Textile printing and dyeing sewage intelligent treatment method and system based on edge calculation

Abstract

The invention provides an intelligent textile printing and dyeing sewage treatment method and system based on edge computing, and relates to the technical field of textile printing and dyeing sewage treatment.The method comprises the steps that edge computing nodes are deployed on multiple treatment devices to collect water quality parameters, a neural network is trained through historical data, and the network structure is optimized according to the fluctuation amplitude; the corresponding relation between the water quality parameters and the equipment efficiency is calculated to predict abnormity, the particle swarm optimization is combined to optimize the operation parameters, and the adjustment strategy is determined based on the effluent standard-reaching rate, the energy consumption and the cost, so that the intelligent treatment of the textile printing and dyeing sewage is realized, the treatment efficiency and the effluent quality are improved, and the energy consumption and the operation cost are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of textile printing and dyeing wastewater treatment, and particularly relates to a textile printing and dyeing wastewater intelligent treatment method and system based on edge computing. BACKGROUND

[0002] Textile printing and dyeing wastewater treatment mainly adopts physical methods, chemical methods and biological methods such as flocculation sedimentation, oxidation-reduction and activated sludge method. With the development of information technology, intelligent wastewater treatment systems are gradually applied to the field of textile printing and dyeing wastewater treatment. Through automatic monitoring, data analysis and intelligent control, the efficiency and stability of wastewater treatment are improved. However, the existing technology still lacks real-time monitoring and intelligent analysis capabilities, and the response to fluctuations in water quality parameters is lagging, making it difficult to adjust the treatment process parameters in a timely manner, resulting in unstable treatment effects, and not fully considering the complexity and variability of textile printing and dyeing wastewater composition, making it impossible to adaptively adjust the treatment parameters according to different water quality characteristics, resulting in low system operation efficiency, high energy consumption and reagent costs, and only focusing on the optimization of a single indicator, ignoring the overall economic and sustainable operation of the system, and failing to achieve multi-objective collaborative optimization control, affecting the production efficiency and environmental protection effect of textile printing and dyeing enterprises. SUMMARY

[0003] The embodiments of the present application provide a textile printing and dyeing wastewater intelligent treatment method and system based on edge computing, which can at least solve some of the problems in the prior art.

[0004] In a first aspect, the embodiments of the present application provide a textile printing and dyeing wastewater intelligent treatment method based on edge computing, comprising: Deploying edge computing nodes in multiple textile printing and dyeing wastewater treatment devices and collecting current water quality parameters; Obtaining historical water quality parameters and historical treatment effect data to establish a training data set, training a pre-set neural network by combining a back propagation algorithm to obtain initial connection weights, calculating the fluctuation amplitude of the current water quality parameters, adjusting the number of hidden layer neurons of the neural network according to the fluctuation amplitude and updating the initial connection weights to obtain optimized connection weights, calculating the corresponding relationship between the current water quality parameters and the device operation efficiency based on the optimized connection weights, judging the abnormal type and the abnormal number according to the corresponding relationship to generate water quality abnormal warning information, and sending the water quality abnormal warning information to a central control platform; The central control platform receives the water quality abnormal warning information sent by each edge computing node; The abnormal water quality early warning information is substituted into an optimization calculation function with the outflow water quality compliance rate, system energy consumption and reagent cost as optimization objectives to obtain an optimization objective function value, operation parameters of the textile printing and dyeing wastewater treatment equipment are converted into optimization calculation variables, and the operation parameter values are calculated by combining a particle swarm algorithm; the adjustment priority level and adjustment range of the operation parameters corresponding to the optimization objective function value are determined, operation parameter adjustment instructions are generated and executed; The operation state of the textile printing and dyeing wastewater treatment equipment is monitored in real time, and the operation parameter adjustment instructions are updated.

[0005] In an alternative implementation, The edge computing nodes are deployed in the textile printing and dyeing wastewater treatment equipment, and the current water quality parameters are collected, including: The location information of the textile printing and dyeing wastewater treatment equipment is collected, and a plurality of equipment groups are divided, each equipment group including at least two textile printing and dyeing wastewater treatment equipment; An edge computing node is deployed in each equipment group by selecting one textile printing and dyeing wastewater treatment equipment, and data collection nodes are deployed in the remaining textile printing and dyeing wastewater treatment equipment in the current equipment group; Raw water quality data of the textile printing and dyeing wastewater treatment equipment are collected by the data collection nodes and sent to the edge computing nodes, and the edge computing nodes obtain the current water quality parameters by averaging and combining different data in the raw water quality data.

[0006] In an alternative implementation, The historical water quality parameters and historical treatment effect data are obtained to establish a training data set, the pre-set neural network is trained by combining a back propagation algorithm to obtain initial connection weights, the fluctuation amplitude of the current water quality parameters is calculated, the number of hidden layer neurons of the neural network is adjusted according to the fluctuation amplitude, and the initial connection weights are updated to obtain optimized connection weights, including: The historical water quality parameters and historical treatment effect data are obtained to establish a training data set, the historical water quality parameters in the training data set are normalized by maximum and minimum values, and the normalized historical water quality parameters and historical treatment effect data are used as training samples; The training samples are input into the pre-set neural network, the network output values are calculated by forward propagation, the loss values are calculated by error calculation between the network output values and the corresponding historical treatment effect data, the error terms of each layer of neurons in the neural network are calculated by a back propagation algorithm, the gradient values of the connection weights are calculated according to the error terms, the connection weights are iteratively updated based on a gradient descent algorithm and the gradient values to obtain initial connection weights; Subtracting the water quality parameter at the last collection time from the pre-acquired current water quality parameter to obtain a parameter difference, dividing the parameter difference by the water quality parameter at the last collection time to obtain a change rate, and obtaining an amplitude of fluctuation by taking an arithmetic average of the change rate; When the amplitude of fluctuation is greater than a pre-set amplitude of fluctuation threshold, calculating the number of newly added neurons by a linear growth function, and when the amplitude of fluctuation is less than the amplitude of fluctuation threshold, calculating the number of reduced neurons by a linear decay function, adjusting the hidden layer structure of the neural network according to the calculated number of neurons, updating the initial connection weight in the adjusted neural network structure by an online learning algorithm, calculating a weight adjustment amount by a stochastic gradient descent algorithm, and obtaining an optimized connection weight by adding the weight adjustment amount to the initial connection weight.

[0007] In an optional implementation, Calculating a corresponding relationship between the current water quality parameter and the equipment operation efficiency based on the optimized connection weight, generating water quality abnormality early warning information according to the corresponding relationship to judge the abnormal type and the abnormal number, and sending the water quality abnormality early warning information to a central control platform include: Taking the current water quality parameter as an input, combining a linear combination of the optimized connection weight and a pre-set bias term, and obtaining the corresponding relationship between the current water quality parameter and the equipment operation efficiency through a sigmoid activation function; Based on the current water quality parameter, a water quality parameter correlation graph is constructed, the current water quality parameter is taken as a vertex set of the correlation graph, the corresponding relationship is taken as an edge set of the water quality parameter correlation graph, and the dynamic correlation strength between the current water quality parameters is calculated in combination with a pre-set attention weight matrix; Based on the dynamic correlation strength, the edge set of the correlation graph is updated, an adjacency matrix is constructed according to the connection relationship between the vertices in the updated correlation graph, and a degree matrix is generated based on the adjacency matrix to calculate the connection number of each vertex, feature extraction is performed on the updated correlation graph through graph convolution operation, local region features are obtained through combined operation of the adjacency matrix and the degree matrix, global state features are obtained through difference pooling operation, and the local region features and the global state features are weighted and fused to obtain comprehensive features; Based on pre-acquired historical data, a water quality parameter reference value under normal working conditions is calculated, a Mahalanobis distance is calculated from the real-time monitoring data and the water quality parameter reference value, the Mahalanobis distance is corrected based on the comprehensive features, an adaptive threshold is calculated according to the change trend of the corrected Mahalanobis distance in combination with the comprehensive features, water quality abnormality early warning information is generated according to the corresponding relationship to judge the abnormal type and the abnormal number, and the information is sent to the central control platform through a secure transmission protocol.

[0008] In an optional implementation, The water quality abnormal early warning information is substituted into an optimization calculation function with the outflow water quality compliance rate, system energy consumption and reagent cost as optimization objectives to obtain an optimization objective function value, operation parameters of the textile printing and dyeing wastewater treatment equipment are converted into optimization calculation variables, and the operation parameter values are calculated by combining a particle swarm algorithm, and the operation parameters include: The water quality abnormal early warning information is substituted into an optimization calculation function with the outflow water quality compliance rate, system energy consumption and reagent cost as optimization objectives to obtain an optimization objective function value; According to the optimization objective function value, the operation parameters of the textile printing and dyeing wastewater treatment equipment are converted into optimization calculation variables and combined into an optimization calculation variable vector, the particle swarm size and the maximum iteration number are initialized based on the optimization calculation variable vector, the initial temperature and the temperature reduction coefficient are set according to the optimization objective function value, the crossover probability and the scaling factor are calculated and set, the initial population distribution is generated, and the operation parameters include the aeration amount adjustment parameter, the reagent dosage parameter, the hydraulic load parameter and the sludge return ratio parameter; According to the initial population distribution, the velocity update equation and the position update equation are determined, the inertia weight in the velocity update equation and the position update equation is dynamically adjusted based on the ratio of the current iteration number to the maximum iteration number to obtain updated velocity values and position values, and the learning factor in the velocity update equation is adaptively updated based on the optimization objective function value to obtain an optimized population distribution; The global optimal solution is obtained by performing the particle swarm algorithm using the optimized population distribution, the global optimal solution is substituted into a simulated annealing operation, the local optimal solution is obtained in combination with the initial temperature and the temperature reduction coefficient, the candidate solution set is obtained by updating the population distribution corresponding to the local optimal solution through a differential evolution operation based on the crossover probability and the scaling factor, the optimal solution that meets the pre-set constraint condition is selected from the candidate solution set, and the operation parameter values are calculated.

