A shear point prediction method and device based on fusion model

Abstract

This application relates to the field of ultrasonic equipment technology, and provides a method and apparatus for predicting the cut point based on a fusion model. By using historical electrical signal parameters, a first model with better predictive accuracy and a second model with better reliability are established. When the ultrasonic scalpel is cutting the object, the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator are obtained; thereby, a first output of the first model and a second output of the second model are obtained; and based on the operator's prediction requirements, according to the prediction requirements, the first output, and the second output, a prediction result for the cut point time is obtained, providing a cut point prediction result that meets the requirements. This avoids situations where the operator, before observing that the object has been cut, causes the ultrasonic scalpel tip to generate high temperatures, reducing the service life of the ultrasonic scalpel.

Description

Technical Field

[0001] This application relates to the field of ultrasonic equipment technology, and in particular to a method and apparatus for predicting shear points based on a fusion model. Background Technology

[0002] The ultrasonic scalpel transmits electrical energy to a piezoelectric transducer via an energy generator. The transducer converts the electrical energy into ultrasonic mechanical energy, thereby generating ultrasonic vibration. The ultrasonic scalpel head then amplifies the ultrasonic vibration and uses it to cut the object being cut.

[0003] When cutting an object, the operator observes to determine whether the ultrasonic scalpel has finished cutting and whether the object has been completely severed. However, in practice, the narrow field of vision in the operating space can lead to significant discrepancies between the operator's observation and the actual cutting situation. Ending the cutting action too early can easily result in cutting failure; ending it too late will cause the ultrasonic scalpel's jaws to continuously wear down the pads, leading to higher temperatures at the scalpel tip and reducing its lifespan. Summary of the Invention

[0004] This application provides a shear point prediction method and solution based on a fusion model. When an ultrasonic scalpel is cutting an object, it provides a predicted shear point time that meets the operator's prediction requirements, thus avoiding situations where the ultrasonic scalpel tip generates a high temperature and reduces the service life of the ultrasonic scalpel because the operator does not observe that the cutting of the object has been completed.

[0005] This application provides a shear point prediction method based on a fusion model, including:

[0006] The variable parameters of the historical electrical signal of the ultrasonic scalpel generator within a preset time period are obtained, and the variable parameters of the historical electrical signal are preprocessed so that the cutting point of the ultrasonic scalpel cutting the object is always within the preset time period.

[0007] The results corresponding to the variable parameters of the historical electrical signal in the first time period are marked as true to obtain the first training set; the results corresponding to the variable parameters of the historical electrical signal in the second time period are marked as true to obtain the second training set. The first time period and the second time period are within a preset time period, and the first time period is earlier than the second time period.

[0008] Construct a first initial model and a second initial model, and train the first initial model with the first training set to obtain the first model; train the second initial model with the second training set to obtain the second model;

[0009] When an ultrasonic scalpel is cutting an object, the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator are obtained.

[0010] The variable parameters of the real-time electrical signal are used as inputs to the first model and the second model to obtain the first output of the first model and the second output of the second model.

[0011] Obtain the forecast demand, and based on the forecast demand, the first output, and the second output, obtain the forecast result at the cutoff point.

[0012] In one possible implementation, the steps for obtaining the predicted demand include:

[0013] Get the thickness of the object being cut;

[0014] Based on the thickness of the object being cut and the preset correspondence, the weight corresponding to the predicted demand is determined. The preset correspondence includes the weight relationship between the thickness of the object being cut and the predicted demand.

[0015] In one possible implementation, the following weighted model is used to obtain the prediction result at the shear point time:

[0016] Y=α″×y1+(1-α″)×y2;

[0017] Where α″ is the weight corresponding to the predicted demand, y1 is the first output, and y2 is the second output.

[0018] In one possible implementation, the step of preprocessing the variable parameters of the historical electrical signal includes:

[0019] Calculate the average value of each parameter in the variable parameters of the historical electrical signal in the third time period, and mark it as the reference parameter of each parameter. The third time period is the start time period in the preset time period.

[0020] Based on the baseline parameters of each parameter, adjust each parameter in the variable parameters of the historical electrical signal.

[0021] In one possible implementation, the step of preprocessing the variable parameters of the historical electrical signal includes:

[0022] Based on preset rules, outliers in the variable parameters of historical electrical signals are removed.

[0023] In one possible implementation, the step of preprocessing the variable parameters of the historical electrical signal includes:

[0024] The exponential moving average model is used to process the variable parameters of historical electrical signals. The exponential moving average model is as follows:

[0025] X t =βX t-1 +(1-β)θ t ;

[0026] Where β is the weighting parameter, θ t Let X be the weight parameter obtained in the t-th update. t Let be the moving average of the variable parameters of the historical electrical signal obtained in the t-th update.

