Stirring machine rotating speed control model, method, device, equipment and medium

Abstract

The invention belongs to the technical field of stirrer control, and discloses a stirrer rotating speed control model, method, device, equipment and medium, and the model comprises a Transform variant module, a multi-mode cross attention module and a full connection module. The Transform variant module is constructed on the basis of a local time window self-attention mechanism, and the local time window self-attention mechanism is used for capturing a short-term local dependency relationship of material characteristic data on different time steps; the multi-modal cross attention module is constructed on the basis of a cross attention mechanism, and the cross attention mechanism is used for performing multi-modal fusion on the feature data output by the Transform variant module so as to capture relevance among different material characteristics; and the full connection module is used for obtaining the rotating speed of the stirrer according to the characteristic data output by the multi-mode cross attention module and outputting the rotating speed of the stirrer. The stirring speed can be flexibly adjusted, and the uniformity of materials and the process stability are ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of mixer control, and in particular relates to a mixer speed control model, method, device, equipment and medium. Background Art

[0002] The process flow of the polyvinyl chloride (PVC) feeding and conveying system is as follows: the material is transported to the designated ton bag unpacking equipment by a crane, and the ton bag unpacking machine unpacks and discharges the material. The butterfly valve controls the discharge switch and the rotary valve controls the discharge amount to the accelerator. Then, the material passes through the accelerator and is conveyed by the Roots blower under dilute phase positive pressure, enters the powder tank, and the discharge amount to the accelerator is controlled by the rotary valve. Subsequently, the material passes through the accelerator and is extracted to the vacuum feeder by the Roots blower under negative pressure, and then is transported to the glue liquid metering tank. The glue liquid metering tank is preheated to T1℃. When the material reaches a certain amount, it starts to heat and stir. The temperature is heated to T2℃. After stirring evenly, the pneumatic butterfly valve below is opened and the material is transported to the next workshop process through the steam heating pipeline.

[0003] In the entire process, the glue metering tank is the key final link. Its stirring process not only directly affects the uniformity of the material, but also is related to the quality control of the subsequent processes. Stirring is not only to make the material reach the target temperature, but also to keep the material uniform during the two-hour heating process. Therefore, the precise control of the stirring speed in the glue metering tank becomes a key factor to ensure the stability of material properties and process consistency.

[0004] In the prior art, the stirring speed is usually adjusted by fixed speed, segmented speed regulation or feedback control. In the fixed speed system, the stirring speed is a constant value preset by the operator based on experience or process requirements, and the whole process does not consider the changes in material properties; the segmented speed regulation system sets multiple predefined speeds according to different stages of the stirring process. For example, a lower speed is used when the viscosity of the material is high, and a higher speed is switched to ensure uniform mixing when the viscosity decreases; the feedback control system dynamically adjusts the stirring speed by monitoring the feedback signal. When the load is too high, the system will automatically reduce the speed to avoid overload, and increase the speed when the load decreases.

[0005] Therefore, the existing stirring system relies on preset rules or a single signal, cannot flexibly respond to changes in material properties, cannot automatically adjust the stirring speed, and has a poor stirring effect. Summary of the invention

[0006] The purpose of the present invention is to provide a mixer speed control model, method, device, equipment and storage medium, which can solve the problems of being unable to flexibly respond to changes in material properties, being unable to automatically adjust the stirring speed, and having poor stirring effect.

[0007] The first aspect of the present invention discloses a mixer speed control model, comprising:

[0008] Transformer variant modules, multimodal cross-attention modules, and fully connected modules;

[0009] The Transformer variant module is constructed based on a local time window self-attention mechanism, which is used to capture the short-term local dependencies of various material characteristic data at different time steps;

[0010] The multimodal cross-attention module is constructed based on a cross-attention mechanism, which is used to perform multimodal fusion on the feature data output by the Transformer variant module to capture the correlation between different material characteristics;

[0011] The fully connected module is used to obtain the mixer speed based on the feature data output by the multimodal cross-attention module and output the mixer speed.