[0009] In an alternative embodiment, According to the optimization objective function value, the adjustment priority level and the adjustment amplitude corresponding to the operation parameters are determined, and the operation parameter adjustment instruction is generated and executed, including: The parameter correlation degree is calculated by collecting the operation time sequence data of the operation parameters, the ratio of the dispersion product of the operation time sequence data to the standard deviation product is calculated, the parameter correlation degree and the optimization objective function value deviation corresponding to the operation parameters are multiplied after being normalized, and the causal relationship weight between the operation parameters is obtained; The direct influence matrix is constructed by multiplying the causal relationship weight between the operation parameters and the operation time sequence data response increment ratio, the indirect influence propagation matrix is obtained by power series summation on the direct influence matrix, and the comprehensive influence degree matrix is calculated by multiplying the indirect influence propagation matrix and the pre-set operation parameter importance weight; The comprehensive influence degree matrix is multiplied by a normalized result of an optimization objective function value deviation to calculate a running parameter adjustment priority level, a weighted average operation is performed on the running parameter adjustment priority level and a causal relationship weight between the running parameters to obtain a synergistic adjustment coefficient; A running parameter adjustment priority level, a synergistic adjustment coefficient and an optimization objective function value deviation are multiplied to calculate an adjustment amplitude, a running parameter value is added to the adjustment amplitude to generate a running parameter adjustment instruction, and the running parameter adjustment instruction is executed in the order of the running parameter adjustment priority level.

[0010] In an optional embodiment, Real-time monitoring of the running state of the textile printing and dyeing wastewater treatment equipment and updating of the running parameter adjustment instruction include: Real-time acquisition of running state data of the textile printing and dyeing wastewater treatment equipment, comparison of the running state data with an expected response value after execution of the running parameter adjustment instruction, and calculation of a running state deviation; According to the running state deviation, a determination is made of an adjustment response degree of the running parameter, and when the running state deviation exceeds a preset deviation threshold, an update coefficient of the running parameter is calculated based on the running state deviation; The update coefficient is multiplied by the original running parameter adjustment instruction to obtain an updated running parameter adjustment instruction, and the running parameter of the textile printing and dyeing wastewater treatment equipment is adjusted according to the updated running parameter adjustment instruction.

[0011] In a second aspect of the embodiment of the application, a textile printing and dyeing wastewater intelligent treatment system based on edge computing is provided, which includes: A first unit is configured to deploy edge computing nodes in a plurality of textile printing and dyeing wastewater treatment equipment and acquire current water quality parameters; A second unit is configured to acquire historical water quality parameters and historical treatment effect data to establish a training data set, train a pre-set neural network by combining a back propagation algorithm to obtain an initial connection weight, calculate a fluctuation amplitude of the current water quality parameters, adjust the number of hidden layer neurons of the neural network according to the fluctuation amplitude and update the initial connection weight to obtain an optimized connection weight, calculate a corresponding relationship between the current water quality parameters and the equipment running efficiency based on the optimized connection weight, judge an abnormal type and an abnormal number according to the corresponding relationship to generate water quality abnormality early warning information, and send the water quality abnormality early warning information to a central control platform; A third unit is configured to receive, by the central control platform, water quality abnormality early warning information sent by each edge computing node; The fourth unit is configured to substitute the water quality abnormality early warning information into an optimization calculation function with the water quality compliance rate, system energy consumption and reagent cost as optimization objectives to obtain an optimization objective function value, convert the operation parameters of the textile printing and dyeing wastewater treatment equipment into optimization calculation variables, calculate the operation parameter values by combining a particle swarm algorithm, determine the adjustment priority level and adjustment range of the operation parameters according to the optimization objective function value, generate an operation parameter adjustment instruction and execute the operation parameter adjustment instruction. The fifth unit is configured to monitor the operation state of the textile printing and dyeing wastewater treatment equipment in real time and update the operation parameter adjustment instruction.

[0012] In a third aspect, an electronic device is provided, including: a processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0013] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0014] In the present application, by deploying edge computing nodes in multiple textile printing and dyeing wastewater treatment equipment to collect water quality parameters, training a neural network combined with historical data and dynamically adjusting the network structure according to the water quality parameter fluctuation range, the water quality abnormality can be accurately identified and early warning can be timely issued, the perception ability and early warning accuracy of water quality changes are improved, the optimal operation parameters are calculated by a particle swarm algorithm, the adjustment priority level and adjustment range are determined, the intelligent regulation and control of the wastewater treatment process are realized, the treatment effect and operation cost are effectively balanced, the equipment operation state is monitored in real time and the operation parameter adjustment instruction is dynamically updated, the self-adaptive ability and operation efficiency of the textile printing and dyeing wastewater treatment are significantly improved, the energy consumption and reagent usage are reduced, and the stability and compliance of the effluent water quality are ensured. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 Fig. 1 is a flowchart of a textile printing and dyeing wastewater intelligent treatment method based on edge computing according to an embodiment of the present application; Figure 2 Fig. 2 is a wastewater treatment parameter optimization adjustment flowchart of the textile printing and dyeing wastewater intelligent treatment method based on edge computing according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0017] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and some embodiments may not be described again for the same or similar concepts or processes.

[0018] Figure 1 The flowchart of the method for intelligent treatment of textile printing and dyeing wastewater based on edge computing according to the embodiments of the present application is shown in FIG. 1, which comprises the following steps. Figure 1 Deploying edge computing nodes in a plurality of textile printing and dyeing wastewater treatment devices and collecting current water quality parameters; Obtaining historical water quality parameters and historical treatment effect data to establish a training data set, training a pre-set neural network by combining a back propagation algorithm to obtain initial connection weights, calculating the fluctuation amplitude of the current water quality parameters, adjusting the number of hidden layer neurons of the neural network according to the fluctuation amplitude and updating the initial connection weights to obtain optimized connection weights, calculating the corresponding relationship between the current water quality parameters and the device operation efficiency based on the optimized connection weights, judging the abnormal type and the abnormal number according to the corresponding relationship to generate water quality abnormal warning information, and sending the water quality abnormal warning information to a central control platform; The central control platform receives the water quality abnormal warning information sent by each edge computing node; Substituting the water quality abnormal warning information into an optimization calculation function with the discharge water quality compliance rate, system energy consumption and reagent cost as optimization objectives to obtain an optimization objective function value, converting the operation parameters of the textile printing and dyeing wastewater treatment device into optimization calculation variables, calculating the operation parameter values by combining a particle swarm algorithm, determining the adjustment priority level and adjustment amplitude corresponding to the operation parameters according to the optimization objective function value, generating an operation parameter adjustment instruction and executing the same; Real-time monitoring the operation state of the textile printing and dyeing wastewater treatment device and updating the operation parameter adjustment instruction.

[0019] In an alternative embodiment, Deploying edge computing nodes in a plurality of textile printing and dyeing wastewater treatment devices and collecting current water quality parameters comprises: Collecting the location information of the textile printing and dyeing wastewater treatment device and dividing a plurality of device groups, each device group comprising at least two textile printing and dyeing wastewater treatment devices;​ select one textile printing and dyeing wastewater treatment equipment in each equipment group to deploy an edge computing node, and deploy data collection nodes in the remaining textile printing and dyeing wastewater treatment equipment in the current equipment group; Collect the original water quality data of the textile printing and dyeing wastewater treatment equipment through the data collection node and send it to the edge computing node. The edge computing node obtains the current water quality parameter by averaging and combining different data in the original water quality data.

[0020] The position information of the textile printing and dyeing wastewater treatment equipment is collected by using the global positioning system or indoor positioning technology. After the position information collection is completed, the equipment group is divided according to the geographical position and physical distribution characteristics of the equipment. The equipment group division is based on the physical distance between the equipment. Generally, the equipment within 100 meters is divided into the same group, or the equipment is divided according to the correlation of the process flow, such as the equipment in the same process stage of pretreatment, biochemical treatment and advanced treatment is divided into the same group. Each equipment group contains at least two textile printing and dyeing wastewater treatment equipment. For example, a printing and dyeing plant can divide 6 wastewater treatment equipment into 2 equipment groups. The first group contains 3 equipment in the pretreatment area, and the second group contains 3 equipment in the biochemical treatment area.

[0021] After the equipment group is divided, one equipment with the best performance in each equipment group is selected to deploy an edge computing node. The selection criteria of the edge computing node include processing capability, storage capacity, network connection stability and other factors. Generally, the equipment with a processor performance of not less than four cores 2.0GHz, a memory capacity of not less than 8GB, and a storage space of not less than 128GB is selected as the edge computing node. For example, in the first group of equipment mentioned above, the wastewater pretreatment equipment A with a processing capability of four cores 2.4GHz and a memory of 12GB can be selected as the edge computing node, and in the second group, the biochemical treatment equipment D with a processing capability of eight cores 3.0GHz and a memory of 16GB is selected as the edge computing node.

[0022] After the edge computing node is determined, data collection nodes are deployed on the remaining textile printing and dyeing wastewater treatment equipment in the current equipment group. The data collection node is usually composed of a sensor module, a data collection module and a communication module. The sensor module includes pH sensor, temperature sensor, turbidity sensor, conductivity sensor, dissolved oxygen sensor, etc. The data collection module is responsible for obtaining the original data from the sensor and performing preliminary processing, such as for the pH sensor, the sampling frequency is set to once per minute, and a set of data is output after continuous sampling for 5 times. The communication module is responsible for transmitting the processed data to the edge computing node, usually using ZigBee, LoRa or Wi-Fi wireless communication technology, and the transmission frequency can be set to once every 5 minutes.

[0023] The raw water quality data collected by the data collection nodes include, but are not limited to, pH value, temperature, turbidity, conductivity, dissolved oxygen, chemical oxygen demand (COD), biochemical oxygen demand (BOD), total suspended solids (TSS), ammonia nitrogen content, etc. Taking the pH value as an example, a data collection node may collect the following raw data within one hour: 6.8, 7.0, 6.9, 7.1, 7.0, 6.8, 7.2, 7.1, 6.9, 7.0, 6.8, 7.2. These data are sent to the edge computing node of the device group through wireless transmission.

[0024] After receiving the raw water quality data, the edge computing node performs data processing and analysis. For multiple groups of data of the same parameter, abnormal value detection and filtering are performed, such as eliminating data that deviates obviously from the normal range. Taking the above pH value data as an example, if the normal working range is 6.5-7.5, all the data are within the valid range and do not need to be eliminated. Then the mean value of the valid data is calculated, i.e. (6.8+7.0+6.9+7.1+7.0+6.8+7.2+7.1+6.9+7.0+6.8+7.2) / 12=7.0. Similar methods are used for different water quality parameters, and finally a group of current water quality parameters are obtained, including pH value mean 7.0, temperature mean 35.2℃, turbidity mean 28 NTU, conductivity mean 420 μS / cm, dissolved oxygen mean 3.2 mg / L, etc.

