A dynamic modeling and simulation method and system for the energy efficiency ratio of a solar-powered seawater desalination system

Abstract

This application provides a dynamic modeling and simulation method and system for the energy efficiency ratio (EER) of a solar-powered seawater desalination system, belonging to the technical field of seawater desalination and system modeling. First, this application acquires data on membrane surface resistance, selective permeability, and DC bus voltage ripple during the photovoltaic electrodialysis process. Second, the ripple data is Fourier transformed and concatenated with membrane parameters to generate an input matrix. This matrix is ​​then imported into a deep belief network, and a restricted Boltzmann machine is used to extract the unsteady-state ion impedance vector reflecting the influence of voltage fluctuations. Subsequently, a nonlinear regression model of this vector and unit water production energy consumption is established using a least-squares support vector machine. Finally, the water production rate and EER are calculated based on the predicted energy consumption and photovoltaic power, generating a dynamic simulation curve. This application can quantify the nonlinear influence of photovoltaic voltage ripple on membrane impedance through deep learning, significantly improving the accuracy of EER prediction for seawater desalination systems under fluctuating power supply conditions.

Description

Technical Field

[0001] This application belongs to the field of seawater desalination and system modeling technology, and in particular relates to a dynamic modeling and simulation method and system for the energy efficiency ratio of a solar-powered seawater desalination system. Background Technology

[0002] Solar photovoltaic-driven seawater desalination technology is an important means of solving the shortage of freshwater resources. Dynamic modeling and simulation of its energy efficiency ratio can effectively guide engineering design and operation optimization. Accurate simulation models can help identify energy loss points, which has significant application value for improving the utilization rate of photovoltaic energy and reducing water production costs.

[0003] Existing dynamic modeling and simulation methods for energy efficiency ratios are typically based on electrochemical steady-state equations or simplified empirical formulas, treating the output of photovoltaic modules as an ideal DC power source for calculation. These methods often neglect high-frequency interference generated during the power conversion process when constructing the model, estimating the resistive characteristics of the ion exchange membrane and the overall energy consumption solely based on average voltage or current.

[0004] However, the DC bus voltage output by a real photovoltaic inverter includes a significant voltage ripple. This fluctuating electric field causes complex nonlinear changes in the surface resistance of the ion exchange membrane, thereby altering the resistance to ion migration across the membrane. Existing models, lacking a description of the coupling relationship between the voltage ripple spectral characteristics and the dynamic response of the membrane impedance, struggle to accurately reflect the degree of ion transport obstruction under unsteady current conditions. Therefore, existing technologies suffer from the technical problem of low accuracy in energy efficiency ratio simulations due to neglecting the nonlinear effects of voltage ripple on membrane performance. Summary of the Invention

[0005] The purpose of this application is to provide a dynamic modeling and simulation method and system for the energy efficiency ratio of a solar-powered seawater desalination system, so as to solve the problem of low accuracy in the simulation and prediction of energy efficiency ratio in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a dynamic modeling and simulation method for the energy efficiency ratio of a solar-powered seawater desalination system, comprising:

[0007] Acquire membrane surface resistance and selective permeability data of ion exchange membranes under different concentration gradients during photovoltaic-driven electrodialysis, and simultaneously collect DC bus voltage ripple data at the DC output terminal of the photovoltaic inverter.

[0008] Fast Fourier transform is performed on the DC bus voltage ripple data to obtain the spectral energy distribution characteristics. The spectral energy distribution characteristics and the normalized mapped film surface resistance data are then aligned with the selected permeability data in the time dimension and vector-stitched to generate the input matrix.

[0009] The input matrix is ​​imported into the deep belief network model as input layer data. Through the layer-by-layer unsupervised pre-training mechanism of the multilayer restricted Boltzmann machine in the deep belief network model, the influence weight of the voltage ripple frequency component on the nonlinear change of the membrane resistance is learned. The ion impedance vector under the unsteady current, which represents the degree of dynamic obstruction when ions pass through the membrane channel under the action of the fluctuating electric field, is extracted from the input matrix.

[0010] Using the ion impedance vector as a regression operator, a nonlinear regression relationship between the ion impedance vector and the energy consumption per unit of water production is established through a least squares support vector machine to generate an energy consumption prediction model.

[0011] Based on the ion impedance vector at different times, the predicted value of unit water production energy consumption at each time is obtained using the energy consumption prediction model. The water production rate is calculated based on the photovoltaic input power and the predicted value of unit water production energy consumption at each time. The energy efficiency ratio is obtained by calculating the ratio of water production rate to photovoltaic input power, and a dynamic simulation curve of energy efficiency ratio is generated.

[0012] Optionally, the method further includes:

[0013] Extract the time data sequence from the ion impedance vector and calculate the average value of the modulus in the time data sequence;

[0014] When the average value increases monotonically over time and exceeds the preset pollution baseline threshold, the difference ratio between the average value and the pollution baseline threshold is calculated to obtain the correction coefficient.

[0015] Based on the ion impedance vectors at different times, the predicted energy consumption per unit of produced water at each time point is obtained using an energy consumption prediction model, including:

[0016] Based on the ion impedance vector and correction coefficient at different times, the predicted value of unit water production energy consumption at each time is obtained using the energy consumption prediction model.

[0017] Optionally, based on the ion impedance vector and correction coefficient at different times, the predicted energy consumption per unit of produced water at each time point is obtained using an energy consumption prediction model, including:

[0018] The ion impedance vectors at different times are input into the least squares support vector machine in the energy consumption prediction model. The kernel function maps the ion impedance vectors to a high-dimensional feature space and solves the linear equations to obtain the initial unit water production energy consumption value at different times.

[0019] The energy consumption gain value is obtained by calculating the product of the correction coefficient and the preset pollution sensitivity factor.

[0020] The predicted energy consumption per unit of water production is obtained by summing the initial energy consumption and energy gain values ​​per unit of water production at different times.

[0021] Optionally, a fast Fourier transform is performed on the DC bus voltage ripple data to obtain the spectral energy distribution characteristics. These spectral energy distribution characteristics and the normalized mapped film surface resistance data are then time-aligned and vector-joined with the selected permeability data to generate an input matrix, including:

[0022] Voltage signal segments are extracted from DC bus voltage ripple data according to a preset sampling time sequence. The voltage signal segments are then discretized to obtain a frequency response sequence. The spectral energy distribution characteristics are obtained by calculating the sum of squares of the magnitudes at each frequency point in the frequency response sequence.

[0023] The spectral energy distribution characteristics and the normalized mapped film surface resistance data are numerically combined with the selective permeability data in a preset order to obtain a joint vector;

[0024] The input matrix is ​​obtained by stacking the joint vectors of the continuous time series as row elements in an ordered manner.

[0025] Optionally, the input matrix is ​​imported as input layer data into the deep belief network model. Through the layer-by-layer unsupervised pre-training mechanism of the multilayer restricted Boltzmann machine in the deep belief network model, the influence weights of the voltage ripple frequency component on the nonlinear change of the membrane surface resistance are learned. From the input matrix, the ion impedance vector under unsteady current, representing the degree of dynamic hindrance encountered by ions passing through the membrane channel under the action of a fluctuating electric field, is extracted, including:

[0026] The input matrix is ​​mapped to the set of visible nodes of the underlying restricted Boltzmann machine in the deep belief network model. The inter-layer connection weights are updated by iteratively calculating the joint probability distribution between the set of visible nodes and the set of hidden nodes, and the primary feature code is output.

[0027] The primary feature encoding is used as the input data for the next layer of the Restricted Boltzmann Machine. The joint probability distribution calculation and connection weight update are repeated layer by layer until the data is passed to the top layer of the Restricted Boltzmann Machine.

[0028] The correlation information between the voltage ripple frequency component and the nonlinear change of the film surface resistance is encoded into a high-order abstract feature, and the connection weights of each layer of the restricted Boltzmann machine are adjusted according to the high-order abstract feature.

