Ink-jet printing liquid drop physical quality prediction method based on BP neural network optimized by improved dung beetle optimization algorithm
By improving the dung beetle optimization algorithm to optimize the BP neural network and constructing the IDBO-BP model, the nonlinear problem of predicting the physical quality of inkjet printing droplets was solved, high-precision and stable droplet control was achieved, and the quality and efficiency of inkjet printing were improved.
Patent Information
- Application Number
- CN202510665682.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-22
- Publication Date
- 2025-09-19
AI Technical Summary
The physical quality of droplets in the inkjet printing process is affected by many factors. Traditional mathematical models are difficult to accurately describe their nonlinear relationships, resulting in insufficient droplet control and prediction accuracy. Existing research mainly focuses on droplet volume and velocity, and lacks effective droplet physical quality prediction methods.
The improved dung beetle optimization algorithm is used to optimize the BP neural network. The weights and thresholds of the BP neural network are optimized through dynamic reverse learning strategy and golden sine factor. The IDBO-BP prediction model is constructed. Combined with data preprocessing and model training, high-precision prediction of droplet physical mass can be achieved.
It significantly improves the prediction accuracy of droplet physical mass and the stability of the model, enhances the adaptability to complex working conditions, improves the control effect of inkjet printing, and ensures the realization of high-quality printing.
Smart Images
Figure CN120671712A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of printing quality control, and in particular to a method for predicting the physical quality of inkjet printing droplets based on an improved dung beetle optimization algorithm and a BP neural network. Background Art
[0002] Due to its flexibility, inkjet printing technology is one of the most promising printing systems currently available. If the sprayed liquid is replaced with a different type of liquid, inkjet printing technology will become a viable alternative to many advanced manufacturing technologies. For example, replacing the sprayed liquid with conductive ink allows for the printing of circuit boards; changing the sprayed liquid to an edible liquid medicine allows for the printing of medicines, which will promote the development of various industries. In the manufacture of electronic devices such as touch screens, the precise preparation of transparent electrodes is very important. Precise droplet control enables the printing of fine circuits and patterns, which is crucial for improving the conductivity of transparent electrodes. If the droplets cannot be precisely controlled, the conductivity of the transparent electrode will be greatly reduced, which in turn affects the performance of the entire electronic device. The physical quality of the droplets in inkjet printing is crucial to improving the performance of printed electronic devices, and accurately predicting the physical quality of the droplets is the key to achieving precise control.
[0003] However, during the inkjet printing process, the physical quality of the droplets is affected by many factors, such as ink inlet velocity, nozzle outlet diameter, ink viscosity and surface tension, etc. There is a complex nonlinear relationship between these parameters, which makes it difficult for traditional mathematical models to accurately describe. Therefore, the use of machine learning methods to predict the physical quality of inkjet printed droplets has become an effective solution. In addition, most current studies have mainly focused on the volume of droplets, the velocity of droplets and the droplet ejection behavior in inkjet printing, while there are fewer studies on finding a method that can accurately predict the physical quality of droplets in the inkjet printing process. Applying machine learning to the field of inkjet printing is also a new attempt. Therefore, the present invention is based on machine learning technology and combines BP neural network to construct a prediction model for the physical quality of droplets to accurately predict the physical quality of inkjet printed droplets, so as to achieve the purpose of controlling droplets and high-quality printing. Summary of the Invention
[0004] The purpose of the present invention is to provide a method for predicting the physical quality of inkjet printing droplets based on an improved dung beetle optimization algorithm and a BP neural network, so as to achieve the purpose of controlling the droplets and high-quality printing.
