A soil mix control method and system

Abstract

The application discloses a kind of earth material stirring control method and system, it is related to production control field, comprising: collecting several kinds of materials, material time sequence, several kinds of materials to be put, current stirring parameter and current stirring image;The initial feeding time of each material and the material to be put is obtained based on multiple linear regression models prediction, and the initial feeding time of each material to be put is corrected to obtain the feeding time interval according to the initial feeding time of each material and the material time sequence;The feeding time interval of all materials to be put is sampled and combined, and multiple time sequences of materials to be put are obtained, and the curing strength of multiple time sequences of materials to be put predicted according to stacked bidirectional neural network is used to screen multiple time sequences of materials to be put to obtain the first feeding sequence;According to the first stirring feature identified by current stirring image segmentation, determine the preset stirring parameter, and correct the current stirring parameter to obtain the first stirring parameter, and control in combination with the first feeding sequence.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of production control, in particular to a soil mixing control method and system. BACKGROUND

[0002] Flow solidified soil is a kind of solidified soil formed by mixing and stirring to have specific fluidity and strength, which is widely used in municipal engineering and other fields. With the increasing requirements of the engineering field, the production efficiency and solidification strength of flow solidified soil are gradually improved. The primary way to achieve the improvement of the production efficiency and solidification strength of flow solidified soil is to improve and optimize the production process, especially the stirring control process, to achieve more accurate and efficient stirring control.

[0003] At present, the stirring control in the production process of flow solidified soil is mostly based on experience to set fixed control parameters and combined with manual adjustment of the process for control. This method relies on human experience, often leading to lack of basis for optimization and inaccurate control. Therefore, there is also a stirring control method in the prior art to predict the solidification strength under different control parameters or processes and select the control parameter or process with the highest solidification strength for optimization. However, this method can only be determined at the beginning of soil mixing, ignoring the mutual influence between the materials during the mixing process, lacking a certain flexibility, so that the stirring control accuracy is difficult to achieve the expected. Therefore, how to improve the accuracy of stirring control of flow solidified soil to improve the production efficiency of flow solidified soil is still a technical problem to be solved in the prior art. SUMMARY

[0004] The present application provides a soil mixing control method and system to solve the technical problem of insufficient stirring control accuracy of existing flow solidified soil.

[0005] According to a first aspect of the embodiments of the present application, a soil mixing control method is provided, comprising:

[0006] Collecting process execution data of a current stirring process and a current stirring image; wherein the process execution data includes a plurality of materials, a time sequence of material feeding, a plurality of materials to be fed and a current stirring parameter;

[0007] According to the plurality of materials and the plurality of materials to be fed, a plurality of linear regression models are predicted based on a plurality of linear regression models to obtain the initial feeding time of each material and the initial feeding time of each material to be fed, and the initial feeding time of each material to be fed is corrected according to the initial feeding time of each material and the time sequence of material feeding to determine the feeding time interval of each material to be fed;

[0008] The feeding time intervals of all materials to be fed are sampled and combined to obtain a plurality of feeding time sequences, the curing strength of the plurality of feeding time sequences is predicted based on a preset stacked bidirectional neural network, and the first feeding sequence is obtained by screening the plurality of feeding time sequences according to the curing strength.

[0009] The current stirring image is segmented and recognized to obtain a first stirring feature, the corresponding preset stirring parameter is determined according to the similarity of the first stirring feature, and the current stirring parameter is corrected according to the preset stirring parameter to obtain a first stirring parameter.

[0010] The current stirring process is controlled according to the first feeding sequence and the first stirring parameter.

[0011] The process execution data and the current stirring image of the current stirring process are collected first, and the initial feeding time of each fed material and each material to be fed is predicted based on a plurality of linear regression models. The initial feeding time of each fed material and the feeding time sequence are combined to correct the initial feeding time of each material to be fed to obtain a corresponding feeding time interval. The mutual influence of the feeding time of each material in the stirring process is considered, and the flexibility of determining the feeding time of the material to be fed is improved. Then, the feeding time interval of each material to be fed is sampled and combined to obtain a plurality of feeding time sequences. The curing strength is predicted based on a stacked bidirectional neural network, and the first feeding sequence is obtained by screening according to the curing strength. A more flexible feeding time interval can be determined to obtain a more comprehensive curing strength. The first feeding sequence is screened to be more accurate, thereby improving the accuracy when the current stirring process is controlled according to the first feeding sequence. Then, the current stirring image is segmented and recognized to obtain a first stirring feature and determine a preset stirring parameter. The current stirring parameter is corrected according to the preset stirring parameter to obtain a first stirring parameter. The influence of the current stirring on the subsequent control is considered, thereby improving the accuracy of the control when the current stirring process is controlled according to the first feeding sequence and the first stirring parameter.

[0012] In some embodiments of the present application, the initial feeding time of each fed material and the initial feeding time of each material to be fed are predicted based on a plurality of preset linear regression models according to the plurality of fed materials and the plurality of materials to be fed, and the initial feeding time of each material to be fed is corrected according to the initial feeding time of each fed material and the feeding time sequence to determine the feeding time interval of each material to be fed, specifically including:

[0013] According to the several kinds of materials, the initial feeding time of each kind of material is obtained by prediction based on the multiple linear regression models; wherein, the multiple linear regression models are one-to-one corresponding to the material types of the materials and the materials to be fed; when predicting, the linear regression model corresponding to the material type of the material to be fed is selected for prediction;

[0014] The time difference between the initial feeding time of each kind of material and the actual feeding time of each kind of material in the feeding time sequence is calculated to obtain the time deviation of each kind of material;

[0015] According to the several kinds of materials to be fed, the initial feeding time of each kind of material to be fed is obtained by prediction based on the multiple linear regression models; wherein, when predicting, the linear regression model corresponding to the material type of the material to be fed is selected for prediction;

[0016] According to the time deviation of each kind of material, the time fluctuation average value of the several kinds of materials is calculated, and the feeding time interval of each kind of material to be fed is determined based on the initial feeding time of each kind of material to be fed and the time fluctuation average value.

[0017] The application first predicts the initial feeding time of the materials and the materials to be fed based on the multiple linear regression models, then calculates the time difference between the initial feeding time of each kind of material and the actual feeding time obtained from the feeding time sequence, calculates the time fluctuation average value according to the time difference and corrects the initial feeding time of the materials to be fed to determine the feeding time interval, corrects the feeding time of the materials to be fed by the feeding time deviation of the materials, considers the mutual influence of the feeding time in the stirring process, improves the flexibility of the determination of the feeding time of the materials to be fed, and then the first feeding sequence is more accurate when determined according to the feeding time interval, and the accuracy of the subsequent stirring control is improved.

[0018] In some embodiments of the application, the construction process of the multiple linear regression models is as follows:

[0019] A plurality of historical stirring data is obtained; wherein, each piece of historical stirring data includes material type, feeding time and measured curing strength;

[0020] The plurality of historical stirring data is repeatedly classified according to the material type to obtain multiple groups of model training data; wherein, when repeatedly classifying, the feeding interval between the feeding time of the current classified material type and the feeding time of the previous material in each piece of historical stirring data is calculated, and the feeding interval is taken as a new data column of the model training data;

[0021] Based on the multiple groups of model training data, the multiple linear regression models corresponding are constructed by taking the feeding interval and the measured curing strength as independent variables and the feeding time as dependent variable.

