Emulsified asphalt quality control method and system

Abstract

The invention discloses an emulsified asphalt quality control method and system, and relates to the technical field of quality control in the road material production process, and the method comprises the following steps: in the emulsified asphalt production process, online collecting production process parameters and raw material characteristic parameters to form a multi-dimensional real-time parameter set; according to the method, a multi-dimensional real-time parameter set is adopted; inputting the multi-dimensional real-time parameter set into a soft measurement model to obtain a predicted value of the off-line detection key quality index data, and comparing and judging whether the predicted value of the off-line detection key quality index data is in an out-of-tolerance state or not; if the multi-dimensional real-time parameter set is in the out-of-tolerance state, a targeted deviation correction instruction is generated based on the multi-dimensional real-time parameter set, and the design of real-time deviation correction in the production process is achieved, so that the control lag problem that in the prior art, most of key quality indexes of emulsified asphalt production are detected in an offline mode, and deviation correction cannot be conducted in time when disqualification is found is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of quality control technology in the production process of road materials, and in particular to a method and system for quality control of emulsified asphalt. Background Technology

[0002] Emulsified asphalt is a key material in road construction and maintenance, and its quality directly affects the service life of road projects and driving safety. Therefore, quality control during the production process is crucial. In existing emulsified asphalt production technology, quality control mainly relies on offline sampling inspection. That is, after production is completed, samples of the finished product are taken to test key quality indicators such as particle size distribution, residue, sieve residue, viscosity, and sedimentation stability, and then the batch is judged as qualified based on the test results.

[0003] However, this control method has the following problems: First, the detection results of key quality indicators have a significant lag. When the indicators are found to be unqualified, the corresponding batch of emulsified asphalt has already been produced, making it impossible to correct the production process in time. The only recourse is rework or scrapping, resulting in a significant waste of raw materials and energy, as well as extended production cycles and increased downtime. Second, offline sampling inspection cannot cover the entire production process and cannot capture in real time the impact of raw material fluctuations and equipment operating status changes on quality indicators, leading to a high rate of out-of-tolerance batches and further increasing production costs. Therefore, how to achieve real-time prediction and judgment of key quality indicators in the emulsified asphalt production process, avoid the lag of offline sampling inspection, and correct deviations in a timely manner to reduce the rate of out-of-tolerance batches has become a pressing technical problem to be solved in the field of emulsified asphalt quality control. Summary of the Invention

[0004] To address the technical problems existing in the background art, the present invention proposes a method and system for quality control of emulsified asphalt.

[0005] The present invention proposes a method for quality control of emulsified asphalt, comprising the following steps: Step 1: During the production of emulsified asphalt, online collection of production process parameters and raw material characteristic parameters forms a multi-dimensional real-time parameter set; Step 2: Obtain qualified and unqualified batch data from historical production of emulsified asphalt and construct a model training sample library; based on the model training sample library, use an ensemble learning algorithm to train a soft measurement model, which is used to output predicted values ​​of key quality indicators for offline detection. Step 3: Obtain the acceptable range of offline detection key quality indicator data. If the predicted value of the offline detection key quality indicator data exceeds the acceptable range, the predicted value of the offline detection key quality indicator data is determined to be in an out-of-tolerance state. Step 4: If the predicted value of the key quality indicator data of offline detection is determined to be out of tolerance, then a targeted correction instruction is generated based on the multi-dimensional real-time parameter set.

[0006] Preferably, step 1 further includes the following step: The raw material characteristic parameters are preprocessed. Abnormal data in the raw material characteristic parameters are removed by the Laida criterion, and missing data in the raw material characteristic parameters are supplemented by the median filling method and the linear interpolation method.

[0007] Preferably, in step 2, the qualified batch data and unqualified batch data of emulsified asphalt in historical production are obtained, and a model training sample library is constructed as follows: Obtain qualified and unqualified batch data from historical production of emulsified asphalt, extract historical production process parameters, historical raw material characteristic parameters, and corresponding offline key quality indicator data, and construct a model training sample library. Offline testing of key quality indicators includes particle size distribution, residue, sieve residue, viscosity, and sedimentation stability.

