of operating cost) in constructed wetlands, ensuring the feasibility and accuracy of the optimization results.

[0062] 2. High-precision performance prediction of neural network

[0063] The deep feedforward neural network (DNN) is used to model the non-linear characteristics of the wetland system, significantly improving the fitting ability of the relationship between input parameters and performance indicators, providing fast and high-precision performance evaluation for the optimization process, and reducing the time cost of complex calculations in traditional methods.

[0064] 3. Global optimization ability of evolutionary algorithm

[0065] By simulating the evolutionary process of process combination through the evolutionary algorithm, the global optimal solution is achieved in a large-scale parameter search space, effectively avoiding the defect that traditional optimization methods are prone to falling into local optima.

[0066] 4. Comprehensiveness of multi-objective optimization

[0067] By comprehensively dealing with multi-objective optimization problems (such as improving pollutant removal efficiency, reducing operating costs and energy consumption), combined with a weight adjustment method, the balanced optimization of performance indicators is achieved to meet the diverse needs of different application scenarios.

[0068] 5. Fast Convergence and High Efficiency

[0069] The prediction function of the neural network is combined with the search ability of the evolutionary algorithm, which greatly improves the convergence speed of the optimization process, shortens the solution time, and provides support for rapid decision-making of complex problems in practical engineering.

[0070] 6. Flexibility and Expandability

[0071] The present invention is applicable to constructed wetlands of different scales and types (such as medium-sized wetlands, small decentralized wetlands, etc.), and supports extension to more variables and complex scenarios (such as sewage treatment optimization under low-temperature conditions), having strong engineering application value and promotion potential. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a schematic diagram of the constructed wetland process design system based on neural network with multi-boundary conditions of the present invention.

[0073] Figure 2 It is a schematic diagram of a deep feedforward neural network.

[0074] Figure 3 It is a schematic diagram of three hidden layers of a deep feedforward neural network. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0075] The present invention will be described in detail below according to the drawings and preferred embodiments. The purpose and effects of the present invention will become more apparent. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0076] On the one hand, the constructed wetland process design system based on neural network with multi-boundary conditions of the present invention, as Figure 1 shown, includes the following four modules:

[0077] 1. Data Preprocessing Module

[0078] Feature extraction: Multidimensional data collected related to the design and operation of constructed wetlands, covering all key variables affecting wetland performance, including influent water quality parameters, hydraulic parameters, environmental conditions, and wetland structure characteristics, are used to perform feature extraction on these multidimensional data by using the principal component analysis method to remove redundant information and reduce data complexity.

[0079] Among them, the influent water quality parameters include COD concentration, ammonia nitrogen concentration, and total phosphorus concentration;

[0080] Hydraulic parameters include the average daily sewage flow and peak flow;

[0081] Environmental conditions include the environmental temperature range and temperature control scheme of the area where the wetland is located;

[0082] The wetland structure characteristics include the wetland area, the heat conduction coefficient of the filler, the plant configuration scheme, and the water flow velocity in the wetland.

[0083] Data standardization: Normalize data with different dimensions to improve the convergence speed and prediction accuracy of subsequent model training.

[0084] 2. Neural network model construction module

[0085] The goal of the neural network model construction module of the present invention is to establish a Deep Feedforward Neural Network (DNN), with the complex non-linear relationship between input parameters and wetland operation performance (such as COD removal rate, ammonia nitrogen removal rate, operation cost, etc.) as the core, to establish a high-precision prediction model for predicting the operation performance indicators (such as COD removal rate, ammonia nitrogen removal rate, operation energy consumption, etc.) of the constructed wetland system under different conditions.

[0086] As Figure 2 shown, the deep feedforward neural network includes an input layer, hidden layers, and an output layer. The input layer is used to receive the data normalized by the data preprocessing module; as Figure 3 shown, there are 3 hidden layers, and the ReLU (Rectified Linear Unit) activation function is used to enhance the non-linear fitting ability. The number of nodes in each layer is dynamically adjusted according to the data dimension and task complexity. The output layer outputs the key performance indicators of the operation of the wetland system. When the model is retrained, supervised learning is performed based on historical operation data, and the mean square error (MSE) is used as the loss function, and the model parameters are optimized through the backpropagation algorithm.