[0027] In one possible implementation, the step of preprocessing the variable parameters of the historical electrical signal includes:

[0028] The variable parameters of historical electrical signals are scaled and standardized according to a preset ratio. The standardization model is as follows:

[0029]

[0030] Where μ is the mean of each variable parameter of the historical electrical signal, and σ is the standard deviation of each variable parameter of the historical electrical signal.

[0031] In one possible implementation, the step of preprocessing the variable parameters of the historical electrical signal includes:

[0032] Obtain the timeliness value at the warning cutoff point;

[0033] Based on the timeliness value, the training data in the first and second training sets are cut.

[0034] In one possible implementation, after constructing the first and second initial models, the following is also included:

[0035] The first and second initial models are pre-trained using the variable parameters of historical electrical signals to obtain basic weights. The batch size of the training input is set to 1024, and the learning rate is set to 0.05.

[0036] In one possible implementation, the steps for training a first initial model using a first training set to obtain the first model are as follows:

[0037] Load the basic weights, train the first initial model using the first training set, set the batch size of the training input to 254, set the decay parameter to 0.01, set the learning rate to 0.005, and set the cross-entropy loss function as follows:

[0038] in, p represents the label values ​​of the real samples in the first training set. t The predicted probability of the true shearing is the output of the first initial model. p represents the label value of the fake samples in the first training set. fα represents the predicted probability of spurious pruning output by the first initial model, and α is the influence factor on the label values ​​of the real samples in the first training set.

[0039] In one possible implementation, the steps for training a second initial model using a second training set to obtain the second model are as follows:

[0040] Load the base weights, train the second initial model using the second training set, set the batch size of the training input to 254, set the decay parameter to 0.01, set the learning rate to 0.005, and set the cross-entropy loss function as follows:

[0041] in, p represents the label values ​​of the real samples in the second training set. t ′ represents the predicted probability of the true shearing from the output of the second initial model. p represents the label value of the fake samples in the second training set. f ' represents the predicted probability of spurious pruning from the output of the second initial model, and α' represents the influence factor on the real samples in the second training set.

[0042] In one possible implementation, the variable parameters of the historical electrical signal and the variable parameters of the real-time electrical signal include: output current value, output voltage value, current-voltage phase difference, the difference between the current-voltage phase difference and the target phase difference, and the first derivative of the resonant frequency.

[0043] This application also provides a shear point prediction device based on a fusion model, which is used to execute a shear point prediction method based on a fusion model. The shear point prediction device based on a fusion model includes:

[0044] The variable parameter acquisition module for historical electrical signals is used to acquire the variable parameters of historical electrical signals of the ultrasonic scalpel generator within a preset time period, and to preprocess the variable parameters of the historical electrical signals so that the cutting point of the ultrasonic scalpel cutting the object being cut is always within the preset time period.

[0045] The variable parameter marking module is used to mark the results corresponding to the variable parameters of the historical electrical signals in the first time period as true, so as to obtain the first training set; and to mark the results corresponding to the variable parameters of the historical electrical signals in the second time period as true, so as to obtain the second training set. The first time period and the second time period are within a preset time period, and the first time period is earlier than the second time period.

[0046] The model building module is used to build a first initial model and a second initial model, and to train the first initial model with a first training set to obtain a first model; and to train the second initial model with a second training set to obtain a second model;

[0047] The real-time electrical signal variable parameter acquisition module acquires the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator when the ultrasonic scalpel is cutting the object.

[0048] The prediction module is used to take the variable parameters of the real-time electrical signal as input to the first model and the second model to obtain the first output of the first model and the second output of the second model.

[0049] The fusion module is used to obtain the prediction requirements and, based on the prediction requirements, the first output, and the second output, to obtain the prediction result at the cutoff point.

[0050] This application provides a shear point prediction method and apparatus based on a fusion model. By using historical electrical signal parameters, a first model with better predictive accuracy and a second model with better reliability are established. When an ultrasonic scalpel is cutting an object, the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator are acquired. This yields the first output of the first model and the second output of the second model. Based on the operator's prediction requirements, and according to the prediction requirements, the first output, and the second output, a prediction result for the shear point time is obtained, providing a shear point prediction result that meets the requirements. This avoids situations where the operator, anticipating that the object has been cut, experiences excessively high temperatures in the ultrasonic scalpel tip, thus reducing the ultrasonic scalpel's lifespan. Attached Figure Description

[0051] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This application provides a flowchart illustrating a shear point prediction method based on a fusion model.

[0053] Figure 2 This is a schematic diagram of the resonant frequency and current-voltage phase difference as a function of shear, provided in an embodiment of this application.

[0054] Figure 3 This is a schematic diagram of the output current and output voltage as a function of shear, provided in an embodiment of this application.