[0022] The application first acquires a plurality of historical stirring data, and repeatedly classifies the data according to material types to obtain a plurality of model training data. When repeatedly classifying, the feeding interval of the feeding time of the current classified material type and the feeding time of the previous material in each historical data is calculated, and the feeding interval is taken as a new data column. Through preprocessing of the data, the data characteristics can be focused, and a basis for subsequent model construction is provided. Then, based on the plurality of model training data, a plurality of linear regression models are constructed with the feeding interval and the measured curing strength as independent variables and the feeding time as a dependent variable. By constructing a linear regression model corresponding to each material type, the feeding time of different types of materials can be predicted, the predicted feeding time of the corresponding type of material can be determined more quickly and conveniently, and a data basis is provided for subsequent steps.

[0023] In some embodiments of the application, the feeding time intervals of all materials to be fed are sampled and combined to obtain a plurality of feeding time sequences. Based on a preset stacked bidirectional neural network, the curing strength of the plurality of feeding time sequences is predicted, and the plurality of feeding time sequences are filtered according to the curing strength to obtain a first feeding sequence. Specifically, the method comprises:

[0024] The feeding time interval of each material to be fed is sampled multiple times to obtain a plurality of feeding sampling times for each material to be fed.

[0025] Based on the plurality of feeding sampling times of each material to be fed, a full combination of the feeding time of each material to be fed is constructed to obtain a plurality of feeding time sequences.

[0026] Based on the stacked bidirectional neural network, the curing strength of each feeding time sequence is predicted to obtain the corresponding predicted curing strength.

[0027] The feeding time sequence corresponding to the highest predicted curing strength is determined as the first feeding sequence.

[0028] The application first samples the feeding time interval of each material to be fed multiple times to obtain a plurality of feeding sampling times, and then constructs a full combination of the feeding time of each material to be fed to obtain a plurality of feeding time sequences. This can determine more abundant feeding time sequences based on more flexible feeding time intervals, and then predict the predicted curing strength of each feeding time sequence through the stacked bidirectional neural network, and determine the first feeding sequence according to the predicted curing strength. This can obtain more comprehensive predicted curing strength through more abundant feeding time sequences, and then more accurately filter the first feeding sequence, thereby improving the accuracy of control according to the first feeding sequence.

[0029] In some embodiments of the present application, the stacked bidirectional neural network is obtained by stacking two layers of bidirectional neural networks, and the stacked bidirectional neural network comprises an input layer, a first bidirectional network layer, a second bidirectional network layer, a fully connected prediction layer, and an output layer; the curing strength of each to-be-cast time series is predicted based on the stacked bidirectional neural network to obtain the corresponding predicted curing strength, specifically comprising:

[0030] The input layer of the stacked bidirectional neural network extracts features of the current input to-be-cast time series to obtain time series input features;

[0031] The first bidirectional network layer of the stacked bidirectional neural network characterizes the time series input features into a first forward hidden state sequence and a first reverse hidden state sequence, and fuses the first forward hidden state sequence and the first reverse hidden state sequence to obtain a first hidden state sequence;

[0032] The second bidirectional network layer of the stacked bidirectional neural network characterizes the first hidden state sequence into a second forward hidden state and a second reverse hidden state, and fuses the second forward hidden state and the second reverse hidden state to obtain a second hidden state;

[0033] The fully connected prediction layer of the stacked bidirectional neural network predicts based on the second hidden state to obtain a prediction feature;

[0034] The output layer of the stacked bidirectional neural network restores the prediction feature to obtain the predicted curing strength corresponding to the current input to-be-cast time series.

[0035] The present application extracts features through the input layer of the stacked neural network, obtains a first hidden state sequence through the first bidirectional network layer, and then obtains a second hidden state through the second bidirectional network layer, which can respectively map, analyze and process the corresponding input data into the required target form, thereby improving the accuracy of the predicted curing strength obtained by predicting and restoring the features of the second hidden state through the fully connected prediction layer and the output layer.

[0036] In some embodiments of the present application, the current stirring image is segmented and identified to obtain a first stirring feature, specifically comprising:

[0037] The current stirring image is subjected to main image segmentation based on a preset image segmentation model to obtain a main region image and a sub-region image of the current stirring image; wherein the main region image is the main body identified by the image segmentation model from the current stirring image, and the sub-region image is a part of the current stirring image other than the main region image;

[0038] perform image enhancement and image denoising on the main region image to obtain a main region enhanced image, and perform image feature extraction on the main region enhanced image to obtain a main feature corresponding to the current stirring image;

[0039] perform image denoising on the auxiliary region image to obtain an auxiliary region denoised image, and perform image feature extraction on the auxiliary region denoised image to obtain a region feature corresponding to the current stirring image;

[0040] splice the main feature and the region feature to obtain a first stirring feature corresponding to the current stirring image.

[0041] The application first performs main image segmentation on the current stirring image based on an image segmentation model to obtain a main region image and an auxiliary region image, performs image enhancement, image denoising, and image feature extraction on the main region image to obtain a main feature, and performs image denoising and image feature extraction on the auxiliary region image to obtain a region feature. The respective processing of images of different segmentation types can meet the image precision requirements of images of different segmentation types, so that the first stirring feature obtained by splicing meets the current control task requirements better, and the accuracy when subsequently correcting and controlling according to the first stirring feature is improved.

[0042] In some embodiments of the application, the first stirring parameter is obtained by determining a preset stirring parameter according to the similarity of the first stirring feature, and correcting the current stirring parameter according to the preset stirring parameter, specifically including:

[0043] calculating the similarity of the first stirring feature and each historical stirring feature in the historical stirring database based on a preset historical stirring database, and taking the historical stirring parameter corresponding to the historical stirring feature with the highest similarity as the preset stirring parameter;

[0044] calculating the weighted average value of each parameter in the preset stirring parameter and the current stirring parameter based on a preset correction weight to obtain the first stirring parameter.

[0045] The application first calculates the similarity of the first stirring feature and each historical stirring feature in the historical stirring database, and determines the preset stirring parameter according to the similarity, thereby improving the accuracy of searching and screening the preset stirring parameter through the similarity. Then, the weighted average value of the current stirring parameter and the preset stirring parameter is calculated based on a preset correction weight to obtain the first stirring parameter, which can meet the parameter correction requirements in different situations, thereby improving the accuracy of the obtained first stirring parameter and improving the accuracy when controlling according to the first stirring parameter.

[0046] According to a second aspect of the embodiments of the present application, a soil mixing control system is provided, comprising a process data acquisition module, a feeding time determination module, a feeding sequence determination module, a mixing parameter determination module, and a mixing process control module;

[0047] The process data acquisition module is configured to acquire process execution data and a current mixing image of a current mixing process, wherein the process execution data comprises a plurality of materials that have been fed, a feeding time sequence, a plurality of materials to be fed, and a current mixing parameter;

[0048] The feeding time determination module is configured to, based on the plurality of materials that have been fed and the plurality of materials to be fed, perform prediction based on a plurality of preset linear regression models to obtain an initial feeding time of each of the materials that have been fed and an initial feeding time of each of the materials to be fed, and correct the initial feeding time of each of the materials to be fed based on the initial feeding time of each of the materials that have been fed and the feeding time sequence to determine a feeding time interval of each of the materials to be fed.

[0049] The feeding sequence determination module is configured to sample and combine the feeding time intervals of all the materials to be fed to obtain a plurality of feeding time sequences, predict a solidification strength of the plurality of feeding time sequences based on a preset stacked bidirectional neural network, and screen the plurality of feeding time sequences based on the solidification strength to obtain a first feeding sequence.

[0050] The mixing parameter determination module is configured to perform segmentation and identification on the current mixing image to obtain a first mixing feature, determine a preset mixing parameter corresponding to the first mixing feature based on a similarity of the first mixing feature, and correct the current mixing parameter based on the preset mixing parameter to obtain a first mixing parameter.

[0051] The mixing process control module is configured to control the current mixing process based on the first feeding sequence and the first mixing parameter.