[0005] In order to achieve the above technical objectives and the above technical effects, the present invention is implemented through the following technical solutions:
[0006] A method for predicting the physical quality of inkjet printing droplets based on a BP neural network optimized by an improved dung beetle optimization algorithm comprises the following steps:
[0007] S1: Collect parameters related to the physical quality of inkjet printing droplets, including ink inlet velocity, nozzle outlet diameter, ink viscosity and surface tension, construct a data set and remove outliers;
[0008] S2: Normalize the dataset and divide it into training set and test set in proportion;
[0009] S3: Initialize the dung beetle population using a dynamic reverse learning strategy and generate reverse solutions to enhance population diversity and uniformity;
[0010] S4: Introducing the golden sine factor into the dung beetle optimization algorithm, optimizing the position updates of the rolling ball dung beetle and the small dung beetle through sinusoidal perturbations, balancing global exploration and local exploitation capabilities, forming the Improved Dung Beetle Optimizer (IDBO) algorithm, and conducting a performance evaluation of the IDBO algorithm;
[0011] S5: Use the IDBO algorithm to optimize the weights and thresholds of the back propagation (BP) neural network and build an IDBO-BP prediction model;
[0012] S6: Input the training set into the IDBO-BP model for training and output the optimal network parameters;
[0013] S7: Use the test set to verify the model performance and output the droplet physical mass prediction results.
[0014] Furthermore, the step S1 includes the following sub-steps:
[0015] S1.1: Establish an inkjet printing simulation model based on COMSOL and obtain a data set.
[0016] S1.2: Use the quartile method to remove outliers from the dataset. The definition of outliers is as follows:
[0017] First, the upper quartile U (25%) and lower quartile L (75%) of all samples were determined, and the interquartile range (IQR) was calculated as the difference between U and L. Values exceeding U + 1.5 IQR or below L - 1.5 IQR were classified as outliers; the identified outliers were then removed.
[0018] Furthermore, in step S2, the data is scaled to between 0 and 1 to achieve data normalization, and is divided into a training set and a test set in a ratio of 8:2.
[0019] Furthermore, step S3 includes the following sub-steps:
[0020] S3.1: The dynamic reverse learning strategy generates the reverse solution through the formula, which is as follows:
[0021] oP i,j =lb j +rand·(ub j -lb j )
[0022]
[0023] r1=rand(1)
[0024] Among them, oP is the initial population; Op DO is the reverse population; N is the population size; D is the dimension; lb and ub are parameter boundaries; the rand() function is a random number generation function provided by MATLAB, which is used to generate random numbers that meet a specific distribution. The 1 in rand(1) means generating a 1-dimensional random number vector; r 1i and r 2i is a random number between the interval [0, 1].
[0025] S3.2: Based on the formula Check the boundaries. Finally, from oP and Op DO Select the first N individuals with the highest fitness as the initial population.
[0026] Furthermore, step S4 includes the following sub-steps:
[0027] S4.1: The formula for updating the position of the rolling ball dung beetle is as follows:
[0028]
[0029] In the formula, τ is the golden section number, which is α is a natural coefficient that indicates whether the algorithm has deviated from its original direction. Its value is 1 or -1, with 1 indicating no deviation from the original direction and -1 indicating no deviation from the original direction. The value of α is determined probabilistically, meaning that in each iteration, α is 1 or -1 with a certain probability. This allows us to simulate the various complex situations that dungbeetle encounters in real-world environments and prevent the algorithm from falling into local optimal solutions.
[0030] S4.2: The formula for updating the position of the dung beetle is as follows:
[0031]
[0032] Where, represents the position information of the i-th dung beetle at the t-th iteration; X brepresents the current local optimal position; β is a random number from the standard normal distribution; γ is the local search factor with a value of 0.5; x op Inverse solution generated by dynamic learning strategy.
[0033] Furthermore, in steps S4.1 and S4.2, the disturbance position of the dung beetle is selected by comparing the accuracy of four models to determine the optimal disturbance position. The four models are as follows:
[0034] (1) Simply add dynamic reverse learning to initialize the original population of dung beetles;
[0035] (2) A dynamic reverse learning strategy is added to initialize the original population of dung beetles and the position of the rolling dung beetles is disturbed by introducing a golden sine factor;
[0036] (3) A dynamic reverse learning strategy is added to initialize the population of dung beetles and the positions of small dung beetles are disturbed by introducing a golden sine factor;
[0037] (4) A dynamic reverse learning strategy is added to initialize the population of dung beetles and the positions of rolling dung beetles and small dung beetles are disturbed by introducing the golden sine factor.