[0012] In some embodiments, a preprocessing module is further included, wherein the preprocessing module is used to preprocess the input material characteristic data and then input it into the Transformer variant module, wherein the preprocessing module includes a missing value filling unit, an abnormal value replacement unit and a normalization unit;

[0013] The missing value filling unit is used to fill the missing values ​​in the material characteristic data by using a spline interpolation method;

[0014] The abnormal value replacement unit is used to calculate the standard deviation of the material characteristic data, perform abnormal value detection according to the standard deviation, and obtain the abnormal value in the material characteristic data; use a moving smoothing method to obtain the mean value corresponding to the abnormal value, and replace the abnormal value with the mean value;

[0015] The normalization unit is used to scale the material characteristic data to a target range.

[0016] In some embodiments, the Transformer variant module further includes a linear mapping unit, which is used to linearly map all input material characteristic data to a high-dimensional space.

[0017] In some embodiments, a pooling layer is provided in the multimodal cross-attention module, and the pooling layer is used to perform a pooling operation on the feature data obtained using the cross-attention mechanism to obtain the feature data output by the multimodal cross-attention module.

[0018] In some embodiments, the material characteristic data includes material temperature, material humidity, material viscosity and material level; the expression of the cross attention mechanism is:

[0019]

[0020] in, Respectively represent the temperature characteristic data, humidity characteristic data, viscosity characteristic data and material level characteristic data output by the Transformer variant module; [] represents the splicing operation; is the weight matrix; is the fused feature data.

[0021] In some embodiments, the global self-attention in the Transformer is adjusted to local self-attention to obtain the local time window self-attention mechanism, and the expression of the local time window self-attention mechanism is:

[0022]

[0023] Among them, Q i =W Q1 z i , K i =W K1 z i , V i =W V1 z i ;d k is the dimension of the K vector; Softmax is the activation function; LMask is a local window matrix; z i Material characteristic data.

[0024] A second aspect of the present invention discloses a method for controlling the speed of a mixer, comprising:

[0025] Pre-construct any one of the above-mentioned mixer speed control models;

[0026] Collect the material characteristics of the tank where the mixer is located in real time to obtain a variety of material characteristic data;

[0027] Inputting all the material characteristic data into the mixer speed control model to obtain the mixer speed;

[0028] The speed of the mixer is adjusted according to the mixer speed.

[0029] The third aspect of the present invention discloses a mixer speed control device, comprising:

[0030] A construction module for constructing any one of the above-mentioned mixer speed control models;

[0031] The acquisition module is used to collect the material characteristics of the tank where the mixer is located in real time to obtain a variety of material characteristic data;

[0032] A prediction module, used for inputting all the material characteristic data into the mixer speed control model to obtain the mixer speed;

[0033] The control module is used to adjust the rotation speed of the mixer according to the rotation speed of the mixer.

[0034] The fourth aspect of the present invention discloses an electronic device, comprising a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the mixer speed control method disclosed in the second aspect.

[0035] A fifth aspect of the present invention discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the mixer speed control method disclosed in the second aspect.

[0036] The beneficial effect of the present invention is that a local time window self-attention mechanism is designed to capture the fluctuation of material properties in the short term, respond to the change of material state in time, flexibly adjust the stirring speed, and ensure the uniformity of the material and process stability. Through the cross-attention mechanism, multimodal data such as temperature, humidity, viscosity, and material level are fully integrated to capture the correlation between various material properties, and the influence of various material properties on the stirring speed can be comprehensively considered, thereby improving the efficiency and accuracy of controlling the speed of the mixer. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings herein show specific examples of the technical solutions described in the present invention, and together with the specific implementation methods constitute a part of the specification, and are used to explain the technical solutions, principles and effects of the present invention.

[0038] Unless otherwise specified or defined, the same reference numerals in different drawings represent the same or similar technical features, and the same or similar technical features may also be represented by different reference numerals.

