Filter pressure drop prediction method and device, storage medium and electronic equipment

Abstract

The application provides a filter pressure drop prediction method and device, a storage medium and an electronic device, which can obtain feature data of a filter to be predicted; the feature data comprises material features and operating environment features of the filter; the feature data is input into a pressure drop prediction model to obtain a pressure drop prediction result of the filter; the pressure drop prediction model is obtained by training a machine learning model by using a preset training sample set, model parameters of the machine learning model are calculated by using a target optimization algorithm, and the target optimization algorithm comprises a Northern Goshawk optimization algorithm improved based on a subtractive optimizer and a Cauchy distribution function. By using the method provided in the application, the pressure drop of the filter can be quickly and accurately determined.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, storage medium, and electronic device for predicting the pressure drop of a filter. Background Technology

[0002] Against the backdrop of the global energy structure transitioning towards a green and low-carbon model, natural gas, as a cleaner, low-carbon fossil fuel, is playing an increasingly prominent role. As the bridge connecting natural gas extraction and end-users, the safe and efficient operation of natural gas transmission systems is crucial. To ensure the quality and stability of natural gas during pipeline transportation and to prevent the formation of harmful impurity droplets such as condensate due to environmental changes, temperature fluctuations, and pressure differences, purification is an indispensable step. Gas-liquid coalescers, as core equipment in key stages such as natural gas purification, liquefied natural gas (LNG) processing, pipeline transportation, and fuel gas processing, are of paramount importance.

[0003] Gas-liquid coalescers, through their unique coalescing mechanism, effectively remove tiny liquid droplets entrained in the gas stream, thereby ensuring the purity of natural gas and the long-term stable operation of the transmission system. This function is not only crucial for the efficient utilization of natural gas but also directly related to the safety of pipeline transportation, the protection of critical equipment, and human health and environmental protection. Therefore, the rigorous evaluation and optimized design of gas-liquid coalescer filtration performance is a continuous focus of attention within the industry.

[0004] When evaluating the performance of gas-liquid coalescers, pressure drop is a key indicator of efficiency and energy consumption, and its accurate prediction is crucial for equipment selection, design optimization, and operation management. However, existing technologies have several limitations in pressure drop prediction. Traditional experimental methods, while intuitive and reliable, are costly, time-consuming, and unable to cover all possible operating scenarios, limiting their widespread application in practice. On the other hand, while mathematical calculation methods based on physical models can theoretically simulate complex physical processes, the simplification of model assumptions, uncertainties in parameter selection, and the complex variations in physical phenomena under different operating conditions often lead to significant deviations in the calculation results, making it difficult to meet the accuracy requirements of different industrial applications. Summary of the Invention

[0005] The technical problem this application aims to solve is to provide a method, apparatus, storage medium, and electronic device for predicting the pressure drop of a filter, capable of quickly and accurately determining the pressure drop of the filter. The specific solution is as follows:

[0006] A method for predicting the pressure drop of a filter, comprising:

[0007] Obtain the feature data of the filter to be predicted; the feature data includes the material characteristics and operating environment characteristics of the filter;

[0008] The feature data is input into the pressure drop prediction model to obtain the pressure drop prediction result of the filter. The pressure drop prediction model is obtained by training a machine learning model using a preset training sample set. The model parameters of the machine learning model are calculated by an objective optimization algorithm, which includes the Northern Eagle optimization algorithm based on a subtraction optimizer and a Cauchy distribution function.

[0009] Optionally, the process of calculating the model parameters of the machine learning model using a target optimization algorithm, as described above, includes:

[0010] Generate an initial population; the initial population includes a preset number of goshawk individuals, each goshawk individual representing a candidate model parameter value scheme for a machine learning model;

[0011] The initial population is identified as the target population, and the fitness value of each individual goshawk in the target population is calculated.

[0012] Determine the target fitness value of the prey location for each individual goshawk in the current target population;

[0013] For each individual goshawk, if its fitness value is less than the target fitness value of the corresponding prey location, a random number is generated. Based on the best goshawk in the target population and the currently generated random number, the first updated position of the goshawk is determined. If the fitness value of the goshawk is not less than the target fitness value of the corresponding prey location, the first updated position of the goshawk is determined by a preset subtraction optimizer. The best goshawk in the target population is the goshawk with the best fitness value in the target population.

[0014] Based on the fitness value of each individual goshawk, the first update position, and the fitness value of the first update position, the target population is updated in the first stage.

[0015] For each goshawk individual in the target population after the first-stage position update, a mutation random number is generated for that goshawk individual. If the mutation random number is less than a preset Cauchy mutation threshold, the second updated position of the goshawk individual is determined based on the best goshawk individual in the target population after the first-stage position update and a preset Cauchy distribution function. If the mutation random number is not less than the preset Cauchy mutation threshold, the second updated position of the goshawk individual is determined based on the hunting radius, upper and lower limits of parameters, and the random number. The hunting radius is determined based on the current iteration number and a preset iteration number threshold.

[0016] Based on the fitness value of each individual goshawk in the target population after the first-stage position update, the second updated position, and the fitness value of the second updated position, the target population after the first-stage position update is updated in the second stage to obtain a new target population.

[0017] If the current iteration number is not equal to the iteration number threshold, increment the current iteration number by one and return to the step of determining the target fitness value of the prey location corresponding to each individual goshawk in the current target population;

[0018] If the current iteration count is equal to the iteration count threshold, the best individual goshawk in the new target population is used as the model parameter of the machine learning model.

[0019] Optionally, in the above method, calculating the fitness value of each individual goshawk in the target population includes:

[0020] The fitness value of each individual goshawk in the target population is calculated based on the k-fold cross-validation algorithm and the training sample set.

[0021] Optionally, the process of training a machine learning model based on a preset training sample set to obtain a pressure drop prediction model, as described above, includes:

[0022] Obtain a training sample set; the training sample set includes multiple training samples; the training samples include filter feature data and voltage drop information;

[0023] The machine learning model is trained using training samples from the training sample set to obtain a pressure drop prediction model.

[0024] Optionally, the above method includes obtaining the training sample set, which may include:

[0025] Acquire each raw sample data, where each raw sample data includes feature data of multiple candidate feature types and voltage drop information;

[0026] Based on the preset Spearman correlation analysis algorithm and the original sample data, the correlation coefficient of each candidate feature type is determined; the correlation coefficient of each candidate feature type characterizes the degree of correlation between the feature data of each candidate feature type and the voltage drop information.