[0087] The principle of the hidden layer of the present invention is similar to a multi-layer filter. There are many "nodes" working in each layer, and these "nodes" will extract important information in the data layer by layer. As the information gradually becomes refined, the number of "nodes" required will decrease, so the number of nodes in each layer will gradually decrease. The first layer is the starting stage of receiving and processing, equivalent to a coarse sieve. It is used to comprehensively analyze all input data and extract preliminary features, such as determining which features of the influent water quality may be related to the removal efficiency. The second layer further screens the features extracted by the first layer. It focuses on processing the most valuable information, such as the main influence of the influent COD concentration and environmental temperature on the system operation. The third layer is the refined analysis stage, deeply mining the key patterns in the data, such as finding the influence of the optimal wetland operation conditions (such as the type of filler and plant configuration) on the removal rate.

[0088] The activation function of the present invention adopts the ReLU activation function, and its mathematical expression is: f(x) = max(0, x). ReLU can effectively avoid the vanishing gradient and improve the training efficiency and model performance.

[0089] The present invention adopts the He initialization method to maintain the numerical stability of the forward propagation and realize the initialization of the connection weights. It is a method of assigning a suitable initial value to each "path" (weight) of the neural network. Its main purpose is to enable the information of each layer to be stably transmitted to the next layer when the network starts to learn, without becoming too large or too small. This random value is calculated according to the formula, and the initialization formula is: Among them, n is the number of nodes in the previous layer. Where W: the initial value of the weight, n: the number of nodes in the previous layer. The core idea of the formula is that the more nodes in the previous layer, the smaller the range of the initial value of the weight should be to prevent the information from becoming too strong.

[0090] The output layer of the present invention is the key performance indicators of the wetland operation after the output model operation, including the removal rates of COD, ammonia nitrogen, total phosphorus, the total operating cost, and the recommended best process of the wetland.

[0091] 3. Multi-boundary constraint processing module

[0092] The Lagrange multiplier method and the penalty function method are used to manage the constraint conditions of the constructed wetland involved in the optimization process, ensuring that the solution result not only satisfies the constraint conditions but also can achieve the global optimal solution.

[0093] Among them, the constraint conditions include:

[0094] Influent water quality constraint: the maximum and minimum values of the pollutant concentrations including COD, ammonia nitrogen, and total phosphorus;

[0095] Environmental condition constraint: the temperature range tolerated by the wetland system;

[0096] Operating economy constraint: the upper limits of the operating energy consumption and the total construction cost;

[0097] Effluent standard constraint: meeting the effluent water quality indicators stipulated by national or regional regulations;

[0098] Spatial constraint: the upper limit of the wetland area.

[0099] The multi-boundary constraint processing method is one of the cores of the present invention, which is used to effectively manage the complex constraint conditions of the constructed wetland during the optimization process, ensuring that the solution result not only meets the boundary requirements but also can achieve the global optimal solution.

[0100] 3.1 Lagrange multiplier method: The present invention introduces the constraint conditions of the constructed wetland (such as the influent water quality range, effluent water quality standard, temperature fluctuation range, etc.) into the objective function to construct the Lagrangian function:

[0101]

[0102] Among them, f(x) is a single (merged) objective function or other objective functions that incorporate multiple performance metrics (removal rate, cost, etc.), and ɡ i (x) represents the constraint condition, and λ i is the Lagrange multiplier, and m is the number of constraint conditions. By solving the partial derivatives of the Lagrange function, the optimization process can simultaneously satisfy the optimality of the objective function and the feasibility of the constraint conditions. This method is applicable to linear and non-linear constraints, ensuring that the solution path is always within the feasible solution range and guaranteeing that the optimization result meets the boundary conditions.

[0103] 3.2 Penalty function method:

[0104] In the present invention, a penalty value is set for solutions that do not satisfy the constraints, and a penalty term is added to the objective function to guide the optimization process to converge to the feasible solution range, thereby increasing the cost of violating the constraint conditions. The penalty function form is as follows:

[0105]

[0106] F(x) represents the objective function corrected by the penalty function, also known as the "penalized objective function"; P is the penalty factor, and the more times the constraint is violated or the greater the deviation from the range, the higher the penalty value. By gradually increasing the penalty factor, the algorithm is guided to gradually converge to the solution region that satisfies the constraint conditions.