[0039] Figure 1 It is a schematic diagram of the architecture of the mixer speed control model disclosed in the embodiment of the present invention;

[0040] Figure 2 is a flow chart of a method for controlling the speed of a mixer disclosed in an embodiment of the present invention;

[0041] Figure 3 is a structural schematic diagram of a mixer speed control device according to an embodiment of the present invention;

[0042] Figure 4 It is a structural schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0043] Unless otherwise specified or defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art. In the case of combining the technical solution of the present invention with realistic scenarios, all technical and scientific terms used herein may also have meanings corresponding to the purpose of implementing the technical solution of the present invention. The "first, second..." used herein is only used to distinguish the names and does not represent a specific quantity or order. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0044] It should be noted that when a component is considered to be "fixed to" another component, it can be directly fixed to the other component or there can be a central component; when an component is considered to be "connected to" another component, it can be directly connected to the other component or there can be a central component at the same time; when an component is considered to be "installed on" another component, it can be directly installed on the other component or there can be a central component at the same time. When an component is considered to be "set on" another component, it can be directly set on the other component or there can be a central component at the same time.

[0045] Unless otherwise specified or defined, the "said" and "the" used in this document refer to the technical features or technical contents mentioned or described before the corresponding position, and the technical features or technical contents may be the same as or similar to the technical features or technical contents mentioned therein. In addition, the terms "including" and "having" and any variations thereof used in this document are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0046] In order to be able to flexibly adjust the speed of the mixer according to the real-time changes in material properties, ensure the uniformity of the material and process stability, and avoid the problem of poor stirring effect caused by the inability to dynamically adjust the mixer speed. The present invention controls the mixer speed based on multimodal fusion and local time window self-attention mechanism. First, the key material property data such as temperature, humidity, viscosity, and material level of the material in the glue metering tank are collected in real time by sensors, and then the Transformer model based on local self-attention is used to capture the short-term local dependency of each material property data at different time steps, and then the cross-attention mechanism is used to deeply fuse these multimodal information; then, the mixer speed is accurately predicted and adjusted. It is possible to flexibly adjust the stirring speed when the material properties are complex and changeable, ensuring the uniformity and process stability of the material during the entire heating and stirring process, thereby improving production efficiency and product quality.

[0047] Based on the above concept, the embodiment of the present invention designs a network model Multimodal and LocalAttention Transformer (MLAT), which is referred to as the mixer speed control model in this article. Figure 1 As shown in the figure, the model first captures the local dependencies between the characteristics of the materials at different time steps in the short term (i.e., short-term local dependencies) through the local time window self-attention mechanism. Then, the multimodal data is fused through the cross-attention mechanism to obtain the correlation and dependency between the characteristics of the materials, and finally the mixer speed is output through the fully connected layer.

[0048] Specifically, the mixer speed control model consists of a Transformer variant module, a multimodal cross-attention module, and a fully connected module. The Transformer variant module is built based on the local time window self-attention mechanism, which is used to capture the short-term local dependencies of various material characteristic data at different time steps; the multimodal cross-attention module is built based on the cross-attention mechanism, which is used to perform multimodal fusion on the feature data output by the Transformer variant module to capture the correlation between different material characteristics; the fully connected module obtains and outputs the mixer speed based on the feature data output by the multimodal cross-attention module.

[0049] The input material characteristic data is x, and the shape is [B, L, D], where B represents the batch size, L represents the time series length, and D represents the number of material characteristics.

[0050] Considering that the dimensions of each material characteristic data in this embodiment are not uniform, for example, the different collection time intervals of the sensor result in different material characteristic numbers D for each material characteristic data. Therefore, a linear mapping unit is also set in the Transformer variant module to linearly map all input material characteristic data to a high-dimensional space through linear changes, so that all material characteristic data can be represented and processed in a unified manner in the model. The linear change formula is:

[0051] z i =W i x i +b i i∈{η,h,θ,l},

[0052] Among them, x i W represents the input data, with a shape of [B,L,1]. i 、b i are the weight matrix and bias term respectively, z i Represents the feature embedding vector after linear transformation, with a shape of [B, L, E], where E is the dimension after mapping.

