Membrane pollution robust identification method and system based on punishment mechanism

Abstract

The invention relates to computer system construction based on a specific calculation model, in particular to a membrane pollution robust recognition method and system based on a penalty mechanism, and the method comprises the steps: obtaining sample data, and dividing the sample data into a training sample and a test sample; establishing an input layer, a membership function layer, a normalization layer and an output layer through a fuzzy neural network to obtain a membrane pollution identification model; parameters of the membrane pollution identification model are updated through a gradient descent algorithm in combination with the parameter penalty term, and an updated membrane pollution identification model is obtained; and inputting the sample data into the updated membrane pollution identification model for training, and inputting the water production flow, the water production pressure, the single-pool membrane scrubbing gas flow and the water production turbidity of the to-be-detected water sample into the trained membrane pollution identification model to obtain a predicted value of the membrane permeability. Good prediction precision and performance can still be achieved under external disturbance, and safe and stable operation of the sewage treatment process is guaranteed.

Description

Technical Field

[0001] The present application relates to the technical field of computer system construction based on a specific computing model, and in particular to a method and system for robustly identifying membrane fouling based on a penalty mechanism. Background Art

[0002] The recycling of urban sewage can effectively alleviate the problems of water shortage and water waste. Improving the level of urban sewage treatment and realizing the sustainable use of water resources are important means to alleviate water shortage and build a water-saving society.

[0003] MBR (Membrane Bio-Reactor) is a new type of sewage treatment process that combines membrane separation technology with biological treatment technology. It has significant advantages such as good solid-liquid separation, low sludge load and small footprint. It is an important means to achieve the recycling of urban sewage. However, in the MBR sewage treatment process, as the amount of membrane treated sewage accumulates, membrane pollution will occur, resulting in a decrease in effluent quality, an increase in operating costs, and even the collapse of the sewage treatment process.

[0004] In the prior art, membrane pollution is usually monitored by monitoring methods such as physical inspection, pressure monitoring and water quality analysis. Among them, the physical inspection method determines membrane pollution by directly observing the appearance of the membrane elements, the pressure monitoring method determines membrane pollution by the pressure at the inlet and outlet of the membrane system, and the water quality analysis method determines membrane pollution by monitoring the water quality of the produced water. However, in actual applications, the above technical solutions are easily affected by external disturbances in the accurate monitoring and identification of membrane pollution, and are unable to predict the trend of membrane pollution.

[0005] Therefore, it is necessary to improve one or more problems existing in the above-mentioned related technical solutions.

[0006] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0007] The purpose of the embodiments of the present disclosure is to provide a method and system for robustly identifying membrane fouling based on a penalty mechanism, thereby overcoming one or more problems caused by limitations and defects of related technologies at least to a certain extent.

[0008] In a first aspect, the present application provides a membrane fouling robust identification method based on a penalty mechanism, comprising: Acquire sample data, and divide the sample data into training samples and test samples, wherein the sample data includes produced water flow rate, produced water pressure, single pool membrane scrubbing gas volume, produced water turbidity and membrane permeability; The input layer, membership function layer, normalization layer and output layer are established through fuzzy neural network to obtain the membrane pollution identification model, in which the membership function layer performs fuzzy processing on the clear data transmitted from the input layer, and the normalization layer uses the preset fuzzy logic rules to process the fuzzy information output by the membership function layer; Updating the parameters of the membrane fouling identification model by combining a gradient descent algorithm with a parameter penalty term to obtain an updated membrane fouling identification model; The sample data is input into the updated membrane pollution identification model for training, and the water production flow rate, water production pressure, single pool membrane scrubbing gas volume and water production turbidity of the water sample to be tested are input into the trained membrane pollution identification model to obtain the predicted value of membrane permeability.

[0009] In a possible implementation, the step of obtaining sample data and dividing the sample data into training samples and test samples, wherein the sample data includes produced water flow rate, produced water pressure, single pool membrane scrubbing gas volume, produced water turbidity and membrane permeability, comprises: Acquire the sample data from the membrane system, wherein the produced water flow rate, the produced water pressure, the single pool membrane scrubbing gas volume and the produced water turbidity are input variables, the membrane permeability is an output variable, and the input variables are normalized to [0,1]; The sample data is divided into the training samples and the test samples, wherein the number of the training samples is greater than the number of the test samples.

