A method and system for controlling millikan oil drop experiment based on multi-voltage sampling

Abstract

This invention discloses a Millikan oil drop experiment control method and system based on multi-voltage sampling. The method includes window sampling of the oil drop trajectory of candidate oil droplets under various preset electric field voltages to obtain the velocity of candidate oil droplets and construct a feature vector. A pre-trained dual-output neural network model is used to activate randomly deactivated neurons to perform K random forward propagation cycles on the feature vector, establishing a mapping relationship between the multi-voltage response characteristics of candidate oil droplets and the predicted equilibrium voltage and predicted free fall velocity. Based on the prediction results and uncertainty statistics, gated screening of target oil droplets is performed. The average predicted equilibrium voltage is used as the reference starting voltage. This invention aims to solve the problems in traditional Millikan oil drop experiments, such as reliance on manual experience for candidate oil droplet screening, reliance on repeated trials for setting the initial equilibrium voltage, long time required to reach a stable and measurable state, and significant impact of operator differences on repeatability, thereby improving experimental quality and efficiency.

Description

Technical Field

[0001] This invention relates to the field of oil drop measurement and control technology in the Millikan oil drop experiment, specifically to a control method and system for the Millikan oil drop experiment based on multi-voltage sampling. Background Technology

[0002] The Millikan oil drop experiment is a classic physics experiment. Its core objective is to accurately determine the elementary charge of an electron and verify the quantization of charge—that is, the charge of any charged body is an integer multiple of the elementary charge. The experimental procedure of the Millikan oil drop experiment includes: (1) Atomizing insulating oil (such as watch oil) into micron-sized droplets through a sprayer, some of which become charged due to friction. (2) The charged oil droplets pass through a small hole between two horizontal parallel metal plates. (3) Controlling the balance between the electric field and gravity: When the electric field is turned off, the oil droplets fall at a constant speed under the action of gravity and air resistance, and their terminal velocity can be measured. After the electric field is turned on, the voltage is adjusted to make the oil droplets stand still (static method) or rise / fall at a constant speed. (4) Calculating the charge: Using Stokes' law, the mechanical equilibrium equation and known parameters (such as the distance between the plates, voltage, air viscosity coefficient, etc.), the charge carried by the oil droplets is calculated. From the experimental procedure, this experiment not only involves clear physical constraints but also two highly empirical pre-decision decisions: first, whether the current candidate oil droplet is worth entering for precise measurement; and second, from which voltage range should subsequent voltage adjustment begin. Improper handling of these two decisions will significantly increase invalid droplet finding, invalid voltage adjustment, and invalid measurements. Traditional teaching and manual operation of the Millikan oil drop experiment generally suffer from time-consuming droplet finding, selection, and voltage balancing adjustments, and the results are highly dependent on the operator's experience. Existing automation routes typically focus on improving single steps such as oil droplet detection, trajectory extraction, or data recording, but for the two key empirical decisions of "whether it's worth measuring" and "where to start adjusting," a unified closed-loop scheme that can directly drive subsequent control execution is still lacking. Summary of the Invention

[0003] The technical problem this invention aims to solve is to provide a Millikan oil drop experiment control method and system based on multi-voltage sampling, addressing the aforementioned problems in existing technologies. This invention aims to move the prediction and judgment decisions in the Millikan oil drop experiment, which originally relied on human experience, forward through a closed-loop technical chain of "three-voltage short-time sampling → feature vector → dual-output prediction → gating screening based on prediction results and their uncertainty statistics → prediction voltage initial value control." This solves the problems of traditional Millikan oil drop experiments, such as reliance on human experience for candidate oil drop selection, repeated trial-and-error for initial equilibrium voltage setting, long time required for a single high-quality oil drop to reach a stable and measurable state, and significant impact of operator differences on repeatability. Ultimately, this improves the quality and efficiency of the Millikan oil drop experiment.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A Millikan oil drop experiment control method based on multi-voltage sampling includes the following steps: window sampling of the trajectory of each candidate oil droplet under various preset electric field voltages in the Millikan oil drop experiment to obtain the velocity of the candidate oil droplets; extracting velocity features that characterize the response law of the candidate oil droplets under different preset electric field conditions based on the velocity of the candidate oil droplets and constructing a feature vector; keeping the network parameters of the pre-trained dual-output neural network model unchanged and turning on the randomly deactivated neurons, performing K random forward propagation on the feature vector of each candidate oil droplet through the pre-trained dual-output neural network model, thereby establishing a dual-output mapping relationship between the feature vector and the predicted equilibrium voltage and the predicted free fall velocity, and obtaining K sets of predicted equilibrium voltage and predicted free fall velocity for the candidate oil droplet; selecting target oil droplets from the candidate oil droplets based on the K sets of predicted equilibrium voltage and predicted free fall velocity and their uncertainty statistics; calculating the mean of the predicted equilibrium voltage of all target oil droplets as the reference electric field voltage or the initial set voltage for the subsequent equilibrium voltage regulation stage of the Millikan oil drop experiment to control the electric field voltage of the Millikan oil drop experiment so that the oil droplets in the electric field remain stationary or move at a uniform speed.