[0027] Based on the correlation coefficient of the feature data of each candidate feature type, feature data of the target feature type are selected from each candidate feature type of each original sample data to obtain the initial sample data corresponding to each original sample data; the initial sample data includes the feature data of the target feature type and voltage drop information.

[0028] The initial sample data are normalized to obtain the final sample data corresponding to each initial sample data.

[0029] The final sample data are divided according to a preset ratio to obtain a training sample set and a test sample set.

[0030] A pressure drop prediction device for a filter, comprising:

[0031] An acquisition unit is used to acquire feature data of the filter to be predicted; the feature data includes the material characteristics and operating environment characteristics of the filter;

[0032] The prediction unit is used to input the feature data into the pressure drop prediction model to obtain the pressure drop prediction result of the filter. The pressure drop prediction model is obtained by training a machine learning model using a preset training sample set. The model parameters of the machine learning model are calculated by an objective optimization algorithm, which includes the Northern Eagle optimization algorithm based on a subtraction optimizer and Cauchy distribution function.

[0033] Optionally, in the aforementioned apparatus, the prediction unit includes:

[0034] A subunit is generated to generate an initial population; the initial population includes a preset number of goshawk individuals, and each goshawk individual represents a candidate model parameter value scheme for a machine learning model;

[0035] The first determining subunit is used to determine the initial population as the target population and calculate the fitness value of each individual goshawk in the target population;

[0036] The second determining subunit is used to determine the target fitness value of the prey location corresponding to each individual goshawk in the current target population.

[0037] The first execution subunit is configured to, for each individual goshawk, generate a random number if the fitness value of the individual goshawk is less than the target fitness value of the corresponding prey location, and determine the first updated position of the individual goshawk based on the best individual goshawk in the target population and the currently generated random number; and determine the first updated position of the individual goshawk through a preset subtraction optimizer if the fitness value of the individual goshawk is not less than the target fitness value of the corresponding prey location; the best individual goshawk in the target population is the individual goshawk with the best fitness value in the target population.

[0038] The first update subunit is used to perform a first-stage position update on the target population based on the fitness value of each individual goshawk, the first update position, and the fitness value of the first update position.

[0039] The second execution subunit is used to generate a mutation random number for each goshawk individual in the target population after the first-stage position update; if the mutation random number is less than a preset Cauchy mutation threshold, determine the second updated position of the goshawk individual based on the best goshawk individual in the target population after the first-stage position update and a preset Cauchy distribution function; if the mutation random number is not less than the preset Cauchy mutation threshold, determine the second updated position of the goshawk individual based on the hunting radius, upper and lower limits of parameters, and the random number; the hunting radius is determined based on the current iteration number and a preset iteration number threshold.

[0040] The second update subunit is used to perform a second-stage position update on the target population after the first-stage position update based on the fitness value of each individual goshawk in the target population after the first-stage position update, the second update position, and the fitness value of the second update position, so as to obtain a new target population.

[0041] The third execution subunit is used to increment the current iteration number by one and return the target fitness value that triggers the second determination subunit to determine the prey location corresponding to each individual goshawk in the current target population when the current iteration number is not equal to the iteration number threshold.

[0042] The fourth execution subunit is used to take the best individual goshawk in the new target population as the model parameter of the machine learning model when the current iteration number is equal to the iteration number threshold.

[0043] Optionally, in the aforementioned apparatus, the first determining subunit includes:

[0044] The calculation module is used to calculate the fitness value of each individual goshawk in the target population based on the k-fold cross-validation algorithm and the training sample set.

[0045] A storage medium comprising stored instructions, wherein, when the instructions are executed, the device in which the storage medium resides executes a filter voltage drop prediction method as described above.

[0046] An electronic device includes a memory and one or more instructions, wherein one or more instructions are stored in the memory and configured to be executed by one or more processors as described above for voltage drop prediction of a filter.

[0047] Based on the above-described embodiments of this application, a method, apparatus, storage medium, and electronic device for predicting the pressure drop of a filter are provided. The method includes: acquiring feature data of the filter to be predicted; the feature data includes material characteristics and operating environment characteristics of the filter; inputting the feature data into a pressure drop prediction model to obtain a pressure drop prediction result for the filter; the pressure drop prediction model is obtained by training a machine learning model using a preset training sample set, and the model parameters of the machine learning model are calculated using an objective optimization algorithm, including the Northern Eagle optimization algorithm based on a subtraction optimizer and a Cauchy distribution function. Applying the method provided in this application, the pressure drop of a filter can be determined quickly and accurately. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in 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 embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0049] Figure 1 A flowchart of a filter pressure drop prediction method provided in this application;

[0050] Figure 2 A flowchart illustrating the process of calculating model parameters of a machine learning model using a target optimization algorithm, as provided in this application;

[0051] Figure 3 An example graph of a feature correlation coefficient provided for this application;

[0052] Figure 4 An example diagram of prediction result comparison provided in this application;

[0053] Figure 5 A schematic diagram of the structure of a pressure drop prediction device for a filter provided in this application;

[0054] Figure 6 This is a schematic diagram of the structure of an electronic device provided in this application. Detailed Implementation

[0055] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] In this application, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0057] When evaluating the performance of gas-liquid coalescers, pressure drop is a key indicator of efficiency and energy consumption, and its accurate prediction is crucial for equipment selection, design optimization, and operation management. However, existing technologies have several limitations in pressure drop prediction. Traditional experimental methods, while intuitive and reliable, are costly, time-consuming, and unable to cover all possible operating scenarios, limiting their widespread application in practice. On the other hand, while mathematical calculation methods based on physical models can theoretically simulate complex physical processes, the simplification of model assumptions, uncertainties in parameter selection, and the complex variations in physical phenomena under different operating conditions often lead to significant deviations in the calculation results, making it difficult to meet the accuracy requirements of different industrial applications.

[0058] Based on this, this application provides a method for predicting the pressure drop of a filter. This method can be applied to electronic devices, such as computers, tablets, smartphones, and smart wearable devices. The flowchart of the method is shown below. Figure 1 As shown, it specifically includes:

[0059] S101: Obtain the feature data of the filter to be predicted; the feature data includes the material characteristics of the filter and the operating environment characteristics.