[0010] In a possible implementation, the step of establishing a membrane pollution identification model, establishing an input layer, a membership function layer, a normalization layer and an output layer through a fuzzy neural network to obtain a membrane pollution identification model, wherein the membership function layer performs fuzzy processing on clear data transmitted from the input layer, and the normalization layer uses preset fuzzy logic rules to process fuzzy information output by the membership function layer, includes: Establishing the input layer, the membership function layer, the normalization layer and the output layer; The input layer, the membership function layer, the normalization layer and the output layer are connected through an initial connection method with a neuron number ratio of 4:10:10:1, wherein the initial value of the connection weight between the membership function layer and the normalization layer is 1, and the initial value of the connection weight between the normalization layer and the output layer is 1.

[0011] In a possible implementation, the input layer is composed of 4 neurons, and the output of each neuron in the input layer is: ; in, x i ( t ) is the tTime input layer i The output of a neuron; u i ( t ) is the t Time input layer i The input of neurons, u i ( t )include u 1( t ), u 2( t ), u 3( t )and u 4( t ),in, u 1( t )express t Water flow rate at any time, u 2( t )express t Water pressure at all times, u 3( t )express t Single pool membrane scrubbing gas volume at any time, u 4( t )express t Water turbidity at all times.

[0012] In a possible implementation, the membership function layer is composed of 10 neurons, and the output of each membership function layer neuron is: ; in, j ( t )for t The moment membership function layer j The output value of a neuron, e is a natural constant, m ij ( t )for t Moment j The membership function layer neurons i The center of the membership function, σ ij (t) is t Moment j The membership function layer neurons i The width of the membership function.

[0013] In a possible implementation, the normalization layer is composed of 10 neurons, and the output of each normalization layer neuron is: ; in, v j (t )for t Time normalization layer j The output value of a neuron, w vj ( t )for t Moment j The weighted connection between the membership function layer and the normalization layer.

[0014] In a possible implementation, the output layer is composed of 1 neuron, and the output of each output layer neuron is: ; in, w ( t )for t The weighted connection between the neurons in the time normalization layer and the output layer, for t The output of the membrane contamination identification model at this moment, w j ( t )for t Moment j The weighted connections between the normalization layer neurons and the output layer.

[0015] In a possible implementation, the step of updating the parameters of the membrane fouling identification model by using a gradient descent algorithm combined with a parameter penalty term to obtain an updated membrane fouling identification model includes: The loss function is defined as ;in, E ( t ) is the loss function, , e ( t ) indicates that the model is t The output error at time g ( t )express t The expected output at the moment; The sensitivity of the parameter to the disturbance is defined as ; in, ; , represents the sensitivity of the center of the j-th membership function layer neuron to disturbance at time t; , represents the sensitivity of the width of the j-th membership function layer neuron to disturbance at time t; , Indicates the sensitivity of the weight parameters between the membership function layer and the normalization layer to disturbances at time t; , Indicates the sensitivity of the weight parameters between the normalization layer and the output layer to disturbances at time t; A parameter penalty term is designed and added to the parameter update strategy. According to the sensitivity of the disturbance, the parameter penalty term is: Among them, Θ( t ) is the model parameter set, ( t )=[ m ij ( t ), σ ij ( t ), w vj ( t ), w j ( t )]; η The learning rate for updating model parameters; λ is the penalty parameter; ; I is the identity matrix; The model parameters are updated by combining the parameter penalty term with the gradient descent algorithm, and the gradient descent algorithm is: ; ; ; .

[0016] In a possible implementation, the step of inputting the sample data into the updated membrane pollution identification model for training, and obtaining the predicted value of membrane water permeability by inputting the produced water flow, produced water pressure, single pool membrane scrubbing gas volume and produced water turbidity of the water sample to be tested into the trained membrane pollution identification model includes: The membrane pollution identification model is trained by the training samples, and the membrane pollution identification model is tested by the test samples to obtain a trained membrane pollution identification model; The produced water pressure, produced water flow rate, produced water turbidity and single-cell membrane scrubbing gas volume corresponding to the membrane permeability to be predicted are input into the trained membrane pollution identification model to obtain the predicted value of membrane permeability.