[0107] 3.3 Fitness function calculation

[0108] To comprehensively consider the various performances of the process design, a fitness function is defined:

[0109] F(x) = w1·(1 - COD out / COD in ) + w2·(1 - Cost / Cost max )

[0110] Among them, COD in represents the chemical oxygen demand concentration in the raw water before entering the constructed wetland system, and COD out represents the COD concentration in the effluent after wetland treatment. Cost represents the actual cost, and Cost max represents the maximum allowable cost, and Cost / Cost max is used to measure the proportion of the cost in the budget. If this ratio is less than 1, it means that the actual cost is within the budget; if it is equal to 1, it just reaches the budget ceiling; if it is greater than 1, it exceeds the allowable cost range; w1 and w2 are weights representing the balance between the removal efficiency and the economic cost.

[0111] Heuristic Rules and Dynamic Constraint Adjustment: The present invention combines the characteristics of constructed wetlands and designs heuristic rules to assist the optimization process.

[0112] (1) Initial Stage:

[0113] Randomly generate initial process combinations, and perform normalization preprocessing and performance prediction on all combinations;

[0114] Set the initial penalty factor P0 to a relatively low value, allowing some solutions to slightly violate the constraints, thereby expanding the search scope;

[0115] At the same time, use heuristic rules to preferentially retain combinations with better performance in key indicators.

[0116] (2) Mid-term Iteration:

[0117] Calculate the fitness function (including the objective function and penalty term) for each generation of solutions, and monitor the constraint violation situation of the current optimal solution;

[0118] If it is found that the optimal solution violates the constraints, dynamically adjust the penalty factor P and the constraint weights according to the degree of violation;

[0119] The adjusted objective function makes the next generation of process combinations more inclined to satisfy all constraint conditions while maintaining a certain global search ability.

[0120] (3) Late Convergence:

[0121] When the fitness change rate drops to the set threshold, it indicates that the algorithm has tended to converge. At this time, the penalty factor P has been adjusted to a sufficiently high level, so that the finally obtained optimal solution satisfies all preset constraint conditions;

[0122] The output design scheme is the global optimal scheme under the comprehensive objective (such as the fitness calculated by weighted COD removal rate, ammonia nitrogen removal rate and operating cost) and multiple constraint conditions.

[0123] Specifically, when the optimal solution output after the iteration termination of the evolutionary process does not meet the preset requirements, the weights and priorities of the constraint conditions can be adjusted according to the actual conditions, and the iterative process of the evolutionary algorithm can be repeated until the optimal solution meets the preset requirements. Through the comprehensive application of heuristic rules and dynamic constraint adjustment mechanisms, the system can fully explore possible process combinations in the initial stage and effectively converge to the optimal solution that meets all constraints in the later stage, ensuring that the engineering design not only achieves the goal of efficient sewage treatment but also meets various limiting conditions such as economy, environment and space.

[0124] Comprehensiveness of multi-objective constraint handling: In multi-objective optimization (such as simultaneously optimizing processing efficiency and operating cost), the present invention incorporates each constraint condition into the objective function in a weighted manner to achieve balanced optimization of multiple objectives. The weighting factor can be dynamically adjusted based on user requirements or the operating environment, thereby achieving comprehensive management of multi-boundary constraints.

[0125] 4. Optimal solution output module

[0126] The optimal solution output module is used to randomly generate an initial constructed wetland process combination and input it into the trained deep feedforward neural network. Based on the key performance indicators of the operation of the constructed wetland output by the trained deep feedforward neural network, calculate the fitness value of the randomly generated initial constructed wetland process combination; based on these fitness values and the constraint conditions of the constructed wetland, use an evolutionary algorithm to correct the process combinations that do not meet the constraint conditions. When the fitness change rate of the process combination reaches the set threshold or the maximum number of iterations is reached, the optimization process terminates, and the process combination with the highest fitness is output as the optimal solution.

[0127] (1) Initial process combination

[0128] Randomly generate an initial constructed wetland process combination, and each combination represents a set of wetland design and operation parameter combinations (such as wetland area, filler type, plant configuration, and flow rate, etc.).

[0129] (2) Fitness evaluation

[0130] Use the trained neural network model to calculate the key performance indicators of the operation of the constructed wetland corresponding to each combination, and further calculate the fitness value based on these key performance indicators. The fitness value reflects the performance of the wetland system under this parameter combination (such as the weighted value of COD removal rate and operating cost).