[0060] Optionally, material characteristics may include one or more of the following: filter media thickness, packing density, filter media basis weight, and average fiber diameter; operating environment characteristics may include one or more of the following: aerosol concentration, surface tension, and liquid viscosity.

[0061] In this embodiment, obtaining the feature data of the filter to be predicted includes: obtaining the original sample data of the filter to be predicted; extracting features from the original sample data of the filter to obtain initial feature data; and normalizing the initial feature data to obtain the feature data of the filter.

[0062] Optionally, the filter can be a coalescing filter, which can be used to filter natural gas.

[0063] S102: Input the feature data into the pressure drop prediction model to obtain the pressure drop prediction result of the filter; the pressure drop prediction model is obtained by training a machine learning model using a preset training sample set, and the model parameters of the machine learning model are calculated by a target optimization algorithm, which includes the Northern Eagle optimization algorithm based on a subtraction optimizer and a Cauchy distribution function.

[0064] In this embodiment, the machine learning model can be a Gaussian process regression (GPR) model.

[0065] The method provided in the embodiments of this application can quickly and accurately determine the pressure drop of the filter.

[0066] Optionally, the machine learning model can be pre-built. The process of building a machine learning model is as follows:

[0067] First, given the input parameter vector X=[x1, x2, ..., x] for predicting gas-liquid coalescence filtration pressure drop, i , ..., x n ] T The corresponding output vector is Y=[y1, y2, ..., y]. i , ..., y n ] T The training sample set consists of The predicted output variable f(x) is a stochastic process, represented as f(x) = {f(x1), f(x2), ..., f(x3)}. i ), ..., f(x) n In reality, measured data is inevitably affected by noise. Therefore, it is necessary to consider adding a certain amount of noise to the model. In this embodiment, the operational relationship between the input vector X and the target output Y in the GPR model is as follows:

[0068]

[0069] in, Let f(x) be independent Gaussian noise, independent of f(x), with a mean of 0 and a variance of 1. The normal distribution is denoted as . . The noise standard deviation is used to quantify the noise in the observed data. The larger the noise standard deviation, the more noise there is in the data.

[0070] Gaussian process regression consists of a mean function (basis function) m(x) and a covariance function (kernel function) k(x). i ,x j The input vector X is uniquely determined. Given an input vector X and a random variable f(x) that follows a Gaussian distribution, f(x) can be expressed as:

[0071]

[0072] Next, the kernel function of GPR is defined. In this embodiment, a mother kernel (ARD matern32) with a parameter of 3 / 2 and a separate length scale is selected as the kernel function of GPR, and its specific form is as follows:

[0073]

[0074]

[0075] in, A separate length scale, l=1,2,…,d, is used for each predictor to control the smoothness of the function. The larger the length scale, the smoother the model, and vice versa. The signal variance determines the fluctuation range of the predicted values ​​in the Gaussian process model. The larger the signal variance, the wider the range of data variation allowed by the model.

[0076] Since the random variable f(x) follows a Gaussian distribution, the output vector Y also follows a Gaussian distribution. According to Bayesian theory, in the pre-defined test sample set... If a prior function is established, then the output vector Y of the test sample set... * The joint Gaussian prior distribution satisfied by the output vector Y of the training sample set can be expressed as:

[0077]

[0078] In the formula, I n Let K(X, X) be the identity matrix. * )=K (X * , X) T Input vector X for the test sample set * The covariance matrix between Y and the input vector X in the training sample set. Therefore, according to the Bayesian posterior formula, Y is calculated. * The posterior condition distribution that is satisfied is:

[0079] After determining the expression for the test sample set used for prediction based on the posterior distribution, the hyperparameter combination for Gaussian process regression is defined as follows: The selection of its numerical value has a significant impact on the model's performance. In this embodiment, the Northern Eagle optimization algorithm, improved based on a subtraction optimizer and Cauchy distribution function, is used to optimize the model. It is used as an optimization target for optimization.

[0080] In one embodiment provided in this application, based on the above implementation process, optionally, the process of calculating the model parameters of the machine learning model through a target optimization algorithm is as follows: Figure 2As shown, it includes:

[0081] S201: Generate the initial population; the initial population includes a preset number of goshawk individuals, each goshawk individual representing a candidate model parameter value scheme for the machine learning model.

[0082] In this embodiment, a set of solutions is first randomly generated as the initial population; these solutions represent the locations of the eagles. That is:

[0083]

[0084] Where U represents the current population matrix of Northern Goshawks; U i (i=1, 2, …, N) represents the initial position of the i-th Northern Goshawk; u i,j Let represent the j-th position of the i-th Northern Goshawk; N represents the number of Northern Goshawks in the population; m represents the dimension of the problem to be solved, i.e., the number of hyperparameters to be solved, with a value of 3.

[0085] In this embodiment, the objective function of the Northern Goshawk population can be represented by an objective function vector, as follows:

[0086]

[0087] Where F is the objective function value vector; F i (i=1, 2, …, N) represents the objective function value of the i-th Northern Goshawk.

[0088] S202: Determine the initial population as the target population and calculate the fitness value of each individual goshawk in the target population.

[0089] In this embodiment, the performance of the constructed model under each set of parameters can be evaluated through the objective function. This application selects k-fold cross-validation to determine the objective function of the model. K-fold cross-validation can be used to estimate the model's performance and generalization ability. Specifically, during training, the training sample set is divided into k equal subsets, where k-1 subsets are used to train the model, and the remaining subset is used to test the model. This process is repeated k times, each time selecting a different subset as the test set, and the rest as the training set. Finally, the average of the k performance evaluation results is taken to obtain the final evaluation result. The fitness value obtained from each test set is the root mean square error (RMSE), and after traversing k times, the average RMSE is used as the final fitness value. The formula is as follows:

[0090]

[0091] Among them, RMSE k denoted as root mean square error, and M is the number of subsets of the training samples used for testing; y is the predicted value of the i-th sample in the j-th cross-validation; ij Output the actual value for the i-th sample in the j-th cross-validation; k is the number of cross-validations.

[0092] In one embodiment provided in this application, based on the above implementation process, optionally, the step of calculating the fitness value of each individual goshawk in the target population includes: calculating the fitness value of each individual goshawk in the target population according to the k-fold cross-validation algorithm and the training sample set.

[0093] S203: Determine the target fitness value of the prey location for each individual goshawk in the current target population.

[0094] In this embodiment, the target fitness value of the prey location corresponding to each individual goshawk in the current target population can be calculated using an objective function.