[0008] Preferably, in step 2, a soft measurement model is trained using an ensemble learning algorithm based on the model training sample library. The soft measurement model is used to output predicted values ​​of key quality indicator data for offline detection, as follows: The data in the model training sample library is normalized, and the core parameters associated with the key quality indicators of offline detection are screened out through feature importance analysis, and a core parameter set is constructed. Based on the core parameter set and the corresponding offline detection key quality index data, an ensemble learning algorithm is used to train a soft measurement model.

[0009] Preferably, in step 2, the data in the model training sample library is normalized, and core parameters associated with key quality indicators of offline detection are selected through feature importance analysis, constructing a core parameter set as follows: The historical production process parameters and historical raw material characteristic parameters in the model training sample library are normalized. Feature importance analysis was performed using the random forest algorithm. Normalized historical production process parameters and historical raw material characteristic parameters are used as input features for the random forest algorithm, and the corresponding offline detection key quality indicator data are used as output labels to train the random forest model. The importance score of each input feature to the output label is obtained through the random forest model. The importance score is quantified by the information gain of the input feature in the decision tree splitting process of the random forest algorithm. Input features with importance scores higher than the feature importance threshold are selected as core parameters, and all core parameters are obtained to form a core parameter set. The importance scores of the input features of each random forest algorithm to the output label are obtained by using the random forest model, forming a mapping table of historical production process parameters, historical raw material characteristic parameters and corresponding offline detection key quality indicator data.

[0010] Preferably, in step 2, the feature importance threshold is: obtaining the importance scores of the input features of all random forest algorithms to the output labels to form an importance score set; obtaining the average value of the importance scores in the importance score set; and using the average value of the importance scores in the importance score set as the feature importance threshold.

[0011] Preferably, in step 2, based on the core parameter set and the corresponding offline detection key quality index data, an ensemble learning algorithm is used to train a soft measurement model, as follows: The core parameter set is used as the input feature of the ensemble learning algorithm, and the offline detection key quality index data that corresponds one-to-one with the core parameter set is used as the output label of the ensemble learning algorithm. Based on the input features and output labels of the ensemble learning algorithm, the ensemble learning algorithm is used for model training. The mapping relationship between core parameters and key quality indicators is learned through the ensemble learning algorithm to obtain the initial model. Cross-validation was used to optimize the parameters of the initial model, and outlier parameters related to overfitting and underfitting were removed to obtain the optimal combination of model parameters. The optimal combination of model parameters was then substituted into the initial model to obtain the optimized soft sensor model. The soft measurement model takes core parameters as input and outputs predicted values ​​of key quality indicators for offline detection corresponding to the core parameters.

[0012] Preferably, in step 2, the ensemble learning algorithm is gradient boosting tree. Gradient boosting tree, also known as GBT, is an existing ensemble learning method that iteratively trains multiple weak learners to gradually optimize the model's predictive performance. It is divided into two types: gradient boosting decision tree and gradient boosting regression tree. Gradient boosting decision tree is called GBDT, and gradient boosting regression tree is called GBRT.

[0013] Preferably, in step 4, if the predicted value of the offline detection key quality indicator data is determined to be out of tolerance, a targeted correction instruction is generated based on a multi-dimensional real-time parameter set, as follows: When the predicted value of the key quality indicator data of offline detection is determined to be out of tolerance, the target parameter associated with the key quality indicator data of offline detection is extracted from the multi-dimensional real-time parameter set. Based on historical production process parameters, historical raw material characteristic parameters, and corresponding offline key quality indicator data, an association rule mining algorithm is used to extract the association relationship between historical production process parameters, historical raw material characteristic parameters, and offline key quality indicator data to form parameter-quality association rules. All parameter-quality association rules are obtained to form a parameter-quality association rule library. Based on the parameter-quality correlation rule base, a quantitative correlation relationship is generated between the deviation of the target parameter and the key quality indicators of offline detection of out-of-tolerance, as follows: The first step is to retrieve the parameter-quality association rules corresponding to the offline detection key quality indicator data and target parameters that are currently out of tolerance from the parameter-quality association rule library; The second step is to obtain the mean and standard deviation of the target parameters for qualified batches corresponding to the target parameters based on the historical qualified batch data of emulsified asphalt production. Obtain the current target parameter value, and then calculate the deviation quantification value by dividing (current target parameter value - mean target parameter value of qualified batches) by the standard deviation of target parameters of qualified batches. The third step is to take the deviation range of the offline key quality indicator data as the value of the predicted value of the offline key quality indicator data if it exceeds the qualified range of the offline key quality indicator data. Based on the retrieved parameter-quality correlation rules, a mapping relationship is established between the quantified deviation value and the out-of-tolerance range of key quality indicator data in offline detection. Based on the mapping relationship between the quantified deviation value and the out-of-tolerance range of key quality indicator data in offline detection, the adjustment direction and adjustment range of the target parameters are determined, and preliminary correction instructions are generated.