[0131] (3) Constraint correction

[0132] In the evolution of each generation of process combinations, correct the process combinations that do not meet the boundary conditions, and ensure that the optimization process is always carried out within the feasible solution space through the Lagrange multiplier method and the penalty function method. Use an evolutionary algorithm to iteratively optimize the process combination. During the iteration process, ensure that the solution meets the constraints through the following methods:

[0133] Use L(x,λ) constructed by the Lagrange multiplier method to ensure that the constraint conditions are considered during the solution process;

[0134] Increase the penalty for the solution that violates the constraints through the penalty function method, so that the optimization process tends to the feasible region that satisfies all constraints.

[0135] (4) Termination condition

[0136] When the change rate of the fitness of the process combination is less than the set threshold or the maximum number of iterations is reached, the optimization process terminates, and the process combination with the highest fitness is output as the optimal solution. After the optimization is completed, the combination of wetland design and operation parameters that meet the multi-boundary constraint conditions is output, including:

[0137] Design parameters: wetland area, filler type, plant configuration plan, etc.

[0138] Operation parameters: influent flow rate, operating temperature, energy regulation plan, etc.

[0139] Performance indicators: predicted COD removal rate, ammonia nitrogen removal rate, total operating cost, and energy consumption level.

[0140] The optimization solution output module realizes the seamless cooperation between the prediction function and the evolutionary algorithm:

[0141] (1) Fast evaluation: The efficient prediction ability of the neural network significantly reduces the time complexity of each fitness evaluation, enabling the evolutionary algorithm to quickly iterate in the large-scale parameter search space.

[0142] (2) Real-time feedback: The evolutionary algorithm updates the individual fitness in real time through the neural network prediction results, ensuring that the search direction always converges towards the optimal solution.

[0143] (3) Dynamic update: In a dynamic environment (such as changes in influent water quality and temperature), the dynamic adaptation of the optimization solution is achieved by real-time adjusting the neural network model parameters and constraint condition weights.

[0144] On the other hand, the method for designing an artificial wetland process based on a neural network under multi-boundary conditions of the present invention includes the following steps:

[0145] S1: Collect data related to the design and operation of the artificial wetland, including influent water quality parameters, hydraulic parameters, environmental conditions, and wetland structure characteristics; use the principal component analysis method to extract features from these data and extract the key influencing parameters; finally, normalize the data with different dimensions of the key influencing parameters extracted by feature extraction.

[0146] S2: Construct a deep feedforward neural network including an input layer, a hidden layer, and an output layer. The input of the deep feedforward neural network is the data normalized in step S1, and the output is the key performance indicators of the operation of the artificial wetland; use the historical data normalized in S1 to train the deep feedforward neural network.

[0147] S3: Randomly generate initial constructed wetland process combinations, where each combination represents a set of wetland design and operation parameter combinations; after performing the same feature extraction and normalization processing on the parameters of the initial constructed wetland process combinations as in step S1, input them into the deep feedforward neural network trained in step S2 to obtain the key performance indicators of the operation of the constructed wetland corresponding to each combination;

[0148] S4: Embed multi-boundary constraint conditions into the evolutionary algorithm, and manage the constraint conditions of the constructed wetland involved in the evolutionary process through the Lagrange multiplier method and penalty function method to ensure that the optimization process is carried out in the feasible solution space; when the fitness change rate of the process combination is less than the set threshold or reaches the maximum number of iterations, the optimization process terminates; based on the key performance indicators of the operation of the constructed wetland corresponding to each process combination obtained in S3, calculate the fitness value of each process combination obtained when the optimization terminates; output the combination with the highest fitness as the optimal solution.