[0029] The output value of the hidden node set is extracted from the top-level restricted Boltzmann machine after the connection weights are adjusted, and used as the ion impedance vector under unsteady current.

[0030] Optionally, using the ion impedance vector as a regression operator, a nonlinear regression relationship between the ion impedance vector and the energy consumption per unit of produced water is established through a least-squares support vector machine to generate an energy consumption prediction model, including:

[0031] The ion impedance vector is mapped to a high-dimensional feature space using a radial basis function kernel to obtain the kernel matrix;

[0032] Based on the kernel matrix and the energy consumption per unit water production, a system of linear equations satisfying the least squares criterion is constructed, and the regression weight coefficients and bias terms are obtained by solving the system of linear equations.

[0033] A nonlinear regression function is established based on the regression weight coefficients and bias terms, and this nonlinear regression function is used as an energy consumption prediction model.

[0034] Optionally, the water production rate is calculated based on the photovoltaic input power and the predicted energy consumption per unit of produced water at each moment. The energy efficiency ratio is obtained by calculating the ratio of the water production rate to the photovoltaic input power, and a dynamic simulation curve of the energy efficiency ratio is generated, including:

[0035] Calculate the ratio of photovoltaic input power to the predicted energy consumption per unit of water production at each time point to obtain the water production rate at each time point;

[0036] Calculate the ratio of water production rate to photovoltaic input power to obtain the energy efficiency ratio at each time point;

[0037] The energy efficiency ratios at multiple consecutive moments are mapped onto a preset two-dimensional coordinate system in chronological order to obtain a dynamic simulation curve of the energy efficiency ratio that changes continuously with time.

[0038] Secondly, this application provides a dynamic modeling and simulation system for the energy efficiency ratio of a solar-powered seawater desalination system, comprising:

[0039] The acquisition module is used to acquire membrane surface resistance data and selective permeability data of ion exchange membrane under different concentration gradients during the electrodialysis process driven by photovoltaics, and at the same time, to collect DC bus voltage ripple data at the DC output terminal of the photovoltaic inverter.

[0040] The generation module is used to perform fast Fourier transform on DC bus voltage ripple data to obtain spectral energy distribution characteristics. The spectral energy distribution characteristics and the normalized mapped film surface resistance data are then aligned with the time dimension and vector-stitched with the selective permeability data to generate the input matrix.

[0041] The extraction module is used to import the input matrix as input layer data into the deep belief network model. Through the layer-by-layer unsupervised pre-training mechanism of the multilayer restricted Boltzmann machine in the deep belief network model, the influence weight of the voltage ripple frequency component on the nonlinear change of the membrane resistance is learned. The ion impedance vector under the unsteady current, which represents the degree of dynamic hindrance encountered by ions when passing through the membrane channel under the action of the fluctuating electric field, is extracted from the input matrix.

[0042] The generation module is also used to establish a nonlinear regression relationship between the ion impedance vector and the unit water production energy consumption by using the ion impedance vector as a regression operator and the least squares support vector machine, thereby generating an energy consumption prediction model.

[0043] The calculation module is used to obtain the predicted energy consumption per unit of water production at each time based on the ion impedance vector at different times using the energy consumption prediction model. It also calculates the water production rate based on the photovoltaic input power and the predicted energy consumption per unit of water production at each time. The energy efficiency ratio is obtained by calculating the ratio of the water production rate to the photovoltaic input power, and a dynamic simulation curve of the energy efficiency ratio is generated.

[0044] Thirdly, this application provides an electronic device, comprising:

[0045] Memory, used to store computer programs;

[0046] A processor is used to execute the computer program to implement the steps of the dynamic modeling and simulation method for the energy efficiency ratio of the solar desalination system as described in the first aspect above.

[0047] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the dynamic modeling and simulation method for the energy efficiency ratio of a solar-powered seawater desalination system as described in the first aspect above.

[0048] The dynamic modeling and simulation method for the energy efficiency ratio of a solar-powered seawater desalination system provided in this application first acquires DC bus voltage ripple data from a photovoltaic inverter and performs a fast Fourier transform. Second, it integrates the obtained spectral energy distribution characteristics with membrane resistance and selective permeability data in a multi-dimensional manner. Subsequently, it utilizes a deep belief network to deeply mine the influence weights of voltage ripple frequency components on the nonlinear changes in membrane resistance, thereby accurately extracting the ion impedance vector under unsteady current that represents the degree of dynamic impediment to ion transmembrane movement under a fluctuating electric field.

[0049] Finally, a nonlinear regression model of the impedance vector and unit water production energy consumption is constructed by combining least squares support vector machine, which makes up for the shortcomings of traditional methods that only rely on steady-state assumptions and ignore the influence of power supply ripple on membrane microstructure. This application realizes accurate dynamic simulation of seawater desalination energy efficiency ratio under unsteady photovoltaic input conditions, effectively solving the technical problem of low simulation prediction accuracy caused by the failure to consider the nonlinear coupling effect of voltage ripple in existing technologies.

[0050] Furthermore, this application identifies the fouling characteristic of monotonically increasing membrane surface resistance over time by monitoring the temporal evolution trend of the ion impedance vector, and uses the calculated correction coefficient to dynamically compensate the energy consumption prediction model. This method breaks through the limitation of traditional models that only focus on short-term electrical characteristics and ignore long-term physicochemical fouling, and can correct the additional energy consumption increase caused by ion exchange membrane fouling or contaminant deposition in real time.

[0051] By incorporating correction coefficients into the prediction calculations, the model maintains high predictive reliability even when membrane performance undergoes slow, time-varying degradation. Therefore, this application further addresses the technical problem in existing technologies where the accuracy of energy efficiency ratio simulation predictions decreases over operating time due to the failure to consider the long-term impact of membrane fouling accumulation on energy consumption. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 A flowchart illustrating a dynamic modeling and simulation method for the energy efficiency ratio of a solar-powered seawater desalination system provided in this application embodiment;

[0054] Figure 2 A flowchart illustrating a method for determining an ion impedance vector provided in an embodiment of this application;

[0055] Figure 3 A flowchart illustrating a method for generating an energy consumption prediction model provided in an embodiment of this application;

[0056] Figure 4 A schematic diagram of the structure of a dynamic modeling and simulation system for the energy efficiency ratio of a solar-powered seawater desalination system provided in this application embodiment;

[0057] Figure 5 This is a schematic diagram of the hardware structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0058] In the field of solar photovoltaic-driven seawater desalination technology, accurate dynamic modeling and simulation of the energy efficiency ratio is crucial for optimizing engineering design and improving operational efficiency. However, most existing modeling methods are based on idealized steady-state assumptions, simply treating the output of photovoltaic modules as a constant DC power source, ignoring the unavoidable DC bus voltage ripple at the inverter output.

[0059] In reality, this high-frequency fluctuating electric field induces complex nonlinear changes in the surface resistance of the ion exchange membrane, thereby altering the resistance to ion migration across the membrane. Because existing models fail to capture the deep coupling between the voltage ripple spectrum characteristics and the dynamic response of the membrane impedance, they cannot accurately reflect the dynamic obstacles to ion transport under unsteady current conditions. This results in a significant discrepancy between the simulated energy efficiency ratio predictions and actual operating data, making it difficult to meet the requirements of refined control.

[0060] To address the aforementioned issues, this application proposes a dynamic modeling and simulation method for the energy efficiency ratio of solar-powered seawater desalination. The core of this method lies in capturing and quantifying the microscopic impact of power supply fluctuations on membrane performance. First, DC bus voltage ripple data is collected, and its spectral energy distribution characteristics are analyzed using Fast Fourier Transform. This characteristic is then aligned and fused with membrane surface resistance and selective permeability data in both time and space. Subsequently, using a multilayer restricted Boltzmann machine mechanism within a deep belief network, the influence weights of the voltage ripple frequency components on the nonlinear changes in membrane surface resistance are learned, thereby extracting a nonsteady-state ion impedance vector that accurately represents the degree of dynamic obstruction encountered by ions passing through membrane channels under the influence of a fluctuating electric field.