[0052] In some embodiments of the present application, the feeding time determination module comprises a fed time prediction unit, a time deviation calculation unit, a to-be-fed time prediction unit, and a time interval determination unit.

[0053] The fed time prediction unit is configured to, based on the plurality of materials that have been fed, perform prediction based on the plurality of linear regression models to obtain an initial feeding time of each of the materials that have been fed, wherein the plurality of linear regression models correspond one-to-one to material types of the materials that have been fed and the materials to be fed; and when performing prediction, a linear regression model corresponding to a material type of the material that has been fed is selected for prediction.

[0054] The time deviation calculation unit is configured to calculate a time difference between the initial feeding time of each fed material and the actual feeding time of each fed material in the feeding time sequence, to obtain a time deviation of each fed material.

[0055] The to-be-fed time prediction unit is configured to predict, based on the plurality of linear regression models, the initial feeding time of each to-be-fed material according to the plurality of to-be-fed materials; and when predicting, a linear regression model corresponding to the material type of the to-be-fed material is selected for prediction.

[0056] The time interval determination unit is configured to calculate a time fluctuation average value of the plurality of fed materials according to the time deviation of each fed material, and determine a feeding time interval of each to-be-fed material based on the initial feeding time of each to-be-fed material according to the time fluctuation average value.

[0057] In some embodiments of the present application, the construction process of the plurality of linear regression models is specifically as follows:

[0058] A plurality of historical mixing data are obtained; each piece of historical mixing data includes a material type, a feeding time and a measured curing strength;

[0059] The plurality of historical mixing data are repeatedly classified according to the material type, to obtain a plurality of groups of model training data; when repeatedly classifying, a feeding interval between the feeding time of the material type of the current classification in each piece of historical mixing data and the feeding time of the previous material is calculated, and the feeding interval is taken as a new data column of the model training data;

[0060] Based on the plurality of groups of model training data, a plurality of linear regression models corresponding thereto are constructed, with the feeding interval and the measured curing strength as independent variables, and with the feeding time as a dependent variable.

[0061] In some embodiments of the present application, the feeding sequence determination module includes a time sampling unit, a feeding combination unit, a strength prediction unit and a sequence determination unit.

[0062] The time sampling unit is configured to sample the feeding time interval of each to-be-fed material multiple times respectively, to obtain a plurality of feeding sampling times of each to-be-fed material.

[0063] The feeding combination unit is configured to construct a full combination of the feeding time of each to-be-fed material according to the plurality of feeding sampling times of each to-be-fed material, to obtain a plurality of to-be-fed time sequences.

[0064] The strength prediction unit is configured to predict the curing strength of each to-be-fed time sequence based on the stacked bidirectional neural network, to obtain a corresponding predicted curing strength.

[0065] The sequence determination unit is configured to determine a time sequence of the highest predicted solidification intensity as a first feeding sequence.

[0066] In some embodiments of the present application, the stacked bidirectional neural network is obtained by stacking two layers of bidirectional neural networks, and the stacked bidirectional neural network comprises an input layer, a first bidirectional network layer, a second bidirectional network layer, a fully connected prediction layer, and an output layer; the feeding sequence determination module comprises a feature extraction unit, a first feature representation unit, a second feature representation unit, a feature prediction unit, and a feature output unit;

[0067] The feature extraction unit is configured to perform feature extraction on the currently input time sequence of the feeding time through the input layer of the stacked bidirectional neural network to obtain time sequence input features.

[0068] The first feature representation unit is configured to represent the time sequence input features as a first forward hidden state sequence and a first reverse hidden state sequence through the first bidirectional network layer of the stacked bidirectional neural network, and fuse the first forward hidden state sequence and the first reverse hidden state sequence to obtain a first hidden state sequence.

[0069] The second feature representation unit is configured to represent the first hidden state sequence as a second forward hidden state and a second reverse hidden state through the second bidirectional network layer of the stacked bidirectional neural network, and fuse the second forward hidden state and the second reverse hidden state to obtain a second hidden state.

[0070] The feature prediction unit is configured to perform prediction based on the second hidden state through the fully connected prediction layer of the stacked bidirectional neural network to obtain predicted features.

[0071] The feature output unit is configured to perform feature restoration on the predicted features through the output layer of the stacked bidirectional neural network to obtain predicted solidification intensity corresponding to the currently input time sequence of the feeding time.

[0072] In some embodiments of the present application, the stirring parameter determination module comprises a stirring image recognition unit; the stirring image recognition unit comprises a stirring image segmentation subunit, a first feature extraction subunit, a second feature extraction subunit, and a stirring feature acquisition subunit.

[0073] The stirring image segmentation subunit is configured to perform main image segmentation on a current stirring image based on a preset image segmentation model to obtain a main region image and a sub-region image of the current stirring image; wherein the main region image is a main body obtained by the image segmentation model from the current stirring image, and the sub-region image is a part of the current stirring image other than the main region image.

[0074] The first feature extraction subunit is configured to perform image enhancement and image denoising on the main region image to obtain a main region enhanced image, and perform image feature extraction on the main region enhanced image to obtain a main feature corresponding to the current stirring image.

[0075] The second feature extraction subunit is configured to perform image denoising on the auxiliary region image to obtain an auxiliary region denoised image, and perform image feature extraction on the auxiliary region denoised image to obtain a region feature corresponding to the current stirring image.

[0076] The stirring feature acquisition subunit is configured to splice the main feature and the region feature to obtain a first stirring feature corresponding to the current stirring image.

[0077] In some embodiments of the present application, the stirring parameter determination module comprises a stirring parameter correction unit; the stirring parameter correction unit comprises a preset parameter determination subunit and a weighted parameter correction subunit.

[0078] The preset parameter determination subunit is configured to calculate the similarity between the first stirring feature and each historical stirring feature in a preset historical stirring database based on the preset historical stirring database, and take the historical stirring parameter corresponding to the historical stirring feature with the highest similarity as a preset stirring parameter.

[0079] The weighted parameter correction subunit is configured to calculate the weighted average value of each parameter in the preset stirring parameter and the current stirring parameter based on a preset correction weight to obtain a first stirring parameter.

[0080] The present application first collects the process execution data and the current stirring image of the current stirring process, and predicts the initial feeding time of each fed material and each to-be-fed material based on multiple linear regression models, then combines the initial feeding time of the fed material and the initial feeding time of the to-be-fed material to correct the initial feeding time of the to-be-fed material to obtain a corresponding feeding time interval, considers the mutual influence of the feeding time of each material in the stirring process, and can improve the flexibility of determining the feeding time of the to-be-fed material; then the feeding time interval of each to-be-fed material is sampled and combined to obtain multiple to-be-fed time sequences, the solidification strength is predicted based on the stacked bidirectional neural network, and the first feeding sequence is obtained by screening according to the solidification strength, which can determine more abundant feeding time sequences based on more flexible feeding time intervals, can predict more comprehensive solidification strength, and then screen the first feeding sequence more accurately, thereby improving the accuracy when controlling according to the first feeding sequence; then the current stirring image is segmented and recognized to obtain a first stirring feature and determine a preset stirring parameter, and then the preset stirring parameter is corrected to obtain a first stirring parameter, which considers the influence of the current stirring situation on subsequent control, thereby improving the accuracy of control when the current stirring process is controlled according to the first feeding sequence and the first stirring parameter. Attached Figure Description

[0081] Figure 1 This is a schematic flowchart illustrating a soil mixing control method according to certain embodiments of this application.

[0082] Figure 2 This is a block diagram of a soil mixing control system according to certain embodiments of this application.

[0083] Figure 3 This is a network structure diagram of a stacked bidirectional neural network shown in some embodiments of this application. Detailed Implementation

[0084] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below in conjunction with the accompanying drawings are exemplary and are only used to explain some embodiments of this application, and should not be construed as limiting the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments shown in this application without inventive effort are within the protection scope of this application.