[0014] Preferably, in step 4, when it is determined that the predicted value of a key quality indicator exceeds the tolerance, the target parameters associated with the deviation indicator are extracted from the multi-dimensional real-time parameter set, as follows: When the predicted value of a certain offline key quality indicator data exceeds the error, the association mapping table between historical production process parameters, historical raw material characteristic parameters and the corresponding offline key quality indicator data is called. Historical production process parameters and / or historical raw material characteristic parameters with importance scores higher than the feature importance threshold corresponding to the offline key quality indicator data are selected. Then, the corresponding production process parameters and / or raw material characteristic parameters are extracted from the multi-dimensional real-time parameter set as target parameters associated with the offline key quality indicator data.

[0015] An emulsified asphalt quality control system, comprising: Data acquisition module: During the production of emulsified asphalt, online acquisition of production process parameters and raw material characteristic parameters forms a multi-dimensional real-time parameter set; Key quality indicator data prediction module: acquire qualified and unqualified batch data of emulsified asphalt in historical production, and build a model training sample library; based on the model training sample library, use an ensemble learning algorithm to train a soft measurement model, which is used to output the predicted values ​​of key quality indicator data for offline detection; Out-of-tolerance state judgment module: Obtain the acceptable range of offline detection key quality indicator data. If the predicted value of the offline detection key quality indicator data exceeds the acceptable range of the offline detection key quality indicator data, it is determined that the predicted value of the offline detection key quality indicator data is in an out-of-tolerance state. Correction instruction generation module: If it is determined that the predicted value of the key quality indicator data of offline detection is out of tolerance, a targeted correction instruction is generated based on the multi-dimensional real-time parameter set.

[0016] The emulsified asphalt quality control method and system proposed in this invention have the following beneficial technical effects: This application employs online acquisition of production process parameters and raw material characteristic parameters to form a multi-dimensional real-time parameter set. This multi-dimensional real-time parameter set is input into a soft sensing model, which then predicts key quality indicators of emulsified asphalt in real time, obtaining predicted values ​​of offline key quality indicator data. The predicted values ​​of the key quality indicators are compared with the acceptable range of the offline key quality indicator data to determine whether the predicted values ​​are out of tolerance. If out of tolerance is determined, a targeted correction instruction is generated based on the multi-dimensional real-time parameter set and issued to the emulsified asphalt production equipment for execution. This design realizes real-time correction of the production process, transforming traditional post-production sampling inspection into process early warning and real-time correction. This significantly reduces the occurrence rate of out-of-tolerance batches, reduces rework / scrapping and repeated sampling and testing, and enables early detection and handling of anomalies to reduce energy consumption and downtime. It effectively alleviates the control lag problem in existing technologies where key quality indicators in emulsified asphalt production are mostly detected offline, making timely correction impossible when non-compliance is detected. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for quality control of emulsified asphalt according to the present invention; Figure 2 This is a schematic diagram of the principle of an emulsified asphalt quality control system according to the present invention. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar symbols denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0019] like Figure 1 The method for quality control of emulsified asphalt, as shown, includes the following steps: Step 1: During the production of emulsified asphalt, online collection of production process parameters and raw material characteristic parameters forms a multi-dimensional real-time parameter set; In an optional embodiment, step 1 further includes the following step: The raw material characteristic parameters are preprocessed. Abnormal data in the raw material characteristic parameters are removed by the Laida criterion, and missing data in the raw material characteristic parameters are supplemented by the median filling method and the linear interpolation method.

[0020] Step 2: Obtain qualified and unqualified batch data from historical production of emulsified asphalt and construct a model training sample library; based on the model training sample library, use an ensemble learning algorithm to train a soft measurement model, which is used to output predicted values ​​of key quality indicators for offline detection. Step 3: Obtain the acceptable range of offline detection key quality indicator data. If the predicted value of the offline detection key quality indicator data exceeds the acceptable range, the predicted value of the offline detection key quality indicator data is determined to be in an out-of-tolerance state. Step 4: If the predicted value of the key quality indicator data detected offline is found to be out of tolerance, a targeted correction instruction is generated based on the multi-dimensional real-time parameter set. The targeted correction instruction is then sent to the emulsified asphalt production equipment and executed to achieve real-time correction of the production process.