[0061] Based on this vector, an energy consumption prediction model is constructed using least-squares support vector machines, achieving a precise mapping from electrical fluctuation characteristics to physicochemical energy consumption characteristics. This scheme abandons the traditional steady-state power supply assumption and effectively solves the simulation distortion problem caused by neglecting the nonlinear coupling effect of voltage ripple in existing technologies through artificial intelligence algorithms, significantly improving the accuracy of energy efficiency ratio prediction under dynamic operating conditions.

[0062] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0063] To address the problems of existing technologies, embodiments of this application provide a method, apparatus, equipment, computer storage medium, and computer program product for dynamic modeling and simulation of the energy efficiency ratio of a solar-powered seawater desalination system. The method for dynamic modeling and simulation of the energy efficiency ratio of a solar-powered seawater desalination system provided in this application embodiment will be described below.

[0064] Figure 1 A flowchart illustrating a dynamic modeling and simulation method for the energy efficiency ratio of a solar-powered seawater desalination system according to an embodiment of this application is shown. Figure 1 As shown, the method includes:

[0065] S101. Acquire the membrane surface resistance data and selective permeability data of the ion exchange membrane under different concentration gradients during the electrodialysis process driven by photovoltaics, and simultaneously collect the DC bus voltage ripple data at the DC output terminal of the photovoltaic inverter.

[0066] Membrane surface resistance data refers to the resistive characteristic parameters of ion exchange membranes that hinder ion migration during electrodialysis. Selective permeability data refers to the performance index of ion exchange membranes that preferentially allow the permeation of ions of a specific charge while blocking ions of the opposite charge. DC bus voltage ripple data refers to the high-frequency AC component data superimposed on the steady-state DC voltage, collected at the DC output terminal of the photovoltaic inverter.

[0067] During the implementation of the scheme, electrochemical sensors were configured at key nodes of the electrodialysis stack, and a high-frequency voltage acquisition probe was connected to the output port of the photovoltaic inverter to achieve data acquisition. First, the concentration of the feed water entering the electrodialysis membrane stack was adjusted to create different concentration gradient environments. Second, the potential difference and current across the membrane were measured using the four-electrode method or electrochemical impedance spectroscopy, and the membrane surface resistance and ion transport number at the corresponding concentrations were calculated as selective permeability indicators. Simultaneously, a high-frequency data acquisition card was used to continuously record the voltage waveform of the inverter's DC bus at a preset sampling rate, and the ripple signal including the complete cycle was captured.

[0068] For example, in a seawater desalination project, suppose three concentration gradient environments, C1, C2, and C3, are set up, and the corresponding membrane resistance datasets are obtained respectively. and selection of permeability datasets Simultaneously, DC bus voltage ripple data sequences of the photovoltaic inverter under specific illumination conditions were collected. .

[0069] S102. Perform a fast Fourier transform on the DC bus voltage ripple data to obtain the spectral energy distribution characteristics. Align the spectral energy distribution characteristics and the normalized mapped film surface resistance data with the selective permeability data in the time dimension and perform vector splicing to generate the input matrix.

[0070] Optionally, step S102, which involves performing a fast Fourier transform on the DC bus voltage ripple data to obtain spectral energy distribution characteristics, and then aligning the spectral energy distribution characteristics and the normalized mapped film resistance data with the selected permeability data in the time dimension and performing vector concatenation to generate the input matrix, may specifically include:

[0071] S1021. Extract voltage signal segments from DC bus voltage ripple data according to the preset sampling time sequence, perform discrete transformation on the voltage signal segments to obtain frequency response sequences, and obtain spectral energy distribution characteristics by calculating the sum of squared magnitudes of each frequency point in the frequency response sequence.

[0072] A voltage signal segment refers to a discrete time sequence extracted from a continuously acquired DC bus voltage ripple data stream according to a specific sampling time window. This segment includes complete waveform information that can reflect the characteristics of electric field fluctuations at the current moment.

[0073] A frequency response sequence is a complex sequence generated by performing a discrete Fourier transform on a voltage signal segment in the time domain, including the amplitude and phase information of each frequency component. Spectral energy distribution characteristics refer to the eigenvectors obtained by calculating the energy intensity of each frequency component in the frequency response sequence, typically represented as the sum of the squares of the magnitudes at each frequency point.

[0074] In the implementation process, a fixed sampling time window length is first set as a sliding window, and voltage signal segments are extracted by sliding along the time axis on the original DC bus voltage ripple data sequence. Next, a Fast Fourier Transform (FFT) algorithm is applied to convert the extracted time-domain voltage signal segments into a frequency response sequence in the frequency domain. Subsequently, for each frequency point in this sequence, the square of its complex modulus is calculated to obtain a value representing the energy intensity of that frequency band. The energy values ​​of all frequency bands of interest are then aggregated to form a spectral energy distribution feature vector.

[0075] For example, suppose the voltage ripple data is obtained... Extracting time voltage signal segment The frequency response sequence is obtained by fast Fourier transform. Then, the spectral energy distribution characteristics are calculated. ,in This represents the number of effective frequency points.

[0076] S1022. Combine the spectral energy distribution characteristics and the normalized mapped film surface resistance data with the selective permeability data in a preset order to obtain a joint vector.

[0077] A joint vector is a one-dimensional data array formed by splicing and combining feature data from different sources and with different physical attributes at the same time point. It is a multi-dimensional hybrid feature descriptor that describes the system's operating state at that moment.

[0078] In the implementation process, the acquired membrane surface resistance and selective permeability data are first processed using the max-min normalization method, mapping them to a standard range. Then, based on the timestamp index, the spectral energy distribution feature vector at the same moment is aligned with the normalized membrane surface resistance and selective permeability values. To eliminate the influence of different physical dimensions on model training, the spectral energy distribution feature is also normalized before concatenation, ensuring that the final input matrix is ​​dimensionless. Finally, according to a preset order, the above data types are concatenated end-to-end to form a long vector encompassing both electrical fluctuation characteristics and membrane physicochemical properties.

[0079] For example, for time Assuming the normalized film surface resistance is The normalized selectivity permeability value is Spectral energy distribution characteristics for By concatenating these three elements, we obtain the joint vector at that moment. for .

[0080] S1023. The input matrix is ​​obtained by stacking the joint vectors of the continuous time series as row elements in an ordered manner.

[0081] The input matrix is ​​a two-dimensional data structure composed of joint vectors of multiple time steps stacked in chronological order. As the input layer data of a deep belief network, each row represents the complete state features at a specific sampling time, and each column represents the evolution of a specific feature dimension over time.

[0082] During the implementation of the scheme, all time points within the entire sampling period are traversed, and the joint vector generated at each time point is used as a row vector of a matrix. These row vectors are then arranged sequentially from top to bottom according to chronological order, thereby constructing the input matrix for training and inference of the deep learning model. Specifically, it is assumed that there exists a starting point from time... arrive Continuous joint vector sequence The final input matrix is ​​generated through ordered stacking. The structure is as follows:

[0083]

[0084] Among them, matrix Each row corresponds to a joint vector at a given time point. The first column shows the spectral energy distribution characteristics of voltage ripple, and the last two columns show the normalized film surface resistance and selective permeability data, respectively.

[0085] This embodiment analyzes the frequency domain energy characteristics of voltage ripple using fast Fourier transform and then aligns and vector-stitches them with membrane physical parameters. This achieves multi-source heterogeneous fusion of electrical fluctuations and membrane performance data, ensuring that the microscopic effects of voltage fluctuations on membrane performance can be accurately captured from a multi-dimensional perspective.