[0038] Furthermore, in step S4.3, the IDBO algorithm is tested using 9 test functions in the CEC2017 test function set. All test functions have been rotated and shifted. Among them, F1 and F3 are unimodal functions with only global minima but no local minima; F4 to F10 are multimodal functions with local extreme points, which are used to test the algorithm's ability to escape from the local optimal solution and its global search and exploration capabilities.
[0039] Furthermore, step S5 includes the following sub-steps:
[0040] S5.1: The number of hidden layers in the BP neural network is determined by comparing the root mean square error: when the number of hidden layers is 2, the training time is 57 and the RMSE is 1.679. When the number of hidden layers is 4, the training time is 61 and the RMSE is 1.4399. When the number of hidden layers is increased from 2 to 4, the time increases by 6.56%, but when the number of hidden layers is increased from 2 to 4, the accuracy increases by 23.91%.
[0041] S5.2: Combine the IDBO algorithm and BP neural network to construct the IDBO-BP model.
[0042] Furthermore, in step S6, the hyperparameters are set as follows: the number of training iterations is set to 1000, the learning rate is adjusted to 0.01, the target error threshold is set to 0.0000001, and the number of hidden layers is determined to be 4.
[0043] Furthermore, the model performance evaluation indicators in step S7 include root mean square error (RMSE), mean absolute error (MAE) and goodness of fit (R 2 ), the specific formula is as follows:
[0044]
[0045] Where i=1, 2…n is the test sample; is the predicted value; y i is the true value; is the average value.
[0046] Beneficial effects of the present invention:
[0047] The present invention introduces an improved dung beetle optimization algorithm (IDBO) with a dynamic reverse learning strategy and a golden sine factor, which significantly improves the global optimization capability for BP neural network weights and thresholds compared to traditional optimization methods. The dynamic reverse learning strategy effectively enhances the diversity and uniformity of the population by generating a reverse solution for the dung beetle population in the initialization phase, expands the parameter search space, and reduces the risk of falling into local optimality in the initial phase. In further optimization iterations, the golden sine factor is introduced to dynamically update the individual positions of dung beetles through a sinusoidal perturbation mechanism, effectively balancing the capabilities of global search and local development. This innovative design is based on the fusion of the golden section and periodic perturbation, and can automatically switch between large-scale global exploration and high-precision local mining during the algorithm iteration process. In view of the fact that the physical quality of inkjet printing droplets is affected by the strong nonlinear coupling of multiple parameters such as ink inlet velocity, nozzle outlet diameter, viscosity, and surface tension, the IDBO algorithm can continuously jump out of local minimum points and continuously approach the global optimal solution, making the BP neural network parameter distribution more reasonable, and ultimately significantly improving the model's high-precision fitting and prediction capabilities for the complex physical processes of printing droplets, while greatly accelerating the model convergence speed and improving the stability of the training process.
[0048] To address the inherent problems of traditional BP neural networks, such as their reliance on gradient descent, susceptibility to initial value fluctuations, slow convergence, and tendency to fall into local optima, the present invention utilizes the IDBO algorithm to globally optimize network weights and thresholds. Based on dynamic reverse learning, the IDBO algorithm can generate solutions from multiple perspectives in the current parameter space. This not only achieves a better population distribution but also promotes multidimensional jumps across the entire parameter space by continuously introducing periodic perturbations of the golden sine factor. Because of this, when faced with the prediction of inkjet printing droplet physical mass, a nonlinear system influenced by multiple complex factors, the BP network can leverage IDBO's global exploration capabilities in high-dimensional space, overcoming the traditional BP network's weakness of being highly sensitive to initial parameters and significantly alleviating the local optimal dilemma. The underlying mechanism of the present invention relies on swarm intelligence collaboration, reverse sample detection, and dynamic adjustment of periodic perturbations, making the optimization process more diverse and efficient. Ultimately, this not only accelerates the BP network's convergence process but also effectively improves its adaptability to ink parameter perturbations and changes in different experimental conditions, resulting in greater industrial applicability and robustness.