[0149] In the method of the present invention in step S4, step S4 further introduces heuristic rules and a dynamic constraint adjustment mechanism to improve the optimization ability of the evolutionary algorithm under multi-boundary constraint conditions and ensure that the final design scheme meets all preset requirements. When the evolutionary process iteration terminates, if the output optimal solution still does not meet all constraint conditions, fitness re-optimization, iterative expansion, and heuristic constraint adjustment will be performed to further optimize the feasibility of the solution. Through the fitness re-optimization strategy, perform constraint sensitivity analysis on the solutions that do not meet the constraints, identify the main violated constraint items, and dynamically adjust the fitness function to enhance the optimization intensity for key constraints. The adjustment method includes increasing the weight of key constraints to make the impact of violating the constraints more significant, and the calculation formula is as follows:

[0150]

[0151] where, w j is the weight corresponding to the violated constraint g j (x), and β is the adjustment coefficient (usually 0.1 ≤ β ≤ 0.5). At the same time, the system dynamically increases the penalty term in the objective function:

[0152]

[0153] where, P new = αP (α > 1) is the adjusted penalty factor to enhance the exclusivity of solutions that violate the constraints, thereby guiding the search towards the feasible solution region.

[0154] Adopt an iterative expansion mechanism to enhance the global search ability of the algorithm and prevent it from falling into local optima by increasing the population size and mutation rate. Combine a heuristic constraint adjustment strategy to dynamically adjust according to the importance of different constraints. When some secondary constraints (such as economy) are slightly violated but the key constraints (such as water quality compliance) are already satisfied, appropriately relax the secondary constraints to avoid affecting the overall optimization process.

[0155] The adjusted optimization process recalculates the fitness value. If the optimal solution satisfies all constraints, the final optimized solution is output; if not, repeat the above optimization steps until the constraint requirements are met or the maximum allowable number of iterations is reached. This mechanism ensures that the finally output process design solution is optimal in terms of water quality compliance, economic feasibility, spatial adaptability, etc. and meets the requirements of engineering practical applications by adaptively adjusting the penalty factor, optimizing the fitness function, expanding the iteration scale, and dynamically adjusting the constraint conditions in combination with heuristic strategies.

[0156] The following gives a specific constructed wetland purification project requirement to build a constructed wetland and presents the specific design process.

[0157] 1. Multi-boundary constraint conditions and objectives

[0158] First of all, due to the large fluctuations in influent water quality and obvious seasonal climate changes in the constructed wetland, the wetland design needs to meet the following multi-boundary constraint conditions:

[0159] Effluent standard: COD ≤ 50 mg / L, ammonia nitrogen ≤ 5 mg / L, total phosphorus ≤ 0.5 mg / L.

[0160] Economic constraint: Construction cost ≤ 35 million yuan, operating cost ≤ 0.2 yuan / m 3 .

[0161] Environmental constraint: The annual minimum temperature is -5°C, and the wetland operating temperature ≥ 4°C.

[0162] Spatial constraint: The wetland area ≤ 5000 ㎡.

[0163] The goal of this design is to adopt a constructed wetland process design method based on neural networks under multi-boundary conditions to optimize the wetland design and operation parameters, so that the constructed wetland can achieve efficient sewage treatment and minimize the operating cost under multi-constraint conditions.

[0164] 2. Data collection and preprocessing

[0165] The multi-dimensional data collected for this constructed wetland design related to the constructed wetland design and operation specifically includes:

[0166] Influent water quality parameters: COD 60 - 80 mg / L, ammonia nitrogen ≤ 8 mg / L, total phosphorus ≤ 0.7 mg / L;

[0167] Hydraulic parameters: The average daily sewage flow is 3000 m 3 / d, and the peak flow is 5000 m 3 / d;

[0168] Environmental conditions: The temperature range is -5°C to 30°C, and the temperature control scheme;

[0169] Wetland structure characteristics: The filler types are gravel, ceramsite, zeolite, etc., the wetland type is horizontal subsurface flow wetland, and the plant configuration scheme is aquatic plants such as reed, cattail, iris, etc.;

[0170] Using the principal component analysis (PCA) method, extract the key influencing parameters from the collected data: wetland area, filler thermal conductivity, plant species, water flow velocity in the wetland, temperature control scheme. The input parameters include the following four categories:

[0171] 1) Influent water quality parameters:

[0172] COD concentration (mg / L): Describes the content of organic matter in water;

[0173] Ammonia nitrogen concentration (mg / L): Describes the concentration of nitrogen pollutants in water;

[0174] Total phosphorus concentration (mg / L): Describes the concentration of phosphorus pollutants in water.

[0175] 2) Hydraulic parameters:

[0176] Daily sewage flow (m 3 / d): Represents the amount of water that the wetland system needs to treat daily;

[0177] Peak flow (m 3 / d): Represents the maximum influent load situation.