[0053] Since the material properties in the glue metering tank will fluctuate in a short period of time, it may directly affect the stirring speed. Therefore, the Transformer variant module first captures the short-term local dependencies of the material properties at different time steps through the local time window self-attention mechanism. Specifically, the global self-attention in the Transformer is adjusted to local self-attention, that is, the self-attention mechanism is applied within the local time window, so that the model only pays attention to the time steps around it, and obtains the local time window self-attention mechanism. The formula of the local time window self-attention mechanism is:

[0054]

[0055] Among them, Q i =W Q1 z i , K i =W K1 z i , V i =W V1 z i ;d k is the dimension of the K vector, used for scaling; Softmax is the activation function, and its mathematical expression is Used to normalize similarity; LMask is a local window matrix. The elements inside the window are 0, that is, the values ​​at these positions do not change, and the elements outside the window are -∞, that is, these positions are masked.

[0056] The stirring speed in the glue metering tank is jointly affected by multiple material characteristics. In order to more accurately predict and adjust the mixer speed, the interaction and dependency between these material characteristics need to be considered. Since the stirring process of this embodiment not only requires the material to reach the target temperature, but also requires the material to be kept uniform during the two-hour heating process, there is an important correlation between the material temperature and other material characteristics (material humidity, material viscosity, and material level), so the temperature and other material characteristics are selected for multimodal fusion. Specifically, a cross-attention mechanism is used for multimodal fusion to capture the correlation between different material characteristics, and by dynamically adjusting the weights between the characteristics, the model can focus on the characteristics that have a greater impact on the stirring speed. The formula for the cross-attention mechanism is as follows:

[0057]

[0058] in, Respectively represent the temperature characteristic data, humidity characteristic data, viscosity characteristic data and material level characteristic data output by the Transformer variant module; [] represents the splicing operation; is the weight matrix; through Q η and K hθl The dot product is used to calculate the similarity between the features, and then the Softmax normalization is used to obtain the weights, and these weights are used to weight V hθl , and finally obtain the fused characteristic data

[0059] This embodiment obtains the fused characteristic data After that, a pooling operation is performed to obtain the feature data output by the multimodal cross attention module. The formula is: Among them, mPool represents the average pooling of the fused feature data obtained by cross-attention calculation.

[0060] Finally, the feature data output by the multimodal cross-attention module is passed to the fully connected module for the final output prediction. The fully connected module compresses the high-dimensional features into the predicted value of the target variable (mixer speed). The formula is:

[0061]

[0062] Among them, W fc is the weight matrix of the fully connected layer; b fc is the bias; ReLU is the activation function, expressed as ReLU(x)=max(0,x). The final output of the mixer speed control model is y^, with a shape of [B,1], which represents the mixer speed.

[0063] To summarize, the mixer speed control model of this embodiment is composed of a local self-attention module, a multimodal cross-attention module and a fully connected module. A local time window self-attention mechanism is designed, which can capture short-term fluctuations in material properties, respond to changes in material state in a timely manner, and flexibly adjust the stirring speed to ensure the uniformity of the material and process stability, thereby avoiding the problem of poor stirring effect caused by the inability to dynamically adjust the speed in the existing method; through the cross-attention mechanism, multimodal data such as temperature, humidity, viscosity, and material level are fully integrated to capture the interrelationships between various material properties, and can comprehensively consider the influence of multiple material properties on the stirring speed, thereby improving the accuracy of predicting the mixer speed.

[0064] In some embodiments, the mixer speed control model also includes a preprocessing module for preprocessing the input material characteristic data and then inputting it into the Transformer variant module. It includes: a missing value filling unit, an outlier replacement unit and a normalization unit. Among them, the missing value filling unit is used to fill the missing values ​​in the material characteristic data using the spline interpolation method; the outlier replacement unit is used to calculate the standard deviation of the material characteristic data, perform outlier detection based on the standard deviation, and obtain the outliers in the material characteristic data; the moving smoothing method is used to calculate the mean corresponding to the outlier, and the mean replaces the outlier; the normalization unit is used to scale the material characteristic data to the same target interval.

[0065] Specifically, the collected material characteristic data is cleaned, missing values, outliers and other problems are handled, and the data is normalized or standardized to ensure data quality and consistency.