[0049] The present invention significantly improves the purity of the data set by using the quartile method to eliminate outliers in the inkjet printing simulation data. It can automatically identify and remove extreme values, effectively avoiding the disturbance of abnormal data caused by equipment errors, outlier working conditions, etc. on the distribution of model parameters and the optimization process, and ensuring the representativeness and consistency of the model training data. In addition, all input features are normalized and different physical quantities are uniformly scaled to the range of 0 to 1, which not only eliminates the dimensional influence between variables, but also promotes the optimization algorithm to search for equal weights for each feature dimension, making the entire training process smoother and more efficient. After the industrial data is processed as described above, the IDBO-BP model can make full use of high-quality and evenly distributed data to optimize weights and thresholds, and improve the generalization ability and robustness of the prediction of droplet physical mass under different working conditions. This mechanism ensures that the model has excellent stability in various complex or sudden situations in actual industrial environments, providing a solid data foundation for high-quality printing control systems.
[0050] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0052] Figure 1 This is the outlier removal graph;
[0053] Figure 2 Schematic diagram of the fitness curve for selecting the best position of the golden sine factor;
[0054] Figure 3 Schematic diagram of the benchmark function convergence curve;
[0055] Figure 4 This is the flow chart of the IDBO-BP model;
[0056] Figure 5 Schematic diagram for determining the number of hidden layers in BP network;
[0057] Figure 6 It is the structure diagram of BP neural network;
[0058] Figure 7 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION
[0059] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.
[0060] Example 1
[0061] The method for predicting the physical mass of inkjet printing droplets based on the BP neural network optimized by the improved dung beetle optimization algorithm described in this embodiment includes the following steps:
[0062] S1: Collect parameters related to the physical quality of inkjet printing droplets, including ink inlet velocity, nozzle outlet diameter, ink viscosity, and surface tension, construct a data set, and remove outliers, including:
[0063] The data was collected by using COMSOL to establish an inkjet printing simulation model, with the ink inlet velocity of 0.7 m / s; nozzle outlet diameter of 60 μm-80 μm; ink viscosity of 0.002 Pa·s-0.02 Pa·s; and surface tension of 0.03 N / m-0.08 N / m.
[0064] The physical mass of the droplet is calculated using the following formula:
[0065]
[0066] ρ represents the density of the ink, Φ represents the level set function, Ω represents the integration region, and z1 represents the spatial coordinate. The function δ is used to select the integration region where z1 > 0.7 mm, and then integrate this integration region to obtain the physical mass of the droplet.
[0067] Data outliers were eliminated by quartile method and data were processed by Origin software. The processing results are shown in the figure below. Figure 1 As shown, there are 3 outliers in total.
[0068] Outliers were defined as follows: the upper quartile U (25%) and the lower quartile L (75%) of all samples were first determined, and the interquartile range (IQR) was calculated as the difference between U and L. Values exceeding U + 1.5 IQR or below L - 1.5 IQR were classified as outliers.
[0069] S2: Normalize the dataset and divide it into training and test sets in proportion, where:
[0070] A total of 400 sets of data are included, which are divided into training sets and test sets in a ratio of 8:2.
[0071] S3: Use a dynamic reverse learning strategy to initialize the dung beetle population and generate a reverse solution to enhance population diversity and uniformity: The dynamic reverse learning strategy generates a reverse solution through the formula, which is as follows:
[0072] oP i,j =lb j +rand·(ub j -lb j )
[0073]
[0074] Among them, oP is the initial population, Op DO is the reverse population, N is the population size, D is the dimension, lb and ub are the parameter boundaries, and r 1i and r 2i is a random number between the interval [0, 1].
[0075] Then according to the formula Check the boundaries. Finally, from oP and Op DO Select the first N individuals with the highest fitness as the initial population.