[0017] In a second aspect, the present application provides a membrane fouling robust identification system based on a penalty mechanism, the system is used to execute the above-mentioned membrane fouling robust identification method based on a penalty mechanism, the system comprising: A sample data acquisition module is used to acquire sample data and divide the sample data into training samples and test samples, wherein the sample data includes produced water flow rate, produced water pressure, single pool membrane scrubbing gas volume, produced water turbidity and membrane permeability; The recognition model establishment module is used to establish the input layer, membership function layer, normalization layer and output layer through the fuzzy neural network to obtain the membrane pollution recognition model, wherein the membership function layer performs fuzzy processing on the clear data transmitted from the input layer, and the normalization layer uses the preset fuzzy logic rules to process the fuzzy information output by the membership function layer; A model parameter updating module, used to update the parameters of the membrane fouling identification model by combining a gradient descent algorithm with a parameter penalty term to obtain an updated membrane fouling identification model; The membrane permeability prediction module is used to input the sample data into the updated membrane pollution identification model for training, and obtain the membrane permeability prediction value by inputting the water production flow, water production pressure, single pool membrane scrubbing gas volume and water production turbidity of the water sample to be tested into the trained membrane pollution identification model.

[0018] The technical solution provided by this application may have the following beneficial effects: Through the membrane fouling robust identification method and system based on the penalty mechanism of the present application, it is possible to update the parameters of the membrane fouling identification model through the characteristics of the fuzzy neural network according to the dynamic characteristics of the fuzzy neural network through the gradient descent algorithm combined with the parameter penalty term, reduce the sensitivity to external disturbances, improve the robustness of the fuzzy neural network, and reduce the impact of external disturbances on the prediction performance, so that the method still has good prediction accuracy and performance under external disturbances, thereby ensuring the safe and stable operation of the sewage treatment process.

[0019] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0021] Figure 1 A flowchart showing a method for robustly identifying membrane fouling based on a penalty mechanism in an exemplary embodiment of the present disclosure is shown; Figure 2 A detailed flow chart showing a method for robustly identifying membrane fouling based on a penalty mechanism in an exemplary embodiment of the present disclosure is shown; Figure 3 A schematic diagram showing membrane water permeability prediction results of a membrane fouling robust identification method based on a penalty mechanism in an exemplary embodiment of the present disclosure; Figure 4 A schematic diagram showing membrane permeability prediction error of a membrane fouling robust identification method based on a penalty mechanism in an exemplary embodiment of the present disclosure is shown; Figure 5 A structural diagram of a membrane fouling robust identification system based on a penalty mechanism in an exemplary embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0022] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0023] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0024] During the MBR sewage treatment process, mechanical or physical and chemical reactions occur between the sludge flocs, particles, colloidal particles and other substances in the mixed liquid, and between these substances and the membrane, and they adhere to and deposit on the membrane surface and in the membrane pores, causing membrane fouling. Membrane fouling will lead to a decrease in effluent quality and an increase in operating energy consumption, seriously affecting the stable operation of the MBR, and even causing the collapse of the sewage treatment process, causing great harm to production. Therefore, it is crucial to accurately identify membrane fouling online.

[0025] In this example implementation, a robust identification method for membrane fouling based on a penalty mechanism is first provided, a robust identification model for membrane fouling based on a penalty mechanism is established, a parameter learning algorithm with a disturbance penalty mechanism is proposed, model parameters are corrected in the presence of external disturbances, the robustness of the membrane fouling identification model is improved, and online accurate identification of membrane fouling is achieved. This method can be applied to a terminal device, such as a mobile phone, desktop computer, personal digital assistant, laptop, tablet computer, smart watch, or other mobile terminal. Reference Figure 1 As shown in , the method may include the following steps: Step S100: acquiring sample data, and dividing the sample data into training samples and test samples, the sample data including produced water flow, produced water pressure, single pool membrane scrubbing gas volume, produced water turbidity and membrane permeability.

[0026] Step S200: An input layer, a membership function layer, a normalization layer and an output layer are established through a fuzzy neural network to obtain a membrane pollution identification model, wherein the membership function layer performs fuzzy processing on the clear data transmitted from the input layer, and the normalization layer uses preset fuzzy logic rules to process the fuzzy information output by the membership function layer.