[0005] Optionally, the velocity characteristics of the candidate oil droplets include velocities under various preset electric field voltages, and some or all of the following: power terms, absolute value terms, pairwise product terms, pairwise ratio terms, statistical terms, difference terms, and relative change rates. The power terms are powers of a specified number of velocities under various preset electric field voltages; the pairwise product terms are pairwise products of velocities under various preset electric field voltages; the pairwise ratio terms are pairwise ratios of velocities under various preset electric field voltages, and the denominator includes a summation term to avoid the denominator being zero; the statistical terms include some or all of the following: average, standard deviation, maximum, minimum, and range; the range is the difference between the maximum and minimum values; the difference terms are the differences in velocities under adjacent preset electric field voltages; and the relative change rate is the relative change in velocities under adjacent preset electric field voltages. These velocity characteristics are used to characterize the response patterns of the candidate oil droplets under different preset electric field conditions.

[0006] Optionally, the step of selecting target oil droplets from candidate oil droplets based on K sets of predicted equilibrium voltages and predicted free fall velocities includes: calculating the mean, dispersion or standard deviation of the predicted equilibrium voltage, and predicted fall time of each candidate oil droplet based on K sets of predicted equilibrium voltages and predicted free fall velocities, respectively; and statistically determining whether each candidate oil droplet meets the following gating condition based on the prediction results and their uncertainties: and and ; in, The dispersion or standard deviation of the predicted equilibrium voltage for candidate oil droplets. The mean of the predicted equilibrium voltages of the candidate oil droplets. The predicted falling time of the candidate oil droplets. For uncertainty-based gating threshold, and These are the lower and upper limits of the balance voltage, respectively. and To predict the lower and upper limits of the fall time, candidate oil droplets that meet the gating conditions are selected as target oil droplets.

[0007] Optionally, the calculation function expression for the mean of the predicted equilibrium voltage of each candidate oil droplet is: ; in, The mean of the predicted equilibrium voltages of the candidate oil droplets. For each candidate oil droplet, the number of groups for predicted equilibrium voltage and predicted free fall velocity is given. Let k be the predicted equilibrium voltage of the candidate oil droplet; the function expression for calculating the dispersion or standard deviation of the predicted equilibrium voltage is: ; in, This represents the dispersion or standard deviation of the predicted equilibrium voltage for candidate oil droplets.

[0008] Optionally, the calculation function expression for the predicted fall time of the candidate oil droplet is: , ; in, The predicted falling time of the candidate oil droplets. The size of the gap in the electric field. To obtain the maximum value, To predict the mean free fall velocity, As a baseline free fall speed, For each candidate oil droplet, the number of groups for predicted equilibrium voltage and predicted free fall velocity is given. Let be the predicted free fall velocity of the k-th candidate oil droplet.

[0009] Optionally, when the pre-trained dual-output neural network model performs K random forward propagation on the feature vector of each candidate oil droplet, the functional expressions for the K sets of predicted equilibrium voltages and predicted free fall velocities of the candidate oil droplets are obtained as follows: ; in, Predict the equilibrium voltage and free fall velocity for the k-th group of candidate oil droplets. The k-th predicted equilibrium voltage of the candidate oil droplet. Let k be the predicted free fall velocity of the candidate oil droplet. For the k-th random forward propagation of the pre-trained dual-output neural network model, These are the network parameters for a pre-trained dual-output neural network model. is the feature vector of the candidate oil droplet.