[0055] Figure 4 This is a schematic diagram of the structure of a fully connected neural network model provided in an embodiment of this application;

[0056] Figure 5 This is a schematic diagram of the structure of the recurrent neural network model provided in the embodiments of this application;

[0057] Figure 6 This application provides a schematic diagram of a combined structure of the first model and the second model in an embodiment;

[0058] Figure 7 This application provides a schematic diagram of the structure of a shear point prediction device based on a fusion model. Detailed Implementation

[0059] In order to accurately predict the time when the object is cut by the ultrasonic scalpel and avoid the operator ending the cutting action too late, which would cause the jaws of the ultrasonic scalpel to wear down the pad, this application provides a shear point prediction method and device based on a fusion model.

[0060] This application provides a shear point prediction method based on a fusion model, such as... Figure 1 As shown, it includes S110 to S160.

[0061] S110: Obtain the variable parameters of the historical electrical signal of the ultrasonic scalpel generator within a preset time period, and preprocess the variable parameters of the historical electrical signal to ensure that the cutting point of the ultrasonic scalpel cutting the object is within the preset time period.

[0062] When an ultrasonic scalpel performs shearing operations, the electrical signal parameters of the ultrasonic scalpel generator exhibit distinct characteristics: the output current I is a constant current, while the output voltage changes with the shearing of the object being cut. For example... Figure 2 The figure shows the curves of the resonant frequency and the phase difference between current and voltage as a function of shearing. To ensure the ultrasonic scalpel operates at its highest efficiency, the phase difference between the current and voltage is typically set to 0. This places the ultrasonic scalpel's operating circuit at the resonant point. As the shearing process proceeds, as... Figure 3 As shown, the output current and output voltage curves change with shear as indicated by the shear point. The actual impedance changes with the load, and the piezoelectric crystal also changes in real time with the operating temperature.

[0063] After being processed in real time by the ultrasonic scalpel's processor, the transducer's resonant frequency is adjusted to ensure that the transducer always operates in its highest efficiency resonant region. As the shearing process completes, the aforementioned voltage and current parameters, such as output voltage, output current, resonant frequency, and phase difference, will undergo subtle changes, as collected in real time by sensors. Therefore, these electrical signal parameters can be sent to the processor for further analysis.

[0064] During the shearing process, the output voltage of an ultrasonic scalpel changes depending on the material the ultrasonic scalpel head contacts. Specifically, the output voltage is relatively low when the ultrasonic scalpel is shearing the object being cut, and increases as the shearing progresses and the scalpel head contacts the Teflon gasket.

[0065] The electrical parameters used in this application—current phase value, voltage phase value, target phase value, and resonant frequency—are closely related to changes in the object being cut. Specifically, during the stable shearing process of the ultrasonic scalpel, the difference between the detected current and voltage phase values ​​and the target phase value is approximately zero, and the rate of change of the resonant frequency is relatively constant. As shearing progresses and the scalpel tip contacts the Teflon gasket, the difference between the current and voltage phase values ​​and the target phase value deviates from zero, the rate of change of the resonant frequency increases, and the waveform exhibits significant fluctuations. Thus, using the resonant frequency as the acquired electrical signal parameter provides a more distinct distinguishing characteristic.

[0066] The historical electrical signal refers to the output current value, output voltage value, current phase value, voltage phase value, target phase value, and resonant frequency. To obtain training data with higher information content and better feature representation, the historical electrical signal can be further processed to obtain its variable parameters. For example, the current phase value and voltage phase value can be subtracted to obtain the current-voltage phase difference; the current-voltage phase difference and the target phase value can be subtracted to obtain the difference between the current-voltage phase difference and the target phase difference; and the first derivative of the resonant frequency can be calculated. Thus, the variable parameters of the historical electrical signal are: output current value, output voltage value, current-voltage phase difference, the difference between the current-voltage phase difference and the target phase difference, and the first derivative of the resonant frequency.

[0067] It should be noted that this application collects the current phase value, voltage phase value, target phase value, and resonant frequency of the ultrasonic scalpel generator because these electrical signal parameters are closely related to changes in the object being cut. In practical applications, it is not limited to collecting only these electrical signal parameters. For example, given the differences in the circuit characteristics of ultrasonic scalpel generators, if other electrical signal parameters closely related to changes in the object being cut exist, they can also be used as collected electrical signal parameters. Furthermore, the collected electrical signal parameters can also include multi-dimensional variables such as temperature, resistance, ultrasonic scalpel parameters, or generator parameters collected by sensors.

[0068] In addition, when determining the variable parameters of historical electrical signals, the ratio of voltage to current can be calculated to obtain the impedance value, and the product of voltage and current can be calculated to obtain the power value. The impedance value and the power value can then be used as variable parameters of historical electrical signals.