[0095] S204: For each individual goshawk, if the fitness value of the individual goshawk is less than the target fitness value of the corresponding prey position, a random number is generated, and the first updated position of the individual goshawk is determined based on the best fitness value of the best individual goshawk in the target population and the currently generated random number; if the fitness value of the individual goshawk is not less than the target fitness value of the corresponding prey position, the first updated position of the individual goshawk is determined by a preset subtraction optimizer; the best individual goshawk in the target population is the individual goshawk with the best fitness value in the target population.

[0096] In this embodiment, the behavior of an eagle circling in the air searching for prey is simulated, and new solutions are explored through a global search of the solution space. The mathematical expression for this stage is as follows:

[0097]

[0098]

[0099] In the formula, P1 represents the first hunting stage; Let P be the new position of the i-th eagle in the j-th dimension during the first hunting phase, which is also the first updated position; i is the location of the prey corresponding to the i-th eagle; w is a random integer in the range [1, N]; r is a random number between [0, 1]; and I is either 1 or 2, which are used to generate random numbers for random NGO behavior during search and update. F represents the fitness value of the location of the prey corresponding to the i-th eagle; i Let u be the fitness value of the i-th eagle. best U represents the optimal value for an individual goshawk found after each iteration. wLet v be the position vector of the w-th Northern Goshawk; v is a vector of dimension m, which is a random number generated from [1, 2].

[0100] In this embodiment, corresponding optimization strategies are introduced to improve the existing Northern Eagle algorithm's NGO first stage, addressing its shortcomings. When <F i While the NGO algorithm can promote rapid population concentration in a specific region and accelerate the convergence process, it is also prone to blindness due to a lack of directionality, sometimes leading to inefficiency or getting trapped in non-optimal solution regions. To mitigate this, this improvement employs optimal value-guided position updates. The improved algorithm no longer relies on randomly selected individuals but uses the best solution found so far to guide the population's position updates. This optimal value-based guidance strategy enhances the algorithm's targeting and accuracy during the search process, effectively reduces the risk of getting trapped in local optima, and helps to more stably and effectively approach the global optimum.

[0101] In this embodiment, when ≥F i In this case, a subtraction optimizer (SABO) is used for position updates. The core of the subtraction optimizer is to guide the search process by using the average subtraction operation among multiple intelligent agents. It is based on a special operation "-v", which is the subtraction of search agent A by the v-subtraction of search agent B.

[0102]

[0103] By calculating each search agent A j The arithmetic mean of v-subtraction is used to determine any search agent A in the search space. i This strategy significantly reduces the blindness and randomness of the original NGO algorithm during the search process by adjusting the displacement. The formula is as follows:

[0104]

[0105] The location of the search proxy will be updated as follows:

[0106]

[0107] in, Let be the fitness value of the i-th search agent; Let be the new fitness value for the i-th search agent.

[0108] Introducing this method into the original NGO algorithm, the displacement of any given eagle in the search space is calculated by arithmetically averaging the positional differences among multiple individual eagles after subtraction, thus enabling fine-tuning of the search direction. It considers the distribution of the entire population, performing directed updates to individuals globally. This not only enhances information exchange between individuals but also increases the likelihood of finding better solutions. This global update mechanism trains each individual throughout the population search process, helping to prevent the algorithm from prematurely getting trapped in local optima and increasing the chance of finding the global optimum. Furthermore, this method improves the algorithm's convergence speed and stability, making it more efficient and reliable in solving complex optimization problems.

[0109] S205: Based on the fitness value of each individual goshawk, the first update position, and the fitness value of the first update position, perform the first phase of position update for the target population.

[0110] In this embodiment, the first-stage location update of the target population is performed as follows:

[0111]

[0112] in, Let be the fitness value of the i-th eagle at the first update position.

[0113] S206: For each goshawk individual in the target population after the first-stage position update, generate a mutation random number for the goshawk individual; if the mutation random number is less than a preset Cauchy mutation threshold, determine the second updated position of the goshawk individual based on the best goshawk individual in the target population after the first-stage position update and a preset Cauchy distribution function; if the mutation random number is not less than the preset Cauchy mutation threshold, determine the second updated position of the goshawk individual based on the hunting radius, upper and lower limits of parameters, and the random number; the hunting radius is determined based on the current iteration number and a preset iteration number threshold.

[0114] In this embodiment, the Northern Eagle algorithm can be improved using the Cauchy distribution function. Specifically, a certain probability factor can be added for Cauchy mutation, and the Cauchy probability density is:

[0115]

[0116] Where z is the location parameter defining the peak position of the distribution; γ is the scale parameter representing half the width at half the maximum value. When z0=0 and γ=1, it is a standard Cauchy distribution, and the Cauchy probability density function is expressed as:

[0117]

[0118] Cauchy mutation possesses superior perturbation properties. Applying this algorithm to optimization algorithms involving target position updates can improve their randomness and diversity, thus enhancing the population's exploration capabilities. Furthermore, Cauchy mutation may discover high-quality solution regions far from the current position during the search process. This strategy makes the NGO algorithm more flexible and adaptable, improving its performance in solving complex optimization problems. Substituting the Cauchy probability density function into the Northern Eagle optimization algorithm yields the following formula for the optimal value of each generation after Cauchy mutation:

[0119]

[0120] in, This is the second update position, where Cauchy(0,1) is a random number generated by Cauchy mutation.

[0121] Subsequently, upper and lower limits are introduced and gradually reduced to achieve dynamic updates of the position. This dynamic adjustment strategy effectively manages the search range of the Northern Eagle optimization algorithm, causing the search radius R to gradually decrease and the search space to gradually shrink as the number of iterations t increases, thus focusing more precisely on the region where the potential optimal solution is located. By gradually reducing the search range, the algorithm can converge to a better solution more quickly, while reducing invalid searches in vast regions that may not contain the optimal solution. This improves the algorithm's local search capability and enhances the accuracy and quality of the solution. Through this dynamic update mechanism, the NGO algorithm can maintain its global exploration capability while strengthening its local development capability, balancing the relationship between exploration and development, and improving the overall optimization efficiency. Therefore, the improved formula for the second stage is as follows:

[0122]

[0123]

[0124] Where R is the hunting radius; t is the current iteration number; T is the iteration threshold; and P2 is the second hunting phase. Let be the new position of the i-th eagle in the j-th dimension during the second hunting phase, i.e., the second updated position; σ is the mutation probability, which can be set to 0.3 in this embodiment; u b For the upper limit value, l b σ is the lower bound value. rand is a random number between [0, 1]. When rand < σ, Cauchy mutation is used; when rand ≥ σ, the NGO algorithm gradually reduces the search range with the number of iterations, accelerating the convergence speed of the algorithm. The improved NGO algorithm is named SCVNGO.