[0010] Optionally, the dual-output neural network model is used to establish a dual-output mapping relationship between the feature vector and the predicted equilibrium voltage and the predicted free fall velocity. The dual-output neural network model is an MLP regression network, which includes an input mapping layer, a residual backbone network, an output correction layer, a multi-model unit, an attention fusion module, and an output layer. The function expression for the input mapping layer is: ; in, The input mapping layer maps the output features to a high-dimensional latent space. For the random deactivation layer of the input mapping layer, The activation function of the input mapping layer, For the layer normalization operation of the input mapping layer, and The input mapping layer contains weights and biases; the residual backbone network includes three sets of residual blocks and two levels of intermediate mapping modules located between the three sets of residual blocks. The first set of residual blocks comprises three cascaded residual blocks, the second set comprises two cascaded residual blocks, and the third set comprises one cascaded residual block. The functional expression of each residual block is: ; in, and These are the output and input features of the l-th residual block, respectively. For the l-th residual block, a random deactivation layer is formed. Let be the activation function for the l-th residual block. For the layer normalization operation of the l-th residual block, and Here are the weights and biases of the l-th residual block; the function expression of the intermediate mapping module is: ; in, and These are the output and input features of the intermediate mapping module, respectively. For the random deactivation layer of the intermediate mapping module, This is the activation function for the intermediate mapping module. For the layer normalization operation of the intermediate mapping module, and The weights and biases of the intermediate mapping module are defined; the output correction layer includes multiple cascaded output correction units for mapping, progressive compression, and shaping of input features. The function expression of the output correction unit is as follows: ; in, and The output and input characteristics of the output correction unit. For the random deactivation layer of the output correction unit, The activation function for the output correction unit. For the layer normalization operation of the output correction unit, and The weights and biases of the output correction unit are defined; the multi-model unit comprises multiple sub-models composed of a multilayer sensing mechanism, and the output layer of each sub-model includes two nodes for outputting an output vector composed of the predicted equilibrium voltage and the predicted free fall velocity of the candidate oil droplets, respectively; the attention fusion module is used to convert the candidate prediction representation output by the output correction layer into a single output vector. Mapped to a weight vector: ; ; in, For the weight vector, For the Softmax function, The output features obtained by the attention mapping module, The weights for each sub-model, Let represent the weight of any m-th sub-model, and T in the superscript indicates the transpose operation; the attention mapping module includes two mapping layers. and The weights and biases of the second mapping layer in the attention mapping module. This is the activation function for the attention mapping module. and The weights and biases of the first mapping layer in the attention mapping module are defined; the output layer is used to generate the final predicted equilibrium voltage and predicted free fall velocity from the output vectors obtained by each sub-model based on the weight vectors. ; in, The output vector is composed of the final predicted equilibrium voltage and the predicted free fall velocity. For the summation operation, This is the output vector obtained from the m-th sub-model. and The final predicted equilibrium voltage and predicted free fall velocity; the functional expression of the random deactivation layer is: ; ; in, For random deactivation layers of input features The obtained output features To match input features Same-dimensional random deactivation mask, For element-wise multiplication, For inactivation rate, for For any i-th dimension, Let represent a Bernoulli distribution, with a probability of 1 for the value 1. The probability of taking the value 0 is .

[0011] The present invention also provides a Millikan oil drop experiment control system based on multi-voltage sampling, including a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the Millikan oil drop experiment control method based on multi-voltage sampling.

[0012] The present invention also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute the Millikan oil drop experiment control method based on multi-voltage sampling by a processor.

[0013] The present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the Millikan oil drop experiment control method based on multi-voltage sampling via a processor.

[0014] Compared with existing technologies, this invention mainly achieves the following beneficial effects: This invention does not merely perform general prediction of candidate oil droplets, but rather, based on the Millikan oil drop experiment scenario, constructs a feature vector from the response information obtained by short-time sampling of the three voltages. A dual-output neural network is then used to establish a mapping relationship between this feature vector and the predicted equilibrium voltage and predicted free-fall velocity. Combined with the prediction results and their uncertainty statistics, gating screening is completed. Finally, the average predicted equilibrium voltage is used as the initial setting for the subsequent equilibrium voltage regulation stage, thus forming a closed-loop technical link of "prediction—judgment—initial value control." This invention moves the two key empirical decisions—whether a candidate oil droplet enters the precise measurement and the starting point of subsequent voltage regulation—to a unified prediction stage. This solves the problems in traditional Millikan oil drop experiments: candidate oil droplet screening relies on manual experience, the initial setting of the equilibrium voltage relies on repeated trials, the time required for a single high-quality oil droplet to enter a stable and measurable state is long, and repeatability is significantly affected by operator differences. It can reduce ineffective voltage regulation and ineffective measurements, improving the quality and efficiency of the Millikan oil drop experiment. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the basic process of the method in an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of the network structure of the MLP regression network in an embodiment of the present invention.