[0069] In the subsequent training of the first and second initial models, in order to enable the first and second initial models to better learn the deep features of the training data, this application performs scale scaling on the input training data to scale the variable parameters of the historical electrical signals to a similar scale.

[0070] For example, if the resonant frequency is around 55000Hz and the current-voltage phase difference is within ±180°, it is difficult for the first and second initial models to learn effective features if the first derivative of the resonant frequency and the current-voltage phase difference are directly input into the first and second initial models for training.

[0071] To address this, embodiments of this application employ a scaling method to process the historical electrical signal or its variable parameters. For example, the resonant frequency is first subtracted by 55000Hz, and then multiplied by a scaling factor of 0.1 to obtain a new frequency variable. Thus, the new frequency variable has the same order of magnitude as the phase difference variable.

[0072] Furthermore, during data acquisition, the electrical signal parameters of the acquired ultrasonic scalpel generator exhibit a certain degree of offset. To avoid the influence of this offset on data characteristics, in some embodiments, firstly, the standard mean and standard deviation of each parameter in the historical electrical signal are calculated. Then, the standard mean and standard deviation are used to standardize each parameter in the historical electrical signal. The corresponding standardization model is as follows:

[0073]

[0074] The standardized processing model is as follows: x represents the parameters in the historical electrical signal, μ represents the mean of the parameters in the historical electrical signal, and σ represents the standard deviation of the parameters in the historical electrical signal.

[0075] The variable parameters of the historical electrical signals obtained in this application embodiment are used for model training. In actual applications, a certain amount of data will be acquired. For example, relevant parameters are collected during multiple shearing processes with an ultrasonic scalpel. The preset time needs to include the time when the object being cut is severed (the moment of the cutting point). For example, if the data acquisition time interval is 10ms, the data collected during the preset time is the data points from 400 data points before the cutting point to 50 data points after the cutting point.

[0076] In this embodiment, the acquisition of the electrical signal from the ultrasonic scalpel generator is accompanied by a frequency locking process. Therefore, the acquired electrical signal from the ultrasonic scalpel generator is unstable in the initial stage of acquisition. To address this issue, the preprocessing step for the variable parameters of the historical electrical signal includes: calculating the average value of each parameter in the variable parameters of the historical electrical signal during a third time period, and labeling it as the baseline parameter for each parameter; the third time period is the beginning time period within a preset time period; and adjusting each parameter in the variable parameters of the historical electrical signal based on the baseline parameters. For example, the average value of the parameters acquired in the first 0.5 seconds is calculated, and the average value of the parameters acquired in the first 0.5 seconds is set as the baseline. The parameters acquired after 0.5 seconds are subtracted from this baseline to obtain the valid data.

[0077] Because the electrical signal of the ultrasonic scalpel generator is susceptible to various uncertainties such as temperature and load, resulting in an unstable waveform, some embodiments include a preprocessing step for the variable parameters of the historical electrical signal, which involves using a second exponential moving average model to process the variable parameters of the historical electrical signal. The corresponding exponential moving average model is as follows:

[0078] X t =βX t-1 +(1-β)θ t .

[0079] Where β is the weighting parameter, θ t Let X be the weight parameters of each parameter in the historical electrical signal obtained in the t-th update. t Let be the moving average of the variable parameters of the historical electrical signal obtained from the t-th update. For example, using an average sliding window with a window size of 30 to smooth the signal does not change the original trend, but it is smoother after removing high-frequency small-amplitude fluctuations. In addition, the variable parameters of the historical electrical signal can also be processed by moving average, low-pass filtering, polynomial fitting, locally weighted scatter smoothing, or Kalman filtering.

[0080] S120, mark the results corresponding to the variable parameters of the historical electrical signal in the first time period as true to obtain the first training set; mark the results corresponding to the variable parameters of the historical electrical signal in the second time period as true to obtain the second training set. The first time period and the second time period are within a preset time period, and the first time period is earlier than the second time period.

[0081] In this embodiment, the data collection time for the first training set is earlier; for example, the variable parameters of the historical electrical signal for the first time period are data points from 200 to 100 before the cut-off point. Thus, the model trained using the first training set has better predictive accuracy. The data collection time for the second training set is later; for example, the variable parameters of the historical electrical signal for the second time period are data points from 50 to 50 after the cut-off point. Thus, the model trained using the second training set has more reliable prediction results.