[0125] S207: Based on the fitness value of each individual goshawk in the target population after the first-stage position update, the second updated position, and the fitness value of the second updated position, perform a second-stage position update on the target population after the first-stage position update to obtain a new target population.

[0126] In this embodiment, a second-stage location update can be performed on the target population after the first-stage location update, as follows:

[0127]

[0128] in, Let be the fitness of the i-th eagle at the second update position.

[0129] S208: Determine whether the current iteration count is equal to the iteration count threshold.

[0130] S209: Increment the current iteration number by one and return to the step of determining the target fitness value of the prey location for each individual goshawk in the current target population.

[0131] S210: If the current iteration number is equal to the iteration number threshold, the best individual goshawk in the new target population is used as the model parameter of the machine learning model.

[0132] In one embodiment provided in this application, based on the above implementation process, optionally, the process of training a machine learning model based on a preset training sample set to obtain a pressure drop prediction model includes:

[0133] Obtain a training sample set; the training sample set includes multiple training samples; the training samples include filter feature data and voltage drop information;

[0134] The machine learning model is trained using training samples from the training sample set to obtain a pressure drop prediction model.

[0135] In some embodiments, a training sample set can be obtained, and target sample data for training can be selected from each training sample in the training sample set; the target sample data can be input into a machine learning model to obtain the output result of the machine learning model; the objective function value corresponding to the target sample data can be obtained based on the output result and the voltage drop information contained in the target sample data; the network parameters of the machine learning model can be updated based on the objective function value; if the machine learning model after updating the network parameters meets the preset training conditions, the machine learning model is determined as a fault identification model; if the machine learning model after updating the network parameters does not meet the preset training conditions, the process returns to the step of selecting target sample data in the training sample set.

[0136] Optionally, the training conditions can be: the objective function converges, the number of training iterations reaches a preset training iteration threshold, or the prediction accuracy of the machine learning model is greater than a preset accuracy threshold.

[0137] Optionally, the prediction accuracy of the machine learning model can be determined based on test samples in a pre-defined test sample set.

[0138] In one embodiment provided in this application, based on the above implementation process, optionally, obtaining the training sample set includes:

[0139] Acquire each raw sample data, where each raw sample data includes feature data of multiple candidate feature types and voltage drop information;

[0140] Based on the preset Spearman correlation analysis algorithm and the original sample data, the correlation coefficient of each candidate feature type is determined; the correlation coefficient of each candidate feature type characterizes the degree of correlation between the feature data of each candidate feature type and the voltage drop information.

[0141] Based on the correlation coefficient of the feature data of each candidate feature type, feature data of the target feature type are selected from each candidate feature type of each original sample data to obtain the initial sample data corresponding to each original sample data; the initial sample data includes the feature data of the target feature type and voltage drop information.

[0142] The initial sample data are normalized to obtain the final sample data corresponding to each initial sample data.

[0143] The final sample data are divided according to a preset ratio to obtain a training sample set and a test sample set.

[0144] Before using the raw dataset to build a model, data preprocessing is necessary to improve data quality and usability. In gas-liquid coalescence filtration experiments, the steady-state pressure drop is affected by the filter medium parameters, droplet properties, and operating conditions. However, different features have varying degrees of influence; changes in some features do not cause significant fluctuations in others. Furthermore, too many features can increase computation time and affect the efficiency of model training and prediction. Therefore, it is crucial to perform correlation analysis on the collected data and remove features with low correlation.

[0145] In practical engineering applications of coalescing filters for natural gas purification, it is necessary to measure and analyze the gas-liquid separation performance of the coalescing filters to ensure their high efficiency and reliability in practical applications. Therefore, in the material testing, fiber filter media was used as the coalescing filter. Sixty-one sets of gas-liquid coalescing filtration experiments were conducted under conditions of 25℃, different gas velocities, and different atomization concentrations. Seventeen different fiber filter media were selected, varying the filter media thickness, packing density, contact angle, basis weight, and fiber diameter, and different liquids were used for atomization to obtain the filtration pressure drop under nine parameter variations. The actual pressure drop data collected during the experiments were combined with the corresponding varying parameters to form the original dataset. After correlation analysis and data normalization, this dataset served as the analysis sample for the prediction model. The varying parameters—filtration gas velocity (feature 1), aerosol concentration (feature 2), filter media thickness (feature 3), packing density (feature 4), contact angle (feature 5), liquid viscosity (feature 6), surface tension (feature 7), filter media basis weight (feature 8), and average fiber diameter (feature 9)—were used as inputs, and the pressure drop was used as the model output.

[0146] The Spearman correlation coefficient ρ ranges from -1 to 1. It is used to assess the dependency between two variables. A Spearman correlation coefficient ρ > 0 indicates a positive correlation; a Spearman correlation coefficient ρ < 0 indicates a negative correlation; and a Spearman correlation coefficient ρ = 0 indicates that Y shows no trend when X increases or decreases. For gas-liquid coalescing filtration, the closer the absolute value of the correlation coefficient between a variable and the output pressure drop is to 1, the stronger the correlation between that variable and the output pressure drop. Based on Spearman correlation analysis, the correlation coefficient between each variable and the output pressure drop is calculated individually to explore the intrinsic relationship between the data variables. This helps identify which variables are relatively independent and select variables with a strong correlation to the output pressure drop, forming a feature set for gas-liquid coalescing filtration. This step removes redundant features, thereby reducing the complexity of subsequent processing.

[0147] Sort each variable and the steady-state pressure drop in ascending (or descending) order, respectively, and then assign them corresponding ranks. If there are identical values ​​among the variable data, take the average of the ranks as the rank of each data point. Further calculate the rank difference for each pair of data points in the two variables. Finally, substitute the rank difference into the formula to obtain:

[0148]

[0149] Where ρ represents the correlation coefficient between the input characteristic variable of the gas-liquid coalescing filter to be predicted and the output pressure drop of the gas-liquid coalescing filter; x i y represents the rank of the i-th gas-liquid coalescence filtration feature data to be predicted; idenoted by rank, i represents the rank of the output pressure drop of the i-th gas-liquid coalescing filter; n represents the number of gas-liquid coalescing filter data to be predicted.