[0017] Figure 3 The image shows the loss function curves of the MLP regression network during the training and validation phases in this embodiment of the invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0019] like Figure 1 As shown, the Millikan oil drop experiment control method based on multi-voltage sampling in this embodiment includes the following steps: S1, window sampling is performed on the trajectory of each candidate oil droplet under various preset electric field voltages in the Millikan oil drop experiment to obtain the velocity of the candidate oil droplets; S2, based on the velocity of the candidate oil droplets, extract velocity features that characterize the response law of the candidate oil droplets under different preset electric field conditions and construct feature vectors. Under the condition of keeping the network parameters of the pre-trained dual-output neural network model unchanged and turning on the random deactivated neurons, perform K random forward propagation on the feature vector of each candidate oil droplet through the pre-trained dual-output neural network model, thereby establishing the dual-output mapping relationship between the feature vector and the predicted equilibrium voltage and the predicted free fall velocity, and obtaining K sets of predicted equilibrium voltage and predicted free fall velocity of the candidate oil droplet. S3, select the target oil droplet from the candidate oil droplets based on the predicted equilibrium voltage and predicted free fall velocity of K groups and their uncertainty statistics; S4. Calculate the average of the predicted equilibrium voltages of all target oil droplets as the starting reference electric field voltage for the subsequent equilibrium voltage regulation stage of the Millikan oil drop experiment, so as to control the electric field voltage of the Millikan oil drop experiment so that the oil droplets in the electric field remain stationary or move at a constant speed.

[0020] In step S1 of this embodiment, when window sampling is performed on the trajectories of each candidate oil droplet in the Millikan oil drop experiment under various preset electric field voltages to obtain the velocity of the candidate oil droplets, including window sampling of the oil droplet trajectories under three preset electric field voltages U1=200V, U2=250V, and U3=300V to obtain the displacement increment of the oil droplet trajectory, the velocity of the candidate oil droplet can be obtained according to the following formula: ; in, Let be the velocity of the candidate oil droplet under the i-th preset electric field voltage. Let be the displacement increment of the candidate oil droplet under the i-th preset electric field voltage. This represents the sampling window duration for the candidate oil droplet under the i-th preset electric field voltage. It should be noted that the number of preset electric field voltages can be specified as needed.

[0021] In this embodiment, the velocity characteristics of the candidate oil droplets include velocities under various preset electric field voltages (which can be expressed as: ) and the power term and absolute value term of velocity ( The terms include pairwise product terms, pairwise ratio terms, statistical terms, difference terms, and relative rates of change (some of which can be selected). The power terms are the velocity under various preset electric field voltages raised to a specified power; for example, the 2nd and 3rd powers can be expressed as: The pairwise product terms are the pairwise products of velocities under various preset electric field voltages, i.e.: The pairwise ratio terms are the pairwise ratios of velocities under various preset electric field voltages, and the velocities in the denominator include a summation term of a constant used to avoid the denominator being zero. ,Right now: , and The velocities are denoted by the i-th and j-th preset electric field voltages, respectively; the statistical terms include the average, standard deviation, maximum, minimum, and range (some of which can be selected), where the range is the difference between the maximum and minimum values; the difference term is the difference between velocities under adjacent preset electric field voltages, i.e.: ; ; ; in, , and The relative change rate is the difference in velocity under three adjacent preset electric field voltages; the relative change rate is the relative change in velocity under adjacent preset electric field voltages, i.e.: ； ； ; in, , and These represent the relative changes in velocity under three adjacent preset electric field voltages.

[0022] Specifically, in this embodiment, velocity features of candidate oil droplets are extracted based on their velocities, and a feature vector is constructed. The function expression is: ; The feature vector is a 12-dimensional vector. Among them: the first three terms are basic velocity terms; terms 4-6 are product terms; terms 7-9 are square terms; and terms 10-12 are absolute value terms. This way of writing can cover the basic idea of ​​basic observation features + polynomial features + combined interactive features, and avoid introducing the ambiguity of negative square roots. This feature vector is used to express the response law of the same candidate oil droplet under multi-voltage short-time sampling conditions.

[0023] In this embodiment, the step of selecting target oil droplets from candidate oil droplets based on K sets of predicted equilibrium voltages and predicted free fall velocities includes: calculating the mean, dispersion or standard deviation of the predicted equilibrium voltage, and predicted fall time of each candidate oil droplet based on K sets of predicted equilibrium voltages and predicted free fall velocities, and determining whether each candidate oil droplet meets the following gating condition: and and ; in, The dispersion or standard deviation of the predicted equilibrium voltage for candidate oil droplets. The mean of the predicted equilibrium voltages of the candidate oil droplets. The predicted falling time of the candidate oil droplets. For uncertainty-based gating threshold, and These are the lower and upper limits of the balance voltage, respectively. and To predict the lower and upper limits of the fall time, candidate oil droplets that meet the gating conditions are selected as target oil droplets.

[0024] In this embodiment, the calculation function expression for the mean of the predicted equilibrium voltage of each candidate oil droplet is as follows: ; in, The mean of the predicted equilibrium voltages of the candidate oil droplets. For each candidate oil droplet, the number of groups for predicted equilibrium voltage and predicted free fall velocity is given. Let k be the predicted equilibrium voltage of the candidate oil droplet; the function expression for calculating the dispersion or standard deviation of the predicted equilibrium voltage is: ; in, This represents the dispersion or standard deviation of the predicted equilibrium voltage for candidate oil droplets.