[0082] It should be noted that after marking the results of the variable parameters corresponding to a portion of the historical electrical signals as true, it is also necessary to mark the results of the variable parameters corresponding to the historical electrical signals in other stages as false. Specifically, a result of a variable parameter corresponding to a historical electrical signal being true means that the object being cut was indeed cut; a result of a variable parameter corresponding to a historical electrical signal being false means that the object being cut was not indeed cut. It should be noted that after marking the result of a variable parameter as true, the corresponding variable parameter is considered a real sample; after marking the result of a variable parameter as false, the corresponding variable parameter is considered a spurious sample.

[0083] In this embodiment, a large amount of collected real and valid data needs to be cut and then used to train the model. To take advantage of the training equipment's ability to perform batch training, the training input data needs to have a uniform length. The preprocessing step for the variable parameters of historical electrical signals includes: obtaining the time-sensitive value at the warning cutoff point; and cutting the training data from the first and second training sets based on the time-sensitive value.

[0084] For example, based on the characteristics of the object being cut, the timeliness of the shearing warning must be less than 1 second. Simultaneously, the electrical signal of the ultrasonic scalpel generator exhibits strong temporal correlation. This application sets the shearing time of the training data in the first and second training sets to 1 second, i.e., sets the amount of data with a shearing time of 1 second as the uniform length of the training data. Furthermore, the length of the training data is set to 2... n When the device used to train the model has better parallelism, this application uses 128 as the uniform length of the training input data. On the other hand, the present invention slides and truncates segments of length 128 in the original data with a step size of 16 as the training input for the subsequent first and second initial models, which can effectively increase the number of data samples in the training set, enabling the fusion model to fully learn the deep data features.

[0085] S130, construct a first initial model and a second initial model, and train the first initial model with the first training set to obtain the first model; train the second initial model with the second training set to obtain the second model.

[0086] In this application, the first and second models are structurally similar but with different weights. To obtain the first and second models with different weights, after constructing the first and second initial models, the first and second initial models are first pre-trained using variable parameters from historical electrical signals to obtain basic weights. During the pre-training process, the batch size of the training input is set to 1024, and the learning rate is set to 0.05. Then, the basic weights are loaded, and the first initial model with the loaded basic weights is trained using training data from the first training set. During the training of the first initial model after loading the basic weights, the batch size of the training input is set to 254, the decay parameter is set to 0.01, the learning rate is set to 0.005, and the cross-entropy loss function is set as follows: in, p represents the label values ​​of the real samples in the first training set. t The predicted probability of the true shearing is the output of the first initial model. p represents the label value of the fake samples in the first training set. f α represents the predicted probability of spurious pruning output by the first initial model, and α is the influence factor on the label values ​​of the real samples in the first training set.

[0087] Furthermore, the basic weights are loaded, and the second initial model with the loaded basic weights is trained using the training data from the second training set. During the training of the second initial model with the loaded basic weights, the batch size of the training input is set to 254, the decay parameter is set to 0.01, the learning rate is set to 0.005, and the cross-entropy loss function is set as follows: in, p represents the label values ​​of the real samples in the second training set. t ′ represents the predicted probability of the true shearing from the output of the second initial model. p represents the label value of the fake samples in the second training set. f ' represents the predicted probability of spurious pruning from the output of the second initial model, and α' represents the influence factor on the real samples in the second training set.

[0088] Among them, using the cross-entropy loss function with an impact factor can increase the loss of the label values ​​of real samples, affect the model's learning of real samples, improve the overall accuracy, and avoid the situation where real samples occupy a very small part of the training data, leading to an imbalance in training labels.

[0089] In this embodiment of the application, when α = 1.2 and α′ = 1.2, the first initial model and the second initial model have good training effects. In addition, the loss function is not limited to the cross-entropy loss function, and the mean squared error loss function, the squared loss function, or the absolute value loss function can also be used.

[0090] S140: When the ultrasonic scalpel is cutting the object, acquire the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator.

[0091] It should be noted that before inputting the variable parameters of the real-time electrical signal into the first model and the second model, the variable parameters of the real-time electrical signal need to be preprocessed. The preprocessing process is the same as the processing process for the variable parameters of the historical electrical signal when training the first initial model and the second initial model. That is, referring to the above-mentioned preprocessing process for the variable parameters of the historical electrical signal, the same preprocessing process is performed on the variable parameters of the real-time electrical signal before inputting them into the first model or the second model.

[0092] S150, the variable parameters of the real-time electrical signal are used as inputs to the first model and the second model to obtain the first output of the first model and the second output of the second model.

[0093] The first output and the second output are the probability values ​​that the predicted cut point of the object being cut is true. The probability value that is true can be a probability value in the range of 0-1 or a set of probability values ​​that sum to 1.

[0094] The first output of the first model has better predictive advance capability, meaning it can provide a prediction of the shear point earlier than the actual shear point. The second output of the second model has higher reliability, meaning it provides a more reliable shear point prediction.