[0076] S4: Introducing the golden sine factor into the dung beetle optimization algorithm, optimizing the position updates of the rolling ball dung beetle and the small dung beetle through sinusoidal perturbations, balancing global exploration and local development capabilities, forming an improved dung beetle optimization algorithm (Improved Dung Beetle Optimizer, IDBO), and performing a performance evaluation of the IDBO algorithm, including:
[0077] The formula for updating the position of the rolling dung beetle is as follows:
[0078]
[0079] In the formula, τ is the golden section number, which is
[0080] The formula for updating the position of the dung beetle is as follows:
[0081]
[0082] Where, represents the position information of the i-th dung beetle at the t-th iteration; X b represents the current local optimal position; β is a random number from standard normal distribution; γ The local search factor value is 0.5; x op Inverse solution generated by dynamic learning strategy.
[0083] The disturbance location selection of dung beetles is determined by comparing the accuracy of four models to determine the optimal disturbance location. The four models are as follows. The fitness curve of the optimal disturbance location is as follows: Figure 2 As shown:
[0084] (1) Simply add dynamic reverse learning to initialize the original population of dung beetles;
[0085] (2) A dynamic reverse learning strategy is added to initialize the original population of dung beetles and the position of the rolling dung beetles is disturbed by introducing a golden sine factor;
[0086] (3) A dynamic reverse learning strategy is added to initialize the population of dung beetles and the positions of small dung beetles are disturbed by introducing a golden sine factor;
[0087] (4) A dynamic reverse learning strategy is added to initialize the population of dung beetles and the positions of rolling dung beetles and small dung beetles are disturbed by introducing the golden sine factor.
[0088] The IDBO algorithm was tested using nine test functions from the CEC2017 test function set. The performance of the improved algorithm was evaluated by comparing the optimal values and average values of the four algorithms: IDBO, DBO, PSO, and ABC. The optimization results of the benchmark functions are shown in Table 1.
[0089] Table 1 Benchmark function optimization results
[0090]
[0091]
[0092] Figure 3 The convergence curves of the IDBO algorithm and other comparison algorithms are shown. In the tests of single-peak functions F1 and F3, the convergence speed of the IDBO algorithm is better than that of other models. However, in terms of convergence speed, the IDBO algorithm performs best, followed by the DBO algorithm. The convergence curve of the IDBO algorithm is relatively smooth and can converge to the optimal value quickly and stably, while the other algorithms all show certain oscillation characteristics. This shows that the optimization performance and stability of the IDBO algorithm exceed those of the other swarm optimization algorithms and are also improved compared to the original algorithm. For multi-peak functions F4 to F10, the convergence speed of the IDBO algorithm is still ahead of the other three algorithms. Except for functions F8 and F10, the convergence curve of the IDBO algorithm does not show oscillation, indicating that the algorithm is very stable, while the convergence curves of the other three algorithms are stepped.
[0093] Overall, the IDBO algorithm demonstrated excellent performance across all tests. It not only converges quickly and accurately, but also produces a smoother and more stable convergence curve. More importantly, the IDBO algorithm strikes a good balance between exploitation and exploration capabilities, further demonstrating its superior overall performance among metaheuristic algorithms.
[0094] S5: Use the IDBO algorithm to optimize the weights and thresholds of the back propagation (BP) neural network and build the IDBO-BP prediction model. The model flow chart is as follows: Figure 4 ,in:
[0095] The number of hidden layers in the BP neural network is determined by comparing the root mean square error: when the number of hidden layers is 2, the training time is 57 and the RMSE is 1.679. When the number of hidden layers is 4, the training time is 61 and the RMSE is 1.4399. When the number of hidden layers is changed from 2 to 4, the time increases by 6.56%, but when the number of hidden layers is changed from 2 to 4, the accuracy increases by 23.91%. The results are as follows: Figure 5 shown.
[0096] BP neural network usually consists of three layers of neurons: input layer, hidden layer and output layer. The learning process of the network is divided into two stages: forward propagation and back propagation. Through these two stages, the network weights are continuously adjusted to reduce the output error. The topology of the BP neural network is as follows: Figure 6 shown
[0097] S6: Input the training set into the IDBO-BP model for training and output the optimal network parameters, where:
[0098] The hyperparameters are set as follows: the number of training iterations is set to 1000, the learning rate is adjusted to 0.01, the target error threshold is set to 0.0000001, and the number of hidden layers is determined to be 4.