[0025] In this embodiment, the calculation function expression for the predicted fall time of the candidate oil droplet is: , ; in, The predicted falling time of the candidate oil droplets. The size of the gap in the electric field. To obtain the maximum value, To predict the mean free fall velocity, As a baseline free fall speed, For each candidate oil droplet, the number of groups for predicted equilibrium voltage and predicted free fall velocity is given. Let be the predicted free fall velocity of the k-th candidate oil droplet.

[0026] In this embodiment, when the pre-trained dual-output neural network model performs K random forward propagation operations on the feature vector of each candidate oil droplet, the resulting K sets of predicted equilibrium voltages and predicted free fall velocities of the candidate oil droplets are expressed as follows: ; in, Predict the equilibrium voltage and free fall velocity for the k-th group of candidate oil droplets. The k-th predicted equilibrium voltage of the candidate oil droplet. Let k be the predicted free fall velocity of the candidate oil droplet. For the k-th random forward propagation of the pre-trained dual-output neural network model, These are the network parameters for a pre-trained dual-output neural network model. is the feature vector of the candidate oil droplet.

[0027] The dual-output neural network model can, as needed, employ a dual-output regression network suitable for establishing the mapping relationship between the aforementioned feature vectors and the predicted equilibrium voltage and predicted free fall velocity. As an optional implementation, in this embodiment, the dual-output neural network model is an MLP regression network, such as... Figure 2 As shown, the MLP regression network includes an input mapping layer, a residual backbone network, an output correction layer, a multi-model unit, an attention fusion module, and an output layer; The function expression for the input mapping layer is: ; in, The input mapping layer maps the output features to a high-dimensional latent space. For the random deactivation layer of the input mapping layer, The activation function for the input mapping layer can be SiLU, GELU, Mish, LeakyReLU, PReLU, etc. For the layer normalization operation of the input mapping layer, and The input mapping layer contains weights and biases; the residual backbone network includes three sets of residual blocks and two levels of intermediate mapping modules located between the three sets of residual blocks. The first set of residual blocks includes three cascaded residual blocks (residual block 1 to residual block 3), the second set includes two cascaded residual blocks (residual block 4 and residual block 5), and the third set includes one cascaded residual block (residual block 6). The functional expression of the residual blocks is: ; in, and These are the output and input features of the l-th residual block, respectively. For the l-th residual block, a random deactivation layer is formed. is the activation function for the l-th residual block (such as SiLU, GELU, Mish, LeakyReLU, PReLU, etc. can be used). For the layer normalization operation of the l-th residual block, and Here are the weights and biases of the l-th residual block; the function expression of the intermediate mapping module is: ; in, and These are the output and input features of the intermediate mapping module, respectively. For the random deactivation layer of the intermediate mapping module, The activation function for the intermediate mapping module (such as SiLU, GELU, Mish, LeakyReLU, PReLU, etc.) can be used. For the layer normalization operation of the intermediate mapping module, and For the weights and biases of the intermediate mapping module; such as Figure 2 As shown, in this embodiment, the activation functions of the first group of residual blocks are SiLU, GELU, and Mish activation functions, respectively; the activation function of the first intermediate mapping module is LeakyReLU activation function; the activation functions of the second group of residual blocks are SiLU and GELU activation functions, respectively; the activation function of the second intermediate mapping module is PReLU activation function; and the activation functions of the third group of residual blocks are Mish activation functions. The output correction layer includes multiple cascaded output correction units for mapping, progressive compression, and shaping of input features. The function expression of the output correction unit is: ; in, and The output and input characteristics of the output correction unit. For the random deactivation layer of the output correction unit, The activation function for the output correction unit (such as SiLU, GELU, Mish, LeakyReLU, PReLU, etc.) can be used. For the layer normalization operation of the output correction unit, and The weights and biases of the output correction unit are defined; the multi-model unit comprises multiple sub-models composed of a multilayer sensing mechanism, and the output layer of each sub-model includes two nodes for outputting an output vector composed of the predicted equilibrium voltage and the predicted free fall velocity of the candidate oil droplets, respectively; the attention fusion module is used to convert the candidate prediction representation output by the output correction layer into a single output vector. Mapped to a weight vector: ; ; in, For the weight vector, For the Softmax function, The output features obtained by the attention mapping module, The weights for each sub-model, Let represent the weight of any m-th sub-model, and T in the superscript indicates the transpose operation; the attention mapping module includes two mapping layers. and The weights and biases of the second mapping layer in the attention mapping module. This is the activation function for the attention mapping module. and Here are the weights and biases of the first mapping layer in the attention mapping module; in this embodiment, there are a total of 5 sub-models, therefore... : ; according to The function yields: ; in, The output features obtained by the attention mapping module for the m-th sub-model , ~ These are the output features obtained by the attention mapping module for the 1st to 5th sub-models, respectively. The output layer is used to generate the final predicted equilibrium voltage and predicted free fall velocity from the output vectors obtained from each sub-model based on the weight vector. ; in, The output vector is composed of the final predicted equilibrium voltage and the predicted free fall velocity. For the summation operation, This is the output vector obtained from the m-th sub-model. and The final predicted equilibrium voltage and predicted free fall velocity; the functional expression of the random deactivation layer is: ; ; in, For random deactivation layers of input features The obtained output features To match input features Same-dimensional random deactivation mask, For element-wise multiplication, For inactivation rate, for For any i-th dimension, Let represent a Bernoulli distribution, with a probability of 1 for the value 1. The probability of taking the value 0 is .like Figure 2 As shown, in this embodiment, different random deactivation layers are configured with different deactivation rates p. The random deactivation layers activate randomly deactivated neurons, and a pre-trained dual-output neural network model performs K random forward propagations on the feature vector of each candidate oil droplet to obtain K sets of predicted equilibrium voltages and predicted free-fall velocities for that candidate oil droplet. It should be noted that the residual block is an existing network module.