[0095] S160: Obtain the prediction requirement, and based on the prediction requirement, the first output, and the second output, obtain the prediction result at the cutoff point.

[0096] Among these, the prediction requirements mainly reflect the operator's expectations for the prediction results. For example, the operator's prediction requirement is to know the predicted cut point time earlier, so as to control the adjustment of the ultrasonic scalpel parameters in advance. Another example is that the operator's requirement is for the cut point prediction results to be more reliable, so as to ensure more accurate operation of the ultrasonic scalpel.

[0097] In practical applications, after using the variable parameters of the real-time electrical signal as input to the first and second models to obtain the first output of the first model and the second output of the second model, the result of the first output can be judged independently. For example, if the first output represents the successful identification of the signal characteristics at the time of cutting, the predicted result of the cutting point time is given, and prompts are given to the operator directly through sound, light, electricity, etc., so that the operator can implement protective measures such as limiting the output power of the ultrasonic scalpel generator.

[0098] Alternatively, after the first output characterizing the signal feature identification at the time of shearing is successful, the result of the second output is judged. If the second output characterizing the signal feature identification at the time of shearing is successful, then the prediction result of the shearing point time is given in a weighted manner according to the prediction requirements.

[0099] The weighted model is: Y = α″ × y1 + (1 - α″) × y2, where α″ is the weight corresponding to the predicted requirement, y1 is the first output, y2 is the second output, and the weight α″ is a set of adjustable values ​​depending on the application scenario. Operators can determine the weight values ​​reflecting the predicted requirement based on the characteristics of the object being cut. For example, setting α″ = 0.9 provides stronger predictability when dealing with thicker objects. Alternatively, setting α″ = 0.7 provides stronger reliability when dealing with thinner objects.

[0100] In other words, the weight of the predicted demand can be determined by the thickness of the object being cut. For example, first, the thickness of the object being cut is obtained, and then, based on the thickness of the object being cut and the preset correspondence, the weight corresponding to the predicted demand is determined, that is, the value of the predicted demand α″ is determined. The preset correspondence includes the weight relationship between the thickness of the object being cut and the predicted demand.

[0101] This application provides a shear point prediction method based on a fusion model. By using historical electrical signal parameters, a first model with better predictive accuracy and a second model with better reliability are established. When an ultrasonic scalpel is cutting an object, the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator are obtained. This yields the first output of the first model and the second output of the second model. Based on the operator's prediction requirements, and according to the prediction requirements, the first output, and the second output, a predicted shear point time is obtained, providing a shear point prediction result that meets the requirements. This avoids situations where the operator, before observing that the object has been cut, causes the ultrasonic scalpel tip to generate high temperatures, reducing the ultrasonic scalpel's lifespan.

[0102] The shear point prediction method based on a fusion model provided in this application embodiment can be a combination of one or more models based on machine learning algorithms or deep learning algorithms. More specifically, the machine learning algorithm may include linear regression, support vector machine, nearest neighbor, decision tree, random forest, or Naive Bayes algorithms, and the deep learning algorithm may include fully connected neural networks, convolutional neural networks, generative adversarial networks, or recurrent neural networks.

[0103] For example, the first model uses a combination of a fully connected neural network model formed from the aforementioned optional range and a recurrent neural network model formed from the above-mentioned recurrent neural network.

[0104] Fully connected neural network models are mathematical models inspired by neurons, consisting of multiple interconnected nodes, and can be used to model complex relationships between data. Fully connected neural network models have a multi-layered structure, with all neurons in each layer interconnected. For example, ... Figure 4 As shown, the fully connected neural network model used in this application embodiment includes an input layer, a hidden layer, and an output layer.

[0105] Furthermore, the fitting ability of a fully connected neural network model is highly correlated with the number of layers and neurons; the more layers and the more neurons, the stronger the mathematical fitting ability of the fully connected neural network model. In the embodiments of this application, the following settings are provided: Figure 4 The two hidden layers shown correspond to the number of output units in each layer of the recurrent neural network. Generally, the number of neurons is set to 8, 16, 32, 64, or 128. In this embodiment, the number of neurons is set to 16.

[0106] In some embodiments, considering that the acquired ultrasonic scalpel electrical signals are highly time-dependent, in order to better handle the temporal information relationships, such as... Figure 5 As shown, a recurrent neural network (RNN) model with memory capabilities can also be used to construct the first or second initial model. RNN models can memorize preceding signals and have a stronger ability to extract features from time-series signals. Thus, the first or second initial model can be constructed using both RNN and fully connected neural network models.

[0107] For example, considering the temporal continuity of the electrical signals of an ultrasonic generator in practical applications, and the stronger correlation of these signals over a short period, in some embodiments, the time interval for acquiring the ultrasonic generator's electrical signals is set to 10 milliseconds. Therefore, the number of loop nodes in the recurrent neural network is designed to be 40. This allows for the simultaneous processing of electrical signals within a 400-millisecond time interval, enhancing the temporal correlation of the output results.