[0086] S103. The input matrix is ​​imported into the deep belief network model as input layer data. Through the layer-by-layer unsupervised pre-training mechanism of the multilayer restricted Boltzmann machine in the deep belief network model, the influence weight of the voltage ripple frequency component on the nonlinear change of the membrane resistance is learned. The ion impedance vector under the unsteady current, which represents the degree of dynamic obstruction encountered by ions when passing through the membrane channel under the action of the fluctuating electric field, is extracted from the input matrix.

[0087] Deep Belief Network (DBN) is a deep neural network architecture based on a probabilistic generative model, designed to extract high-order abstract features by learning deep statistical patterns in data. The core framework of this model consists of several layers of Restricted Boltzmann Machines (RBMs) stacked in a bottom-up manner, typically connected to a supervised learning layer for classification or regression, such as a Softmax classifier or a linear regression layer. In this application, a least squares support vector machine is subsequently connected. This hierarchical structure allows DBN to capture non-linear features from low to high levels in the input data layer by layer.

[0088] Specifically, the structure of a deep belief network (RBM) mainly consists of a visible layer and multiple hidden layers. The bottom visible layer directly receives the original input data, i.e., the input matrix in this application, and its number of nodes matches the dimension of the input vector. Each subsequent RBM layer consists of the visible nodes of the current layer and the hidden nodes of the previous layer. However, unlike traditional neural networks, the nodes within an RBM layer are not interconnected; only the nodes between layers have fully connected weights. This special two-layer undirected graph structure allows the RBM to define the joint probability distribution of visible and hidden variables through an energy function.

[0089] During the pre-training phase, each layer of the Restricted Block Model (RBM) undergoes independent unsupervised learning, using the output of the previous layer as the input of the current layer, greedily optimizing the weights layer by layer to initialize the parameters of the entire deep network. This mechanism effectively solves the training difficulty caused by random initialization of deep networks, enabling DBNs to possess powerful nonlinear mapping and feature extraction capabilities, making them particularly suitable for handling complex nonlinear coupling relationships between voltage ripple and membrane impedance, as described in this application.

[0090] Optionally, step S103, which imports the input matrix as input layer data into the deep belief network model and learns the influence weights of the voltage ripple frequency component on the nonlinear change of the membrane surface resistance through the layer-by-layer unsupervised pre-training mechanism of the multilayer restricted Boltzmann machine in the deep belief network model, and extracts the ion impedance vector under the unsteady current representing the degree of dynamic hindrance encountered by ions when passing through the membrane channel under the action of the fluctuating electric field from the input matrix, can specifically include:

[0091] Figure 2 A flowchart illustrating a method for determining an ion impedance vector according to an embodiment of this application is shown. Figure 2 As shown, the input matrix is ​​first mapped to the set of visible nodes in the bottom-level Restricted Boltzmann Machine (RBM) of the deep belief network model. The inter-layer connection weights are then updated by iteratively calculating the joint probability distribution between visible and hidden nodes, thus outputting a primary feature code. This primary feature code is then used as the input data for the next layer of the RBM, and the joint probability distribution calculation and connection weight update are repeated layer by layer until the data is passed to the top-level RBM.

[0092] Based on this, the correlation information between the voltage ripple frequency component and the nonlinear change in film resistance is encoded into a high-order abstract feature, and the connection weights of each layer of the confined Boltzmann machine are adjusted according to this high-order abstract feature. Finally, the output value of the hidden node set is extracted from the weight-adjusted top-layer confined Boltzmann machine and used as the ion impedance vector under unsteady current. This method includes:

[0093] S1031. Map the input matrix to the set of visible nodes of the underlying restricted Boltzmann machine in the deep belief network model. Update the inter-layer connection weights by iteratively calculating the joint probability distribution between the set of visible nodes and the set of hidden nodes, and output the primary feature code.

[0094] The visible node set refers to the group of input layer neurons in a Restricted Boltzmann Machine (RBM) that receives raw external data; their states directly correspond to the feature values ​​in the input matrix. The hidden node set refers to the group of unconnected latent layer neurons in a RBM, used to capture the higher-order correlations implicit in the visible node data.

[0095] Joint probability distribution refers to a probability measure based on an energy function that describes the simultaneous occurrence of visible node states and hidden node states. Primary feature encoding refers to the activation probabilities or state vectors output by the hidden node set in response to input data after the underlying Restricted Boltzmann Machine (RBM) has completed training.

[0096] In the implementation process, the input matrix generated in step S102, which includes joint vectors from multiple time points, is first imported into the underlying restricted Boltzmann machine as the training set. Secondly, unsupervised training is performed using the contrastive divergence algorithm: for each row vector in the input matrix, i.e., a sample at a certain time point, it is assigned to a visible node, the activation probability of the hidden node is calculated, and the hidden node state is sampled. Then, the visible node is reconstructed through the hidden node. This process is iterated repeatedly to update the connection weights and biases between the visible and hidden nodes, enabling the network to reconstruct the input data.

[0097] After training, each row of the input matrix is ​​fed back into the network to calculate the output value of the hidden nodes, thereby obtaining the primary feature encoding vector corresponding to each time step. The encoding vectors of all time steps are combined in sequence to form the input dataset for the next layer.

[0098] For example, it will include The input matrix of the joint vector at each time step Import the model, for the first element in the matrix row joint vector That is, a sample, after being mapped by a low-level restricted Boltzmann machine, outputs a corresponding primary feature encoding vector. For the entire matrix The final output is a feature dataset consisting of vectors from all time points. .

[0099] S1032. The primary feature encoding is used as the input data for the next layer of the Restricted Boltzmann Machine. The joint probability distribution calculation and connection weight update are repeated layer by layer until the data is passed to the top layer of the Restricted Boltzmann Machine.

[0100] In the implementation of the scheme, firstly, a layer-by-layer greedy pre-training strategy is adopted, using the feature dataset output by the previous layer's Restricted Boltzmann Machine (RBM) as the input training set for the current layer's RBM. This maintains the feedforward connections between layers, but treats each layer as an independent energy model during training. Secondly, the training process in step S1031 is repeated, calculating the joint probability distribution of the input features and hidden layer features and updating the weights, so that each layer can further extract higher-order features based on the output of the previous layer.

[0101] For example, the feature dataset output in step S1031 This serves as the input data for the visual layer of the second-layer Restricted Boltzmann Machine. After training, for the dataset... Each vector in The second layer outputs the corresponding secondary feature encoding vector. This generates a new feature dataset. If the network includes more layers, this process continues until it reaches the top-level restricted Boltzmann machine.

[0102] S1033. Encode the correlation information between the voltage ripple frequency component and the nonlinear change of the film surface resistance into a high-order abstract feature, and adjust the connection weights of each layer of restricted Boltzmann machine according to the high-order abstract feature.

[0103] High-order abstract features refer to the feature representation output by the top hidden node of the deep belief network. This feature integrates the results of multi-layer nonlinear transformation and can encode the deep implicit correlation between the voltage ripple frequency component and the nonlinear change of the film resistance.

[0104] In the implementation of the scheme, a fine-tuning strategy using deep neural networks is employed to achieve global optimization of the weights. First, the pre-trained multi-layer restricted Boltzmann machine is expanded into a complete deep autoencoder network. The joint vector in the input matrix is ​​used as the network input. Through forward propagation and nonlinear activation of neurons in each layer, high-order abstract features are generated at the top layer, thus completing the encoding process of the associated information. Next, the generated high-order abstract features are used to reconstruct the original input data through a decoding network, and the error between the reconstructed data and the original input data is calculated.

[0105] Finally, based on this error, the gradient is calculated using the backpropagation algorithm. From the top to the bottom, the connection weights of each layer of the restricted Boltzmann machine are then slightly modified and adjusted, allowing the network to more accurately represent the relationship between voltage ripple and membrane impedance. For example, suppose that for time... joint vector First, the input is fed into the network, and then high-order features are obtained at the top layer. Secondly, utilize Reverse reconstruction is performed to obtain the estimated value. Then, the error loss between the two is calculated. Finally, based on this loss The weight matrices of each layer are updated using gradient descent. This completes the adjustment of the weights.