[0027] Step S300: updating the parameters of the membrane fouling identification model by combining a gradient descent algorithm with a parameter penalty term to obtain an updated membrane fouling identification model.

[0028] Step S400: input the sample data into the updated membrane pollution identification model for training, and obtain the predicted value of membrane permeability by inputting the water production flow, water production pressure, single pool membrane scrubbing gas volume and water production turbidity of the water sample to be tested into the trained membrane pollution identification model.

[0029] Through the above method, the network parameters can be updated through a parameter learning algorithm with a disturbance penalty mechanism, the sensitivity to external disturbances can be reduced, the robustness of the fuzzy neural network can be improved, and the impact of external disturbances on the prediction performance can be reduced, so that the method still has good prediction accuracy and performance under external disturbances, ensuring the safe and stable operation of the sewage treatment process.

[0030] Next, we will refer to Figure 1 to Figure 2 Each step of the above method in this example implementation is described in more detail.

[0031] In step S100, sample data is obtained and divided into training samples and test samples, wherein the sample data includes produced water flow, produced water pressure, single cell membrane scrubbing gas volume, produced water turbidity and membrane permeability.

[0032] It should be noted that, through characteristic analysis of the sewage treatment process, process variables related to membrane fouling are selected as the input of the identification model: produced water flow, produced water pressure, single pool membrane scrubbing gas volume, produced water turbidity, and the obtained input variables are normalized to [0,1]; membrane permeability is used as the output of the identification model; all sample data are divided into two groups, one group contains N training samples, and the other group contains M test samples, where N>M.

[0033] In one embodiment, step S100 may include the following sub-steps: Step S110: Acquire the sample data from the membrane system, wherein the produced water flow, the produced water pressure, the single pool membrane scrubbing gas volume and the produced water turbidity are input variables, the membrane permeability is the output variable, and the input variables are normalized to [0,1].

[0034] Step S120: Divide the sample data into the training samples and the test samples, wherein the number of the training samples is greater than the number of the test samples.

[0035] It should be noted that the input variables of the above-mentioned membrane system: water production flow rate, water production pressure, single pool membrane scrubbing gas volume, water production turbidity and membrane permeability can be collected by sensors installed in the membrane system, and the collected input variables are transmitted to the processor for normalization processing.

[0036] In step S200, an input layer, a membership function layer, a normalization layer and an output layer are established through a fuzzy neural network to obtain a membrane pollution identification model, wherein the membership function layer fuzzifies the clear data transmitted from the input layer, and the normalization layer uses preset fuzzy logic rules to process the fuzzy information output by the membership function layer.

[0037] In one embodiment, step S200 may include the following sub-steps: Step S210: Establish the input layer, the membership function layer, the normalization layer and the output layer.

[0038] Step S220: Connect the input layer, the membership function layer, the normalization layer and the output layer through an initial connection method with a neuron number ratio of 4:10:10:1, wherein the initial value of the connection weight between the membership function layer and the normalization layer is 1, and the initial value of the connection weight between the normalization layer and the output layer is 1.

[0039] Furthermore, in one embodiment, the input layer is composed of 4 neurons, and the output of each input layer neuron is: ; in, x i ( t ) is the t Time input layer i The output of a neuron; u i ( t ) is the t Time input layer i The input of a neuron, u 1( t )express t Water flow rate at any time, u 2(t )express t Water pressure at all times, u 3( t )express t Single pool membrane scrubbing gas volume at any time, u 4( t )express t Water turbidity at all times.

[0040] Understandably, the moment t= (1, ..., N), the input layer neurons are 4, so i = (1, 2, 3, 4), through the input layer, directly receive the above variables, such as water production flow, water production pressure, single pool membrane scrubbing gas volume, water production turbidity, etc. These variables are normalized to [0, 1], which can unify data of different magnitudes into the same scale range, facilitating effective calculation and processing in subsequent layers. This enables the model to obtain the most direct and standardized information from the sewage treatment process, laying the foundation for accurate identification of membrane fouling.

[0041] Furthermore, in one embodiment, the membership function layer is composed of 10 neurons, and the output of each membership function layer neuron is: ; in, j ( t )for t The moment membership function layer j The output value of a neuron, e is a natural constant, m ij ( t )for t Moment j The membership function layer neurons i The center of the membership function, σ ij (t) is t Moment j The membership function layer neurons i The width of the membership function.