[0025] The edge computing node evaluates the processed water quality parameters according to the preset threshold. For example, if the normal range of pH value is 6.5-8.5, when the calculated mean value of pH value is 7.0, the system determines that the pH value is within the normal range; if the temperature warning threshold is 40℃, when the calculated mean value of temperature is 35.2℃, the system determines that the temperature is within the safe range. When a parameter exceeds the preset threshold, the edge computing node will trigger the corresponding early warning mechanism and send a notification to the relevant operator.

[0026] For example, in the raw water quality data collected by the first group of devices of a certain textile printing and dyeing plant, the COD values of three devices are 480 mg / L, 460 mg / L and 500 mg / L, and the edge computing node calculates the COD mean value as (480+460+500) / 3=480 mg / L. The data collected by the second group of devices show that the ammonia nitrogen contents of three devices are 18 mg / L, 20 mg / L and 17 mg / L, and the edge computing node calculates the ammonia nitrogen mean value as (18+20+17) / 3=18.3 mg / L.

[0027] In this embodiment, by reasonably dividing the equipment groups in the multiple textile printing and dyeing wastewater treatment equipment, and selecting part of the equipment in each equipment group to deploy the edge computing node and the rest of the equipment to deploy the data acquisition node, distributed collection and nearby processing of water quality data are realized. The original water quality data is collected by the data acquisition node and transmitted to the edge computing node, which can avoid the communication bandwidth pressure and delay problem caused by centralized uploading of all data. The edge computing node calculates and combines the received original water quality data to obtain the current water quality parameter, so as to ensure the timeliness of data processing while improving the accuracy and stability of water quality monitoring, and realize efficient monitoring and intelligent management of the operation state of the wastewater treatment equipment.

[0028] In an optional implementation, The historical water quality parameters and the historical treatment effect data are acquired to establish a training data set, an initial connection weight is obtained by training a pre-set neural network combined with a back propagation algorithm, a fluctuation amplitude of the current water quality parameter is calculated, the number of hidden layer neurons of the neural network is adjusted according to the fluctuation amplitude, and the initial connection weight is updated to obtain an optimized connection weight, which includes: The historical water quality parameters and the historical treatment effect data are acquired to establish a training data set, an initial connection weight is obtained by training a pre-set neural network combined with a back propagation algorithm, a fluctuation amplitude of the current water quality parameter is calculated, the number of hidden layer neurons of the neural network is adjusted according to the fluctuation amplitude, and the initial connection weight is updated to obtain an optimized connection weight, which includes: The training sample is input into the pre-set neural network, and a network output value is calculated by forward propagation. An error value is calculated by comparing the network output value with the corresponding historical treatment effect data. The error terms of each layer of neurons in the neural network are calculated by a back propagation algorithm. The gradient value of the connection weight is calculated according to the error terms. The connection weight is iteratively updated based on the gradient descent algorithm and the gradient value to obtain an initial connection weight. The parameter difference is obtained by subtracting the water quality parameter at the last acquisition time from the pre-acquired current water quality parameter. The change rate is obtained by dividing the parameter difference by the water quality parameter at the last acquisition time. The fluctuation amplitude is obtained by calculating the arithmetic mean of the change rate. When the fluctuation amplitude is greater than a pre-set fluctuation amplitude threshold, the number of newly added neurons is calculated by a linear growth function. When the fluctuation amplitude is less than the fluctuation amplitude threshold, the number of neurons is reduced by a linear decay function. The hidden layer structure of the neural network is adjusted according to the calculated number of neurons. The initial connection weight in the adjusted neural network structure is updated by an online learning algorithm. The weight adjustment amount is calculated by a stochastic gradient descent algorithm. The optimized connection weight is obtained by adding the weight adjustment amount to the initial connection weight.

[0029] The historical water quality parameters and historical treatment effect data are obtained and a training data set is established. The historical water quality parameters include pH value, chemical oxygen demand (COD), biochemical oxygen demand (BOD), total suspended solids (TSS), color, turbidity, conductivity and the like. The historical treatment effect data include effluent water quality indicators and treatment efficiency after sewage treatment. For example, in the historical data of a certain textile printing and dyeing plant, the influent COD is 520 mg / L, the effluent COD is 95 mg / L, and the treatment efficiency is 81.7%; the influent BOD is 280 mg / L, the effluent BOD is 35 mg / L, and the treatment efficiency is 87.5%. The foregoing data is sorted in chronological order to form a training data set containing multiple groups of input parameters and corresponding output results.

[0030] The historical water quality parameters in the training data set are subjected to maximum and minimum normalization processing, so that each parameter value is mapped to a range of 0 to 1. The original parameter value is subtracted from the minimum value of the parameter in the data set, and then divided by the difference between the maximum and minimum values of the parameter. Taking chemical oxygen demand as an example, if the minimum value of chemical oxygen demand in the data set is 300 mg / L and the maximum value is 800 mg / L, then the normalized value of chemical oxygen demand of 520 mg / L is (520-300) / (800-300)=0.44. The same processing is performed on the biochemical oxygen demand and total suspended solids and the like. The normalized historical water quality parameters and the corresponding historical treatment effect data together constitute the training samples.

[0031] The training samples are input into a pre-set neural network for training. The neural network adopts a multi-layer perceptron structure, including an input layer, a hidden layer and an output layer. The number of input layer neurons is equal to the number of water quality parameters, such as 7 parameters of pH value, COD, BOD, TSS and the like corresponding to 7 input neurons; the hidden layer is initially set to two layers, each containing 10 neurons; the number of output layer neurons is equal to the number of treatment effect indicators, such as 2 output neurons corresponding to 2 indicators of effluent water quality and treatment efficiency. The neurons are connected through connection weights, and the initial connection weights are randomly set between -0.5 and 0.5.

[0032] The network output value is calculated by forward propagation. In the forward propagation process, each neuron receives weighted input from all neurons in the previous layer. The weighted input is the sum of the product of the output value of the neuron in the previous layer and the connection weight. Each neuron uses an activation function to process the weighted input to obtain the output value. The hidden layer uses a ReLU (Rectified Linear Unit) activation function, which outputs equal to the input when the input is positive, and outputs zero when the input is negative. The output layer uses a Sigmoid activation function to map any real number to a range of 0 to 1.

[0033] The error between the network output value and the corresponding historical treatment effect data is calculated to obtain a loss value, and the loss value adopts mean square error, i.e., the square sum of the difference between the network output value and the actual value divided by the sample number. For example, the neural network predicts that the COD removal rate is 83.2%, while the actual treatment efficiency is 81.7%, and the error is (83.2%-81.7%) 2 =2.25. The total loss value is obtained by accumulating the errors of all output neurons.

[0034] The error term of each layer of neurons in the neural network is calculated by the back propagation algorithm. The back propagation starts from the output layer, calculates the error term of each neuron, and then propagates back to the hidden layer layer by layer. For the output layer neurons, the error term is the difference between the network output value and the actual value multiplied by the derivative of the activation function; for the hidden layer neurons, the error term is the weighted sum of the error terms of all connected neurons in the next layer multiplied by the derivative of the activation function.

[0035] The gradient value of the connection weight is calculated according to the error term of each layer of neurons, and the gradient value of the connection weight is equal to the output value of the neuron in the previous layer connected with the weight multiplied by the error term of the corresponding neuron in the next layer. Based on the gradient descent algorithm and the gradient value, the connection weight is iteratively updated, and the update formula is: the new weight is equal to the old weight minus the learning rate multiplied by the gradient value. The learning rate is set to 0.01, and the iteration number is 1000 times. Through multiple iterations, the connection weight that minimizes the loss value, i.e., the initial connection weight, is obtained.

[0036] The parameter change is calculated using the current water quality parameters and the water quality parameters at the last collection time. For example, the current COD is 550 mg / L, and the last COD is 520 mg / L, so the parameter difference is 550-520=30 mg / L. The change rate is obtained by dividing the parameter difference by the water quality parameter at the last collection time, i.e., 30 / 520=5.77%. The arithmetic mean of the change rates of all water quality parameters is obtained, and the fluctuation amplitude is obtained. Assuming that the change rates of pH, COD, BOD, TSS, color, turbidity, and conductivity are 2.1%, 5.77%, 4.8%, 3.5%, 6.2%, 5.1%, and 2.8%, respectively, the fluctuation amplitude is (2.1%+5.77%+4.8%+3.5%+6.2%+5.1%+2.8%) / 7=4.32%.

[0037] According to the comparison between the fluctuation amplitude and the preset fluctuation amplitude threshold, the hidden layer structure of the neural network is dynamically adjusted. The fluctuation amplitude threshold is set to 3%. When the fluctuation amplitude is greater than the threshold, the number of newly added neurons is calculated by a linear growth function. The linear growth function is: the number of newly added neurons is equal to the basic growth number (set to 2) plus the difference between the fluctuation amplitude and the threshold multiplied by the amplification coefficient (set to 10). In this embodiment, the number of newly added neurons is 2+(4.32%-3%)*10=2+13.2=15.2, rounded to 15. At this time, 15 neurons are added to each hidden layer, and the size of the hidden layer is increased from the original 10 neurons per layer to 25 neurons per layer.

[0038] When the fluctuation amplitude is less than the threshold, the number of neurons to be reduced is calculated by a linear decay function. The linear decay function is: the number of neurons to be reduced is equal to the basic reduction number (set to 1) plus the difference between the threshold and the fluctuation amplitude multiplied by the reduction coefficient (set to 5). For example, if the fluctuation amplitude is 2.5%, the number of neurons to be reduced is 1+(3%-2.5%)*5=1+2.5=3.5, rounded to 4. At this time, 4 neurons are reduced from each hidden layer, and the size of the hidden layer is reduced from the original 10 neurons per layer to 6 neurons per layer.

[0039] The initial connection weights in the adjusted neural network structure are updated using an online learning algorithm. Each time a new set of water quality parameters is received, the neural network is updated once using the set of parameters and the corresponding treatment effect data. The weight adjustment amount is calculated by a stochastic gradient descent algorithm, which is equal to the learning rate multiplied by the gradient value of a single sample. The weight adjustment amount is added to the initial connection weight to obtain the optimized connection weight.