[0106] S1034. Extract the output value of the hidden node set from the top-level restricted Boltzmann machine after the connection weight adjustment as the ion impedance vector under unsteady current.

[0107] The ion impedance vector under unsteady current refers to the feature vector generated by the deep belief network in the final output layer. This vector quantifies the degree of dynamic resistance encountered by ions when passing through the membrane pores under the action of photovoltaic fluctuation electric field. Mathematically, it is a dimensionless vector used to represent the relative change trend of physical impedance rather than being directly equivalent to the resistance value with the dimension of ohms.

[0108] During the implementation of the scheme, firstly, the connection weights and biases, after fine-tuning in step S1033, are fixed. Secondly, the initial input matrix generated in step S102 is... Again, using this as input, perform forward propagation computation: each row in the matrix represents a time step. joint vector The signal is passed sequentially through each layer of Restricted Boltzmann Machines (RBMs), and processed by weighted summation and nonlinear activation functions. Finally, the output value is extracted from the hidden node set of the top-level RBM as the ion impedance vector at that moment.

[0109] For example, for time input joint vector Assuming the three-layer network, after fine-tuning, sequentially generates intermediate feature vectors, and finally outputs a vector with dimension at the top layer. eigenvectors This vector That is, time. The ion impedance vector under unsteady current. For the entire time series, a matrix consisting of the impedance vectors at all times is obtained. .

[0110] This embodiment utilizes the multilayer restricted Boltzmann machine structure and layer-by-layer pre-training mechanism of deep belief networks to deeply explore the complex nonlinear coupling relationship between voltage ripple spectrum and membrane resistance. It effectively extracts the unsteady-state ion impedance vector representing the transmembrane transport characteristics of ions under fluctuating electric fields, overcomes the defect of shallow models that cannot capture deep dynamic laws, and significantly improves the accuracy of representing dynamic membrane impedance characteristics.

[0111] S104. Using the ion impedance vector as a regression operator, a nonlinear regression relationship between the ion impedance vector and the energy consumption per unit of produced water is established through a least squares support vector machine to generate an energy consumption prediction model.

[0112] Optionally, step S104, which uses the ion impedance vector as a regression operator and establishes a nonlinear regression relationship between the ion impedance vector and the energy consumption per unit of produced water through a least-squares support vector machine to generate an energy consumption prediction model, may specifically include:

[0113] Figure 3 A flowchart illustrating a method for generating an energy consumption prediction model according to an embodiment of this application is shown. Figure 3 As shown, the method includes:

[0114] S1041. The ion impedance vector is mapped to a high-dimensional feature space using the radial basis function kernel to obtain the kernel matrix.

[0115] A radial basis function kernel is a mathematical function that takes the Euclidean distance between vectors as its independent variable. It can map low-dimensional nonlinear data to a high-dimensional feature space to achieve linear separability and is often used to measure the similarity between samples. A kernel matrix is ​​a symmetric square matrix formed by calculating the values ​​of any two ion impedance vectors in the training dataset using a kernel function. Its element values ​​represent the inner product of the samples in high-dimensional space.

[0116] In the implementation process, the key parameter of the radial basis function kernel, namely the kernel width coefficient, is first determined, as this parameter determines the distribution characteristics of the mapping space. Next, all time-series samples in the training dataset are traversed to extract the corresponding ion impedance vectors. For each pair of sample vectors, the square of their Euclidean distance is calculated and substituted into the radial basis function formula to obtain the corresponding kernel function value. Finally, the kernel function values ​​of all sample pairs are arranged in sample index order to construct a symmetric square matrix, i.e., the kernel matrix.

[0117] For example, the data extracted using step S103 includes... Ion impedance vector dataset at each time step Select the kernel width Secondly, for the first [item] in the dataset... vectors and the vectors Calculate kernel function value The specific calculation formula is shown in formula (1) below:

[0118] (1)

[0119] Finally, traverse all Combining, generating dimensional kernel matrix .

[0120] S1042. Based on the kernel matrix and unit water production energy consumption, construct a system of linear equations that satisfy the least squares criterion, and obtain the regression weight coefficients and bias terms by solving the system of linear equations.

[0121] The regression weight coefficients refer to the Lagrange multipliers in the least squares support vector machine optimization problem, which determine the importance weight of each training sample point in constructing the final regression model. The bias term refers to the scalar intercept parameter in the model, used to correct the position of the regression curve in the output space.

[0122] During the implementation of the scheme, firstly, the actual unit water production energy consumption data corresponding to the ion impedance vector at each moment is obtained to form the target value vector. Secondly, based on the theoretical framework of least squares support vector machine, a system of linear equations satisfying the Karl-Kuntucker conditions is constructed using the kernel matrix, regularization parameters, and unity product matrix. This system of equations is typically composed of block matrices, including the sum of the kernel matrix and regularization terms, as well as an all-one vector. Finally, the system of linear equations is solved using matrix inversion or decomposition algorithms to directly calculate the regression weight coefficient vector and the bias term values.

[0123] For example, assuming the corresponding The actual unit water production energy consumption label vector at each moment nuclear matrix Set regularization parameters Construct the following system of linear equations:

[0124]

[0125] in It is a column vector of all 1s. It is the identity matrix. Let be the kernel matrix. Solving this system of equations yields the bias terms. and regression weight coefficient vector .

[0126] S1043. Establish a nonlinear regression function based on the regression weight coefficients and bias terms, and use the nonlinear regression function as an energy consumption prediction model.

[0127] An energy consumption prediction model refers to a final, established nonlinear regression function entity that includes all deterministic parameters, capable of estimating energy consumption for new, unknown inputs.

[0128] During the implementation of the scheme, the regression weight coefficient vector and bias term obtained in step S1042 are substituted into the decision function formula of the least squares support vector machine. This decision function is expressed in a summation form, where each term is the product of the weight coefficient of the training sample and the kernel function value between the new input vector and the training sample. This establishes a clear nonlinear regression function, which is then encapsulated as an energy consumption prediction model.

[0129] For example, the expression for the nonlinear regression function is shown in the following formula (2):

[0130] (2)

[0131] in The unit water production energy consumption predicted by the model and its dimensions are... This is used to represent the electrical energy consumed in producing one unit volume of fresh water. Let be the ion impedance vector at any time to be predicted. This represents the total number of samples used in model training. For the first in the training set Each ion impedance vector For the corresponding number The regression weight coefficients of each sample. For radial basis kernel functions, For bias. Suppose we input a new time step. ionic impedance vector Immediately Substituting the values ​​into formula (2) above, the predicted unit water production energy consumption value of the model is obtained. , can be represented as .

[0132] This embodiment utilizes least squares support vector machine to map the extracted ion impedance features to a high-dimensional space. By solving a system of linear equations, a nonlinear regression model is quickly constructed. This not only avoids the problem of traditional neural networks getting trapped in local optima, but also significantly reduces computational complexity. It can accurately fit the nonlinear relationship between ion impedance and energy consumption, thereby achieving efficient prediction of energy consumption per unit of water production.

[0133] S105. Based on the ion impedance vector at different times, the predicted value of unit water production energy consumption at each time is obtained using the energy consumption prediction model. The water production rate is calculated based on the photovoltaic input power and the predicted value of unit water production energy consumption at each time. The energy efficiency ratio is obtained by calculating the ratio of water production rate to photovoltaic input power, and a dynamic simulation curve of energy efficiency ratio is generated.

[0134] The predicted energy consumption per unit of produced water refers to the final prediction result after pollution correction. It integrates the initial energy consumption calculated based on real-time fluctuation conditions and the additional energy consumption caused by membrane performance degradation, and can truly reflect the actual operating energy efficiency level of the equipment under the current condition.

[0135] Photovoltaic input power refers to the DC power output by the photovoltaic array at a specific sampling moment and actually supplied to the electrodialysis equipment. Its magnitude exhibits dynamic fluctuations with changes in ambient light intensity and temperature. Water production rate refers to the volume of freshwater meeting desalination standards produced by the device per unit time under the current photovoltaic input power.