[0028] To verify the effectiveness of the Millikan oil drop experiment control method based on multi-voltage sampling in this embodiment, the loss function used during the training of the MLP regression network is a weighted hybrid loss obtained by combining the MSE loss and MAE loss; the AdamW optimizer and cosine annealing learning rate scheduling are used during training. The convergence curves of the loss during the training phase (training loss) and the loss during the validation phase (validation loss) are shown below. Figure 3 As shown. See also Figure 3 As can be seen, after 400 iterations in this embodiment, both the training loss and the validation loss converge, verifying the effectiveness of the Millikan oil drop experiment control method based on multi-voltage sampling and its hybrid loss in this embodiment.

[0029] To further verify the application effect of this embodiment in the actual Millikan oil drop experiment, under the same experimental conditions, the traditional manual screening and voltage adjustment method and the method of this embodiment were compared, and the following indicators were statistically analyzed: the time for a single high-quality oil drop to enter a stable and measurable state, which is used to characterize the comprehensive time spent in finding droplets, screening, and adjusting voltage to enter the measurable state; the standard deviation of the balance voltage repeatability, which is used to characterize the consistency of the balance voltage results under repeated measurement conditions; and the relative uncertainty of a single charge measurement, which is used to characterize the accuracy of a single measurement. The final results are shown in Table 1.

[0030]

[0031] As shown in Table 1, the method of this embodiment can significantly shorten the time for high-quality oil droplets to enter a stable and measurable state, and improve the repeatability of the balance voltage and the accuracy of single charge measurement.

[0032] Furthermore, to illustrate the effects of short-time sampling of three voltages, dual-output prediction, gating based on prediction results and their uncertainty statistics, and initial value control of predicted voltage on the overall effect, a comparative analysis of different module combination schemes can be conducted, and the results are shown in Table 2.

[0033]

[0034] The baseline scheme relies heavily on human experience, resulting in low efficiency in screening and voltage regulation, and its repeatability is significantly affected by the operator. Scheme 1 can more fully characterize the response differences of oil droplets under different electric fields, and its screening stability is better than the baseline scheme. Scheme 2 establishes a dual-output mapping relationship between feature vectors and predicted equilibrium voltage and predicted free fall velocity, which is used to simultaneously support whether a sample is worth entering fine measurement and to determine the starting point of subsequent voltage regulation. Scheme 3 further introduces a random deactivation layer, which enables the network to have better generalization ability during the training phase and to support gating judgment based on prediction results and their uncertainty statistics during the inference phase, thereby reducing the probability of high-risk samples being mistakenly screened into subsequent processes. The final scheme directly uses the average predicted equilibrium voltage obtained through gating to set the starting voltage for the subsequent equilibrium voltage regulation stage, thereby achieving the best overall effect in terms of sample screening, voltage regulation starting point setting, and efficiency in entering a stable and measurable state. The comparison results shown in Table 2 indicate that short-time sampling of three voltages helps to characterize the response differences of candidate oil droplets under different electric field conditions, dual-output prediction can provide a basis for both target oil droplet screening and voltage regulation starting point setting, random deactivation layer provides a basis for subsequent gating screening based on prediction results and its uncertainty statistics, and prediction voltage initial value control can shorten the subsequent voltage regulation convergence time.