[0178] 3) Environmental condition parameters:

[0179] Minimum air temperature (°C): Represents the lowest environmental temperature in the area where the wetland is located;

[0180] Temperature control scheme: Such as whether to use a ground source heat pump for auxiliary heating, etc.

[0181] Wetland structure parameters:

[0182] 4) Wetland area (m 2 ): The available land space size;

[0183] Filler thermal conductivity (W / m·K): The heat preservation and heat conduction performance of different fillers at low temperatures;

[0184] Plant configuration scheme: Such as different plant combination schemes of reed, cattail, water hyacinth, etc.;

[0185] 5) Flow velocity (m / d): The flow velocity of water in the wetland.

[0186] Normalize the above data, that is, normalize all parameters to the interval [0, 1] according to the following formula:

[0187]

[0188] where X norm is the normalized data, X min and X max are the minimum and maximum values of the parameter respectively.

[0189] The normalized data is used as the input data of the neural network to ensure the balanced contribution of each input variable to the model and avoid the influence of parameters with a large numerical range on the training process.

[0190] 3. Construction and training of the deep feedforward neural network

[0191] Input layer: It contains 12 nodes, and each node corresponds to an input parameter.

[0192] Hidden layer: There are 3 hidden layers, and the number of nodes is 128, 64, and 32 respectively; the activation function adopts the ReLU function: f(x) = max(0, x); where the first hidden layer performs a preliminary nonlinear transformation on the input data to extract features; the second hidden layer filters out irrelevant features and focuses on the relationship between key parameters and performance indicators; the third hidden layer further deeply fits the nonlinear mapping relationship between the input and output; the output layer contains 4 nodes, and outputs the performance indicators of the wetland system: COD removal rate, ammonia nitrogen removal rate, total phosphorus removal rate, and total operating cost.

[0193] Divide the data of the key influencing parameters after normalization into a training set (80%) and a validation set (20%), and evaluate the performance of the deep feedforward neural network through the cross-validation method to ensure the accuracy and stability of the prediction results. The loss function adopts the mean square error (MSE) function. Use the Adam optimization algorithm to update the network weights, and the learning rate is set to 0.001.

[0194] 4. Optimization solution and result output

[0195] (1) Initial parameter combination

[0196] Randomly generate a process combination of the initial constructed wetland design parameters, including wetland area, filler type, plant configuration, and flow velocity, etc., as the initial combination.

[0197] (2) Key performance indicators of the initial parameter combination

[0198] Input the process combinations of the randomly generated initial constructed wetland design parameters into the trained deep feedforward neural network to obtain the key performance indicators of the operation of the constructed wetland corresponding to each combination.

[0199] (3) Using the Lagrange multiplier method and penalty function method to manage the constraint conditions

[0200] Boundary condition setting:

[0201] Effluent water quality constraint: COD ≤ 50 mg / L, ammonia nitrogen ≤ 5 mg / L, total phosphorus ≤ 0.5 mg / L;

[0202] Economic constraint: Construction cost ≤ 35 million yuan, operating cost ≤ 0.2 yuan / m 3 ;

[0203] Environmental constraint: The lowest annual temperature is -5°C, and the wetland operating temperature ≥ 4°C;

[0204] Spatial constraint: Wetland area ≤ 5000 ㎡.

[0205] (4) Fitness calculation

[0206] Fitness = w1·(COD removal rate) + w2·(ammonia nitrogen removal rate) - w3·(operating cost)

[0207] COD removal rate = (COD in -COD out )) / COD in ×100%

[0208] Among them, w1, w2, and w3 are weight coefficients.

[0209] (5) Iterative optimization

[0210] Based on the evolutionary algorithm for global search, select the design combination with the highest fitness.

[0211] (6) Result output

[0212] (a) Constructed wetland design parameters:

[0213] Wetland area: 4800 m 2

[0214] Process type: Two-stage vertical subsurface flow wetland + one-stage horizontal subsurface flow wetland combination

[0215] Packing selection: Mixed packing of ceramsite, zeolite, and quartz sand

[0216] Plant configuration: Mixed planting of reed and cattail

[0217] Temperature regulation: In winter, it is recommended to use a ground source heat pump to assist in heating the constructed wetland filter bed, and the wetland temperature should be maintained above 4°C.