[0040] In this embodiment, the historical water quality parameters and treatment effect data are used to construct a training data set, and the normalization preprocessing improves the stability and comparability of data input, ensuring the effectiveness of neural network training. The network parameters are optimized and updated by combining the forward propagation and back propagation algorithms, which improves the accuracy and convergence speed in water quality treatment effect prediction. The online learning algorithm and stochastic gradient descent algorithm are used to update the weights of the optimized structure, which can continuously adapt to the dynamic changes of real-time water quality data, improve the stability and self-adaptive ability of the calculation results, and significantly enhance the intelligent level and real-time response ability of the water quality treatment process.

[0041] In an alternative embodiment, The corresponding relationship between the current water quality parameters and the device operation efficiency is calculated based on the optimized connection weights, the abnormal type and the abnormal number are determined according to the corresponding relationship to generate water quality abnormal warning information, and the water quality abnormal warning information is sent to the central control platform, which comprises: The current water quality parameter is taken as input, a linear combination of the optimized connection weight and a preset bias term is combined, and a corresponding relationship between the current water quality parameter and the equipment operation efficiency is obtained through a sigmoid activation function; Based on the current water quality parameter, a water quality parameter correlation graph is constructed, the current water quality parameter is taken as a vertex set of the correlation graph, the corresponding relationship is taken as an edge set of the water quality parameter correlation graph, and the dynamic correlation strength between the current water quality parameters is calculated in combination with a preset attention weight matrix; Based on the dynamic correlation strength, the edge set of the correlation graph is updated, an adjacency matrix is constructed according to the connection relationship between the vertices in the updated correlation graph, and the number of connections of each vertex is calculated to generate a degree matrix based on the adjacency matrix. Feature extraction is performed on the updated correlation graph through graph convolution operation, local region features are obtained through combined operation of the adjacency matrix and the degree matrix, global state features are obtained through difference pooling operation, and comprehensive features are obtained by weighted fusion of the local region features and the global state features; Based on the historical data obtained in advance, the water quality parameter reference value under normal working conditions is calculated, the Mahalanobis distance is calculated by comparing the real-time monitoring data with the water quality parameter reference value, the Mahalanobis distance is corrected based on the comprehensive features, the adaptive threshold is calculated according to the change trend of the corrected Mahalanobis distance combined with the comprehensive features, the abnormal type and the abnormal number are judged according to the corresponding relationship, the water quality abnormality early warning information is generated, and the information is sent to the central control platform through the secure transmission protocol.

[0042] After the edge computing node obtains the current water quality parameters as input, linear combination operation is performed with the optimized connection weight and the pre-set bias term. Each water quality parameter is multiplied by the corresponding connection weight and then added, and then the bias term is added to obtain the intermediate result. For example, the current water quality parameters include pH value 7.8, COD value 580 mg / L, BOD value 320 mg / L, TSS value 210 mg / L, color value 300 degrees, turbidity value 85 NTU, and conductivity value 420 μS / cm. The corresponding optimized connection weights are 0.32, 0.56, 0.48, 0.38, 0.42, 0.35, and 0.29, respectively, and the bias term is -0.25. The linear combination result is 7.8 x 0.32 + 580 x 0.56 + 320 x 0.48 + 210 x 0.38 + 300 x 0.42 + 85 x 0.35 + 420 x 0.29 - 0.25 = 703.53. The linear combination result is converted by the sigmoid activation function for non-linear conversion. The sigmoid function maps any real number to a value between 0 and 1. The calculation formula is 1 divided by (1 plus the negative input value of the natural base e raised to the power). For the linear combination result 703.53, the result obtained after the sigmoid function conversion is close to 1, indicating that the current water quality parameter combination has a great impact on the device operation efficiency. Repeat the calculation to obtain the corresponding relationship between all water quality parameters and device operation efficiency indicators, such as COD removal rate 85.3%, BOD removal rate 88.7%, color removal rate 92.1%, etc.

[0043] Based on the current water quality parameters, a water quality parameter correlation graph is constructed. In the correlation graph, each water quality parameter is taken as a vertex set, for example, pH value, COD, BOD, TSS, color, turbidity, and conductivity, which form 7 vertices. The corresponding relationship calculated is taken as the edge set of the correlation graph, that is, the connection relationship between parameters. For example, the corresponding relationship between pH value and COD is 0.72, indicating that there is a strong correlation between them, so an edge is established from the pH value vertex to the COD vertex in the correlation graph, and the edge weight is 0.72. The dynamic correlation strength between water quality parameters is calculated in combination with the pre-set attention weight matrix. The attention weight matrix is a 7 x 7 matrix representing the mutual influence degree between parameters. Through the product operation of the current water quality parameters and the attention weight matrix, the dynamic correlation strength reflecting the interaction between parameters under the current state is obtained. For example, the attention weight of pH value to COD is 0.65, and the current pH value is 7.8, so the dynamic correlation strength is 7.8 x 0.65 = 5.07, indicating that the current pH value has a large influence on COD.

[0044] The edge set of the association graph is updated based on the dynamic correlation strength. If the dynamic correlation strength exceeds a preset threshold value 0.5, a corresponding edge is established in the association graph. If it is lower than the threshold value, the original edge is deleted. The updated association graph more accurately reflects the actual correlation between the parameters under the current water quality state. An adjacency matrix is constructed according to the connection relationship between the vertices in the updated association graph. The adjacency matrix is a 7x7 matrix. For example, if there is a connection between parameter i and parameter j, the value at the corresponding position in the matrix is 1, otherwise it is 0. The number of connections of each vertex is calculated based on the adjacency matrix to generate a degree matrix. The degree matrix is also a 7x7 matrix. The elements on the main diagonal line represent the number of connections of the corresponding vertex, and the other positions are 0. For example, if the pH value has a connection with other four parameters, the diagonal element corresponding to the pH value in the degree matrix is 4.

[0045] Feature extraction is performed on the updated association graph through graph convolution operation. The graph convolution operation combines the combination operation of the adjacency matrix and the degree matrix to extract local region features. The negative square root of the degree matrix is multiplied by the adjacency matrix, and then multiplied by the current water quality parameter vector to obtain the local region features. For example, through the graph convolution operation, the local region feature vector reflecting the combination features of the pH value and its adjacent parameters [0.68, 0.72, 0.55, 0.61, 0.48, 0.59, 0.52] can be obtained. Global state features are obtained through differential pooling operation. The differential pooling operation extracts global structure information by calculating the difference between adjacent vertex features. The differential features of all vertices are combined to obtain the global state feature vector [0.15, 0.22, 0.18, 0.09, 0.25, 0.12, 0.20]. The local region features and the global state features are weighted and fused to obtain the comprehensive feature vector [0.53, 0.57, 0.45, 0.46, 0.42, 0.45, 0.43], with weights of 0.7 and 0.3 respectively.

[0046] The water quality parameter baseline value under normal working conditions is calculated based on the historical data obtained in advance. The mean and standard deviation of each water quality parameter are obtained by statistical analysis of historical normal operation data. For example, the baseline mean of the pH value is 7.2, and the standard deviation is 0.3; the baseline mean of COD is 500 mg / L, and the standard deviation is 50 mg / L. The Mahalanobis distance is calculated between the real-time monitoring data and the water quality parameter baseline value. The Mahalanobis distance takes into account the correlation between parameters, and can more accurately measure the deviation of the current state from the normal state. For example, the calculation result of the Mahalanobis distance of the current water quality parameter is 2.8, indicating that the current state deviates from the normal state to a certain extent.

[0047] The Mahalanobis distance is modified based on the comprehensive characteristics, and the elements in the comprehensive characteristic vector are weighted and summed to obtain a modification factor 0.65, the weight being the importance coefficient of each parameter. The modified Mahalanobis distance is obtained by multiplying the modification factor by the Mahalanobis distance, and the modified Mahalanobis distance is 1.82. The adaptive threshold is calculated according to the change trend of the modified Mahalanobis distance combined with the comprehensive characteristics. The calculation of the adaptive threshold takes into account the historical abnormal patterns and the particularity of the current water quality parameters, for example, when the characteristic values of pH and COD in the comprehensive characteristics are relatively high, the abnormal judgment threshold is reduced. The basic threshold is set to 2.0, and the currently calculated adaptive threshold is 1.75.

[0048] If the modified Mahalanobis distance exceeds the adaptive threshold, it is determined to be an abnormal state; according to the magnitude of the threshold exceeded and the distribution characteristics of the comprehensive characteristics, the abnormal type is determined. For example, when the modified Mahalanobis distance is 1.82, which exceeds the adaptive threshold 1.75, and the characteristic values of pH and COD are significantly higher than those of other parameters, it is determined to be two abnormal types of "abnormality of acid-base balance" and "excessive organic load". The water quality abnormality warning information is generated, including the abnormal time, the abnormal type, the abnormal degree and the recommended treatment scheme. The warning information is sent to the central control platform through the TLS encryption protocol to ensure the security and reliability of data transmission.

[0049] In this embodiment, the dynamic correlation strength is calculated through the attention weight matrix, which can accurately capture the change relationship and influence relationship between water quality parameters. The Mahalanobis distance between the real-time monitoring data and the reference value is calculated, and it is modified combined with the comprehensive characteristics, which can effectively reduce the false judgment caused by single index abnormality. The adaptive threshold is calculated based on the change trend of the modified Mahalanobis distance, which can flexibly adjust the judgment standard according to the water quality change, thereby improving the accuracy and sensitivity of abnormal detection. The corresponding relationship is used to identify the abnormal type and quantity, generate water quality abnormality warning information and safely transmit it to the central control platform, realize timely and accurate warning of water quality abnormality, enhance the reliability of monitoring, and improve the adaptability to complex dynamic environment.