[0136] Energy efficiency ratio (EER) is a key performance indicator for measuring the energy utilization efficiency of the seawater desalination process. In this application, it is defined as the ratio of water production rate to photovoltaic input power. Its physical meaning represents the volume of fresh water that can be produced by consuming a unit of electrical energy. The higher the value, the higher the energy conversion efficiency under the current fluctuating power supply.

[0137] The dynamic simulation curve of energy efficiency ratio refers to a visual chart that continuously displays the energy efficiency ratio value calculated at discrete time points in the time domain. It can intuitively reflect the efficiency response trajectory under the coupling effect of multiple factors such as light fluctuation, voltage ripple interference and membrane performance evolution.

[0138] Optionally, the method further includes:

[0139] Extract the time data sequence from the ion impedance vector and calculate the average value of the modulus in the time data sequence.

[0140] A time data series is an ordered set of vectors formed by arranging ion impedance vectors extracted from multiple consecutive moments in chronological order.

[0141] During the implementation of the scheme, firstly, a sliding time window or a fixed monitoring period is set, and a corresponding set of vector sequences is extracted from the ion impedance vector stream output in step S103. Secondly, for each ion impedance vector in the sequence, its vector magnitude is calculated. Subsequently, these magnitudes are accumulated and divided by the number of vectors to obtain the average magnitude over that time period. For example, assume that the monitoring window is set as follows: Extracted within this window The ion impedance vector at each time step Calculate the magnitude of each vector. Then calculate the average value. .

[0142] When the average value increases monotonically over time and exceeds the preset pollution baseline threshold, the difference ratio between the average value and the pollution baseline threshold is calculated to obtain the correction coefficient.

[0143] The fouling baseline threshold is a pre-set upper limit standard for the impedance modulus of an ion exchange membrane in an uncontaminated or slightly contaminated state. When the actual observed value exceeds this standard, it means that fouling or contaminant deposition may have occurred on the membrane surface. The correction factor is a dimensionless scaling factor used to quantify the impact of membrane fouling on the energy consumption model. Its value is determined based on the degree to which the actual impedance exceeds the baseline threshold. The pre-set fouling baseline thresholds are shown in Table 1 below:

[0144] Table 1: Preset Pollution Benchmark Threshold Comparison Table

[0145]

[0146] As shown in Table 1, different dimensionless threshold values ​​were set for cation exchange membrane A and anion exchange membrane B based on different salinity levels (high salinity, low salinity, and organic matter content). For example, the threshold value for cation exchange membrane A was set to 1.0 in a low salinity environment and 1.2 in a high salinity environment; the threshold value for anion exchange membrane B was set to a higher value of 1.5 in a high organic matter environment because it is more susceptible to contamination.

[0147] During the implementation of the scheme, the pollution baseline threshold is first set according to Table 1. Secondly, the changing trend of the average value of the membrane is monitored in real time. If the average value of multiple consecutive monitoring cycles shows a monotonically increasing trend, and the average value of the current cycle exceeds the selected threshold, membrane fouling is determined to have occurred. At this point, the difference between the current average value and the threshold is calculated, and then this difference is divided by the threshold to obtain the correction coefficient.

[0148] For example, referring to Table 1, assuming cation exchange membrane A is used and the system operates in a low-salinity environment, the pollution baseline threshold is set as follows: If the average value of the modulus for the current period is detected. Rise to If the trend is monotonic, then calculate the correction factor. .

[0149] In step S105, based on the ion impedance vectors at different times, the predicted energy consumption per unit of produced water at each time moment is obtained using the energy consumption prediction model, including:

[0150] S1051. Based on the ion impedance vector and correction coefficient at different times, the predicted value of unit water production energy consumption at each time is obtained using the energy consumption prediction model.

[0151] During the implementation of the scheme, the obtained correction coefficients are incorporated into the calculation process of the energy consumption prediction model. Specifically, using the ion impedance vector at the current moment as the basic input, the model calculates the basic energy consumption value. Simultaneously, the correction coefficients are used to compensate for this basic value, such as gain adjustment, to obtain the final corrected predicted energy consumption per unit of produced water. For example, suppose a new time moment is input... ionic impedance vector Correction factor Then, through the compensation logic within the model, it outputs the predicted energy consumption per unit of water production after pollution correction. .

[0152] This embodiment achieves adaptive compensation for model prediction results. This design effectively solves the long-term prediction bias problem caused by the traditional model ignoring slow time-varying factors such as membrane fouling and aging, and ensures high fidelity and reliability of energy efficiency ratio simulation throughout the entire life cycle.

[0153] Optionally, step S1051, which uses the energy consumption prediction model to obtain the predicted unit water production energy consumption value at each time based on the ion impedance vector and correction coefficient at different times, may specifically include:

[0154] S10511. Input the ion impedance vectors at different times into the least squares support vector machine in the energy consumption prediction model. Map the ion impedance vectors to a high-dimensional feature space through the kernel function and solve the linear equation system to obtain the initial unit water production energy consumption value at different times.

[0155] The initial unit water production energy consumption value refers to the theoretical energy consumption value directly calculated by an energy consumption prediction model that does not include pollution correction terms, based solely on the electrical and physical fluctuation characteristics contained in the ion impedance vector input at the current moment.

[0156] In the implementation process, firstly, the trained least squares support vector machine model generated in step S104 is invoked, and its core parameters, including the support vector set, regression weight coefficients, and bias terms, are loaded. Secondly, the ion impedance vector at the current time to be predicted is used as the input variable, and the radial basis function kernel function is used to calculate the kernel function value between this vector and all support vectors in the model, thus completing the mapping to the high-dimensional feature space.

[0157] Finally, based on the calculation logic of the nonlinear regression function, all kernel function values ​​are weighted and summed with their corresponding regression weight coefficients, and a bias term is added to calculate the initial unit water production energy consumption value at that time. For example, suppose time... ionic impedance vector Substituting into formula (2) above, we obtain the initial unit water production energy consumption value. .

[0158] S10512. The energy consumption gain value is obtained by calculating the product of the correction coefficient and the preset pollution sensitivity factor.

[0159] The preset fouling sensitivity factor is an empirical conversion coefficient used to map the dimensionless correction factor to a specific increase in energy consumption. The energy gain value refers to the additional energy consumed due to increased ion migration resistance caused by fouling or contaminant deposition on the membrane surface. The preset fouling sensitivity factors are shown in Table 2 below:

[0160] Table 2: Preset Pollution Sensitivity Factor Comparison Table

[0161]

[0162] As shown in Table 2, different sensitivity factor values ​​were set for cation exchange membrane A and anion exchange membrane B based on their dense and porous structures. For example, cation exchange membrane A, with its dense structure, has a more significant impact on energy consumption due to its smaller pore size, and its sensitivity factor was set at 0.5 kWh / m³; while the sensitivity factor for the same type of membrane with a porous structure was lower, at 0.3 kWh / m³.

[0163] During the implementation of the scheme, firstly, based on the actual type and structural characteristics of the ion exchange membrane used, matching pollution sensitivity factors are retrieved from Table 2. Then, the correction coefficient calculated in step S105 is obtained, and this dimensionless correction coefficient is multiplied by the retrieved pollution sensitivity factor with energy density dimensions. This converts the relative degradation of membrane performance into an absolute energy consumption increment, and the calculation result is the dimensionless correction factor. The energy gain value. For example, assuming a dense cation exchange membrane A, the correction factor... The pollution sensitivity factors can be found in Table 2. Then calculate the energy gain value. .

[0164] S10513. By summing the initial unit water production energy consumption value and energy consumption gain value at different times, the predicted unit water production energy consumption value at each time is obtained.

[0165] During the implementation of the scheme, a simple addition operation is performed. The initial unit water production energy consumption value obtained in step S10511 is then used. The energy gain value obtained in step S10512 Add them together to obtain the final predicted energy consumption per unit of water production. .