[0150] See Figure 3 Table 1 provides an example of the correlation coefficients between features in this embodiment, showing the correlation coefficients between voltage drop and various features. Based on the correlation coefficients between voltage drop and each feature, the target feature type can be determined. As shown in Table 1, feature 2 is negatively correlated with steady-state voltage drop, while other features are positively correlated. The correlation coefficients are sorted by size. Features 3 and 8 have absolute correlation coefficients above 0.5, showing a significant correlation with voltage drop; features 2, 4, 6, 7, and 9 have absolute correlation coefficients between 0.1 and 0.5, showing a moderate correlation; and features 1 and 5 have absolute correlation coefficients close to 0, showing a weak correlation. Therefore, features 3 and 8 play a dominant role in enhancing the model's predictive performance. Features 1 and 5, with the smallest correlation coefficients, are removed, and the remaining seven features are used as model input variables to build the model.

[0151] Table 1

[0152] variable Feature 1 Feature 2 Feature 3 Feature 4 Feature 5 Feature 6 Feature 7 Feature 8 Feature 9 Correlation coefficient 0.057 -0.126 0.564 0.411 0.065 0.243 0.352 0.601 0.161

[0153] In this embodiment, the dataset after feature parameter selection can be standardized. Optionally, a normalization tool like mapminmax can be used to normalize each row of data to a restricted interval [ymin, ymax], ensuring they are of the same magnitude. This eliminates the impact of dimensional differences on model predictions and facilitates subsequent processing and analysis. The standardization formula is as follows:

[0154] (2)

[0155] Where x represents the sample data before standardization, and y represents the sample data after standardization (final sample data); x min and x max These are the minimum and maximum values ​​in the sample data, respectively; y max and y min These are the upper and lower bounds of the standardized range, respectively. By setting the transmission parameters, the restricted range is changed to [0, 1].

[0156] Finally, the normalized sample set is divided into a training set. One set of data is used for training and building the model, and the remaining data is used as the test set. This is used to evaluate the generalization ability of the final model.

[0157] The preprocessed gas-liquid coalescence filtration steady-state pressure drop samples from step one are divided into two parts: 80% is used as the training sample D for the model, and the remaining 20% ​​is used as the test sample D to test the model's generalization ability.* Based on the settings of input and output parameters, a model with a structure of 7 inputs and 1 output is established.

[0158] In this embodiment, after obtaining the optimal hyperparameters using the improved Northern Eagle algorithm, the obtained optimal hyperparameters are used to construct the optimal GPR model, resulting in the final voltage drop prediction model. The network is then retrained and tested using the pre-defined test samples D. * The simulation prediction results are as follows: Figure 4 As shown, the error range is between -1.94 and 0.03, the actual pressure drop value is close to the model prediction value, and the error size meets the requirements.

[0159] Optionally, the prediction accuracy of the model is ultimately analyzed based on the evaluation index results, including the mean absolute error (MAE), mean absolute error percentage (MAPE), mean squared error (MSE), root mean square error (RMSE), and coefficient of determination (R²). 2 The calculation formula for ) is as follows:

[0160]

[0161]

[0162]

[0163]

[0164]

[0165] The evaluation metrics of the SCVNGO-GPR prediction model (pressure drop prediction model) and the NGO-GPR model (original model) established in this embodiment of the invention are shown in Table 2. The table shows that the MAPE of the newly improved model SCVNGO-GPR is within 5%, meeting engineering requirements, and all evaluation metrics are superior to the original NGO-GPR model. This indicates that the improved SCVNGO can effectively enhance the optimization performance of the NGO algorithm, and the SCVNGO-GPR model established based on it has good prediction accuracy and can effectively predict the pressure drop of the filter.

[0166] Table 2

[0167] Model MAE MAPE MSE RMSE <![CDATA[R 2 ]]> NGO-GPR (original model) 0.406 6.02% 0.594 0.771 0.9621 SCNGO-GPR (New Improved Model) 0.302 4.13% 0.368 0.607 0.9766

[0168] This invention proposes a pressure drop prediction model specifically targeting the filtration performance of gas-liquid coalescing filters used in key stages of the oil and gas industry, namely natural gas purification and fuel gas processing. The model is constructed using a Gaussian process regression algorithm, particularly suitable for small sample datasets. To improve prediction accuracy and prevent overfitting, this patent further proposes a Northern Eagle Optimization Algorithm (SCVNGO) that integrates a subtraction optimizer and Cauchy mutation. By introducing an optimal value guidance strategy and a subtraction optimizer algorithm, the first stage (exploration stage) of the original Northern Eagle algorithm is improved. Furthermore, Cauchy mutation is performed by adding a probability factor, and upper and lower limits are introduced for dynamic position updates to improve the second stage (development stage). This algorithm optimizes the hyperparameters of the GPR model, establishing the SCVNGO-GPR model. This enhanced prediction model enables rapid and accurate prediction of filtration pressure drop.

[0169] In this embodiment, considering that process data variables in natural gas pipeline transportation and fuel gas processing do not always approximately conform to a normal distribution, Spearman correlation analysis was employed. This method is used to reasonably analyze the correlation between the performance of the gas-liquid coalescing filter and various influencing parameters, and effectively eliminate parameters that are weakly correlated with filtration performance. This ensures the accuracy of the analysis while simplifying the model and reducing computational complexity.

[0170] In this embodiment, the objective function of the Northern Eagle optimization algorithm has been improved by adopting k-fold cross-validation. Specifically, k-fold cross-validation is used to iteratively calculate the repartitioned training set, and the average root mean square error of the k different subsets is used as the fitness evaluation index. This improvement enables the model to learn data features more comprehensively, thereby reducing evaluation bias caused by different data partitioning methods. In this way, performance evaluation becomes more comprehensive, and the algorithm optimizes hyperparameters more accurately.

[0171] and Figure 1 Corresponding to the method described above, this application also provides a filter pressure drop prediction device for predicting the pressure drop of a filter. Figure 1 The specific implementation of the method is shown in the schematic diagram of the device. Figure 5 As shown, it includes:

[0172] The acquisition unit 501 is used to acquire feature data of the filter to be predicted; the feature data includes the material characteristics and operating environment characteristics of the filter;

[0173] The prediction unit 502 is used to input the feature data into the pressure drop prediction model to obtain the pressure drop prediction result of the filter; the pressure drop prediction model is obtained by training a machine learning model using a preset training sample set, and the model parameters of the machine learning model are calculated by an objective optimization algorithm, which includes the Northern Eagle optimization algorithm based on a subtraction optimizer and Cauchy distribution function.