[0035] Those skilled in the art will understand that the technical solutions provided by this invention can take the form of a method, system, or computer program product. Therefore, this invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. This invention can provide a Millikan oil drop experiment control system based on multi-voltage sampling, including an interconnected microprocessor and a memory, the microprocessor being programmed or configured to execute the Millikan oil drop experiment control method based on multi-voltage sampling. This invention can provide a computer-readable storage medium storing a computer program or instructions programmed or configured to execute the Millikan oil drop experiment control method based on multi-voltage sampling via a processor. This invention can provide a computer program product including a computer program or instructions programmed or configured to execute the Millikan oil drop experiment control method based on multi-voltage sampling via a processor. Furthermore, this invention can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0036] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A Millikan oil drop experiment control method based on multi-voltage sampling, characterized in that, The process includes the following steps: window sampling of the trajectories of each candidate oil droplet under various preset electric field voltages in the Millikan oil drop experiment to obtain the velocity of the candidate oil droplets; extraction of velocity features that characterize the response behavior of the candidate oil droplets under different preset electric field conditions based on the velocity of the candidate oil droplets and construction of feature vectors; keeping the network parameters of the pre-trained dual-output neural network model unchanged and enabling random deactivated neurons, performing K random forward propagation on the feature vector of each candidate oil droplet through the pre-trained dual-output neural network model to establish a dual-output mapping relationship between the feature vector and the predicted equilibrium voltage and predicted free fall velocity, and obtaining K sets of predicted equilibrium voltage and predicted free fall velocity for the candidate oil droplet; screening out target oil droplets from the candidate oil droplets based on the K sets of predicted equilibrium voltage and predicted free fall velocity and their uncertainty statistics; calculating the mean of the predicted equilibrium voltage of all target oil droplets as the starting reference electric field voltage for the subsequent equilibrium voltage regulation stage of the Millikan oil drop experiment to control the electric field voltage of the Millikan oil drop experiment so that the oil droplets in the electric field remain stationary or move at a uniform speed.

2. The Millikan oil drop experiment control method based on multi-voltage sampling according to claim 1, characterized in that, The velocity characteristics of the candidate oil droplets include velocities under various preset electric field voltages, and some or all of the following: power terms, absolute value terms, pairwise product terms, pairwise ratio terms, statistical terms, difference terms, and relative change rates. The power terms are powers of the velocities under various preset electric field voltages; the pairwise product terms are pairwise products of the velocities under various preset electric field voltages; the pairwise ratio terms are pairwise ratios of the velocities under various preset electric field voltages, and the denominator includes a summation term to avoid a zero denominator; the statistical terms include some or all of the following: average, standard deviation, maximum, minimum, and range; the range is the difference between the maximum and minimum values; the difference term is the difference in velocities under adjacent preset electric field voltages; and the relative change rate is the relative change in velocities under adjacent preset electric field voltages. These velocity characteristics are used to characterize the response patterns of the candidate oil droplets under different preset electric field conditions.

3. The Millikan oil drop experiment control method based on multi-voltage sampling according to claim 1, characterized in that, The step of selecting target oil droplets from candidate oil droplets based on K sets of predicted equilibrium voltages and predicted free fall velocities includes: calculating the mean, dispersion or standard deviation of the predicted equilibrium voltage, and predicted fall time of each candidate oil droplet based on K sets of predicted equilibrium voltages and predicted free fall velocities, respectively; and determining whether each candidate oil droplet meets the following gating conditions based on the prediction results and their uncertainty statistics: and and ; in, The dispersion or standard deviation of the predicted equilibrium voltage for candidate oil droplets. The mean of the predicted equilibrium voltages of the candidate oil droplets. The predicted falling time of the candidate oil droplets. For uncertainty-based gating threshold, and These are the lower and upper limits of the balance voltage, respectively. and To predict the lower and upper limits of the fall time, candidate oil droplets that meet the gating conditions are selected as target oil droplets.

4. The Millikan oil drop experiment control method based on multi-voltage sampling according to claim 3, characterized in that, The calculation function expression for the mean of the predicted equilibrium voltage of each candidate oil droplet is as follows: ； in, The mean of the predicted equilibrium voltages of the candidate oil droplets. For each candidate oil droplet, the number of groups for predicted equilibrium voltage and predicted free fall velocity is given. Let k be the predicted equilibrium voltage of the candidate oil droplet; the function expression for calculating the dispersion or standard deviation of the predicted equilibrium voltage is: ； in, This represents the dispersion or standard deviation of the predicted equilibrium voltage for candidate oil droplets.

5. The Millikan oil drop experiment control method based on multi-voltage sampling according to claim 3, characterized in that, The calculation function expression for the predicted fall time of the candidate oil droplets is as follows: ， ； in, The predicted falling time of the candidate oil droplets. The size of the gap in the electric field. To obtain the maximum value, To predict the mean free fall velocity, As a baseline free fall speed, For each candidate oil droplet, the number of groups for predicted equilibrium voltage and predicted free fall velocity is given. Let be the predicted free fall velocity of the k-th candidate oil droplet.