[0099] S7: Use the test set to verify the model performance and output the droplet physical mass prediction results, where:
[0100] Model performance evaluation indicators include root mean square error (RMSE), mean absolute error (MAE) and goodness of fit (R 2 ), the specific formula is as follows:
[0101]
[0102]
[0103] Where i=1, 2…n is the test sample; is the predicted value; y i is the true value; is the average
[0104] The IDBO-BP-based inkjet printing droplet physical mass prediction model was validated on a test set. The IDBO-BP model was compared with the DBO-BP, WOA-BP, SSA-SVM, ACO-SVM, PSO-SVM, GWO-BP, PSO-BP, LSTM, and CNN models. The predictive ability of the IDBO-BP inkjet printing droplet physical mass prediction model was measured using three evaluation metrics. The results are shown in Table 2.
[0105] Table 2 Comparison of prediction model results
[0106]
[0107] As shown in Table 2, the root mean square error (RMSE) of the IDBO-BP model is 2.7127, which is reduced by 26%, 39%, 33%, 42%, 44%, 48%, 22%, 45%, and 46% compared with the prediction models such as DBO-BP, WOA-BP, SSA-SVM, ACO-SVM, PSO-SVM, WOA-BP, LSTM, CNN, and PSO-BP, respectively; the mean absolute error (MAE) of the IDBO-BP model is 1.9252, which is reduced by 15%, 28%, 10%, 11%, 30%, 33%, 27%, 22%, and 38%, respectively; the goodness of fit (R 2) was 0.98752, a maximum improvement of 3%. Experimental results confirm that the IDBO-BP model's predicted values are closer to the true values, and its performance in predicting the physical mass of inkjet printing droplets is superior to other models, achieving better prediction results.
[0108] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. A method for predicting the physical quality of inkjet printing droplets based on an improved dung beetle optimization algorithm and a BP neural network, characterized in that: The following steps are involved: S1: Collect parameters related to the physical quality of inkjet printing droplets, including ink inlet velocity, nozzle outlet diameter, ink viscosity and surface tension, construct a data set and remove outliers; S2: Normalize the dataset and divide it into training set and test set in proportion; S3: Initialize the dung beetle population using a dynamic reverse learning strategy and generate reverse solutions to enhance population diversity and uniformity; S4: Introducing the golden sine factor into the dung beetle optimization algorithm, optimizing the position updates of the rolling ball dung beetle and the small dung beetle through sinusoidal perturbations, balancing global exploration and local development capabilities, forming an improved dung beetle optimization algorithm, and performing a performance evaluation of the IDBO algorithm; S5: Use the IDBO algorithm to optimize the weights and thresholds of the back propagation neural network and build the IDBO-BP prediction model; S6: Input the training set into the IDBO-BP model for training and output the optimal network parameters; S7: Use the test set to verify the model performance and output the droplet physical mass prediction results.
2. The method for predicting the physical mass of inkjet printing droplets based on the BP neural network optimized by the improved dung beetle optimization algorithm according to claim 1, characterized in that: The step S1 includes the following sub-steps: S1.1: Establish an inkjet printing simulation model based on COMSOL and obtain a data set; S1.2: Use the quartile method to remove outliers from the dataset. The definition of outliers is as follows: First, determine the upper quartile U and lower quartile L of all samples and calculate the interquartile range as the difference between U and L; classify values exceeding U + 1.5 IQR or below L - 1.5 IQR as outliers; then remove the identified outliers.
3. The method for predicting the physical mass of inkjet printing droplets based on the BP neural network optimized by the improved dung beetle optimization algorithm according to claim 1, characterized in that: In step S2, the data is scaled to between 0 and 1 to achieve data normalization, and is divided into a training set and a test set in a ratio of 8:
2.