[0042] It is understandable that there are 10 neurons in the membership function layer, so j =(1,…,10), m ij ( t ), σ ij (t) are randomly selected in the interval (0,1], using multiple neurons and randomly initialized membership function centers m ij ( t) and width σ ij (t), fuzzify the clear data from the input layer. Converting precise input values ​​into fuzzy sets, this fuzzification can better handle the uncertainty and ambiguity of data in the sewage treatment process. For example, in actual sewage treatment, the relationship between membrane pollution and each input variable is not a completely accurate linear or deterministic relationship. Fuzzification can simulate the fuzzy thinking mode of human beings, handle this complex relationship more flexibly, and enhance the adaptability of the model to complex systems. Multiple membership functions (each neuron corresponds to multiple membership functions) can characterize the input data from different angles, improve the model's ability to extract data features, and thus more comprehensively capture information related to membrane pollution, providing a richer information basis for subsequent rule-layer decision-making.

[0043] Furthermore, in one embodiment, the normalization layer is composed of 10 neurons, and the output of each normalization layer neuron is: ; in, v j (t )for t Time normalization layer j The output value of a neuron, w vj ( t )for t Moment j The weighted connection between the membership function layer and the normalization layer.

[0044] It is understandable that the normalization layer can process the fuzzy information output by the membership function layer using fuzzy logic rules. Through the fuzzy rules that are pre-set or learned, the fuzzified input data is comprehensively judged, expert knowledge and experience are simulated, and the membrane contamination state is inferred and decided. For example, according to the degree of fuzziness output by different membership functions and the combination relationship between them, the possibility of the current membrane contamination degree is judged, and an effective logical connection is established between the complex input data and the membrane contamination degree, thereby realizing the identification and evaluation of membrane contamination. This processing method based on fuzzy rules has a certain degree of fault tolerance. Even if there are certain errors or inaccuracies in the input data, relatively reasonable conclusions can be drawn according to the fuzzy rules, which helps to improve the reliability and robustness of the model in actual complex environments.

[0045] Furthermore, in one embodiment, the output layer is composed of 1 neuron, and the output of each output layer neuron is: ; in, w ( t )fort The weighted connection between the neurons in the time normalization layer and the output layer, for t The output of the membrane contamination identification model at this moment, w j ( t )for t Moment j The weighted connections between the normalization layer neurons and the output layer.

[0046] It is understandable that the output layer finally outputs the membrane permeability, an important indicator for evaluating the degree of membrane pollution, which directly provides a key decision-making basis for the sewage treatment process. The information processed by the previous layers is converted into a clear output value closely related to membrane pollution, so that operators can intuitively understand the situation of membrane pollution, so as to take corresponding measures in time, such as adjusting sewage treatment process parameters, membrane cleaning, etc., to ensure the safety and stable operation of the sewage treatment process. The singleness of the model output (1 neuron) makes the goal clear, focusing on the prediction of membrane permeability, avoiding the output of too much irrelevant or redundant information, and improving the practicality and pertinence of the model.

[0047] In general, this four-layer model structure, through the collaborative work between each layer and combined with the characteristics of fuzzy neural networks, can effectively deal with the uncertainty in the membrane fouling identification process, improve the robustness and prediction accuracy of the model, adapt to the complex and changeable working conditions in the sewage treatment process, and realize accurate online identification and monitoring of membrane fouling.

[0048] In step S300, the parameters of the membrane fouling identification model are updated by combining a gradient descent algorithm with a parameter penalty term to obtain an updated membrane fouling identification model.

[0049] In one embodiment, step S300 may include the following sub-steps: Step S310: Define the loss function as ;in, E ( t ) is the loss function, , e ( t ) indicates that the model is t The output error at time g ( t )express t Expected output at time.

[0050] Step S320: Define the sensitivity of the parameter to the disturbance as ; in, ; , represents the sensitivity of the center of the j-th membership function layer neuron to disturbance at time t; , represents the sensitivity of the width of the j-th membership function layer neuron to disturbance at time t; , Indicates the sensitivity of the weight parameters between the membership function layer and the normalization layer to disturbances at time t; , It represents the sensitivity of the weight parameters between the normalization layer and the output layer to the disturbance at time t.