[0174] In one embodiment provided in this application, based on the above-described scheme, optionally, the prediction unit 502 includes:

[0175] A subunit is generated to generate an initial population; the initial population includes a preset number of goshawk individuals, and each goshawk individual represents a candidate model parameter value scheme for a machine learning model;

[0176] The first determining subunit is used to determine the initial population as the target population and calculate the fitness value of each individual goshawk in the target population;

[0177] The second determining subunit is used to determine the target fitness value of the prey location corresponding to each individual goshawk in the current target population.

[0178] The first execution subunit is configured to, for each individual goshawk, generate a random number if the fitness value of the individual goshawk is less than the target fitness value of the corresponding prey position, and determine the first updated position of the individual goshawk based on the best individual goshawk in the target population and the currently generated random number; and determine the first updated position of the individual goshawk through a preset subtraction optimizer if the fitness value of the individual goshawk is not less than the target fitness value of the corresponding prey position.

[0179] The first update subunit is used to perform a first-stage position update on the target population based on the fitness value of each individual goshawk, the first update position, and the fitness value of the first update position.

[0180] The second execution subunit is used to generate a mutation random number for each goshawk individual in the target population after the first-stage position update; if the mutation random number is less than a preset Cauchy mutation threshold, determine the second updated position of the goshawk individual based on the best goshawk individual in the target population after the first-stage position update and a preset Cauchy distribution function; if the mutation random number is not less than the preset Cauchy mutation threshold, determine the second updated position of the goshawk individual based on the hunting radius, upper and lower limits of parameters, and the random number; the hunting radius is determined based on the current iteration number and a preset iteration number threshold.

[0181] The second update subunit is used to perform a second-stage position update on the target population after the first-stage position update based on the fitness value of each individual goshawk in the target population after the first-stage position update, the second update position, and the fitness value of the second update position, so as to obtain a new target population.

[0182] The third execution subunit is used to increment the current iteration number by one and return the target fitness value that triggers the second determination subunit to determine the prey location corresponding to each individual goshawk in the current target population when the current iteration number is not equal to the iteration number threshold.

[0183] The fourth execution subunit is used to take the best individual goshawk in the new target population as the model parameter of the machine learning model when the current iteration number is equal to the iteration number threshold.

[0184] In one embodiment provided in this application, based on the above-described solution, optionally, the first determining subunit includes:

[0185] The calculation module is used to calculate the fitness value of each individual goshawk in the target population based on the k-fold cross-validation algorithm and the training sample set.

[0186] In one embodiment provided in this application, based on the above-described solution, optionally, the prediction unit 702 includes:

[0187] A sub-unit is used to acquire a training sample set; the training sample set includes multiple training samples; the training samples include filter feature data and voltage drop information.

[0188] The training subunit is used to train the machine learning model based on the training samples in the training sample set to obtain the voltage drop prediction model.

[0189] In one embodiment provided in this application, based on the above-described solution, optionally, the acquisition subunit includes:

[0190] The acquisition module is used to acquire each raw sample data, and each raw sample data includes feature data of multiple candidate feature types and voltage drop information;

[0191] The determination module is used to determine the correlation coefficient of each candidate feature type based on a preset Spearman correlation analysis algorithm and the original sample data; the correlation coefficient of each candidate feature type characterizes the degree of correlation between the feature data of each candidate feature type and the voltage drop information;

[0192] The selection module is used to select feature data of the target feature type from each candidate feature type in each original sample data according to the correlation coefficient of the feature data of each candidate feature type, so as to obtain the initial sample data corresponding to each original sample data; the initial sample data includes feature data of the target feature type and voltage drop information.

[0193] The normalization module is used to normalize each of the initial sample data to obtain the final sample data corresponding to each of the initial sample data.

[0194] The execution module is used to create a training sample set composed of the final sample data.

[0195] The specific principles and execution processes of each unit and module in the filter pressure drop prediction device disclosed in the above embodiments of this application are the same as those of the filter pressure drop prediction method disclosed in the above embodiments of this application. Please refer to the corresponding parts of the filter pressure drop prediction method provided in the above embodiments of this application, and they will not be repeated here.

[0196] This application embodiment also provides a storage medium, the storage medium including stored instructions, wherein, when the instructions are executed, the device where the storage medium is located is controlled to execute the above-described filter pressure drop prediction method.

[0197] This application also provides an electronic device, the structural schematic diagram of which is shown below. Figure 6 As shown, it specifically includes a memory 601 and one or more instructions 602, wherein one or more instructions 602 are stored in the memory 601 and configured to be executed by one or more processors 603 to perform the following operations:

[0198] Obtain the feature data of the filter to be predicted; the feature data includes the material characteristics and operating environment characteristics of the filter;

[0199] The feature data is input into the pressure drop prediction model to obtain the pressure drop prediction result of the filter. The pressure drop prediction model is obtained by training a machine learning model using a preset training sample set. The model parameters of the machine learning model are calculated by an objective optimization algorithm, which includes the Northern Eagle optimization algorithm based on a subtraction optimizer and a Cauchy distribution function.

[0200] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0201] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0202] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0203] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0204] The pressure drop prediction method for a filter provided in this application has been described in detail above. Specific examples have been used to illustrate the principle and implementation of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​this application. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the pressure drop of a filter, characterized in that, include: Obtain the feature data of the filter to be predicted; the feature data includes the material characteristics and operating environment characteristics of the filter; The feature data is input into the pressure drop prediction model to obtain the pressure drop prediction result of the filter. The pressure drop prediction model is obtained by training a machine learning model using a preset training sample set. The model parameters of the machine learning model are calculated by an objective optimization algorithm, which includes the Northern Eagle optimization algorithm based on a subtraction optimizer and a Cauchy distribution function.