[0066] First, during the data collection process, sensor failure or communication interruption may occur, resulting in some missing data. The filling method can be used to preserve the integrity of the data. For example, the spline interpolation method is used for data filling. The interpolation formula is as follows:

[0067] S k (t) = a k +b k (tt k )+c k (tt k ) 2 +d k (tt k ) 3 ,

[0068] Among them, t is the time point to be interpolated, a k , b k , c k , d k are the coefficients to be determined. In the interpolation process, the interpolation polynomial S k (t) The following conditions must be met:

[0069] 1. At each known data point, S k (t k )=y k , S k (t k+1 )=y k+1 , that is, the data values ​​at that moment are known at the two endpoints of each interval;

[0070] 2. At the boundary point t of the adjacent interval k+1 At, the first-order derivative of the interpolation function S ′ k (t) and the second-order derivative S ′ k ′ (t) must be continuous, that is, S ′ k (t i+1 )=S ′ k+1 (t k+1 ), S ′ k ′ (t k+1 )=S ′ k ′ +1 (t k+1 );

[0071] 3. Set boundary conditions for the entire interpolation function to constrain the behavior of the interpolation curve at the left and right ends. ′ 0 ′ (t0) = 0, S ′ n ′ -1 (t n )=0, n represents the total number of known data points.

[0072] Secondly, outliers may appear due to sensor failure, measurement error, external interference, etc. If the outliers are not handled, it may have an adverse effect on subsequent model training or data analysis. You can use standard deviation for outlier detection and use moving smoothing to replace outliers.

[0073] Among them, the standard deviation method is a commonly used statistical method that identifies outliers by calculating the deviation of data from the mean. The specific steps are as follows:

[0074]

[0075] Among them, x qis the qth data point, N is the total number of data points. μ is the mean of the data, and σ is the standard deviation of the data. The criterion for judging an outlier in this embodiment is: if the value of a data point differs from the mean μ by 3σ times, then the point is considered an outlier. That is: if and only if |x q -μ|>3σ, then x q is an outlier.

[0076] After detecting the outliers, use the moving smoothing method to process the outliers. Moving smoothing replaces the outliers by taking the mean or weighted mean of adjacent data points, thereby eliminating the impact of outliers on the data and making the data smoother and more continuous. The specific steps are: select a moving window size W (set to 3 in this embodiment), for each outlier point, take W data points before and after the point, and calculate the average value of these points. Replace the outliers with the mean, the formula is as follows:

[0077]

[0078] Among them, x p is the data point in the window. If it is at the beginning or end of the data sequence, since there are not enough preceding and following data points for smoothing, the existing data points can be used for weighted mean.

[0079] Then, normalize the data and scale it to the target interval [-1,1]. The formula is as follows:

[0080]

[0081] Among them, x s,t represents the sth material characteristic data (temperature, humidity, viscosity, material level) at the tth time step, x s,max and x s,min are the minimum and maximum values ​​of the sth material characteristic data at all time steps.

[0082] The main process of training the mixer speed control model in this embodiment is: using sensors to collect material property data in the glue liquid metering tank in real time, including temperature, humidity, viscosity, material level and stirring speed, to obtain a variety of material property data x i , where i∈{η,h,θ,l}, η represents material temperature, h represents material humidity, θ represents material viscosity, and l represents material level; and obtain label data y, where y represents the speed of the mixer. These material characteristic data and label data constitute training data. For example, real-time collected data and historical collected data can also be used together as training data to train the mixer speed control model.

[0083] In some embodiments, there is no preprocessing module in the mixer speed control model. In this case, the collected data can be subjected to the same preprocessing operations (missing value filling, outlier replacement and data normalization) as the preprocessing module to obtain training data.

[0084] The training data is divided into a training set and a validation set in a ratio of 8:2 to ensure that the training set can cover the material characteristics under different working conditions, while the validation set is used to evaluate the model performance. During multiple rounds of training, the model parameters are continuously optimized by minimizing the loss function. After the training is completed, the validation set is used to evaluate the model performance, and the hyperparameters are adjusted according to the validation set to ensure that the model can accurately predict the mixer speed under different working conditions.