[0051] Step S330: Design a parameter penalty term and add it to the parameter update strategy. According to the sensitivity of the disturbance, the parameter penalty term is: Among them, Θ(t)=[mij(t),σij(t),wvj(t),wj(t)], Θ(t) is the model parameter set; η is the learning rate of model parameter update; λ is the penalty term parameter; ; I is the identity matrix.

[0052] It should be noted that the parameter penalty term formula is derived from the original parameter penalty term formula: ; in, ; ; Θ( t ) is the neural network parameter set, Θ( t )=[ m ij ( t ), σ ij ( t ), w vj ( t ), w j ( t )]; is a step function, when |S( t )|<1 and |S( t )|=0,Ξ(|S( t )|-1)=0, when |S( t )|≥1,Ξ(|S( t )|-1)=1, η is the learning rate for updating network parameters, which is randomly selected between (0.1, 0.3]; λ is the penalty parameter, which takes a random value between (0,0.1]; After derivation, the original parameter penalty term formula is written as: Finally, according to S( t ) is used to obtain the parameter penalty term formula.

[0053] It can be understood that the parameter penalty term shows that different disturbance sensitivity values ​​and different parameter update strategies. When the disturbance sensitivity is ≥ 1, it indicates that the parameter is highly sensitive to the disturbance, and the penalty term can be used to reduce the parameter.

[0054] Step S340: Update the model parameters by combining the parameter penalty term with the gradient descent algorithm, wherein the gradient descent algorithm is: ; ; ; .

[0055] It should be noted that by adding a penalty term on the basis of the gradient descent algorithm, the updated gradient obtained is the rate of change of the parameter, that is, the partial derivative of the loss function with respect to the parameter, and the parameter is updated according to the rate of change.

[0056] In step S400, the sample data is input into the updated membrane pollution identification model for training, and the water production flow rate, water production pressure, single pool membrane scrubbing gas volume and water production turbidity of the water sample to be tested are input into the trained membrane pollution identification model to obtain the predicted value of membrane permeability.

[0057] It should be noted that the trained fuzzy neural network model is used to predict membrane permeability, and the collected water production pressure, water production flow rate, water production turbidity and single-cell membrane scrubbing gas volume are used as input variables of the model to obtain the membrane permeability prediction output value of the model.

[0058] In one embodiment, step S400 may include the following sub-steps: Step S410: training the membrane fouling identification model using the training samples, and testing the membrane fouling identification model using the test samples to obtain a trained membrane fouling identification model.

[0059] Step S420: input the produced water pressure, produced water flow rate, produced water turbidity and single-cell membrane scrubbing gas volume corresponding to the membrane permeability to be predicted into the trained membrane pollution identification model to obtain the predicted value of membrane permeability.

[0060] It should be noted that the use of the membrane fouling robust identification method based on the penalty mechanism of the present application can reduce the sensitivity of the fuzzy neural network parameters to disturbances, improve the robustness of the model, and ultimately improve the model's prediction accuracy for membrane permeability, accurately monitor the status of the membrane bioreactor in real time, and ensure the smooth operation of the urban sewage treatment process.

[0061] like Figure 3 The following table shows the predicted results of membrane permeability. The X-axis is the test sample, the unit is pieces, and the Y-axis is the membrane permeability, the unit is L / (m 2 *h), the solid line is the actual value of membrane water permeability, and the dotted line is the predicted value of membrane water permeability. It can be seen from the figure that the actual value of membrane water permeability is basically consistent with the predicted value of membrane water permeability.

[0062] like Figure 4 As shown, it is the prediction error of membrane permeability; the X-axis is the test sample, the unit is pieces, and the Y-axis is the prediction error of membrane permeability, the unit is L / (m 2 *h), it can be seen from the figure that the prediction error of membrane permeability does not exceed ±10L / (m 2 *h).

[0063] Furthermore, in this exemplary embodiment, a membrane fouling robust identification system based on a penalty mechanism is also provided. Figure 5 As shown in , the system may include a sample data acquisition module, a recognition model establishment module, a model parameter updating module and a membrane permeability prediction module.

[0064] The sample data acquisition module is used to acquire sample data and divide the sample data into training samples and test samples. The sample data includes produced water flow, produced water pressure, single pool membrane scrubbing gas volume, produced water turbidity and membrane permeability.