[0108] It should be noted that the first and second models in this application can be various combinations of machine learning algorithm models and deep learning algorithm models, or combinations of algorithm models of the same type. The combination method can be parallel, serial, or cross-functional. For example... Figure 6 The diagram illustrates a specific approach employed in an embodiment of this application, comprising a model structure formed by the parallel fusion of the first and second models constructed above. Specifically, while the two basic model structures are identical, they differ in training data, resulting in different weight parameters and different discrimination results derived from forward inference.

[0109] This application also provides a shear point prediction device based on a fusion model, which is used to execute the above-described shear point prediction method based on a fusion model. Figure 7 As shown, the shear point prediction device based on the fusion model includes:

[0110] The variable parameter acquisition module for historical electrical signals is used to acquire the variable parameters of historical electrical signals of the ultrasonic scalpel generator within a preset time period, and to preprocess the variable parameters of the historical electrical signals so that the cutting point of the ultrasonic scalpel cutting the object being cut is always within the preset time period.

[0111] The variable parameter marking module is used to mark the results corresponding to the variable parameters of the historical electrical signals in the first time period as true, so as to obtain the first training set; and to mark the results corresponding to the variable parameters of the historical electrical signals in the second time period as true, so as to obtain the second training set. The first time period and the second time period are within a preset time period, and the first time period is earlier than the second time period.

[0112] The model building module is used to build a first initial model and a second initial model, and to train the first initial model with a first training set to obtain a first model; and to train the second initial model with a second training set to obtain a second model;

[0113] The real-time electrical signal variable parameter acquisition module acquires the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator when the ultrasonic scalpel is cutting the object.

[0114] The prediction module is used to take the variable parameters of the real-time electrical signal as input to the first model and the second model to obtain the first output of the first model and the second output of the second model.

[0115] The fusion module is used to obtain the prediction requirements and, based on the prediction requirements, the first output, and the second output, to obtain the prediction result at the cutoff point.

[0116] This application provides a shear point prediction method and apparatus based on a fusion model. By using historical electrical signal parameters, a first model with better predictive accuracy and a second model with better reliability are established. When an ultrasonic scalpel is cutting an object, the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator are acquired. This yields the first output of the first model and the second output of the second model. Based on the operator's prediction requirements, and according to the prediction requirements, the first output, and the second output, a prediction result for the shear point time is obtained, providing a shear point prediction result that meets the requirements. This avoids situations where the operator, anticipating that the object has been cut, experiences excessively high temperatures in the ultrasonic scalpel tip, thus reducing the ultrasonic scalpel's lifespan.

[0117] The above specific embodiments further illustrate the purpose, technical solution and beneficial effects of this application. It should be understood that the above are only specific embodiments of this application and are not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of this application should be included within the scope of protection of this application.

Claims

1. A shear point prediction method based on a fusion model, characterized in that, include: The variable parameters of the historical electrical signal of the ultrasonic scalpel generator during a preset time period are obtained, and the variable parameters of the historical electrical signal are preprocessed, wherein the cutting point of the ultrasonic scalpel cutting the object being cut is always within the preset time period. The results corresponding to the variable parameters of the historical electrical signal in the first time period are marked as true to obtain the first training set; the results corresponding to the variable parameters of the historical electrical signal in the second time period are marked as true to obtain the second training set. The first time period and the second time period are within a preset time period, and the first time period is earlier than the second time period. Construct a first initial model and a second initial model, and train the first initial model with the first training set to obtain the first model; train the second initial model with the second training set to obtain the second model; When the ultrasonic scalpel is cutting the object, the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator are obtained; The variable parameters of the real-time electrical signal are used as inputs to the first model and the second model to obtain the first output of the first model and the second output of the second model. Obtain the prediction requirement, and based on the prediction requirement, the first output, and the second output, obtain the prediction result at the shearing point time; The step of obtaining the predicted demand includes: Get the thickness of the object being cut; Based on the thickness of the object being cut and a preset correspondence, the weight corresponding to the predicted demand is determined. The preset correspondence includes the weight relationship between the thickness of the object being cut and the predicted demand.

2. The shear point prediction method based on a fusion model according to claim 1, characterized in that, The following weighted model is used to obtain the prediction results for the shear point time. The weighted model is as follows: Y=α″×y1+(1-α″)×y2; Where α″ is the weight corresponding to the predicted demand, y1 is the first output, and y2 is the second output.