[0021] In an optional embodiment, in step 2, qualified batch data and unqualified batch data of emulsified asphalt historical production are obtained, and a model training sample library is constructed as follows: Obtain qualified and unqualified batch data from historical production of emulsified asphalt, extract historical production process parameters, historical raw material characteristic parameters, and corresponding offline key quality indicator data, and construct a model training sample library. Offline detection of key quality indicators includes particle size distribution, residue, sieve residue, viscosity, and sedimentation stability; In an optional embodiment, in step 2, a soft measurement model is trained using an ensemble learning algorithm based on the model training sample library. The soft measurement model is used to output predicted values ​​of key quality indicator data for offline detection, as follows: The data in the model training sample library is normalized, and the core parameters associated with the key quality indicators of offline detection are screened out through feature importance analysis, and a core parameter set is constructed. In an optional embodiment, in step 2, the data in the model training sample library is normalized, and core parameters associated with key quality indicators of offline detection are selected through feature importance analysis, constructing a core parameter set as follows: The historical production process parameters and historical raw material characteristic parameters in the model training sample library are normalized. Feature importance analysis was performed using the random forest algorithm. Normalized historical production process parameters and historical raw material characteristic parameters are used as input features for the random forest algorithm, and the corresponding offline detection key quality indicator data are used as output labels to train the random forest model. The importance score of each input feature to the output label is obtained through the random forest model. The importance score is quantified by the information gain of the input feature in the decision tree splitting process of the random forest algorithm. Input features with importance scores higher than the feature importance threshold are selected as core parameters, and all core parameters are obtained to form a core parameter set.

[0022] In an optional embodiment, in step 2, the feature importance threshold is: obtaining the importance scores of the input features of all random forest algorithms to the output labels to form an importance score set; obtaining the average value of the importance scores in the importance score set; and using the average value of the importance scores in the importance score set as the feature importance threshold. The importance scores of the input features of each random forest algorithm to the output label are obtained by using the random forest model, forming a mapping table of the relationship between historical production process parameters, historical raw material characteristic parameters and corresponding offline detection key quality indicator data; Based on the core parameter set and the corresponding offline detection key quality index data, an ensemble learning algorithm is used to train a soft measurement model. In an optional embodiment, in step 2, based on the core parameter set and the corresponding offline detection key quality index data, an ensemble learning algorithm is used to train a soft measurement model, as follows: The core parameter set is used as the input feature of the ensemble learning algorithm, and the offline detection key quality index data that corresponds one-to-one with the core parameter set is used as the output label of the ensemble learning algorithm. Based on the input features and output labels of the ensemble learning algorithm, the ensemble learning algorithm is used for model training. The mapping relationship between core parameters and key quality indicators is learned through the ensemble learning algorithm to obtain the initial model. Cross-validation was used to optimize the parameters of the initial model, and outlier parameters related to overfitting and underfitting were removed to obtain the optimal combination of model parameters. The optimal combination of model parameters was then substituted into the initial model to obtain the optimized soft sensor model. The soft measurement model takes core parameters as input and outputs predicted values ​​of key quality indicators for offline detection corresponding to the core parameters.

[0023] In an optional embodiment, in step 2, the ensemble learning algorithm is gradient boosting tree. Gradient boosting tree, also known as GBT, is an existing ensemble learning method that iteratively trains multiple weak learners to gradually optimize the model's predictive performance. It is divided into two types: gradient boosting decision tree and gradient boosting regression tree. Gradient boosting decision tree is called GBDT, and gradient boosting regression tree is called GBRT.