[0085] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, unless otherwise explicitly specified, "a plurality of" or "several" means two or more.

[0086] Currently, the production process of fluidized bed solidified soil mostly relies on experience to set fixed control parameters, combined with manual adjustments to the mixing process. This dependence on human experience makes it difficult to optimize the process, and optimization lacks a basis, resulting in imprecise control. To address the control dilemma caused by human experience, existing mixing control methods predict the solidification strength under different control parameters or processes and select the control parameter or process with the highest solidification strength for optimization. However, this method can only be determined at the beginning of soil mixing, ignoring the interaction between materials during mixing and lacking flexibility, thus making it difficult to achieve the expected mixing control accuracy. Therefore, how to improve the accuracy of mixing control for fluidized bed solidified soil to improve its production efficiency remains a technical problem that urgently needs to be solved.

[0087] Based on the above technical background, please refer to Figure 1The embodiment of the present application provides a soil mixing control method, comprising steps S101 to S105, and each step is specifically as follows:

[0088] Step S101: collecting process execution data of a current mixing process and a current mixing image; wherein the process execution data comprises several kinds of materials, a time sequence of material feeding, several kinds of materials to be fed and current mixing parameters.

[0089] Specifically, when the current mixing image of the current mixing process is collected, a high-speed industrial camera can be used for shooting to obtain an image with a resolution meeting the requirements of subsequent tasks.

[0090] Step S102: according to the several kinds of materials and the several kinds of materials to be fed, performing prediction based on a plurality of preset linear regression models to obtain an initial feeding time of each kind of material and an initial feeding time of each kind of material to be fed, and correcting the initial feeding time of each kind of material to be fed according to the initial feeding time of each kind of material and the time sequence of material feeding to determine a feeding time interval of each kind of material to be fed.

[0091] In some embodiments of the present application, according to the several kinds of materials and the several kinds of materials to be fed, performing prediction based on a plurality of preset linear regression models to obtain an initial feeding time of each kind of material and an initial feeding time of each kind of material to be fed, and correcting the initial feeding time of each kind of material to be fed according to the initial feeding time of each kind of material and the time sequence of material feeding to determine a feeding time interval of each kind of material to be fed, specifically comprises:

[0092] According to the several kinds of materials, performing prediction based on the plurality of linear regression models to obtain an initial feeding time of each kind of material; wherein the plurality of linear regression models are one-to-one corresponding to material types of the materials and the materials to be fed; when prediction, a linear regression model corresponding to a material type of the material is selected for prediction;

[0093] Calculating a time difference between the initial feeding time of each kind of material and an actual feeding time of each kind of material in the time sequence of material feeding to obtain a time deviation of each kind of material;

[0094] According to the several kinds of materials to be fed, performing prediction based on the plurality of linear regression models to obtain an initial feeding time of each kind of material to be fed; wherein when prediction, a linear regression model corresponding to a material type of the material to be fed is selected for prediction;

[0095] According to the time deviation of each kind of material, calculating an average value of time fluctuation of the several kinds of materials, and according to the average value of time fluctuation, determining a feeding time interval of each kind of material to be fed based on the initial feeding time of each kind of material to be fed.

[0096] In some embodiments of the present application, the time fluctuation average value of the several materials is calculated according to the time deviation of each material, specifically, the arithmetic average value of the time deviation of the several materials is taken as the time fluctuation average value.

[0097] In some embodiments of the present application, the feeding time interval of each material is determined according to the time fluctuation average value and the initial feeding time of each material, specifically:

[0098] The initial feeding time of each material is taken as the origin of the feeding time interval, and the time fluctuation average value is taken as the boundary value farthest from the origin in the feeding time interval, to obtain the feeding time interval of each material.

[0099] Specifically, if the time fluctuation average value in the current mixing process is , the initial feeding time of the current material predicted by the linear regression model is , the feeding time interval of the current material is , and the interval length is . It is easy to understand that, since the time fluctuation average value in the current mixing process is a unique value, the interval length of the feeding time interval of each material is .

[0100] The present application first predicts the initial feeding time of the materials and the materials to be fed based on multiple linear regression models, then calculates the time difference between the initial feeding time and the actual feeding time of each material, calculates the time fluctuation average value according to the time difference and corrects the initial feeding time of the material to be fed, determines the feeding time interval, and corrects the feeding time of the material to be fed by the feeding time deviation of the material, which considers the mutual influence of the feeding time in the mixing process, improves the flexibility of the determination of the feeding time of the material to be fed, and further improves the accuracy of the subsequent determination of the first feeding sequence, and improves the accuracy of the subsequent mixing control.

[0101] In some embodiments of the present application, the construction process of the multiple linear regression models is specifically:

[0102] Obtain multiple historical mixing data; wherein each historical mixing data includes material type, feeding time and measured curing strength;

[0103] The plurality of historical mixing data is repeatedly classified according to the material types to obtain a plurality of groups of model training data; wherein, when repeatedly classifying, the feeding interval of the feeding time of the current classified material type and the feeding time of the previous material in each historical mixing data is calculated, and the feeding interval is taken as a new data column of the model training data;

[0104] Based on the plurality of groups of model training data, a plurality of linear regression models are constructed with the feeding interval and the measured curing strength as independent variables and the feeding time as a dependent variable.

[0105] In some embodiments of the present application, the historical mixing data further includes material proportions; when obtaining the plurality of historical mixing data, the material proportions column of each historical mixing data is removed.

[0106] The present application first obtains a plurality of historical mixing data, and repeatedly classifies the historical mixing data according to the material types to obtain a plurality of groups of model training data; when repeatedly classifying, the feeding interval of the feeding time of the current classified material type and the feeding time of the previous material in each historical data is calculated, and the feeding interval is taken as a new data column; through the preprocessing of the data, the data characteristics can be focused, and the basis for subsequent model construction is provided, and then based on the plurality of groups of model training data, a plurality of linear regression models are constructed with the feeding interval and the measured curing strength as independent variables and the feeding time as a dependent variable; by constructing the linear regression model corresponding to each material type, the feeding time of different types of materials can be predicted respectively, the predicted feeding time of the corresponding type of material can be determined more quickly and conveniently, and the data basis for the subsequent is provided.

[0107] Step S103: sampling and combining the feeding time intervals of all materials to be fed to obtain a plurality of to-be-fed time sequences, predicting the curing strength of the plurality of to-be-fed time sequences based on a preset stacked bidirectional neural network, and screening the plurality of to-be-fed time sequences according to the curing strength to obtain a first feeding sequence.

[0108] In some embodiments of the present application, the sampling and combining of the feeding time intervals of all materials to be fed to obtain a plurality of to-be-fed time sequences, the prediction of the curing strength of the plurality of to-be-fed time sequences based on a preset stacked bidirectional neural network, and the screening of the plurality of to-be-fed time sequences according to the curing strength to obtain a first feeding sequence specifically include:

[0109] sampling the feeding time interval of each to-be-fed material multiple times to obtain a plurality of feeding sampling times of each to-be-fed material;

[0110] constructing the full combination of the feeding time of each to-be-fed material according to the plurality of feeding sampling times of each to-be-fed material to obtain a plurality of to-be-fed time sequences;

[0111] predict the curing strength of each of the to-be-put time series based on the stacked bidirectional neural network to obtain a corresponding predicted curing strength;

[0112] determine the to-be-put time series corresponding to the highest predicted curing strength as the first feeding sequence.

[0113] In some embodiments of the present application, the preferred value of the sampling number is 10.