2. The method according to claim 1, characterized in that, The process of calculating the model parameters of the machine learning model using a target optimization algorithm includes: Generate an initial population; the initial population includes a preset number of goshawk individuals, each goshawk individual representing a candidate model parameter value scheme for a machine learning model; The initial population is identified as the target population, and the fitness value of each individual goshawk in the target population is calculated. Determine the target fitness value of the prey location for each individual goshawk in the current target population; For each individual goshawk, if its fitness value is less than the target fitness value of the corresponding prey location, a random number is generated. Based on the best goshawk in the target population and the currently generated random number, the first updated position of the goshawk is determined. If the fitness value of the goshawk is not less than the target fitness value of the corresponding prey location, the first updated position of the goshawk is determined by a preset subtraction optimizer. The best goshawk in the target population is the goshawk with the best fitness value in the target population. Based on the fitness value of each individual goshawk, the first update position, and the fitness value of the first update position, the target population is updated in the first stage. For each goshawk individual in the target population after the first-stage position update, a mutation random number is generated for that goshawk individual. If the mutation random number is less than a preset Cauchy mutation threshold, the second updated position of the goshawk individual is determined based on the best goshawk individual in the target population after the first-stage position update and a preset Cauchy distribution function. If the mutation random number is not less than the preset Cauchy mutation threshold, the second updated position of the goshawk individual is determined based on the hunting radius, upper and lower limits of parameters, and the random number. The hunting radius is determined based on the current iteration number and a preset iteration number threshold. Based on the fitness value of each individual goshawk in the target population after the first-stage position update, the second updated position, and the fitness value of the second updated position, the target population after the first-stage position update is updated in the second stage to obtain a new target population. If the current iteration number is not equal to the iteration number threshold, increment the current iteration number by one and return to the step of determining the target fitness value of the prey location corresponding to each individual goshawk in the current target population; If the current iteration count is equal to the iteration count threshold, the best individual goshawk in the new target population is used as the model parameter of the machine learning model.

3. The method according to claim 2, characterized in that, The calculation of the fitness value of each individual goshawk in the target population includes: The fitness value of each individual goshawk in the target population is calculated based on the k-fold cross-validation algorithm and the training sample set.

4. The method according to claim 1, characterized in that, The process of training a machine learning model based on a pre-defined training sample set to obtain a pressure drop prediction model includes: Obtain a training sample set; the training sample set includes multiple training samples; the training samples include filter feature data and voltage drop information; The machine learning model is trained using training samples from the training sample set to obtain a pressure drop prediction model.

5. The method according to claim 4, characterized in that, The acquisition of the training sample set includes: Acquire each raw sample data, where each raw sample data includes feature data of multiple candidate feature types and voltage drop information; Based on the preset Spearman correlation analysis algorithm and the original sample data, the correlation coefficient of each candidate feature type is determined; the correlation coefficient of each candidate feature type characterizes the degree of correlation between the feature data of each candidate feature type and the voltage drop information. Based on the correlation coefficient of the feature data of each candidate feature type, feature data of the target feature type are selected from each candidate feature type of each original sample data to obtain the initial sample data corresponding to each original sample data; the initial sample data includes the feature data of the target feature type and voltage drop information. The initial sample data are normalized to obtain the final sample data corresponding to each initial sample data. The final sample data are divided according to a preset ratio to obtain a training sample set and a test sample set.

6. A pressure drop prediction device for a filter, characterized in that, include: An acquisition unit is used to acquire feature data of the filter to be predicted; the feature data includes the material characteristics and operating environment characteristics of the filter; The prediction unit is used to input the feature data into the pressure drop prediction model to obtain the pressure drop prediction result of the filter. The pressure drop prediction model is obtained by training a machine learning model using a preset training sample set. The model parameters of the machine learning model are calculated by an objective optimization algorithm, which includes the Northern Eagle optimization algorithm based on a subtraction optimizer and Cauchy distribution function.

7. The apparatus according to claim 6, characterized in that, The prediction unit includes: A subunit is generated to generate an initial population; the initial population includes a preset number of goshawk individuals, and each goshawk individual represents a candidate model parameter value scheme for a machine learning model; The first determining subunit is used to determine the initial population as the target population and calculate the fitness value of each individual goshawk in the target population; The second determining subunit is used to determine the target fitness value of the prey location corresponding to each individual goshawk in the current target population. The first execution subunit is configured to, for each individual goshawk, generate a random number if the fitness value of the individual goshawk is less than the target fitness value of the corresponding prey location, and determine the first updated position of the individual goshawk based on the best individual goshawk in the target population and the currently generated random number; and determine the first updated position of the individual goshawk through a preset subtraction optimizer if the fitness value of the individual goshawk is not less than the target fitness value of the corresponding prey location; the best individual goshawk in the target population is the individual goshawk with the best fitness value in the target population. The first update subunit is used to perform a first-stage position update on the target population based on the fitness value of each individual goshawk, the first update position, and the fitness value of the first update position. The second execution subunit is used to generate a mutation random number for each goshawk individual in the target population after the first-stage position update; if the mutation random number is less than a preset Cauchy mutation threshold, determine the second updated position of the goshawk individual based on the best goshawk individual in the target population after the first-stage position update and a preset Cauchy distribution function; if the mutation random number is not less than the preset Cauchy mutation threshold, determine the second updated position of the goshawk individual based on the hunting radius, upper and lower limits of parameters, and the random number; the hunting radius is determined based on the current iteration number and a preset iteration number threshold. The second update subunit is used to perform a second-stage position update on the target population after the first-stage position update based on the fitness value of each individual goshawk in the target population after the first-stage position update, the second update position, and the fitness value of the second update position, so as to obtain a new target population. The third execution subunit is used to increment the current iteration number by one and return the target fitness value that triggers the second determination subunit to determine the prey location corresponding to each individual goshawk in the current target population if the current iteration number is not equal to the iteration number threshold. The fourth execution subunit is used to take the best individual goshawk in the new target population as the model parameter of the machine learning model when the current iteration number is equal to the iteration number threshold.

8. The apparatus according to claim 7, characterized in that, The first determining subunit includes: The calculation module is used to calculate the fitness value of each individual goshawk in the target population based on the k-fold cross-validation algorithm and the training sample set.

9. A storage medium, characterized in that, The storage medium includes stored instructions, wherein, when the instructions are executed, the device containing the storage medium is controlled to perform the pressure drop prediction method for the filter as described in any one of claims 1 to 5.

10. An electronic device, characterized in that, It includes a memory, and one or more instructions, wherein one or more instructions are stored in the memory and configured to be executed by one or more processors of the voltage drop prediction method of the filter as described in any one of claims 1 to 5.

Images