[0218] (b) Operating parameters:

[0219] Hydraulic retention time (HRT): 1.5 days

[0220] Ground source heat pump temperature maintenance: ≥4°C

[0221] Insulation thickness of the covering layer: 30 cm (straw or wood chips)

[0222] (c) Performance of the optimized constructed wetland system:

[0223] COD removal rate: 51.5%

[0224] Ammonia nitrogen removal rate: 77.2%

[0225] Total phosphorus removal rate: 60.5%

[0226] Construction cost: 34 million yuan

[0227] Operating cost: 0.18 yuan / m 3 ;

[0228] By adjusting multiple boundary constraint conditions and combining with the neural network multi-boundary constraint process optimal solution algorithm of the present invention, the efficient operation and economic cost control of the constructed wetland under low-temperature conditions are realized. Through the neural network multi-boundary constraint process optimal solution algorithm of the present invention, the following goals are achieved for the design and operating parameters of the constructed wetland on the premise of meeting the constraint conditions:

[0229] 1. Efficient sewage treatment: The removal rates of COD, ammonia nitrogen, and total phosphorus reach 51.5%, 77.2%, and 60.5% respectively, all of which are better than the effluent standards.

[0230] 2. Cost minimization: The operating cost is reduced to 0.18 yuan / m 3 , and the construction cost is controlled within 34 million yuan.

[0231] 3. Stable operation: Through the temperature regulation scheme in a low-temperature environment, the wetland bed temperature is stably maintained above 4°C to ensure the stability of the system.

[0232] The final design scheme meets the effluent standards within the area limit of 5000 ㎡, effectively reduces the operating cost and construction cost, and realizes the dual optimization of the stability and economy of sewage treatment.

[0233] Those of ordinary skill in the art can understand that the above are only preferred examples of the invention and are not used to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent replacements for some of the technical features. Any modifications, equivalent replacements, etc. made within the spirit and principle of the invention shall be included within the protection scope of the invention.

Claims

1. An artificial wetland process design system based on a neural network under multiple boundary conditions, characterized in that It includes the following modules: A data preprocessing module, which uses the principal component analysis method to extract features from multi-dimensional data related to the design and operation of constructed wetlands, removes redundant information, and reduces data complexity; and normalizes the data with different dimensions after feature extraction; A neural network model construction module, which is used to construct a deep feedforward neural network including an input layer, a hidden layer, and an output layer. The input of the deep feedforward neural network is the data normalized by the data preprocessing module, and the output is the key performance indicators of the operation of the constructed wetland; A multi-boundary constraint processing module, which uses the Lagrange multiplier method and the penalty function method to manage the constraint conditions of the constructed wetland involved in the optimization process, ensuring that the solution result not only meets the constraint conditions but also can achieve the global optimal solution; The constraint conditions include influent water quality constraints, environmental condition constraints, operation economy constraints, effluent standard constraints, and space constraints; An optimal solution output module, which is used to randomly generate an initial constructed wetland process combination, input it into the trained deep feedforward neural network, and calculate the fitness value of the randomly generated initial constructed wetland process combination based on the key performance indicators of the operation of the constructed wetland output by the trained deep feedforward neural network; Based on these fitness values and the constraint conditions of the constructed wetland, an evolutionary algorithm is used to correct the process combinations that do not meet the constraint conditions. When the fitness change rate of the process combination is set to a threshold or the maximum number of iterations is reached, the optimization process terminates, and the process combination with the highest fitness is output as the optimal solution.

2. The artificial wetland process design system based on a neural network under multiple boundary conditions according to claim 1, wherein The Lagrangian function in the Lagrange multiplier method is: Among them, f(x) is the objective function that combines several performance indicators, and ɡ i (x) represents the constraint condition, and λ i is the Lagrange multiplier, and m is the number of boundary conditions; The penalty function in the penalty function method is: Among them, F(x) represents the objective function corrected by the penalty function, P is the penalty factor, and the more times the constraint is violated or the greater the deviation from the range, the higher the penalty value.

3. The artificial wetland process design system based on a neural network under multiple boundary conditions according to claim 1, characterized in that, The calculation formula of the fitness is as follows: Fitness = w1·(COD removal rate) + w2·(ammonia nitrogen removal rate) - w3·(operation cost) Among them, w1, w2, and w3 are the weight coefficients of the COD removal rate, ammonia nitrogen removal rate, and operation cost respectively.