[0050] In an alternative embodiment, The water quality abnormality warning information is substituted into the optimization calculation function with the water quality compliance rate, system energy consumption and reagent cost as the optimization objectives to obtain the optimization objective function value. The operation parameters of the textile printing and dyeing wastewater treatment equipment are converted into optimization calculation variables, and the particle swarm algorithm is used to calculate the operation parameter values, including: The water quality abnormality warning information is substituted into the optimization calculation function with the water quality compliance rate, system energy consumption and reagent cost as the optimization objectives to obtain the optimization objective function value. According to the optimization objective function value, the operation parameters of the textile printing and dyeing wastewater treatment equipment are converted into optimization calculation variables and combined into an optimization calculation variable vector, the particle swarm size and the maximum iteration number are initialized based on the optimization calculation variable vector, the initial temperature and the temperature reduction coefficient are set according to the optimization objective function value, the crossover probability and the scaling factor are calculated and set, the initial population distribution is generated, and the operation parameters include aeration amount adjustment parameters, reagent dosage parameters, hydraulic load parameters and sludge return ratio parameters; According to the initial population distribution, the velocity update equation and the position update equation are determined, the inertia weight in the velocity update equation and the position update equation is dynamically adjusted based on the ratio of the current iteration number to the maximum iteration number to obtain updated velocity values and position values, and the learning factor in the velocity update equation is adaptively updated based on the optimization objective function value to obtain an optimized population distribution; The global optimal solution is obtained by performing the particle swarm algorithm using the optimized population distribution, the global optimal solution is substituted into the simulated annealing operation, the local optimal solution is obtained in combination with the initial temperature and the temperature reduction coefficient, the candidate solution set is obtained by updating the population distribution corresponding to the local optimal solution through the differential evolution operation based on the crossover probability and the scaling factor, and the optimal solution meeting the pre-set constraint condition is selected from the candidate solution set to calculate the operation parameter value.

[0051] The edge computing node receives the water quality abnormal early warning information and substitutes it into the optimization calculation function, the optimization calculation function takes the water quality standard reaching rate, the system energy consumption and the reagent cost as the optimization objectives, adopts the weighted summation method to construct a composite objective function, and exemplarily sets the water quality standard reaching rate weight to 0.5, the system energy consumption weight to 0.3 and the reagent cost weight to 0.2. When the water quality abnormal early warning information shows that the COD is over standard and the pH value is abnormal, the optimization calculation function dynamically adjusts the weights of the objectives according to the abnormal degree, for example, the weight of the water quality standard reaching rate is increased to 0.6, and the weights of the energy consumption and the cost are reduced to 0.25 and 0.15 respectively. For example, under the current water quality abnormal condition, the water quality standard reaching rate is 85%, the system energy consumption is 280 kWh / d, and the reagent cost is 1200 yuan / d, and after being substituted into the optimization calculation function, the optimization objective function value is 0.72.

[0052] The operation parameters of the textile printing and dyeing wastewater treatment equipment are converted into optimization calculation variables, including aeration amount adjustment parameters, reagent dosage parameters, hydraulic load parameters and sludge return ratio parameters, the aeration amount adjustment parameter range is set to 1.5-4.0 m 3 / min, the reagent dosage parameter range is 20-60 mg / L, polyacrylamide 2-8 mg / L, and the hydraulic load parameter range is 15-25 m 3 / (m 2• d), the sludge return ratio parameter range is 30%-80%, the parameter combination is the optimization calculation variable vector, such as [2.5, 40, 5, 20, 60], which respectively corresponds to the aeration amount, the PAC dosage, the polyacrylamide dosage, the hydraulic load and the sludge return ratio.

[0053] The particle swarm size and the maximum iteration number are initialized based on the optimization calculation variable vector. The particle swarm size is set to 30, and the maximum iteration number is set to 100. The initial temperature and the temperature reduction coefficient are set according to the optimization objective function value. The initial temperature T0=100×optimization objective function value is 100×0.72=72, and the temperature reduction coefficient a is set to 0.95, that is, the temperature is reduced by 5% each iteration. The crossover probability and the scaling factor are calculated and set. The crossover probability CR is set to 0.8, and the scaling factor F is set to 0.6. The initial population distribution is generated, and 30 particles are generated by random sampling in the parameter value range, each particle representing a possible combination of operating parameters.

[0054] The velocity update equation and the position update equation are determined according to the initial population distribution. The velocity update equation includes three parts: the inertia term, the cognitive term and the social term. The inertia term represents the ability of the particle to maintain its original motion trend, the cognitive term represents the influence of the particle's own historical optimal position, and the social term represents the influence of the particle's group historical optimal position. The position update equation obtains the new position by adding the current position and the updated velocity. The inertia weight in the velocity update equation is dynamically adjusted based on the ratio of the current iteration number to the maximum iteration number. The initial inertia weight is set to 0.9, which decreases linearly to 0.4 as the iteration proceeds. For example, when the iteration proceeds to the 50th time, the inertia weight is 0.9-(0.9-0.4)×(50 / 100)=0.65. The updated velocity value and position value are calculated using the adjusted inertia weight.

[0055] The learning factor in the velocity update equation is updated adaptively based on the optimization objective function value. The learning factor includes the cognitive learning factor c1 and the social learning factor c2, which are both initially set to 2.0. The learning factor is dynamically adjusted according to the difference between the current solution and the optimal solution. When the difference between the current solution and the optimal solution is large, the cognitive learning factor is increased to 2.5 and the social learning factor is decreased to 1.5 to promote independent exploration of the particle; when the difference is small, the cognitive learning factor is decreased to 1.5 and the social learning factor is increased to 2.5 to accelerate convergence. Through multiple iterations of updating, the optimized population distribution is obtained.

[0056] The particle swarm optimization algorithm is executed using the optimized population distribution to obtain a global optimal solution. In each iteration, the fitness value of each particle is calculated, and the individual optimal position and global optimal position are updated. After 100 iterations, the global optimal solution [3.2, 45, 6, 18, 65] is obtained, and the corresponding optimized objective function value is 0.85. The global optimal solution is substituted into the simulated annealing operation for further optimization. The simulated annealing algorithm can jump out of the local optimal solution and find a better solution by introducing random perturbation and probability acceptance mechanism. The initial solution is set to the global optimal solution obtained by the particle swarm optimization algorithm, the initial temperature is 72, and the cooling coefficient is 0.95.

[0057] In the simulated annealing process, a neighborhood solution is generated in each iteration, and the difference in objective function value between the new solution and the current solution is calculated. If the new solution is better, the new solution is accepted; if the new solution is worse, the new solution is accepted with a certain probability, and the probability value is the negative difference value divided by the current temperature power. For example, the current solution is [3.2, 45, 6, 18, 65], the objective function value is 0.85; the neighborhood solution is [3.3, 43, 7, 19, 63], the objective function value is 0.83, the difference is -0.02; the current temperature is 40.3, and the acceptance probability is e's negative (-0.02) / 40.3 power, which is about 1.0005, greater than the randomly generated 0.7, so the new solution is accepted. After 30 temperature iterations, the temperature decreases to 72 x (0.95)^30 ≈ 10.6, and the local optimal solution [3.5, 42, 7, 19, 68] is obtained, corresponding to the optimized objective function value of 0.89.

[0058] Based on the crossover probability of 0.8 and the scaling factor of 0.6, the population distribution corresponding to the local optimal solution is updated by the differential evolution operation. The differential evolution algorithm generates new candidate solutions through mutation, crossover and selection operations. The mutation operation selects three different individuals in the population, and generates a mutation vector by weighted combination of the difference between the first individual and the second and third individuals. For example, select individuals [3.1, 44, 5, 17, 70], [3.3, 46, 6, 18, 65] and [2.9, 40, 7, 20, 60], calculate the mutation vector [3.1, 44, 5, 17, 70]+0.6x([3.3, 46, 6, 18, 65]-[2.9, 40, 7, 20, 60])=[3.1, 44, 5, 17, 70]+0.6x([0.4, 6, -1, -2, 5])=[3.1, 44, 5, 17, 70]+[0.24, 3.6, -0.6, -1.2, 3]=[3.34, 47.6, 4.4, 15.8, 73].

[0059] The cross operation exchanges information between the mutation vector and the target vector to generate a trial vector. For each component of the mutation vector, a random number between 0 and 1 is generated. If the random number is less than the crossover probability 0.8, the corresponding component of the trial vector is replaced by the component of the mutation vector. For example, the target vector is [3.5, 42, 7, 19, 68] and the mutation vector is [3.34, 47.6, 4.4, 15.8, 73]. Five random numbers between 0 and 1 are generated, i.e., [0.75, 0.92, 0.32, 0.88, 0.65]. Comparing the random numbers with the crossover probability 0.8, the trial vector is [3.34, 42, 4.4, 19, 73]. The selection operation compares the fitness values of the trial vector and the target vector, and the better one is selected to enter the next generation population. Multiple candidate solutions are generated to form a candidate solution set.

[0060] The optimal solution that satisfies the pre-set constraint conditions is selected from the candidate solution set. The constraint conditions include: the aeration amount should meet the COD load requirement, the reagent dosage should be within the effective range, the hydraulic load should consider the area limitation of the sedimentation tank, and the sludge return ratio should ensure the stability of the activated sludge concentration. For example, when the COD load is 500 kg / d, the minimum aeration requirement is 2.8 m 3 / min; the maximum dosages of PAC and polyacrylamide are 60 mg / L and 8 mg / L, respectively; the area of the sedimentation tank is 50 m 2 , and the influent flow is 800 m 3 / d, the hydraulic load should not exceed 16 m 3 / (m 2 ·d).

[0061] After the constraint condition screening, the optimal solution is finally determined as [3.2, 42, 6, 16, 70], i.e., the aeration amount is 3.2 m 3 / min, the PAC dosage is 42 mg / L, the polyacrylamide dosage is 6 mg / L, the hydraulic load is 16 m 3 / (m 2 ·d), and the sludge return ratio is 70%.

[0062] In this embodiment, the operation parameters of the sewage treatment equipment are converted into optimization calculation variables and a vector is constructed, particle swarm initialization and simulated annealing parameter setting are used, and initial population distribution is generated by combining crossover probability and scaling factor, so as to provide reasonable initial conditions for subsequent global search and local search. Through dynamic adjustment of inertia weight and learning factor in the iteration process, the globality of the search can be enhanced in the early stage, the convergence accuracy can be improved in the later stage, the convergence speed of the optimization process and the stability of the solution can be improved, the local search of the solution space is performed through the simulated annealing mechanism, the local extreme value can be effectively avoided, the population update of the local optimal solution is performed through the differential evolution operation, the candidate solution set is formed, and the final optimal solution is obtained through the constraint condition screening, so that the optimization result is more in line with the actual operation demand of the sewage treatment process. Under the premise of ensuring that the effluent quality meets the standard, the system energy consumption and the reagent cost can be effectively reduced, and the economic efficiency, stability and intelligent level of the operation can be improved.