[0114] In some embodiments of the present application, the sampling methods include but are not limited to random sampling, uniform sampling, or sampling based on function-determined sampling time, and the preferred embodiment is uniform sampling.

[0115] To specifically describe the multiple feeding sampling times of each to-be-put material, the full combination of the feeding time of each to-be-put material is constructed to obtain the implementation process of multiple to-be-put time series, taking to-be-put materials A, B and C as examples, wherein the sampling number is 2, the feeding sampling time corresponding to to-be-put material A is , the feeding sampling time corresponding to to-be-put material B is , and the feeding sampling time corresponding to to-be-put material C is Therefore, the full combination of the feeding time of all to-be-put materials includes 8 kinds, which are specifically:

[0116] .

[0117] The present application first samples the feeding time interval of each to-be-put material multiple times to obtain multiple feeding sampling times, then constructs the full combination of the feeding time of each to-be-put material to obtain multiple to-be-put time series, which can determine more abundant to-be-put time series based on more flexible feeding time intervals, and then predict the predicted curing strength of each to-be-put time series through the stacked bidirectional neural network, and determine the first feeding sequence according to the predicted curing strength, which can obtain more comprehensive predicted curing strength through more abundant to-be-put time series, and then more accurately screen the first feeding sequence, thereby improving the accuracy of control according to the first feeding sequence.

[0118] In some embodiments of the present application, the stacked bidirectional neural network is obtained by stacking two layers of bidirectional neural networks, and the stacked bidirectional neural network includes an input layer, a first bidirectional network layer, a second bidirectional network layer, a fully connected prediction layer and an output layer; the curing strength of each to-be-put time series is predicted based on the stacked bidirectional neural network to obtain a corresponding predicted curing strength, specifically including:

[0119] extracting features of the current input to-be-put time series through the input layer of the stacked bidirectional neural network to obtain time series input features;

[0120] characterize the time sequence input feature into a first forward hidden state sequence and a first reverse hidden state sequence through a first bidirectional network layer of the stacked bidirectional neural network, and fuse the first forward hidden state sequence and the first reverse hidden state sequence to obtain a first hidden state sequence;

[0121] characterize the first hidden state sequence into a second forward hidden state and a second reverse hidden state through a second bidirectional network layer of the stacked bidirectional neural network, and fuse the second forward hidden state and the second reverse hidden state to obtain a second hidden state;

[0122] perform prediction based on the second hidden state through a full connection prediction layer of the stacked bidirectional neural network to obtain a predicted feature;

[0123] perform feature restoration on the predicted feature through an output layer of the stacked bidirectional neural network to obtain a predicted solidification strength corresponding to the input time sequence of the to-be-cast.

[0124] In some embodiments of the present application, the first bidirectional network layer and the second bidirectional network layer in the stacked bidirectional neural network are the same type of bidirectional neural network, which can be a bidirectional RNN or a bidirectional LSTM, and the preferred embodiment is a bidirectional LSTM, i.e., Bi-LSTM.

[0125] In some embodiments of the present application, please refer to Figure 3 a network structure diagram of a stacked bidirectional neural network provided by the present application, wherein Input represents an input layer, Bi-LSTM-1 represents a first bidirectional network layer, Bi-LSTM-2 represents a second bidirectional network layer, FC represents a full connection prediction layer, and Output represents an output layer. Exemplarily, the input layer Input extracts the input time sequence feature of the to-be-cast into a time sequence input feature; the time sequence input feature is processed and characterized into a first forward hidden state sequence and a first reverse hidden state sequence through the first bidirectional network layer Bi-LSTM-1, and is spliced and fused to obtain a first hidden state sequence, which is equal in length to the time sequence input feature; the first hidden state sequence is processed and characterized into a second forward hidden state sequence and a second reverse hidden state sequence through the second bidirectional network layer Bi-LSTM-2, and the final hidden states of the second forward hidden state sequence and the second reverse hidden state sequence are taken respectively to obtain a second forward hidden state and a second reverse hidden state, and then spliced and fused to obtain a second hidden state, which is a fixed-length vector; after the second hidden state is predicted through a linear activation function through the full connection prediction layer FC, the predicted feature is restored and output through the output layer Output to obtain a predicted solidification strength.

[0126] The application extracts features through the input layer of the stacked neural network, processes to obtain a first hidden state sequence through a first bidirectional network layer, and processes to obtain a second hidden state through a second bidirectional network layer, so as to respectively map and analyze and process the corresponding input data into a corresponding required target form. When the second hidden state is predicted and the features are restored and output through the full connection prediction layer and the output layer, the accuracy of the obtained predicted solidification strength is improved.

[0127] Step S104: segmenting and identifying the current stirring image to obtain a first stirring feature, determining a corresponding preset stirring parameter according to the similarity of the first stirring feature, and correcting the current stirring parameter according to the preset stirring parameter to obtain a first stirring parameter.

[0128] In some embodiments of the application, the segmenting and identifying the current stirring image to obtain a first stirring feature specifically includes:

[0129] segmenting a main image of the current stirring image based on a preset image segmentation model to obtain a main region image and a sub-region image of the current stirring image; wherein the main region image is a main body obtained by the image segmentation model identifying the current stirring image, and the sub-region image is a part of the current stirring image other than the main region image;

[0130] performing image enhancement and image denoising on the main region image to obtain a main region enhanced image, and performing image feature extraction on the main region enhanced image to obtain a main body feature corresponding to the current stirring image;

[0131] performing image denoising on the sub-region image to obtain a sub-region denoised image, and performing image feature extraction on the sub-region denoised image to obtain a region feature corresponding to the current stirring image;

[0132] splicing the main body feature and the region feature to obtain a first stirring feature corresponding to the current stirring image.

[0133] In some embodiments of the application, the image segmentation model can be CNN, RCNN or YOLO v5, and the preferred embodiment is RCNN model.

[0134] In some embodiments of the application, the image enhancement is based on an image enhancement model, the image enhancement model can be YOLO v5, GAN or UNet, and the preferred embodiment is GAN model; the image denoising is based on an image denoising model, the image denoising model can be CNN or UNet, and the preferred embodiment is UNet; the image feature extraction can be performed by an encoder in the Transformer.

[0135] The application first performs main body image segmentation on the current stirring image based on an image segmentation model to obtain a main region image and a secondary region image, performs image enhancement, image denoising and image feature extraction on the main region image to obtain main body features, and performs image denoising and image feature extraction on the secondary region image to obtain region features. The respective processing of images of different segmentation types can meet the image accuracy requirements of images of different segmentation types, so that the first stirring feature obtained by splicing is more in line with the current control task requirements, and the accuracy during subsequent correction and control according to the first stirring feature is improved.

[0136] In some embodiments of the application, the corresponding preset stirring parameter is determined according to the similarity of the first stirring feature, and the current stirring parameter is corrected according to the preset stirring parameter to obtain a first stirring parameter, specifically including:

[0137] Based on a preset historical stirring database, the similarity of the first stirring feature and each historical stirring feature in the historical stirring database is calculated, and the historical stirring parameter corresponding to the historical stirring feature with the highest similarity is taken as the preset stirring parameter;

[0138] Based on a preset correction weight, a weighted average value of each parameter in the current stirring parameter and the preset stirring parameter is calculated to obtain a first stirring parameter.

[0139] In some embodiments of the application, the similarity is measured based on a similarity algorithm, which can be a cosine similarity algorithm, a peak signal-to-noise ratio (PSNR) algorithm, a structural similarity (SSIM) algorithm or a mean square error (MSE) algorithm, and the preferred embodiment is the structural similarity (SSIM) algorithm.

[0140] In some embodiments of the application, the preset correction weight is used to calculate the weighted average value of each parameter in the current stirring parameter and the preset stirring parameter to obtain a first stirring parameter, specifically: based on the preset correction weight, the weighted average value of each parameter in the current stirring parameter and the preset stirring parameter is taken as the parameter value of each parameter of the first stirring parameter.