[0166] This embodiment not only retains the ability of artificial intelligence models to fit complex and fluctuating operating conditions, but also makes up for the shortcomings of pure data-driven models in adapting to slow variables by using explicit physical correction terms, which significantly improves the accuracy of energy efficiency ratio prediction in long-term operating scenarios.

[0167] Optionally, the process of calculating the water production rate based on the photovoltaic input power and the predicted energy consumption per unit of water production at each moment in step S105, obtaining the energy efficiency ratio by calculating the ratio of the water production rate to the photovoltaic input power, and generating the dynamic simulation curve of the energy efficiency ratio can specifically include:

[0168] S1052. Calculate the ratio of photovoltaic input power to the predicted energy consumption per unit of water production at each moment to obtain the water production rate at each moment.

[0169] During the implementation of the scheme, the photovoltaic input power data corresponding to the ion impedance vector at each moment is first acquired synchronously. This data typically comes from the power monitoring module built into the photovoltaic inverter or a photovoltaic output model based on meteorological data. Next, the pollution-corrected unit water production energy consumption prediction value output in step S1051 is read. Subsequently, a division operation is used to divide the photovoltaic input power at each moment by the corresponding unit water production energy consumption prediction value, thereby calculating the theoretical water production rate at that moment.

[0170] For example, suppose for including First, obtain the photovoltaic input power sequence for each time period. and the corresponding predicted energy consumption per unit of water production Secondly, calculations are performed on each element in the sequence to obtain the water production rate sequence. , at any time Water production rate .

[0171] S1053. Calculate the ratio of water production rate to photovoltaic input power to obtain the energy efficiency ratio at each time point.

[0172] During the implementation of the scheme, firstly, the water production rate sequence calculated in step S1052 and the original photovoltaic input power sequence are invoked. Secondly, an element-wise division operation is performed, dividing the water production rate by the corresponding photovoltaic input power to obtain the instantaneous energy efficiency ratio (EER) value at each moment. Since the water production rate is obtained by dividing the photovoltaic input power by the predicted energy consumption per unit of water production, the instantaneous EER is numerically equivalent to taking the reciprocal of the predicted energy consumption per unit of water production. Finally, the values ​​from all times are compiled into a sequence. For example, this is done using a water production rate sequence. and photovoltaic input power sequence The energy efficiency ratio sequence was calculated. any element .

[0173] S1054. Map the energy efficiency ratios at multiple consecutive moments to a preset two-dimensional coordinate system in chronological order to obtain a dynamic simulation curve of the energy efficiency ratio that changes continuously with time.

[0174] In the implementation process, firstly, a two-dimensional mapping space including time and numerical dimensions is established. Secondly, the entire simulation cycle is traversed, and the time series is... The calculated energy efficiency ratio sequence will be used as the horizontal axis data. The data, represented by the ordinate, is plotted and mapped onto a coordinate system. Subsequently, interpolation algorithms or smoothing techniques are used to connect these discrete points, ultimately generating a continuously changing dynamic curve. .

[0175] This embodiment not only clearly demonstrates the transient impact of power fluctuations on production efficiency, but also reveals the evolution of energy efficiency over time through curve trends, providing engineers with an intuitive and quantitative basis for system optimization design and operational strategy adjustment.

[0176] Figure 4 This is a schematic diagram of a specific implementation of a dynamic modeling and simulation system for the energy efficiency ratio of a solar-powered seawater desalination system provided in this application embodiment. (Refer to...) Figure 4 The system may include:

[0177] The 410 acquisition module is used to acquire membrane surface resistance data and selective permeability data of ion exchange membrane under different concentration gradients during the electrodialysis process driven by photovoltaics, and at the same time, it collects DC bus voltage ripple data at the DC output terminal of the photovoltaic inverter.

[0178] The 420 generation module is used to perform fast Fourier transform on DC bus voltage ripple data to obtain spectral energy distribution characteristics. The spectral energy distribution characteristics and the normalized mapped film surface resistance data are then aligned with the selected permeability data in the time dimension and vector-stitched to generate the input matrix.

[0179] The 430 extraction module is used to import the input matrix as input layer data into the deep belief network model. Through the layer-by-layer unsupervised pre-training mechanism of the multilayer restricted Boltzmann machine in the deep belief network model, the influence weight of the voltage ripple frequency component on the nonlinear change of the membrane resistance is learned. The ion impedance vector under the unsteady current, which represents the degree of dynamic hindrance encountered by ions when passing through the membrane channel under the action of the fluctuating electric field, is extracted from the input matrix.

[0180] The 420 generation module is also used to establish a nonlinear regression relationship between the ion impedance vector and the unit water production energy consumption by using the ion impedance vector as a regression operator and the least squares support vector machine, thereby generating an energy consumption prediction model.

[0181] The 440 calculation module is used to obtain the predicted energy consumption per unit of water production at each time based on the ion impedance vector at different times using the energy consumption prediction model. It also calculates the water production rate based on the photovoltaic input power and the predicted energy consumption per unit of water production at each time. The energy efficiency ratio is obtained by calculating the ratio of the water production rate to the photovoltaic input power, and a dynamic simulation curve of the energy efficiency ratio is generated.

[0182] The dynamic modeling and simulation system for the energy efficiency ratio of a solar desalination system in this application is used to implement the aforementioned dynamic modeling and simulation method for the energy efficiency ratio of a solar desalination system. Therefore, the specific implementation of the dynamic modeling and simulation system for the energy efficiency ratio of a solar desalination system can be found in the embodiment section of the dynamic modeling and simulation method for the energy efficiency ratio of a solar desalination system mentioned above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0183] Figure 5 A schematic diagram of the hardware structure of an electronic device provided in one embodiment of this application is shown.

[0184] The electronic device may include a processor 510 and a memory 520 storing computer program instructions.

[0185] Specifically, the processor 510 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0186] Memory 520 may include mass storage for data or instructions. For example, and not limitingly, memory 520 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 520 may include removable or non-removable (or fixed) media. Where appropriate, memory 520 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 520 is non-volatile solid-state memory.

[0187] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to the first aspect of this disclosure.

[0188] The processor 510 reads and executes computer program instructions stored in the memory 520 to implement any of the dynamic modeling and simulation methods for the energy efficiency ratio of the solar desalination system in the above embodiments.

[0189] In one example, the electronic device may also include a communication interface 530 and a bus 540. Wherein, such as Figure 5 As shown, the processor 510, memory 520, and communication interface 530 are connected through bus 540 and complete communication with each other.

[0190] The communication interface 530 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0191] Bus 540 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 540 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0192] The electronic device can execute the dynamic modeling and simulation method for the energy efficiency ratio of the solar desalination system in the embodiments of this application, thereby realizing the dynamic modeling and simulation method for the energy efficiency ratio of the solar desalination system described in conjunction with the accompanying drawings.

[0193] Furthermore, in conjunction with the dynamic modeling and simulation method for the energy efficiency ratio of the solar-powered seawater desalination system in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the dynamic modeling and simulation methods for the energy efficiency ratio of the solar-powered seawater desalination system in the above embodiments.