[0063] In an alternative embodiment, According to the optimization objective function value, the adjustment priority level and the adjustment amplitude of the operation parameter are determined, the operation parameter adjustment instruction is generated and executed, and the operation parameter adjustment instruction is generated and executed, including: The operation time sequence data corresponding to the operation parameter is collected to calculate the parameter correlation degree, the ratio of the dispersion product to the standard deviation product of the operation time sequence data is calculated, the parameter correlation degree and the optimization objective function value deviation corresponding to the operation parameter are normalized and multiplied to obtain the causal relationship weight between the operation parameters; The causal relationship weight between the operation parameters and the operation time sequence data response increment ratio are multiplied to construct a direct influence matrix, and the direct influence matrix is summed by power series to obtain an indirect influence propagation matrix. The indirect influence propagation matrix and the pre-set operation parameter importance weight are multiplied to calculate a comprehensive influence degree matrix; The comprehensive influence degree matrix and the normalized result of the optimization objective function value deviation are multiplied to calculate the operation parameter adjustment priority level, and the operation parameter adjustment priority level and the causal relationship weight between the operation parameters are weighted and averaged to obtain a cooperative adjustment coefficient; The operation parameter adjustment priority level, the cooperative adjustment coefficient and the optimization objective function value deviation are multiplied to calculate the adjustment amplitude, the operation parameter value and the adjustment amplitude are added to generate the operation parameter adjustment instruction, and the operation parameter adjustment instruction is executed according to the order of the operation parameter adjustment priority level.

[0064] The edge computing node collects the operation parameter corresponding runtime sequence data to calculate the parameter correlation degree. The operation parameters include aeration quantity, reagent dosage, hydraulic load and sludge return ratio, and the operation time sequence data refers to the numerical record of these parameters at continuous multiple time points. For example, the aeration quantity data is recorded once an hour within 24 hours: [3.2, 3.1, 3.3, 3.2, 3.4, 3.5, 3.3, 3.2, 3.1, 3.0, 3.2, 3.3, 3.4, 3.5, 3.6, 3.4, 3.3, 3.2, 3.1, 3.0, 3.1, 3.2, 3.3, 3.4] m3 / min; the PAC dosage data: [42, 43, 44, 45, 46, 45, 44, 43, 42, 41, 40, 41, 42, 43, 44, 45, 46, 45, 44, 43, 42, 41, 40, 41] mg / L. The ratio of the dispersion product of the time sequence data to the standard deviation product is obtained to calculate the parameter correlation degree. The dispersion product refers to the product sum of the deviation of two parameters at the same time from the respective mean value, and the standard deviation product refers to the product of the standard deviations of the two parameters. For the aeration quantity and the PAC dosage, the calculated correlation degree is 0.85, indicating that the two have strong correlation. The correlation degrees between all pairs of parameters are calculated to form a correlation degree matrix.

[0065] The parameter correlation degree is multiplied by the normalized deviation of the operation parameter corresponding to the optimization objective function value to obtain the causal relationship weight between the operation parameters. The deviation of the optimization objective function value refers to the gap between the actual operation effect of the system and the ideal optimization objective. For example, the actual value of the effluent COD is 65 mg / L, the target value is 50 mg / L, and the deviation is 15 mg / L. The actual value of the system energy consumption is 260 kWh / d, the target value is 240 kWh / d, and the deviation is 20 kWh / d. The deviation value is normalized to make the range between 0 and 1. After normalization, the COD deviation is 0.6 and the energy consumption deviation is 0.4. The correlation degree is multiplied by the normalized deviation, for example, the causal relationship weight between the aeration quantity and the PAC dosage is 0.85 x 0.6 = 0.51. The causal relationship weights between all pairs of parameters are calculated to construct a causal relationship weight matrix.

[0066] The causal relationship weight between the operation parameters is multiplied by the response increment ratio of the operation time sequence data to construct a direct influence matrix. The response increment ratio refers to the ratio of the response change amount of one parameter to the original change amount when the other parameter changes. For example, when the aeration quantity increases from 3.2 m 3 / min to 3.5 m 3 / min, the increment is 0.3 m 3 / min; while the dissolved oxygen increased from 2.5 mg / L to 3.2 mg / L, the increment was 0.7 mg / L; the response increment ratio was 0.7 / 0.3 = 2.33. The causal relationship weight was multiplied by the response increment ratio, and the direct influence value of the aeration rate on the dissolved oxygen was 0.51 x 2.33 = 1.19. In this way, the direct influence values between all parameters were calculated, and a direct influence matrix was constructed.

[0067] The indirect influence propagation matrix was obtained by power series summation of the direct influence matrix. The indirect influence propagation considered the cascade effect between parameters, i.e., the change of one parameter indirectly influenced other parameters through intermediate parameters. The power series summation was a weighted summation of the first power, the second power, the third power, etc. of the direct influence matrix, and the weight decreased with the increase of the power. In the specific implementation, the first 5 terms were taken for approximate calculation. For example, the indirect influence value of the aeration rate on the effluent COD in the calculated indirect influence propagation matrix was 1.75, indicating that the change of the aeration rate had a strong influence on the effluent COD through a series of intermediate parameters. The indirect influence propagation matrix was multiplied by the pre-set importance weight of the operating parameters to calculate a comprehensive influence degree matrix. The importance weight of the operating parameters was determined based on pre-obtained expert experience or historical data analysis. For example, the importance of the aeration rate was 0.3, the importance of the PAC dosage was 0.25, the importance of the hydraulic load was 0.2, and the importance of the sludge return ratio was 0.25. After the product operation, the comprehensive influence degree matrix was obtained.

[0068] The operating parameter adjustment priority level was calculated by multiplying the comprehensive influence degree matrix and the normalized result of the optimization objective function value deviation. For example, for the case of effluent COD exceeding the standard, the calculated adjustment priority level of the aeration rate was 0.85, the adjustment priority level of the PAC dosage was 0.72, the adjustment priority level of the hydraulic load was 0.35, and the adjustment priority level of the sludge return ratio was 0.65, indicating that the aeration rate should be adjusted first, followed by the PAC dosage and the sludge return ratio, and finally the hydraulic load. The synergistic adjustment coefficient was obtained by weighted average operation of the operating parameter adjustment priority level and the causal relationship weight between the operating parameters. The synergistic adjustment coefficient considered the mutual influence between parameters, avoiding system instability caused by single parameter adjustment. For example, the synergistic adjustment coefficient of the aeration rate was 0.75, the synergistic adjustment coefficient of the PAC dosage was 0.68, the synergistic adjustment coefficient of the hydraulic load was 0.42, and the synergistic adjustment coefficient of the sludge return ratio was 0.63.

[0069] The adjustment amplitude was calculated by multiplying the operating parameter adjustment priority level, the synergistic adjustment coefficient, and the optimization objective function value deviation. For example, the current aeration rate was 3.2 m 3 / min, the adjustment priority level is 0.85, the synergistic adjustment coefficient is 0.75, and the normalized deviation of the optimization objective function value is 0.6. Therefore, the adjustment range of the aeration amount is 0.85*0.75*0.6=0.38. The specific adjustment amount is obtained by multiplying the adjustment range by the current aeration amount, i.e., 3.2*0.38=1.22 m 3 / min. The running parameter adjustment instruction is generated by adding the running parameter value to the adjustment range, i.e., the new aeration amount set value is 3.2+1.22=4.42 m 3 / min. The adjustment instructions for other running parameters are calculated: the PAC dosage is adjusted from 42 mg / L to 50 mg / L, the hydraulic loading is adjusted from 16 m 3 / (m 2 ·d) to 15 m 3 / (m 2 ·d), and the sludge return ratio is adjusted from 70% to 82%.

[0070] The running parameter adjustment instructions are executed in the order of the running parameter adjustment priority level. First, the aeration amount is adjusted to 4.42 m 3 / min, and the system response is observed for 15 minutes; then the PAC dosage is adjusted to 50 mg / L, and the system response is observed for 10 minutes; the sludge return ratio is adjusted to 82%, and the system response is observed for 20 minutes; and the hydraulic loading is adjusted to 15 m 3 / (m 2 ·d), and the system response is observed for 30 minutes. After the adjustment is completed, the effluent COD decreases from 65 mg / L to 52 mg / L, which is close to the target value of 50 mg / L. The energy consumption increases from 260 kWh / d to 275 kWh / d, which is slightly higher than the target value of 240 kWh / d. However, considering the benefits of water quality improvement, the current energy consumption increase is acceptable.

[0071] In this embodiment, the indirect influence propagation matrix is obtained by power series summation, and the comprehensive influence degree matrix is calculated by combining the preset running parameter importance weight, which realizes comprehensive analysis of the direct and indirect effects of running parameters, helps to more accurately depict the degree of action of each parameter in the overall operation, calculates the adjustment priority level using the normalized result of the comprehensive influence degree matrix and the optimization objective function deviation, and performs weighted averaging with the causal relationship weight to obtain the synergistic adjustment coefficient, so that the parameter adjustment not only considers the importance of a single parameter, but also reflects the synergistic action relationship between multiple parameters. The adjustment range is calculated by the adjustment priority level, the synergistic adjustment coefficient, and the target function value deviation, the running parameter adjustment instructions are generated and executed in sequence, which can realize the ordered and adaptive optimization adjustment of running parameters, effectively identify the causal relationship and comprehensive influence between running parameters, reasonably determine the adjustment sequence and range, ensure the achievement of the optimization target while improving the coordination and execution efficiency of the adjustment, and thus enhance the stability and flexibility of the wastewater treatment process.

[0072] Figure 2 The sewage treatment parameter optimization and adjustment flowchart of the textile printing and dyeing sewage intelligent treatment method based on edge computing of the embodiments of the present application is shown in FIG. 1.

[0073] In an optional embodiment, The real-time monitoring of the running state of the textile printing and dyeing sewage treatment equipment and the updating of the running parameter adjustment instruction include: The running state data of the textile printing and dyeing sewage treatment equipment are collected in real time, the running state data are compared with the expected response value after the execution of the running parameter adjustment instruction, and the running state deviation is calculated; The adjustment response degree of the running parameter is determined according to the running state deviation, and when the running state deviation exceeds a preset deviation threshold, the updating coefficient of the running parameter is calculated based on the running state deviation; The updating coefficient is multiplied by the original running parameter adjustment instruction to obtain the updated running parameter adjustment instruction, and the running parameter of the textile printing and dyeing sewage treatment equipment is adjusted according to the updated running parameter adjustment instruction.