[0141] In some embodiments of the application, the preset correction weight can be set according to the current control requirement, and the weight is preferably set to 1:1, i.e. the weighted average value of each parameter is the arithmetic average value of each parameter in the current stirring parameter and the preset stirring parameter; if the current control requirement is biased towards the reference preset, the preset correction weight can be set to 7:3, at this time the corresponding weighting coefficient of the preset stirring parameter is 0.7 and the corresponding weighting coefficient of the current stirring parameter is 0.3; if the current control requirement is biased towards the reference actual, the preset correction weight can be set to 2:8, at this time the corresponding weighting coefficient of the preset stirring parameter is 0.2 and the corresponding weighting coefficient of the current stirring parameter is 0.8.

[0142] The application first calculates the similarity of the first stirring feature and each historical stirring feature in the historical stirring database, and determines the preset stirring parameter according to the similarity, improves the accuracy of the preset stirring parameter searching and screening through the similarity, and then calculates the weighted average of the current stirring parameter and the preset stirring parameter based on the preset correction weight to obtain the first stirring parameter, which can meet the parameter correction requirements in different situations, thereby improving the accuracy of the obtained first stirring parameter and improving the accuracy when controlling according to the first stirring parameter.

[0143] Step S105: controlling the current stirring process according to the first feeding sequence and the first stirring parameter.

[0144] Specifically, the current stirring process is controlled according to the first feeding sequence and the first stirring parameter, specifically: the feeding time and the feeding type of the current stirring process are controlled according to the first feeding sequence; and the parameters of the stirrer in the current stirring process are set according to the first stirring parameter.

[0145] The application first collects the process execution data and the current stirring image of the current stirring process, and predicts the initial feeding time of each kind of fed material and each kind of to-be-fed material based on multiple linear regression models, and then combines the initial feeding time of the fed material and the feeding time sequence to correct the initial feeding time of the to-be-fed material to obtain the corresponding feeding time interval, which considers the mutual influence of the feeding time of each material in the stirring process, and can improve the flexibility of determining the feeding time of the to-be-fed material; then the feeding time interval of each to-be-fed material is sampled and combined to obtain multiple to-be-fed time sequences, the solidification strength is predicted based on the stacked bidirectional neural network, and the first feeding sequence is screened according to the solidification strength, which can determine more abundant feeding time sequences based on more flexible feeding time intervals, can predict more comprehensive solidification strength, and then screen the first feeding sequence more accurately, thereby improving the accuracy when controlling according to the first feeding sequence; then the first stirring feature is obtained by segmenting and recognizing the current stirring image, and the preset stirring parameter is determined, and then the first stirring parameter is obtained by correcting the current stirring parameter according to the preset stirring parameter, which considers the influence of the current stirring situation on the subsequent control, thereby improving the accuracy of the control when the current stirring process is controlled according to the first feeding sequence and the first stirring parameter.

[0146] Corresponding to the foregoing method, please refer to Figure 2 The embodiment of the application provides a soil material stirring control system, which comprises a process data acquisition module 210, a feeding time determination module 220, a feeding sequence determination module 230, a stirring parameter determination module 240 and a stirring process control module 250.

[0147] The process data collection module 210 is configured to collect process execution data of a current stirring process and a current stirring image, wherein the process execution data comprises several kinds of materials already added, a time sequence of adding materials, several kinds of materials to be added, and current stirring parameters;

[0148] The material adding time determination module 220 is configured to determine an initial adding time of each kind of material already added and an initial adding time of each kind of material to be added based on a plurality of preset linear regression models according to the several kinds of materials already added and the several kinds of materials to be added, correct the initial adding time of each kind of material to be added according to the initial adding time of each kind of material already added and the time sequence of adding materials, and determine an adding time interval of each kind of material to be added.

[0149] The material adding sequence determination module 230 is configured to sample and combine the adding time intervals of all the materials to be added to obtain a plurality of time sequences of adding materials to be added, predict a solidification strength of the plurality of time sequences of adding materials to be added based on a preset stacked bidirectional neural network, and screen the plurality of time sequences of adding materials to be added according to the solidification strength to obtain a first material adding sequence.

[0150] The stirring parameter determination module 240 is configured to perform segmentation and identification on the current stirring image to obtain a first stirring feature, determine a preset stirring parameter corresponding to the first stirring feature according to a similarity of the first stirring feature, and correct the current stirring parameter according to the preset stirring parameter to obtain a first stirring parameter.

[0151] The stirring process control module 250 is configured to control the current stirring process according to the first material adding sequence and the first stirring parameter.

[0152] In some embodiments of the present application, the material adding time determination module 220 comprises an added time prediction unit, a time deviation calculation unit, a to-be-added time prediction unit, and a time interval determination unit.

[0153] The added time prediction unit is configured to determine an initial adding time of each kind of material already added based on the plurality of linear regression models according to the several kinds of materials already added, wherein the plurality of linear regression models are one-to-one corresponding to material types of the materials already added and the materials to be added, and when predicting, a linear regression model corresponding to a material type of the material already added is selected for prediction.

[0154] The time deviation calculation unit is configured to calculate a time difference between the initial adding time of each kind of material already added and an actual adding time of each kind of material already added in the time sequence of adding materials to obtain a time deviation of each kind of material already added.

[0155] The to-be-poured time prediction unit is configured to predict, based on the plurality of linear regression models, the initial pouring time of each to-be-poured material according to the to-be-poured materials; and when predicting, a linear regression model corresponding to the material type of the to-be-poured material is selected for prediction.

[0156] The time interval determination unit is configured to calculate the average time fluctuation of the plurality of poured materials according to the time deviation of each poured material, and determine the pouring time interval of each to-be-poured material based on the initial pouring time of each to-be-poured material according to the average time fluctuation.

[0157] In some embodiments of the present application, the construction process of the plurality of linear regression models is specifically as follows:

[0158] A plurality of historical mixing data are acquired, wherein each historical mixing data comprises a material type, a pouring time and a measured curing strength;

[0159] The plurality of historical mixing data are repeatedly classified according to the material type to obtain a plurality of groups of model training data; wherein when repeatedly classifying, the pouring interval between the pouring time of the current classified material type and the pouring time of the previous material in each historical mixing data is calculated, and the pouring interval is taken as a new data column of the model training data;

[0160] Based on the plurality of groups of model training data, a plurality of linear regression models corresponding thereto are constructed by taking the pouring interval and the measured curing strength as independent variables and taking the pouring time as a dependent variable.

[0161] In some embodiments of the present application, the pouring sequence determination module 230 comprises a time sampling unit, a pouring combination unit, a strength prediction unit and a sequence determination unit.

[0162] The time sampling unit is configured to sample the pouring time interval of each to-be-poured material multiple times respectively to obtain a plurality of pouring sampling times of each to-be-poured material.

[0163] The pouring combination unit is configured to construct the full combination of the pouring time of each to-be-poured material according to the plurality of pouring sampling times of each to-be-poured material to obtain a plurality of to-be-poured time sequences.

[0164] The strength prediction unit is configured to predict the curing strength of each to-be-poured time sequence based on the stacked bidirectional neural network to obtain a corresponding predicted curing strength.

[0165] The sequence determination unit is configured to determine the to-be-poured time sequence corresponding to the highest predicted curing strength as the first pouring sequence.