[0085] The loss function designed in this embodiment according to the error between the prediction result of the model and the actual stirring speed, as well as the regularization requirement of the model parameters is:

[0086]

[0087] Where m is the number of samples, is the predicted value of the model, y i is the true label value, W is the model weight matrix, the first part of the loss function is used to minimize the mean square error (MSE) between the predicted stirring speed and the actual value; the second part of the loss function is the L2 regularization term used to limit the complexity of the model to prevent overfitting.

[0088] Based on the mixer speed control model, the embodiment of the present invention further discloses a mixer speed control method, which is used to control the speed of the mixer in the polyvinyl chloride feeding and conveying system. The method can be implemented by computer programming as an independent control system or a part of the control system in the polyvinyl chloride feeding and conveying system. The execution subject of the method can be an electronic device such as a computer, a laptop computer, a tablet computer, or a control chip embedded in an electronic device, and the present invention does not limit this.

[0089] like Figure 2 As shown, the method comprises the following steps:

[0090] Step S100: pre-constructing a mixer speed control model;

[0091] Please refer to the above description for the specific construction process, which will not be repeated here.

[0092] Step S200: collecting the material properties of the tank where the mixer is located in real time to obtain a variety of material property data;

[0093] In this embodiment, the tank where the mixer is located is a glue liquid metering tank. The material property data in the glue liquid metering tank is collected in real time by sensors, including temperature, humidity, viscosity, material level and stirring speed, to obtain a variety of material property data x i , where i∈{η,h,θ,l}, η represents the material temperature, h represents the material humidity, θ represents the material viscosity, and l represents the material level. It should be noted that the types of material characteristic data are not limited to the above description and can be changed as needed.

[0094] Step S300: inputting all material characteristic data into the mixer speed control model to obtain the mixer speed;

[0095] Step S400: adjusting the speed of the mixer according to the mixer speed.

[0096] All material characteristic data are combined into input data of the mixer speed control model, input into the mixer speed control model, obtain the mixer speed predicted by the mixer speed control model and adjust the mixer speed accordingly.

[0097] For example, the mixer speed is compared with the speed threshold range corresponding to the preset process target. If the mixer speed is found to deviate from the speed threshold range, the mixer speed is adjusted through the feedback control mechanism to make corrections, ensuring that the mixer speed always remains within the optimal range.

[0098] The mixer speed control method of this embodiment introduces sensors to collect key material property data such as temperature, humidity, viscosity, and material level of the material in real time, inputs the data into the mixer speed control model, and performs multimodal fusion through the cross-attention mechanism in the mixer speed control model to capture the interrelationship between the various material properties. It can flexibly adjust the stirring speed according to the real-time changes of the material properties to ensure the uniformity and process stability of the material. Through the local time window self-attention mechanism in the mixer speed control model, it is possible to capture the fluctuations of the material properties in the short term, accurately adjust the stirring speed, and enable the system to respond to changes in the material state in a timely manner. Compared with the limitations of the existing methods that rely on fixed speed, segmented speed regulation or single load feedback, it is possible to adjust the speed more flexibly and accurately during the stirring process, thereby significantly improving the stirring efficiency and quality, and can comprehensively consider the influence of various material properties on the stirring speed, and improve the accuracy of the stirring process.

[0099] like Figure 3 As shown, based on the above-mentioned mixer speed control method, an embodiment of the present invention discloses a mixer speed control device, including:

[0100] A construction module 600 is used to construct any one of the above-mentioned mixer speed control models;

[0101] The acquisition module 610 is used to collect the material characteristics of the tank where the mixer is located in real time to obtain a variety of material characteristic data;

[0102] A prediction module 620, for inputting all the material characteristic data into the mixer speed control model to obtain the mixer speed;

[0103] The control module 630 is used to adjust the speed of the mixer according to the speed of the mixer.

[0104] like Figure 4 As shown, an embodiment of the present invention discloses an electronic device, including a memory 401 storing executable program codes and a processor 402 coupled to the memory 401;

[0105] The processor 402 calls the executable program code stored in the memory 401 to execute the mixer speed control method described in the above embodiments.

[0106] The embodiment of the present invention further discloses a computer-readable storage medium storing a computer program, wherein the computer program enables a computer to execute the mixer speed control method described in the above embodiments.