[0024] In an optional embodiment, in step 4, if it is determined that the predicted value of the offline detection key quality indicator data is out of tolerance, a targeted correction instruction is generated based on a multi-dimensional real-time parameter set, as follows: When the predicted value of the key quality indicator data of offline detection is determined to be out of tolerance, the target parameter associated with the key quality indicator data of offline detection is extracted from the multi-dimensional real-time parameter set. In an optional embodiment, in step 4, when it is determined that the predicted value of a key quality indicator is out of tolerance, the target parameters associated with the out-of-tolerance indicator are extracted from the multi-dimensional real-time parameter set, as follows: When the predicted value of a certain offline key quality indicator data exceeds the error, the association mapping table between historical production process parameters, historical raw material characteristic parameters and the corresponding offline key quality indicator data is called. Historical production process parameters and / or historical raw material characteristic parameters with importance scores higher than the feature importance threshold corresponding to the offline key quality indicator data are selected. Then, the corresponding production process parameters and / or raw material characteristic parameters are extracted from the multi-dimensional real-time parameter set as target parameters associated with the offline key quality indicator data.

[0025] Based on historical production process parameters, historical raw material characteristic parameters, and corresponding offline key quality indicator data, an association rule mining algorithm is used to extract the association relationships between historical production process parameters, historical raw material characteristic parameters, and offline key quality indicator data, forming parameter-quality association rules. All parameter-quality association rules are obtained to form a parameter-quality association rule library.

[0026] Based on the parameter-quality correlation rule base, a quantitative correlation relationship is generated between the deviation of the target parameter and the key quality indicators of offline detection of out-of-tolerance, as follows: The first step is to retrieve the parameter-quality association rules corresponding to the offline detection key quality indicator data and target parameters that are currently out of tolerance from the parameter-quality association rule library; The second step is to obtain the mean and standard deviation of the target parameters for qualified batches corresponding to the target parameters based on the historical qualified batch data of emulsified asphalt production. Obtain the current target parameter value, and then calculate the deviation quantification value by dividing (current target parameter value - mean target parameter value of qualified batches) by the standard deviation of target parameters of qualified batches. The third step is to take the deviation range of the offline key quality indicator data as the value of the predicted value of the offline key quality indicator data if it exceeds the qualified range of the offline key quality indicator data. Based on the retrieved parameter-quality correlation rules, a mapping relationship is established between the quantified deviation value and the out-of-tolerance range of key quality indicator data in offline detection. Based on the mapping relationship between the quantified deviation value and the out-of-tolerance range of key quality indicator data in offline detection, the adjustment direction and adjustment range of the target parameters are determined, and preliminary correction instructions are generated. As an explanation, based on the mapping relationship between the quantified deviation value and the out-of-tolerance range of key quality indicator data detected offline, the adjustment direction and magnitude of the target parameter are determined, and preliminary correction instructions are generated. This can be achieved using existing industrial control technologies such as interpolation algorithms and PID control algorithms, for example: The first step is to extract the corresponding rules between the direction of change of the target parameter and the direction of error correction of the offline key quality indicator data from the mapping relationship between the quantified deviation value and the error range of the offline key quality indicator data: If the mapping relationship shows that the quantified value of the deviation of the target parameter is positive, and the deviation of the offline detection key quality indicator data increases synchronously, then the adjustment direction is determined to be to reduce the value of the target parameter. If the mapping relationship shows that the quantified value of the deviation of the target parameter is negative, and the deviation of the key quality indicator data detected offline increases synchronously, then the adjustment direction is determined to be to increase the value of the target parameter. The second step is to calculate the target parameter adjustment amount required to completely eliminate the deviation amplitude based on the mapping relationship between the deviation quantification value and the out-of-tolerance amplitude of the offline key quality indicator data. Using the existing deviation quantification value corresponding to the out-of-tolerance amplitude data pair in the mapping relationship as the benchmark, the current out-of-tolerance amplitude is substituted, and the target parameter deviation quantification value correction amount required to reduce the out-of-tolerance amplitude to 0 is obtained by reverse interpolation. Then, combined with the calculation formula of deviation quantification value = (current target parameter value - mean of target parameters of qualified batches) / standard deviation of target parameters of qualified batches, the adjustment amplitude of the target parameter is derived in reverse. The third step is to use a weighted PID control algorithm to coordinate and optimize the adjustment direction and magnitude of each target parameter if multiple target parameters are associated with the out-of-tolerance index. The weights can be set based on the importance scores of each target parameter and the out-of-tolerance index to avoid mutual interference between the adjustment of multiple parameters. The fourth step is to convert the determined target parameter adjustment direction and adjustment range into control signals that the production equipment can recognize, thus forming a preliminary correction command; Among them, linear interpolation algorithm and PID control algorithm are both existing mature technologies in the field of industrial process control, which can be directly adapted to the parameter adjustment scenarios of emulsified asphalt production.