4. The method for predicting the physical mass of inkjet printing droplets based on the BP neural network optimized by the improved dung beetle optimization algorithm according to claim 1, characterized in that: The step S3 includes the following sub-steps: S3.1: The dynamic reverse learning strategy generates the reverse solution through the formula, which is as follows: oP i,j =lb j +rand·(ub j -lb j ) Among them, oP is the initial population, Op DO is the reverse population, N is the population size, D is the dimension, lb and ub are the parameter boundaries, and the rand() function is a random number generation function provided by MATLAB, which is used to generate random numbers that meet a specific distribution. The 1 in rand(1) means generating a 1-dimensional random number vector; r 1i and r 2i is a random number between [0, 1]; S3.2: Based on the formula Check the boundaries; finally, from oP and Op DO Select the first N individuals with the highest fitness as the initial population.
5. The method for predicting the physical mass of inkjet printing droplets based on the BP neural network optimized by the improved dung beetle optimization algorithm according to claim 1, characterized in that: The step S4 includes the following sub-steps: S4.1: The formula for updating the position of the rolling ball dung beetle is as follows: In the formula, τ is the golden section number, which is α is a natural coefficient, indicating whether it deviates from the original direction. Its value is 1 or -1. 1 means no deviation from the original direction, and -1 means deviation from the original direction. S4.2: The formula for updating the position of the dung beetle is as follows: Where, represents the position information of the i-th dung beetle at the t-th iteration; X b represents the current local optimal position; β is a random number from standard normal distribution; γ The local search factor value is 0.5; x op Inverse solution generated by dynamic learning strategy.
6. The method for predicting the physical mass of inkjet printing droplets based on the BP neural network optimized by the improved dung beetle optimization algorithm according to claim 5, characterized in that: In steps S4.1 and S4.2, the optimal disturbance location for the dung beetle is determined by comparing the accuracy of four models. The four models are as follows: (1) Simply add dynamic reverse learning to initialize the original population of dung beetles; (2) A dynamic reverse learning strategy is added to initialize the original population of dung beetles and the position of the rolling dung beetles is disturbed by introducing a golden sine factor; (3) A dynamic reverse learning strategy is added to initialize the population of dung beetles and the positions of small dung beetles are disturbed by introducing a golden sine factor; (4) A dynamic reverse learning strategy is added to initialize the population of dung beetles and the positions of rolling dung beetles and small dung beetles are disturbed by introducing the golden sine factor.
7. The method for predicting the physical mass of inkjet printing droplets based on the BP neural network optimized by the improved dung beetle optimization algorithm according to claim 5, characterized in that: In step S4.3, the IDBO algorithm is tested using 9 test functions in the CEC2017 test function set. All test functions have been rotated and shifted. Among them, F1 and F3 are unimodal functions with only global minima but no local minima; F4 to F10 are multimodal functions with local extreme points, which are used to test the algorithm's ability to escape from local optimal solutions and its global search and exploration capabilities.
8. The method for predicting the physical mass of inkjet printing droplets based on the BP neural network optimized by the improved dung beetle optimization algorithm according to claim 1, characterized in that: The step S5 includes the following sub-steps: S5.1: The number of hidden layers in the BP neural network was determined by comparing the root mean square error: when the number of hidden layers was 2, the training time was 57 and the RMSE was 1.679; when the number of hidden layers was 4, the training time was 61 and the RMSE was 1.4399; S5.2: Combine the IDBO algorithm and BP neural network to construct the IDBO-BP model; Furthermore, in step S6, the hyperparameters are set as follows: the number of training iterations is set to 1000, the learning rate is adjusted to 0.01, the target error threshold is set to 0.0000001, and the number of hidden layers is determined to be 4.
9. The method for predicting the physical mass of inkjet printing droplets based on the BP neural network optimized by the improved dung beetle optimization algorithm according to claim 1, characterized in that: The model performance evaluation indicators in step S7 include root mean square error, mean absolute error and goodness of fit, and the specific formulas are as follows: Where i=1, 2…n is the test sample; is the predicted value; y i is the true value; is the average value.