[0107] The purpose of the above embodiments is to exemplarily reproduce and deduce the technical solution of the present invention, and to fully describe the technical solution, purpose and effect of the present invention. Its purpose is to make the public understand the disclosed content of the present invention more thoroughly and comprehensively, and it does not limit the scope of protection of the present invention.

[0108] The above embodiments are not exhaustive enumerations of the present invention, and there may be multiple other implementations not listed. Any replacement and improvement made without violating the concept of the present invention shall fall within the protection scope of the present invention.

Claims

1. A mixer speed control model, characterized in that: include: Transformer variant modules, multimodal cross-attention modules, and fully connected modules; The Transformer variant module is constructed based on a local time window self-attention mechanism, which is used to capture the short-term local dependencies of various material characteristic data at different time steps; The multimodal cross-attention module is constructed based on a cross-attention mechanism, and the cross-attention mechanism is used to perform multimodal fusion on the feature data output by the Transformer variant module to capture the correlation between different material characteristics; The fully connected module is used to obtain the mixer speed based on the feature data output by the multimodal cross-attention module and output the mixer speed.

2. The mixer speed control model according to claim 1, characterized in that: It also includes a preprocessing module, which is used to preprocess the input material characteristic data and then input it into the Transformer variant module, and the preprocessing module includes a missing value filling unit, an abnormal value replacement unit and a normalization unit; The missing value filling unit is used to fill the missing values ​​in the material characteristic data by using a spline interpolation method; The abnormal value replacement unit is used to calculate the standard deviation of the material characteristic data, perform abnormal value detection according to the standard deviation, and obtain the abnormal value in the material characteristic data; use a moving smoothing method to obtain the mean value corresponding to the abnormal value, and replace the abnormal value with the mean value; The normalization unit is used to scale the material characteristic data to a target range.

3. The mixer speed control model according to claim 1, characterized in that: The Transformer variant module also includes a linear mapping unit, which is used to linearly map all input material characteristic data to a high-dimensional space.

4. The mixer speed control model according to claim 1, characterized in that: A pooling layer is provided in the multimodal cross-attention module, and the pooling layer is used to perform a pooling operation on the feature data obtained by adopting the cross-attention mechanism to obtain the feature data output by the multimodal cross-attention module.

5. The mixer speed control model according to claim 1, characterized in that: The material characteristic data includes material temperature, material humidity, material viscosity and material level; the expression of the cross attention mechanism is: in, Respectively represent the temperature characteristic data, humidity characteristic data, viscosity characteristic data and material level characteristic data output by the Transformer variant module; [] represents the splicing operation; is the weight matrix; is the fused feature data.

6. The mixer speed control model according to claim 1, characterized in that: The global self-attention in Transformer is adjusted to local self-attention to obtain the local time window self-attention mechanism. The expression of the local time window self-attention mechanism is: Among them, Q i =W Q1 z i , K i =W K1 z i , V i =W V1 z i ;d k is the dimension of the K vector; Softmax is the activation function; LMask is a local window matrix; z i Material characteristic data.

7. A method for controlling the speed of a mixer, characterized in that: include: Pre-constructing a mixer speed control model as described in any one of claims 1 to 6; Collect the material characteristics of the tank where the mixer is located in real time to obtain a variety of material characteristic data; Inputting all the material characteristic data into the mixer speed control model to obtain the mixer speed; The speed of the mixer is adjusted according to the mixer speed.

8. A mixer speed control device, characterized in that: include: A construction module, used to construct a mixer speed control model as described in any one of claims 1 to 6; The acquisition module is used to collect the material characteristics of the tank where the mixer is located in real time to obtain a variety of material characteristic data; A prediction module, used for inputting all the material characteristic data into the mixer speed control model to obtain the mixer speed; The control module is used to adjust the rotation speed of the mixer according to the rotation speed of the mixer.

9. An electronic device, characterized in that: It comprises a memory storing executable program code and a processor coupled to the memory; the processor calls the executable program code stored in the memory to execute the mixer speed control method according to claim 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program enables a computer to execute the mixer speed control method according to claim 7.

Images