[0074] The edge computing node collects the running state data of the textile printing and dyeing sewage treatment equipment in real time, and the collection frequency is once every 5 minutes. The running state data include dissolved oxygen concentration, mixed liquor suspended solids concentration (MLSS), pH value, oxidation reduction potential (ORP), sludge settling ratio (SV30), effluent turbidity, effluent COD and other indicators. For example, after the aeration amount is adjusted from 3.2 m 3 / min to 4.42 m 3 / min according to the aeration amount adjustment instruction, the dissolved oxygen concentration collected by the edge computing node is 2.8 mg / L, while the expected response value is 3.5 mg / L; the MLSS is 3200 mg / L, and the expected response value is 3500 mg / L; the pH value is 7.2, and the expected response value is 7.3; the ORP is 120 mV, and the expected response value is 150 mV. The actual collection values are compared with the expected response values, and the running state deviation is calculated. The running state deviation is calculated by using the relative error method, that is, the absolute value of the difference between the actual value and the expected value is divided by the expected value. For the dissolved oxygen concentration, the running state deviation is |2.8-3.5| / 3.5=0.2, that is, 20%; for the MLSS, the running state deviation is |3200-3500| / 3500=0.086, that is, 8.6%; for the pH value, the running state deviation is |7.2-7.3| / 7.3=0.014, that is, 1.4%; and for the ORP, the running state deviation is |120-150| / 150=0.2, that is, 20%.

[0075] The degree of adjustment response of operating parameters is determined based on the deviation of operating status. The degree of adjustment response refers to the effect of the equipment on the execution of adjustment commands, and can be divided into three levels: high response, medium response, and low response. When the deviation of operating status is less than 5%, it is judged as high response; when the deviation is between 5% and 15%, it is judged as medium response; when the deviation is greater than 15%, it is judged as low response. In the example above, the deviation of dissolved oxygen concentration and ORP is 20%, which belongs to low response; the deviation of MLSS is 8.6%, which belongs to medium response; and the deviation of pH value is 1.4%, which belongs to high response. Different weights are set for the deviation of different indicators for comprehensive evaluation. For example, the weight of dissolved oxygen concentration is 0.4, the weight of MLSS is 0.3, the weight of pH value is 0.1, and the weight of ORP is 0.2. The weighted average deviation is calculated as 0.2×0.4+0.086×0.3+0.014×0.1+0.2×0.2=0.14, or 14%, which is close to the upper limit of medium response level.

[0076] When the operational deviation exceeds a preset deviation threshold, an update coefficient for the operational parameters is calculated based on the deviation. The preset deviation threshold is typically set to 10%, meaning that when the weighted average deviation exceeds 10%, the operational parameter adjustment command needs to be updated. In this embodiment, the weighted average deviation is 14%, exceeding the preset threshold of 10%, necessitating the calculation of the update coefficient. The calculation of the update coefficient considers both the degree of deviation and the direction of adjustment. When the actual value is less than the expected value, it indicates insufficient adjustment, and the update coefficient should be greater than 1; when the actual value is greater than the expected value, it indicates excessive adjustment, and the update coefficient should be less than 1. For dissolved oxygen concentration, the actual value of 2.8 mg / L is less than the expected value of 3.5 mg / L, so the update coefficient is set to 1 + (3.5 - 2.8) / 3.5 = 1.2; for MLSS, the actual value of 3200 mg / L is less than the expected value of 3500 mg / L, so the update coefficient is set to 1 + (3500 - 3200) / 3500 = 1.086; for pH, the actual value of 7.2 is less than the expected value of 7.3, so the update coefficient is set to 1 + (7.3 - 7.2) / 7.3 = 1.014; for ORP, the actual value of 120 mV is less than the expected value of 150 mV, so the update coefficient is set to 1 + (150 - 120) / 150 = 1.2. A weighted average of the update coefficients yields a comprehensive update coefficient of 1.2 × 0.4 + 1.086 × 0.3 + 1.014 × 0.1 + 1.2 × 0.2 = 1.14.

[0077] The updated operating parameter adjustment command is obtained by multiplying the updated coefficient by the original operating parameter adjustment command. For the aeration rate, the original adjustment command is 4.42 m³ / s. 3 / min, the updated adjustment command is 4.42 × 1.14 = 5.04m 3 / min. Similarly, the adjustment instruction of the PAC dosage is updated from the original instruction of 50 mg / L to 54 mg / L (based on the deviation calculation of the effluent turbidity and COD), the adjustment instruction of the sludge return ratio is updated from the original instruction of 82% to 86.1% (based on the deviation calculation of MLSS and SV30), and the adjustment instruction of the hydraulic load is updated from the original instruction of 15 m 3 / (m 2 ·d) to 14.25 m 3 / (m 2 ·d).

[0078] The running parameters of the textile printing and dyeing wastewater treatment equipment are adjusted according to the updated running parameter adjustment instructions. The adjustment sequence still follows the order of the priority level, that is, the aeration amount is first adjusted to 5.04 m 3 / min, then the PAC dosage is adjusted to 54 mg / L, the sludge return ratio is adjusted to 86.1%, and the hydraulic load is adjusted to 14.25 m 3 / (m 2 ·d). After each adjustment, the system response is observed to ensure that the system is stable before the next adjustment. For example, after adjusting the aeration amount, it is observed for 15 minutes, and it is found that the dissolved oxygen concentration rises to 3.3 mg / L, close to the expected value of 3.5 mg / L, and the deviation is reduced to 5.7%. After adjusting the PAC dosage, it is observed for 10 minutes, and it is found that the effluent turbidity is reduced from 12 NTU to 8 NTU, reaching the expected value. After adjusting the sludge return ratio, it is observed for 20 minutes, and it is found that the MLSS rises to 3450 mg / L, close to the expected value of 3500 mg / L, and the deviation is reduced to 1.4%. After adjusting the hydraulic load, it is observed for 30 minutes, and it is found that the sedimentation effect is significantly improved, and the effluent suspended solids are reduced from 28 mg / L to 15 mg / L, reaching the expected target.

[0079] In this embodiment, the running state of the textile printing and dyeing wastewater treatment equipment is collected in real time, and the actual running state is compared with the expected response value after executing the running parameter adjustment instruction to obtain the running state deviation, which can accurately reflect the execution effect of the adjustment instruction. According to the running state deviation, the adjustment response degree of the running parameter is determined, when the deviation exceeds the preset threshold, the update coefficient is calculated based on the deviation, and the original adjustment instruction is modified using the update coefficient to generate the updated adjustment instruction, which can realize dynamic correction and self-adaptive update of the running parameter adjustment, avoid deviation accumulation caused by static instruction setting, and adjust the equipment according to the updated running parameter adjustment instruction, which can improve the accuracy and timeliness of the running adjustment, and make the running state closer to the expected target faster.

[0080] In a second aspect, the present application provides a textile printing and dyeing wastewater intelligent treatment system based on edge computing, comprising: A first unit is configured to deploy edge computing nodes in a plurality of textile printing and dyeing wastewater treatment devices and collect current water quality parameters; A second unit is configured to obtain historical water quality parameters and historical treatment effect data to establish a training data set, train a pre-set neural network by using a back propagation algorithm to obtain initial connection weights, calculate a fluctuation amplitude of the current water quality parameters, adjust the number of hidden layer neurons of the neural network according to the fluctuation amplitude and update the initial connection weights to obtain optimized connection weights, calculate a corresponding relationship between the current water quality parameters and the device operation efficiency based on the optimized connection weights, determine an abnormal type and an abnormal quantity according to the corresponding relationship to generate water quality abnormality early warning information, and send the water quality abnormality early warning information to a central control platform; A third unit is configured to receive water quality abnormality early warning information sent by each edge computing node by the central control platform; A fourth unit is configured to substitute the water quality abnormality early warning information into an optimization calculation function with an effluent water quality compliance rate, system energy consumption and chemical cost as optimization objectives to obtain an optimization objective function value, convert operation parameters of the textile printing and dyeing wastewater treatment device into optimization calculation variables, calculate the operation parameter values by using a particle swarm algorithm, determine an adjustment priority level and an adjustment amplitude corresponding to the operation parameters according to the optimization objective function value, generate an operation parameter adjustment instruction and execute the operation parameter adjustment instruction; A fifth unit is configured to monitor a running state of the textile printing and dyeing wastewater treatment device in real time and update the operation parameter adjustment instruction.

[0081] In a third aspect, the present application provides an electronic device, comprising: A processor and a memory for storing processor-executable instructions, wherein the processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0082] In a fourth aspect, the present application provides a computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions are executed by a processor to implement the method described above.

[0083] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon for executing various aspects of the present application.

[0084] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent textile printing and dyeing wastewater treatment method based on edge computing, characterized in that, The application relates to a textile printing and dyeing wastewater treatment method and device. An edge computing node is deployed in a plurality of textile printing and dyeing wastewater treatment devices to collect current water quality parameters; Historical water quality parameters and historical treatment effect data are obtained to establish a training data set, an initial connection weight is obtained by training a pre-set neural network by combining a back propagation algorithm, a fluctuation amplitude of the current water quality parameters is calculated, the number of hidden layer neurons of the neural network is adjusted according to the fluctuation amplitude, the initial connection weight is updated to obtain an optimized connection weight, a corresponding relationship between the current water quality parameters and the device operation efficiency is calculated based on the optimized connection weight, an abnormal type and an abnormal quantity are judged according to the corresponding relationship to generate water quality abnormality early warning information, and the water quality abnormality early warning information is sent to a central control platform; The central control platform receives the water quality abnormality early warning information sent by each edge computing node; The water quality abnormality early warning information is substituted into an optimization calculation function with an effluent water quality compliance rate, system energy consumption and reagent cost as optimization objectives to obtain an optimization objective function value, operation parameters of the textile printing and dyeing wastewater treatment device are converted into optimization calculation variables, and the operation parameter values are calculated by combining a particle swarm algorithm, the adjustment priority and adjustment amplitude of the operation parameters corresponding to the optimization objective function value are determined, operation parameter adjustment instructions are generated and executed, and the operation state of the textile printing and dyeing wastewater treatment device is monitored in real time to update the operation parameter adjustment instructions. An edge computing node is deployed in a plurality of textile printing and dyeing wastewater treatment devices to collect current water quality parameters, which includes the following steps:

2. The method of claim 1, wherein, The position information of the textile printing and dyeing wastewater treatment device is collected, and a plurality of device groups are divided, each device group including at least two textile printing and dyeing wastewater treatment devices; An edge computing node is deployed in one textile printing and dyeing wastewater treatment device in each device group, and data collection nodes are deployed in the remaining textile printing and dyeing wastewater treatment devices in the current device group; Raw water quality data of the textile printing and dyeing wastewater treatment device are collected by the data collection nodes and sent to the edge computing node, and the edge computing node obtains the current water quality parameters by combining the mean values of different data in the raw water quality data. The historical water quality parameters and historical treatment effect data are obtained to establish a training data set, an initial connection weight is obtained by training a pre-set neural network by combining a back propagation algorithm, a fluctuation amplitude of the current water quality parameters is calculated, the number of hidden layer neurons of the neural network is adjusted according to the fluctuation amplitude, the initial connection weight is updated to obtain an optimized connection weight, a corresponding relationship between the current water quality parameters and the device operation efficiency is calculated based on the optimized connection weight, an abnormal type and an abnormal quantity are judged according to the corresponding relationship to generate water quality abnormality early warning information, and the water quality abnormality early warning information is sent to a central control platform; 3. The method of claim 1, wherein, The historical water quality parameters and historical treatment effect data are obtained to establish a training data set, the historical water quality parameters in the training data set are subjected to maximum and minimum normalization processing, and the normalized historical water quality parameters and historical treatment effect data are used as training samples. ​ The training sample is input into a pre-set neural network, the network output value is calculated through forward propagation, the loss value is calculated by error calculation on the network output value and the corresponding historical processing effect data, the error term of each layer neuron in the neural network is calculated through the back propagation algorithm, the gradient value of the connection weight is calculated according to the error term, the connection weight is iteratively updated based on the gradient descent algorithm and the gradient value, and the initial connection weight is obtained; The parameter difference value is obtained by subtracting the water quality parameter at the last collection time from the pre-acquired current water quality parameter, the change rate is obtained by dividing the parameter difference value by the water quality parameter at the last collection time, and the fluctuation amplitude is obtained by calculating the arithmetic mean value of the change rate; When the fluctuation amplitude is greater than the pre-set fluctuation amplitude threshold, the number of new neurons is calculated through a linear growth function, and when the fluctuation amplitude is less than the fluctuation amplitude threshold, the number of neurons is reduced through a linear decay function, the hidden layer structure of the neural network is adjusted according to the calculated number of neurons, the initial connection weight in the adjusted neural network structure is updated through an online learning algorithm, the weight adjustment amount is calculated through a stochastic gradient descent algorithm, and the optimized connection weight is obtained by adding the weight adjustment amount to the initial connection weight.

4. The method of claim 1, wherein, Based on the optimized connection weight, the corresponding relationship between the current water quality parameter and the equipment operation efficiency is calculated, the abnormal type and the abnormal number are judged according to the corresponding relationship to generate water quality abnormality early warning information, and the water quality abnormality early warning information is sent to the central control platform, including: The current water quality parameter is taken as input, a linear combination of the optimized connection weight and a pre-set bias term is combined, and the corresponding relationship between the current water quality parameter and the equipment operation efficiency is obtained through a sigmoid activation function; Based on the current water quality parameter, a water quality parameter correlation graph is constructed, the current water quality parameter is taken as the vertex set of the correlation graph, the corresponding relationship is taken as the edge set of the water quality parameter correlation graph, and the dynamic correlation strength between the current water quality parameters is calculated combined with a pre-set attention weight matrix; Based on the dynamic correlation strength, the edge set of the correlation graph is updated, an adjacency matrix is constructed according to the connection relationship between the vertices in the updated correlation graph, a degree matrix is generated based on the adjacency matrix, feature extraction is performed on the updated correlation graph through graph convolution operation, local region features are obtained through combination operation of the adjacency matrix and the degree matrix, global state features are obtained through difference pooling operation, and the local region features and the global state features are weighted and fused to obtain comprehensive features; Based on the pre-acquired historical data, the water quality parameter reference value under normal working conditions is calculated, the Mahalanobis distance is calculated from the real-time monitoring data and the water quality parameter reference value, the Mahalanobis distance is corrected based on the comprehensive features, the adaptive threshold is calculated according to the change trend of the corrected Mahalanobis distance combined with the comprehensive features, the water quality abnormality early warning information is generated according to the abnormal type and the abnormal number, and the information is sent to the central control platform through the secure transmission protocol.

5. The method of claim 1, wherein, The water quality abnormality early warning information is substituted into an optimization calculation function with the outflow water quality compliance rate, system energy consumption and reagent cost as optimization objectives to obtain an optimization objective function value, operation parameters of the textile printing and dyeing wastewater treatment equipment are converted into optimization calculation variables, and the operation parameter values are calculated by combining a particle swarm algorithm, including: The water quality abnormality early warning information is substituted into an optimization calculation function with the outflow water quality compliance rate, system energy consumption and reagent cost as optimization objectives to obtain an optimization objective function value; According to the optimization objective function value, the operation parameters of the textile printing and dyeing wastewater treatment equipment are converted into optimization calculation variables and combined into an optimization calculation variable vector, the particle swarm size and the maximum iteration number are initialized based on the optimization calculation variable vector, the initial temperature and the temperature reduction coefficient are set according to the optimization objective function value, the crossover probability and the scaling factor are calculated and set, the initial population distribution is generated, and the operation parameters include the aeration amount adjustment parameter, the reagent dosage parameter, the hydraulic load parameter and the sludge return ratio parameter; According to the initial population distribution, the velocity update equation and the position update equation are determined, the inertia weight in the velocity update equation and the position update equation is dynamically adjusted based on the ratio of the current iteration number to the maximum iteration number to obtain updated velocity values and position values, and the learning factor in the velocity update equation is adaptively updated based on the optimization objective function value to obtain an optimized population distribution; The global optimal solution is obtained by performing the particle swarm algorithm using the optimized population distribution, the global optimal solution is substituted into a simulated annealing operation, the local optimal solution is obtained in combination with the initial temperature and the temperature reduction coefficient, the candidate solution set is obtained by updating the population distribution corresponding to the local optimal solution through differential evolution operation based on the crossover probability and the scaling factor, the optimal solution that meets the pre-set constraint condition is selected from the candidate solution set, and the operation parameter values are calculated.

6. The method of claim 1, wherein, According to the optimization objective function value, the adjustment priority level and the adjustment amplitude corresponding to the operation parameters are determined, the operation parameter adjustment instruction is generated and executed, including: The parameter correlation degree is calculated by collecting the operation time sequence data corresponding to the operation parameters, the ratio of the dispersion product of the operation time sequence data to the standard deviation product is calculated, the parameter correlation degree is multiplied by the normalized processing result of the optimization objective function value deviation corresponding to the operation parameters to obtain the causal relationship weight between the operation parameters; The direct influence matrix is constructed by multiplying the causal relationship weight between the operation parameters and the operation time sequence data response increment ratio, the indirect influence propagation matrix is obtained by power series summation of the direct influence matrix, and the comprehensive influence degree matrix is calculated by multiplying the indirect influence propagation matrix and the pre-set operation parameter importance weight; The operation parameter adjustment priority level is calculated by multiplying the comprehensive influence degree matrix and the normalized result of the optimization objective function value deviation, and the synergistic adjustment coefficient is obtained by weighted average operation of the operation parameter adjustment priority level and the causal relationship weight between the operation parameters. The adjustment range is calculated by multiplying the running parameter adjustment priority level, the coordination adjustment coefficient, and the optimization target function value deviation, and the running parameter value is added to the adjustment range to generate a running parameter adjustment instruction, and the running parameter adjustment instruction is executed in order of the running parameter adjustment priority level.

7. The method of claim 1, wherein, The real-time monitoring of the running state of the textile printing and dyeing wastewater treatment equipment and the updating of the running parameter adjustment instruction include: The running state data of the textile printing and dyeing wastewater treatment equipment is collected in real time, and the running state data is compared with the expected response value after the execution of the running parameter adjustment instruction to calculate a running state deviation; The adjustment response degree of the running parameter is determined according to the running state deviation, and when the running state deviation exceeds a preset deviation threshold, an update coefficient of the running parameter is calculated based on the running state deviation; The update coefficient is multiplied by the original running parameter adjustment instruction to obtain an updated running parameter adjustment instruction, and the running parameter of the textile printing and dyeing wastewater treatment equipment is adjusted according to the updated running parameter adjustment instruction.

8. The intelligent textile printing and dyeing wastewater treatment system based on edge computing, for implementing the method of any one of the preceding claims 1-7, characterized in that, It includes: A first unit for deploying edge computing nodes in a plurality of textile printing and dyeing wastewater treatment equipment and collecting current water quality parameters; A second unit for obtaining historical water quality parameters and historical treatment effect data to establish a training data set, training a pre-set neural network by combining a back propagation algorithm to obtain initial connection weights, calculating the fluctuation amplitude of the current water quality parameters, adjusting the number of hidden layer neurons of the neural network according to the fluctuation amplitude and updating the initial connection weights to obtain optimized connection weights, calculating the corresponding relationship between the current water quality parameters and the equipment running efficiency based on the optimized connection weights, judging the abnormal type and the abnormal quantity according to the corresponding relationship to generate water quality abnormal warning information, and sending the water quality abnormal warning information to a central control platform; A third unit for the central control platform to receive the water quality abnormal warning information sent by each edge computing node; A fourth unit for substituting the water quality abnormal warning information into an optimization calculation function with the outflow water quality compliance rate, system energy consumption, and reagent cost as optimization objectives to obtain an optimization target function value, converting the running parameters of the textile printing and dyeing wastewater treatment equipment into optimization calculation variables, and calculating the running parameter values by combining a particle swarm algorithm, determining the adjustment priority level and the adjustment range corresponding to the running parameters according to the optimization target function value, generating a running parameter adjustment instruction, and executing the running parameter adjustment instruction; A fifth unit for real-time monitoring of the running state of the textile printing and dyeing wastewater treatment equipment and updating the running parameter adjustment instruction.

9. An electronic device, comprising: It includes: A processor; A memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions are executed by the processor to implement the method of any one of claims 1 to 7.

Images