[0194] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0195] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0196] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0197] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0198] The above provides a detailed description of the dynamic modeling and simulation method and system for the energy efficiency ratio of a solar-powered seawater desalination system. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A dynamic modeling and simulation method for the energy efficiency ratio of a solar-powered seawater desalination system, characterized in that, The method includes: Acquire membrane surface resistance and selective permeability data of ion exchange membranes under different concentration gradients during photovoltaic-driven electrodialysis, and simultaneously collect DC bus voltage ripple data at the DC output terminal of the photovoltaic inverter. The DC bus voltage ripple data is subjected to fast Fourier transform to obtain the spectral energy distribution characteristics. The spectral energy distribution characteristics and the normalized mapped film surface resistance data are aligned in the time dimension and vector-stitched with the selective permeability data to generate an input matrix. The input matrix is ​​imported into the deep belief network model as input layer data. Through the layer-by-layer unsupervised pre-training mechanism of the multilayer restricted Boltzmann machine in the deep belief network model, the influence weight of the voltage ripple frequency component on the nonlinear change of the membrane resistance is learned. The ion impedance vector under the unsteady current, which represents the degree of dynamic obstruction encountered by ions when passing through the membrane channel under the action of the fluctuating electric field, is extracted from the input matrix. Using the ion impedance vector as a regression operator, a nonlinear regression relationship between the ion impedance vector and the energy consumption per unit of produced water is established through a least squares support vector machine to generate an energy consumption prediction model. Based on the ion impedance vector at different times, the predicted energy consumption per unit of water production is obtained using the energy consumption prediction model. The water production rate is calculated based on the photovoltaic input power at each time and the predicted energy consumption per unit of water production. The energy efficiency ratio is obtained by calculating the ratio of the water production rate to the photovoltaic input power, and a dynamic simulation curve of the energy efficiency ratio is generated.

2. The method according to claim 1, characterized in that, The method further includes: Extract the time data sequence from the ion impedance vector and calculate the average value of the modulus in the time data sequence; When the average value increases monotonically over time and is greater than a preset pollution baseline threshold, a correction coefficient is obtained by calculating the ratio of the difference between the average value and the pollution baseline threshold. The method of obtaining the predicted unit water production energy consumption value at each time point based on the ion impedance vector at different times using the energy consumption prediction model includes: Based on the ion impedance vector and the correction coefficient at different times, the predicted value of unit water production energy consumption at each time is obtained using the energy consumption prediction model.

3. The method according to claim 2, characterized in that, The method of obtaining the predicted unit water production energy consumption value at each time point using the energy consumption prediction model based on the ion impedance vector and the correction coefficient at different times includes: The ion impedance vectors at different times are input into the least squares support vector machine in the energy consumption prediction model. The ion impedance vectors are mapped to a high-dimensional feature space through the kernel function and the linear equation system is solved to obtain the initial unit water production energy consumption value at different times. The energy consumption gain value is obtained by calculating the product of the correction coefficient and the preset pollution sensitivity factor. The predicted value of unit water production energy consumption at each time moment is obtained by summing the initial unit water production energy consumption value and the energy consumption gain value at different times.

4. The method according to claim 1, characterized in that, The process involves performing a Fast Fourier Transform on the DC bus voltage ripple data to obtain spectral energy distribution characteristics. These spectral energy distribution characteristics, along with the normalized mapped film surface resistance data, are then time-dimensionally aligned and vector-joined with the selective permeability data to generate an input matrix. This process includes: Voltage signal segments are extracted from the DC bus voltage ripple data according to a preset sampling time sequence. The voltage signal segments are then discretized to obtain a frequency response sequence. The spectral energy distribution characteristics are obtained by calculating the sum of squares of the magnitudes at each frequency point in the frequency response sequence. The spectral energy distribution characteristics and the normalized mapped film surface resistance data are numerically combined with the selective permeability data in a preset order to obtain a joint vector; The input matrix is ​​obtained by stacking the joint vectors from the continuous time series as row elements in an ordered manner.

5. The method according to claim 1, characterized in that, The input matrix is ​​imported as input layer data into a deep belief network model. Through the layer-by-layer unsupervised pre-training mechanism of the multilayer restricted Boltzmann machine in the deep belief network model, the influence weights of the voltage ripple frequency component on the nonlinear change of the membrane surface resistance are learned. From the input matrix, an ion impedance vector representing the degree of dynamic impediment encountered by ions passing through the membrane channel under a fluctuating electric field under unsteady current is extracted, including: The input matrix is ​​mapped to the set of visible nodes of the bottom restricted Boltzmann machine in the deep belief network model. The inter-layer connection weights are updated by iteratively calculating the joint probability distribution between the set of visible nodes and the set of hidden nodes, and the primary feature code is output. The primary feature encoding is used as the input data for the next layer of the Restricted Boltzmann Machine, and the joint probability distribution calculation and connection weight update are repeated layer by layer until they are passed to the top layer of the Restricted Boltzmann Machine. The correlation information between the voltage ripple frequency component and the nonlinear change of the film surface resistance is encoded into a high-order abstract feature, and the connection weights of each layer of restricted Boltzmann machine are adjusted according to the high-order abstract feature. The output value of the hidden node set is extracted from the top-level restricted Boltzmann machine after the connection weight adjustment and used as the ion impedance vector under the unsteady current.

6. The method according to claim 1, characterized in that, The process involves using the ion impedance vector as a regression operator, establishing a nonlinear regression relationship between the ion impedance vector and the energy consumption per unit of produced water through a least-squares support vector machine, and generating an energy consumption prediction model, including: The ion impedance vector is mapped to a high-dimensional feature space using a radial basis function kernel to obtain the kernel matrix; Based on the kernel matrix and the unit water production energy consumption, a system of linear equations satisfying the least squares criterion is constructed, and the regression weight coefficients and bias terms are obtained by solving the system of linear equations. A nonlinear regression function is established based on the regression weight coefficients and the bias term, and the nonlinear regression function is used as the energy consumption prediction model.

7. The method according to claim 1, characterized in that, The process of calculating the water production rate based on the photovoltaic input power at each moment and the predicted energy consumption per unit of water production, obtaining the energy efficiency ratio by calculating the ratio of the water production rate to the photovoltaic input power, and generating a dynamic simulation curve of the energy efficiency ratio includes: The ratio of the photovoltaic input power to the predicted energy consumption per unit of water production is calculated at each time step to obtain the water production rate at each time step. Calculate the ratio of the water production rate to the photovoltaic input power to obtain the energy efficiency ratio at each moment; The energy efficiency ratios at multiple consecutive moments are mapped onto a preset two-dimensional coordinate system in chronological order to obtain a dynamic simulation curve of the energy efficiency ratio that changes continuously with time.

8. A dynamic modeling and simulation system for the energy efficiency ratio of a solar-powered seawater desalination system, characterized in that, include: The acquisition module is used to acquire membrane surface resistance data and selective permeability data of ion exchange membrane under different concentration gradients during the electrodialysis process driven by photovoltaics, and at the same time, to collect DC bus voltage ripple data at the DC output terminal of the photovoltaic inverter. The generation module is used to perform a fast Fourier transform on the DC bus voltage ripple data to obtain the spectral energy distribution characteristics, and to perform time-dimensional alignment and vector concatenation of the spectral energy distribution characteristics and the normalized mapped film surface resistance data with the selective permeability data to generate an input matrix. The extraction module is used to import the input matrix as input layer data into the deep belief network model. Through the layer-by-layer unsupervised pre-training mechanism of the multilayer restricted Boltzmann machine in the deep belief network model, the influence weight of the voltage ripple frequency component on the nonlinear change of the membrane resistance is learned. The ion impedance vector under the unsteady current, which represents the degree of dynamic obstruction encountered by ions when passing through the membrane channel under the action of the fluctuating electric field, is extracted from the input matrix. The generation module is also used to establish a nonlinear regression relationship between the ion impedance vector and the unit water production energy consumption by using the ion impedance vector as a regression operator and least squares support vector machine, thereby generating an energy consumption prediction model. The calculation module is used to obtain the predicted value of unit water production energy consumption at each time based on the ion impedance vector at different times using the energy consumption prediction model, and to calculate the water production rate based on the photovoltaic input power at each time and the predicted value of unit water production energy consumption. The energy efficiency ratio is obtained by calculating the ratio of the water production rate to the photovoltaic input power, and a dynamic simulation curve of the energy efficiency ratio is generated.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the dynamic modeling and simulation method for the energy efficiency ratio of a solar-powered seawater desalination system as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables a dynamic modeling and simulation method for the energy efficiency ratio of a solar-powered seawater desalination system as described in any one of claims 1 to 7.

Images