[0166] In some embodiments of the present application, the stacked bidirectional neural network is obtained by stacking two layers of bidirectional neural networks, and the stacked bidirectional neural network comprises an input layer, a first bidirectional network layer, a second bidirectional network layer, a fully connected prediction layer, and an output layer; the batching sequence determination module 230 comprises a feature extraction unit, a first feature extraction unit, a second feature extraction unit, a feature prediction unit, and a feature output unit;

[0167] The feature extraction unit is configured to perform feature extraction on the current input batching time sequence through the input layer of the stacked bidirectional neural network to obtain time sequence input features;

[0168] The first feature extraction unit is configured to represent the time sequence input features as a first forward hidden state sequence and a first reverse hidden state sequence through the first bidirectional network layer of the stacked bidirectional neural network, and fuse the first forward hidden state sequence and the first reverse hidden state sequence to obtain a first hidden state sequence;

[0169] The second feature extraction unit is configured to represent the first hidden state sequence as a second forward hidden state and a second reverse hidden state through the second bidirectional network layer of the stacked bidirectional neural network, and fuse the second forward hidden state and the second reverse hidden state to obtain a second hidden state;

[0170] The feature prediction unit is configured to perform prediction based on the second hidden state through the fully connected prediction layer of the stacked bidirectional neural network to obtain predicted features;

[0171] The feature output unit is configured to perform feature restoration on the predicted features through the output layer of the stacked bidirectional neural network to obtain a predicted solidification strength corresponding to the current input batching time sequence.

[0172] In some embodiments of the present application, the stirring parameter determination module 240 comprises a stirring image recognition unit; the stirring image recognition unit comprises a stirring image segmentation subunit, a first feature extraction subunit, a second feature extraction subunit, and a stirring feature acquisition subunit;

[0173] The stirring image segmentation subunit is configured to perform main image segmentation on a current stirring image based on a preset image segmentation model to obtain a main region image and a secondary region image of the current stirring image; wherein the main region image is a main body obtained by the image segmentation model from the current stirring image, and the secondary region image is a part of the current stirring image other than the main region image;

[0174] The first feature extraction subunit is configured to perform image enhancement and image denoising on the main region image to obtain a main region enhanced image, and perform image feature extraction on the main region enhanced image to obtain a main feature corresponding to the current stirring image.

[0175] The second feature extraction subunit is configured to perform image denoising on the auxiliary region image to obtain an auxiliary region denoised image, and perform image feature extraction on the auxiliary region denoised image to obtain a region feature corresponding to the current stirring image.

[0176] The stirring feature acquisition subunit is configured to splice the main feature and the region feature to obtain a first stirring feature corresponding to the current stirring image.

[0177] In some embodiments of the present application, the stirring parameter determination module 240 comprises a stirring parameter correction unit; the stirring parameter correction unit comprises a preset parameter determination subunit and a weighted parameter correction subunit.

[0178] The preset parameter determination subunit is configured to calculate the similarity between the first stirring feature and each historical stirring feature in a preset historical stirring database based on the preset historical stirring database, and take the historical stirring parameter corresponding to the historical stirring feature with the highest similarity as a preset stirring parameter.

[0179] The weighted parameter correction subunit is configured to calculate the weighted average value of each parameter in the preset stirring parameter and the current stirring parameter based on a preset correction weight, to obtain a first stirring parameter.

[0180] The present application first collects the process execution data and the current stirring image of the current stirring process, and predicts the initial feeding time of each material that has been fed and each material that is to be fed based on multiple linear regression models, and then combines the initial feeding time of the material that has been fed and the feeding time sequence to correct the initial feeding time of the material to be fed to obtain a corresponding feeding time interval. The mutual influence of the feeding time of each material in the stirring process is considered, which can improve the flexibility of determining the feeding time of the material to be fed. Then, the feeding time interval of each material to be fed is sampled and combined to obtain multiple feeding time sequences. The solidification strength is predicted based on the stacked bidirectional neural network, and the first feeding sequence is selected according to the solidification strength. A more flexible feeding time interval can be determined to obtain a more abundant feeding time sequence, a more comprehensive solidification strength can be predicted, and the first feeding sequence can be more accurately selected, thereby improving the accuracy when the current stirring process is controlled according to the first feeding sequence. Then, the current stirring image is segmented and recognized to obtain a first stirring feature and determine a preset stirring parameter, and then the current stirring parameter is corrected according to the preset stirring parameter to obtain a first stirring parameter. The influence of the current stirring condition on the subsequent control is considered, so that the accuracy of the control is improved when the current stirring process is controlled according to the first feeding sequence and the first stirring parameter.

[0181] It should be understood that the system provided by the embodiments of the present application corresponds to the foregoing method, and the soil material mixing control system provided by the embodiments of the present application can implement the soil material mixing control method provided by any one of the embodiments of the present application.

[0182] Adaptively, the embodiments of the present application further provide a computer device and a computer readable storage medium.

[0183] The computer device comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor;

[0184] When the processor executes the computer program, the soil material mixing control method of the present application is implemented.

[0185] The computer readable storage medium stores a plurality of instructions, which are adapted to be loaded by the processor to execute the soil material mixing control method of the present application.

[0186] The above is part of the embodiments of the present application, which further details the purpose, technical solutions and beneficial effects of the present application. It should be clear that the above part of the embodiments of the present application cannot be understood as a limitation of the present application. It is particularly pointed out that any changes, modifications, equivalent replacements and variations, etc. made by those skilled in the art within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method of controlling soil mixture, characterized by, The application relates to a method for controlling a current stirring process, comprising the following steps: collecting process execution data and a current stirring image of the current stirring process; wherein the process execution data comprises several kinds of materials already added, a time sequence of adding materials, several kinds of materials to be added and current stirring parameters; predicting initial adding time of each kind of material already added and initial adding time of each kind of material to be added based on a plurality of preset linear regression models according to the several kinds of materials already added and the several kinds of materials to be added, correcting the initial adding time of each kind of material to be added according to the initial adding time of each kind of material already added and the time sequence of adding materials, and determining an adding time interval of each kind of material to be added; sampling and combining the adding time intervals of all the materials to be added to obtain a plurality of adding time sequences, predicting a solidification strength of the plurality of adding time sequences based on a preset stacked bidirectional neural network, and screening the plurality of adding time sequences to obtain a first adding sequence according to the solidification strength; segmenting and identifying the current stirring image to obtain a first stirring feature, determining corresponding preset stirring parameters according to the similarity of the first stirring feature, and correcting the current stirring parameters according to the preset stirring parameters to obtain first stirring parameters; controlling the current stirring process according to the first adding sequence and the first stirring parameters; the initial adding time of each kind of material already added and the initial adding time of each kind of material to be added are predicted based on a plurality of preset linear regression models according to the several kinds of materials already added and the several kinds of materials to be added, the initial adding time of each kind of material to be added is corrected according to the initial adding time of each kind of material already added and the time sequence of adding materials, and an adding time interval of each kind of material to be added is determined, and the method specifically comprises the following steps: the initial adding time of each kind of material already added is predicted based on the plurality of linear regression models according to the several kinds of materials already added; wherein the plurality of linear regression models are one-to-one corresponding to material types of the materials already added and the materials to be added; when predicting, a linear regression model corresponding to a material type of the material already added is selected for prediction; a time difference between the initial adding time of each kind of material already added and actual adding time of each kind of material already added in the time sequence of adding materials is calculated to obtain a time deviation of each kind of material already added; the initial adding time of each kind of material to be added is predicted based on the plurality of linear regression models according to the several kinds of materials to be added; wherein when predicting, a linear regression model corresponding to a material type of the material to be added is selected for prediction; the time fluctuation average value of the several kinds of materials already added is calculated according to the time deviation of each kind of material already added, and the adding time interval of each kind of material to be added is determined according to the time fluctuation average value and the initial adding time of each kind of material to be added as a reference; the plurality of adding time sequences are obtained by sampling and combining the adding time intervals of all the materials to be added, the solidification strength of the plurality of adding time sequences is predicted based on a preset stacked bidirectional neural network, and the first adding sequence is obtained by screening the plurality of adding time sequences according to the solidification strength. The feeding time interval of each material to be fed is sampled multiple times to obtain multiple feeding sampling times of each material to be fed; According to the multiple feeding sampling times of each material to be fed, a full combination of the feeding time of each material to be fed is constructed to obtain multiple feeding time sequences; Based on the stacked bidirectional neural network, the curing strength of each feeding time sequence is predicted to obtain the corresponding predicted curing strength; The feeding time sequence corresponding to the highest predicted curing strength is determined as the first feeding sequence; According to the similarity of the first stirring feature, the corresponding preset stirring parameter is determined, and the current stirring parameter is corrected according to the preset stirring parameter to obtain the first stirring parameter, specifically including: Based on the preset historical stirring database, the similarity of the first stirring feature and each historical stirring feature in the historical stirring database is calculated, and the historical stirring parameter corresponding to the historical stirring feature with the highest similarity is taken as the preset stirring parameter; Based on the preset correction weight, the weighted average value of each parameter in the current stirring parameter and the preset stirring parameter is calculated to obtain the first stirring parameter.