6. The Millikan oil drop experiment control method based on multi-voltage sampling according to claim 1, characterized in that, When the pre-trained dual-output neural network model performs K random forward propagation operations on the feature vector of each candidate oil droplet, the functional expressions for the K sets of predicted equilibrium voltages and predicted free fall velocities of the candidate oil droplets are as follows: ； in, Predict the equilibrium voltage and free fall velocity for the k-th group of candidate oil droplets. The k-th predicted equilibrium voltage of the candidate oil droplet. Let k be the predicted free fall velocity of the candidate oil droplet. For the k-th random forward propagation of the pre-trained dual-output neural network model, These are the network parameters for a pre-trained dual-output neural network model. is the feature vector of the candidate oil droplet.

7. The Millikan oil drop experiment control method based on multi-voltage sampling according to claim 1, characterized in that, The dual-output neural network model is used to establish a dual-output mapping relationship between the feature vector and the predicted equilibrium voltage and the predicted free fall velocity. The dual-output neural network model is preferably an MLP regression network, which includes an input mapping layer, a residual backbone network, an output correction layer, a multi-model unit, an attention fusion module, and an output layer. The function expression for the input mapping layer is: ； in, The input mapping layer maps the output features to a high-dimensional latent space. For the random deactivation layer of the input mapping layer, The activation function of the input mapping layer, For the layer normalization operation of the input mapping layer, and The input mapping layer contains weights and biases; the residual backbone network includes three sets of residual blocks and two levels of intermediate mapping modules located between the three sets of residual blocks. The first set of residual blocks comprises three cascaded residual blocks, the second set comprises two cascaded residual blocks, and the third set comprises one cascaded residual block. The functional expression of each residual block is: ； in, and These are the output and input features of the l-th residual block, respectively. For the l-th residual block, a random deactivation layer is formed. Let be the activation function for the l-th residual block. For the layer normalization operation of the l-th residual block, and Here are the weights and biases of the l-th residual block; the function expression of the intermediate mapping module is: ； in, and These are the output and input features of the intermediate mapping module, respectively. For the random deactivation layer of the intermediate mapping module, This is the activation function for the intermediate mapping module. For the layer normalization operation of the intermediate mapping module, and The weights and biases of the intermediate mapping module are defined; the output correction layer includes multiple cascaded output correction units for mapping, progressive compression, and shaping of input features. The function expression of the output correction unit is as follows: ； in, and The output and input characteristics of the output correction unit. For the random deactivation layer of the output correction unit, The activation function for the output correction unit. For the layer normalization operation of the output correction unit, and The weights and biases of the output correction unit are defined; the multi-model unit comprises multiple sub-models composed of a multilayer sensing mechanism, and the output layer of each sub-model includes two nodes for outputting an output vector composed of the predicted equilibrium voltage and the predicted free fall velocity of the candidate oil droplets, respectively; the attention fusion module is used to convert the candidate prediction representation output by the output correction layer into a single output vector. Mapped to a weight vector: ； ; in, For the weight vector, For the Softmax function, The output features obtained by the attention mapping module, The weights for each sub-model, Let represent the weight of any m-th sub-model, and T in the superscript indicates the transpose operation; the attention mapping module includes two mapping layers. and The weights and biases of the second mapping layer in the attention mapping module. This is the activation function for the attention mapping module. and The weights and biases of the first mapping layer in the attention mapping module are defined; the output layer is used to generate the final predicted equilibrium voltage and predicted free fall velocity from the output vectors obtained by each sub-model based on the weight vectors. ； in, The output vector is composed of the final predicted equilibrium voltage and the predicted free fall velocity. For the summation operation, This is the output vector obtained from the m-th sub-model. and The final predicted equilibrium voltage and predicted free fall velocity; the functional expression of the random deactivation layer is: ； ； in, For random deactivation layers of input features The obtained output features To match input features Same-dimensional random deactivation mask, For element-wise multiplication, For inactivation rate, for For any i-th dimension, Let represent a Bernoulli distribution, with a probability of 1 for the value 1. The probability of taking the value 0 is .

8. A Millikan oil drop experiment control system based on multi-voltage sampling, comprising a microprocessor and a memory interconnected, characterized in that, The microprocessor is programmed or configured to execute the Millikan oil drop experiment control method based on multiple voltage sampling as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the Millikan oil drop experiment control method based on multiple voltage sampling as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the Millikan oil drop experiment control method based on multiple voltage sampling as described in any one of claims 1 to 7.

Images