3. The shear point prediction method based on a fusion model according to claim 1, characterized in that, The step of preprocessing the variable parameters of the historical electrical signal includes: Calculate the average value of each parameter in the variable parameters of the historical electrical signal in the third time period, and mark it as the reference parameter of each parameter. The third time period is the start time period in the preset time period. Based on the baseline parameters of each parameter, adjust each parameter in the variable parameters of the historical electrical signal.

4. The shear point prediction method based on a fusion model according to claim 1, characterized in that, The step of preprocessing the variable parameters of the historical electrical signal includes: Based on preset rules, outlier values ​​in the variable parameters of the historical electrical signals are removed.

5. The shear point prediction method based on a fusion model according to claim 1, characterized in that, The step of preprocessing the variable parameters of the historical electrical signal includes: The variable parameters of the historical electrical signal are processed using an exponential moving average model, which is as follows: X t =βX t-1 +(1-β)θ t ; Where β is the weighting parameter, θ t Let X be the weight parameter obtained in the t-th update. t Let be the moving average of the variable parameters of the historical electrical signal obtained in the t-th update.

6. The shear point prediction method based on a fusion model according to claim 1, characterized in that, The step of preprocessing the variable parameters of the historical electrical signal includes: Obtain the timeliness value at the warning cutoff point; Based on the timeliness value, the training data in the first training set and the second training set are cut.

7. The shear point prediction method based on a fusion model according to claim 1, characterized in that, After constructing the first and second initial models, the following steps are also included: The first and second initial models are pre-trained using the variable parameters of historical electrical signals to obtain basic weights. The batch size of the training input is set to 1024, and the learning rate is set to 0.

05.

8. The shear point prediction method based on a fusion model according to claim 7, characterized in that, The step of training the first initial model with the first training set to obtain the first model is as follows: Load the basic weights, train the first initial model using the first training set, set the batch size of the training input to 254, set the decay parameter to 0.01, set the learning rate to 0.005, and set the cross-entropy loss function as follows: in, p represents the label values ​​of the real samples in the first training set. t The predicted probability of the true shearing is the output of the first initial model. p represents the label value of the fake samples in the first training set. f The predicted probability of spurious pruning is the output of the first initial model, and α is the influence factor on the label values ​​of the real samples in the first training set. The step of training the second initial model with the second training set to obtain the second model is as follows: Load the base weights, train the second initial model using the second training set, set the batch size of the training input to 254, set the decay parameter to 0.01, set the learning rate to 0.005, and set the cross-entropy loss function as follows: in, p represents the label values ​​of the real samples in the second training set. t ′ represents the predicted probability of the true shearing from the output of the second initial model. p represents the label value of the fake samples in the second training set. f ' represents the predicted probability of spurious pruning from the output of the second initial model, and α' represents the influence factor on the real samples in the second training set.

9. The shear point prediction method based on a fusion model according to claim 1, characterized in that, The variable parameters of the historical electrical signal and the variable parameters of the real-time electrical signal include: output current value, output voltage value, current-voltage phase difference, the difference between the current-voltage phase difference and the target phase difference, and the first derivative of the resonant frequency.

10. A shear point prediction device based on a fusion model, characterized in that, The shear point prediction device based on the fusion model is used to execute the shear point prediction method based on the fusion model according to any one of claims 1-9, wherein the shear point prediction device based on the fusion model includes: The variable parameter acquisition module for historical electrical signals is used to acquire the variable parameters of historical electrical signals of the ultrasonic scalpel generator within a preset time period, and to preprocess the variable parameters of the historical electrical signals, wherein the cutting point of the ultrasonic scalpel cutting the object being cut is always within the preset time period. The variable parameter marking module is used to mark the results corresponding to the variable parameters of the historical electrical signals in the first time period as true, so as to obtain the first training set; and to mark the results corresponding to the variable parameters of the historical electrical signals in the second time period as true, so as to obtain the second training set, wherein the first time period and the second time period are within a preset time period, and the first time period is earlier than the second time period. The model building module is used to build a first initial model and a second initial model, and to train the first initial model with a first training set to obtain a first model; and to train the second initial model with a second training set to obtain a second model; The real-time electrical signal variable parameter acquisition module acquires the variable parameters of the real-time electrical signal of the ultrasonic scalpel generator when the ultrasonic scalpel is cutting the object. The prediction module is used to take the variable parameters of the real-time electrical signal as input to the first model and the second model to obtain the first output of the first model and the second output of the second model. The fusion module is used to obtain the prediction requirement and, based on the prediction requirement, the first output, and the second output, obtain the prediction result at the shearing point time. The fusion module is also used to obtain the thickness of the object being cut; Based on the thickness of the object being cut and a preset correspondence, the weight corresponding to the predicted demand is determined. The preset correspondence includes the weight relationship between the thickness of the object being cut and the predicted demand.

Images