[0065] The identification model establishment module is used to establish the input layer, membership function layer, normalization layer and output layer through the fuzzy neural network to obtain the membrane pollution identification model, wherein the membership function layer fuzzifies the clear data transmitted from the input layer, and the normalization layer uses preset fuzzy logic rules to process the fuzzy information output by the membership function layer.

[0066] The model parameter updating module is used to update the parameters of the membrane pollution identification model through a gradient descent algorithm combined with a parameter penalty term to obtain an updated membrane pollution identification model.

[0067] The membrane permeability prediction module is used to input the sample data into the updated membrane pollution identification model for training, and obtain the membrane permeability prediction value by inputting the water production flow, water production pressure, single pool membrane scrubbing gas volume and water production turbidity of the water sample to be tested into the trained membrane pollution identification model.

[0068] Regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0069] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of a module or unit described above can be further divided into multiple modules or units for concretization. The components displayed as modules or units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. Those of ordinary skill in the art can understand and implement it without paying creative work.

[0070] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A robust identification method for membrane fouling based on a penalty mechanism, characterized in that: include: Acquire sample data, and divide the sample data into training samples and test samples, wherein the sample data includes produced water flow rate, produced water pressure, single pool membrane scrubbing gas volume, produced water turbidity and membrane permeability; The input layer, membership function layer, normalization layer and output layer are established through fuzzy neural network to obtain the membrane pollution identification model, in which the membership function layer performs fuzzy processing on the clear data transmitted from the input layer, and the normalization layer uses the preset fuzzy logic rules to process the fuzzy information output by the membership function layer; Updating the parameters of the membrane fouling identification model by combining a gradient descent algorithm with a parameter penalty term to obtain an updated membrane fouling identification model; The sample data is input into the updated membrane pollution identification model for training, and the water production flow rate, water production pressure, single pool membrane scrubbing gas volume and water production turbidity of the water sample to be tested are input into the trained membrane pollution identification model to obtain the predicted value of membrane permeability.

2. The robust identification method of membrane fouling based on penalty mechanism according to claim 1 is characterized in that: The step of obtaining sample data and dividing the sample data into training samples and test samples, wherein the sample data includes produced water flow rate, produced water pressure, single pool membrane scrubbing gas volume, produced water turbidity and membrane permeability, comprises: Acquire the sample data from the membrane system, wherein the produced water flow rate, the produced water pressure, the single pool membrane scrubbing gas volume and the produced water turbidity are input variables, the membrane permeability is an output variable, and the input variables are normalized to [0,1]; The sample data is divided into the training samples and the test samples, wherein the number of the training samples is greater than the number of the test samples.

3. The robust identification method of membrane fouling based on penalty mechanism according to claim 1 is characterized in that: The step of establishing an input layer, a membership function layer, a normalization layer and an output layer through a fuzzy neural network to obtain a membrane pollution identification model, wherein the membership function layer performs fuzzy processing on clear data transmitted from the input layer, and the normalization layer uses preset fuzzy logic rules to process the fuzzy information output by the membership function layer, includes: Establishing the input layer, the membership function layer, the normalization layer and the output layer; The input layer, the membership function layer, the normalization layer and the output layer are connected through an initial connection method with a neuron number ratio of 4:10:10:1, wherein the initial value of the connection weight between the membership function layer and the normalization layer is 1, and the initial value of the connection weight between the normalization layer and the output layer is 1.

4. The robust identification method of membrane fouling based on penalty mechanism according to claim 3 is characterized in that: The input layer consists of 4 neurons, and the output of each neuron in the input layer is: ； in, x i ( t ) is the t Time input layer i The output of a neuron; u i ( t ) is the t Time input layer i The input of neurons, u i ( t )include u 1( t ), u 2( t ), u 3( t )and u 4( t ),in, u 1( t )express t Water flow rate at any time, u 2( t )express t Water pressure at all times, u 3( t )express t Single pool membrane scrubbing gas volume at any time, u 4( t )express t Water turbidity at all times.