2. The soil material mixing control method according to claim 1, wherein The construction process of the multiple linear regression models is specifically: Obtain multiple historical stirring data; wherein each historical stirring data includes material type, feeding time and measured curing strength; The multiple historical stirring data are repeatedly classified according to the material type to obtain multiple groups of model training data; wherein when repeatedly classifying, the feeding interval of the material type of the current classification in each historical stirring data and the feeding time of the previous material are calculated, and the feeding interval is taken as a new data column of the model training data; Based on the multiple groups of model training data, the feeding interval and the measured curing strength are taken as independent variables, and the feeding time is taken as dependent variable to construct the corresponding multiple linear regression models.

3. The method of claim 1, wherein The stacked bidirectional neural network is obtained by stacking two layers of bidirectional neural networks, and the stacked bidirectional neural network includes an input layer, a first bidirectional network layer, a second bidirectional network layer, a fully connected prediction layer and an output layer; based on the stacked bidirectional neural network, the curing strength of each feeding time sequence is predicted to obtain the corresponding predicted curing strength, specifically including: Through the input layer of the stacked bidirectional neural network, the current input feeding time sequence is feature extracted to obtain a time sequence input feature; Through the first bidirectional network layer of the stacked bidirectional neural network, the time sequence input feature is represented as a first forward hidden state sequence and a first reverse hidden state sequence, and the first forward hidden state sequence and the first reverse hidden state sequence are fused to obtain a first hidden state sequence; Through the second bidirectional network layer of the stacked bidirectional neural network, the first hidden state sequence is represented as a second forward hidden state and a second reverse hidden state, and the second forward hidden state and the second reverse hidden state are fused to obtain a second hidden state; Through the fully connected prediction layer of the stacked bidirectional neural network, the second hidden state is predicted to obtain a predicted feature; The predicted characteristics are feature-reduced through an output layer of the stacked bidirectional neural network to obtain a predicted solidification strength corresponding to a to-be-cast time sequence of the current input.

4. The method of claim 1, wherein The current stirring image is segmented and identified to obtain first stirring characteristics, and specifically includes: Based on a preset image segmentation model, the current stirring image is subjected to main image segmentation to obtain a main region image and a sub-region image of the current stirring image; the main region image is a main body identified by the image segmentation model from the current stirring image, and the sub-region image is a part of the current stirring image other than the main region image; The main region image is subjected to image enhancement and image denoising to obtain a main region enhanced image, and image feature extraction is performed on the main region enhanced image to obtain main body characteristics corresponding to the current stirring image; The sub-region image is subjected to image denoising to obtain a sub-region denoised image, and image feature extraction is performed on the sub-region denoised image to obtain region characteristics corresponding to the current stirring image; The main body characteristics and the region characteristics are spliced to obtain first stirring characteristics corresponding to the current stirring image.

5. A soil mix control system characterized by, A soil material stirring control method for implementing any one of claims 1 to 4, comprising a process data acquisition module, a casting time determination module, a casting sequence determination module, a stirring parameter determination module, and a stirring process control module; The process data acquisition module is configured to acquire process execution data and a current stirring image of a current stirring process; the process execution data includes a plurality of cast materials, a casting time sequence, a plurality of to-be-cast materials, and a current stirring parameter; The casting time determination module is configured to predict, based on a plurality of preset linear regression models, the initial casting time of each cast material and the initial casting time of each to-be-cast material according to the plurality of cast materials and the plurality of to-be-cast materials, correct the initial casting time of each to-be-cast material according to the initial casting time of each cast material and the casting time sequence, and determine the casting time interval of each to-be-cast material; The casting sequence determination module is configured to sample and combine the casting time intervals of all to-be-cast materials to obtain a plurality of to-be-cast time sequences, predict the solidification strength of the plurality of to-be-cast time sequences based on a preset stacked bidirectional neural network, and select the plurality of to-be-cast time sequences according to the solidification strength to obtain a first casting sequence; The stirring parameter determination module is configured to segment and identify the current stirring image to obtain first stirring characteristics, determine a preset stirring parameter according to the similarity of the first stirring characteristics, correct the current stirring parameter according to the preset stirring parameter, and obtain first stirring parameters; The stirring process control module is configured to control the current stirring process according to the first casting sequence and the first stirring parameters.

6. A soil mix control system according to claim 5, wherein, The casting time determination module includes a cast time prediction unit, a time deviation calculation unit, a to-be-cast time prediction unit, and a time interval determination unit; The cast time prediction unit is configured to predict, according to the cast materials, initial feeding time of each cast material based on the plurality of linear regression models, wherein the plurality of linear regression models correspond to the material types of the cast materials and the to-be-fed materials one by one, and when predicting, a linear regression model corresponding to the material type of the cast material is selected for prediction. The time deviation calculation unit is configured to calculate time difference between the initial feeding time of each cast material and actual feeding time of each cast material in the feeding time sequence, to obtain time deviation of each cast material. The to-be-fed time prediction unit is configured to predict, according to the to-be-fed materials, initial feeding time of each to-be-fed material based on the plurality of linear regression models, wherein when predicting, a linear regression model corresponding to the material type of the to-be-fed material is selected for prediction. The time interval determination unit is configured to calculate time fluctuation average value of the cast materials according to the time deviation of each cast material, and determine feeding time interval of each to-be-fed material based on the initial feeding time of each to-be-fed material according to the time fluctuation average value.

7. A soil mix control system according to claim 5 wherein, The stirring parameter determination module comprises a stirring image recognition unit; the stirring image recognition unit comprises a stirring image segmentation subunit, a first feature extraction subunit, a second feature extraction subunit, and a stirring feature acquisition subunit; The stirring image segmentation subunit is configured to perform main image segmentation on a current stirring image based on a preset image segmentation model, to obtain a main region image and a secondary region image of the current stirring image; wherein the main region image is a main body obtained by the image segmentation model from the current stirring image, and the secondary region image is a part of the current stirring image other than the main region image; The first feature extraction subunit is configured to perform image enhancement and image denoising on the main region image, to obtain a main region enhanced image, and perform image feature extraction on the main region enhanced image, to obtain a main body feature corresponding to the current stirring image; The second feature extraction subunit is configured to perform image denoising on the secondary region image, to obtain a secondary region denoised image, and perform image feature extraction on the secondary region denoised image, to obtain a region feature corresponding to the current stirring image; The stirring feature acquisition subunit is configured to splice the main body feature and the region feature, to obtain a first stirring feature corresponding to the current stirring image.

Images