4. The artificial wetland process design system based on a neural network under multiple boundary conditions according to claim 3, wherein The deep feedforward neural network includes three hidden layers. The first hidden layer includes 128 nodes, the second hidden layer includes 64 nodes, and the third hidden layer includes 32 nodes.

5. The artificial wetland process design system based on a neural network under multiple boundary conditions according to claim 4, characterized in that, The key performance indicators of the operation of the constructed wetland output by the trained deep feedforward neural network include COD removal rate, ammonia nitrogen removal rate, total phosphorus removal rate, and total operation cost.

6. The artificial wetland process design system based on a neural network under multiple boundary conditions according to claim 5, wherein, The influent water quality constraints include the maximum and minimum values of pollutant concentrations including COD, ammonia nitrogen, and total phosphorus; The environmental condition constraints include the temperature range tolerated by the wetland system; The operation economy constraints include the upper limits of operation energy consumption and total construction cost; The effluent standard constraints are the effluent water quality indicators stipulated by national or regional regulations; The space constraints are the upper limit of the wetland area.

7. A method for designing an artificial wetland process under multi-boundary conditions based on a neural network, characterized in that, It includes the following steps: S1: Collect data related to the design and operation of constructed wetlands, including influent water quality parameters, hydraulic parameters, environmental conditions, and wetland structure characteristics; use the principal component analysis method to extract features from these data and obtain the key influencing parameters; finally, normalize the data with different dimensions of the key influencing parameters obtained by feature extraction. S2: Construct a deep feedforward neural network including an input layer, a hidden layer, and an output layer. The input of the deep feedforward neural network is the data normalized in step S1, and the output is the key performance indicators of the operation of the constructed wetland; use the historical data normalized in S1 to train the deep feedforward neural network. S3: Randomly generate an initial constructed wetland process combination, and each combination represents a set of wetland design and operation parameter combinations; after performing the same feature extraction and normalization processing on the parameters of the initial constructed wetland process combination as in step S1, input them into the deep feedforward neural network trained in step S2 to obtain the key performance indicators of the operation of the constructed wetland corresponding to each combination. S4: Embed multi-boundary constraint conditions into the evolutionary algorithm, introduce the Lagrange multiplier method and the penalty function method to manage the multi-boundary constraint conditions, and correct the process combinations that do not meet the constraint conditions to ensure that the optimization process is carried out in the feasible solution space; when the fitness change rate of the process combination is less than the set threshold or the maximum number of iterations is reached, the optimization process terminates. Based on the key performance indicators of the operation of the constructed wetland corresponding to each process combination obtained in S3, calculate the fitness value of each process combination obtained when the optimization terminates, evaluate each combination until the termination condition is met, and output the combination with the highest fitness as the optimal design scheme.

8. The method for designing an artificial wetland process under multi-boundary conditions based on a neural network according to claim 7, characterized in that, In step S4, heuristic rules and a dynamic constraint adjustment mechanism are introduced to improve the optimization ability of the evolutionary algorithm under multi-boundary constraint conditions: when the iteration of the evolutionary process terminates, if the output optimal solution still does not meet all constraint conditions, adaptively adjust the penalty factor, optimize the fitness function, expand the iteration scale, and dynamically adjust the constraint conditions in combination with heuristic strategies to ensure that the finally output process design scheme is optimal in terms of water quality compliance, economic feasibility, and space adaptation, meeting the requirements of engineering practical applications.

9. The method for designing an artificial wetland process under multiple boundary conditions based on a neural network according to claim 8, characterized in that, The influent water quality parameters in S1 include COD concentration, ammonia nitrogen concentration, and total phosphorus concentration; the hydraulic parameters include the daily average sewage flow and peak flow; the environmental conditions include the environmental temperature range and temperature control scheme of the area where the wetland is located; the wetland structure characteristics include wetland area, filler thermal conductivity, plant configuration scheme, and water flow velocity in the wetland.

10. The method for designing an artificial wetland process based on a neural network under multiple boundary conditions according to claim 9, wherein, The key influencing parameters in S1 include wetland area, filler thermal conductivity, plant species, water flow velocity in the wetland, and temperature control scheme.

Images