5. The robust identification method of membrane fouling based on penalty mechanism according to claim 4 is characterized in that: The membership function layer consists of 10 neurons, and the output of each membership function layer neuron is: ； in, j ( t )for t The moment membership function layer j The output value of a neuron, e is a natural constant, m ij ( t )for t Moment j The membership function layer neurons i The center of the membership function, σ ij (t) is t Moment j The membership function layer neurons i The width of the membership function.

6. The robust identification method of membrane fouling based on penalty mechanism according to claim 5 is characterized in that: The normalization layer consists of 10 neurons, and the output of each normalization layer neuron is: ； in, v j (t )for t Time normalization layer j The output value of a neuron, w vj ( t )for t Moment j The weighted connection between the membership function layer and the normalization layer.

7. The robust identification method of membrane fouling based on penalty mechanism according to claim 6 is characterized in that: The output layer consists of 1 neuron, and the output of each output layer neuron is: ； in, w ( t )for t The weighted connection between the neurons in the time normalization layer and the output layer, for t The output of the membrane contamination identification model at this moment, w j ( t )for t Moment j The weighted connections between the normalization layer neurons and the output layer.

8. The robust identification method of membrane fouling based on penalty mechanism according to claim 7 is characterized in that: The step of updating the parameters of the membrane fouling identification model by combining a gradient descent algorithm with a parameter penalty term to obtain an updated membrane fouling identification model comprises: The loss function is defined as ;in, E ( t ) is the loss function, , e(t) Indicates that the model t The output error at time g ( t )express t The expected output at the moment; The sensitivity of the parameter to the disturbance is defined as ; in, ; , represents the sensitivity of the center of the j-th membership function layer neuron to disturbance at time t; , represents the sensitivity of the width of the j-th membership function layer neuron to disturbance at time t; , Indicates the sensitivity of the weight parameters between the membership function layer and the normalization layer to disturbances at time t; , Indicates the sensitivity of the weight parameters between the normalization layer and the output layer to disturbances at time t; A parameter penalty term is designed and added to the parameter update strategy. According to the sensitivity of the disturbance, the parameter penalty term is: Among them, Θ( t )=[ m ij ( t ), σ ij ( t ), w vj ( t ), w j ( t )],Θ( t ) is a set of model parameters; η The learning rate for updating model parameters; λ is the penalty parameter; ; I is the identity matrix; The model parameters are updated by combining the parameter penalty term with the gradient descent algorithm, and the gradient descent algorithm is: ； ； ； 。 9. The robust identification method of membrane fouling based on penalty mechanism according to claim 1 is characterized in that: The step of inputting the sample data into the updated membrane pollution identification model for training, and obtaining the predicted value of membrane water permeability by inputting the produced water flow, produced water pressure, single pool membrane scrubbing gas volume and produced water turbidity of the water sample to be tested into the trained membrane pollution identification model includes: The membrane pollution identification model is trained by the training samples, and the membrane pollution identification model is tested by the test samples to obtain a trained membrane pollution identification model; The produced water pressure, produced water flow rate, produced water turbidity and single-cell membrane scrubbing gas volume corresponding to the membrane permeability to be predicted are input into the trained membrane pollution identification model to obtain the predicted value of membrane permeability.

10. A robust identification system for membrane fouling based on a penalty mechanism, characterized in that: The system is used to execute the membrane fouling robust identification method based on the penalty mechanism according to any one of claims 1 to 9, and the system comprises: A sample data acquisition module is used to acquire sample data and divide the sample data into training samples and test samples, wherein the sample data includes produced water flow rate, produced water pressure, single pool membrane scrubbing gas volume, produced water turbidity and membrane permeability; The recognition model establishment module is used to establish the input layer, membership function layer, normalization layer and output layer through the fuzzy neural network to obtain the membrane pollution recognition model, wherein the membership function layer performs fuzzy processing on the clear data transmitted from the input layer, and the normalization layer uses the preset fuzzy logic rules to process the fuzzy information output by the membership function layer; A model parameter updating module, used to update the parameters of the membrane fouling identification model by combining a gradient descent algorithm with a parameter penalty term to obtain an updated membrane fouling identification model; The membrane permeability prediction module is used to input the sample data into the updated membrane pollution identification model for training, and obtain the membrane permeability prediction value by inputting the water production flow, water production pressure, single pool membrane scrubbing gas volume and water production turbidity of the water sample to be tested into the trained membrane pollution identification model.

Images