[0027] This application employs online acquisition of production process parameters and raw material characteristic parameters to form a multi-dimensional real-time parameter set. This multi-dimensional real-time parameter set is input into a soft sensing model, which then predicts the key quality indicators of emulsified asphalt in real time, obtaining predicted values ​​of offline key quality indicator data. The predicted values ​​of the key quality indicators are compared with the acceptable range of the offline key quality indicator data to determine whether the predicted values ​​are out of tolerance. If out of tolerance is determined, a targeted correction instruction is generated based on the multi-dimensional real-time parameter set and sent to the emulsified asphalt production equipment for execution. This design realizes real-time correction of the production process, transforming traditional post-production sampling inspection into process early warning and real-time correction. This significantly reduces the occurrence rate of out-of-tolerance batches, reduces rework / scrapping and repeated sampling and testing, and enables early detection and handling of anomalies to reduce energy consumption and downtime. It effectively alleviates the control lag problem in existing technologies where key quality indicators in emulsified asphalt production are mostly detected offline, making timely correction impossible when non-compliance is detected.

[0028] like Figure 2 The emulsified asphalt quality control system shown includes: Data acquisition module: During the production of emulsified asphalt, online acquisition of production process parameters and raw material characteristic parameters forms a multi-dimensional real-time parameter set; Key quality indicator data prediction module: acquire qualified and unqualified batch data of emulsified asphalt in historical production, and build a model training sample library; based on the model training sample library, use an ensemble learning algorithm to train a soft measurement model, which is used to output the predicted values ​​of key quality indicator data for offline detection; Out-of-tolerance state judgment module: Obtain the acceptable range of offline detection key quality indicator data. If the predicted value of the offline detection key quality indicator data exceeds the acceptable range of the offline detection key quality indicator data, it is determined that the predicted value of the offline detection key quality indicator data is in an out-of-tolerance state. Correction instruction generation module: If the predicted value of the key quality indicator data of offline detection is determined to be out of tolerance, a targeted correction instruction is generated based on a multi-dimensional real-time parameter set.

[0029] For clarification, "acquisition" in this application refers to obtaining the required content or data using existing technical means.

[0030] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0031] In the embodiments provided by this invention, it should be understood that the disclosed system or method can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0032] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0033] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0034] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the basic characteristics of the present invention.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for quality control of emulsified asphalt, characterized in that, Includes the following steps: Step 1: During the production of emulsified asphalt, online collection of production process parameters and raw material characteristic parameters forms a multi-dimensional real-time parameter set; Step 2: Obtain qualified and unqualified batch data from historical production of emulsified asphalt and construct a model training sample library; based on the model training sample library, use an ensemble learning algorithm to train a soft measurement model, which is used to output predicted values ​​of key quality indicators for offline detection. Step 3: Obtain the acceptable range of offline detection key quality indicator data. If the predicted value of the offline detection key quality indicator data exceeds the acceptable range, the predicted value of the offline detection key quality indicator data is determined to be in an out-of-tolerance state. Step 4: If the predicted value of the key quality indicator data of offline detection is determined to be out of tolerance, then a targeted correction instruction is generated based on the multi-dimensional real-time parameter set.

2. The method for quality control of emulsified asphalt according to claim 1, characterized in that, Step 1 also includes the following steps: The raw material characteristic parameters are preprocessed. Abnormal data in the raw material characteristic parameters are removed by the Raida criterion, and missing data in the raw material characteristic parameters are supplemented by the median filling method and the linear interpolation method.

3. The method for quality control of emulsified asphalt according to claim 1, characterized in that, In step 2, obtain historical production data of qualified and unqualified batches of emulsified asphalt, and construct a model training sample library as follows: Obtain qualified and unqualified batch data from historical production of emulsified asphalt, extract historical production process parameters, historical raw material characteristic parameters, and corresponding offline key quality indicator data, and construct a model training sample library. Offline testing of key quality indicators includes particle size distribution, residue, sieve residue, viscosity, and sedimentation stability.

4. The method for quality control of emulsified asphalt according to claim 3, characterized in that, In step 2, based on the model training sample library, an ensemble learning algorithm is used to train a soft measurement model. The soft measurement model is used to output predicted values ​​of key quality indicator data for offline detection, as follows: The data in the model training sample library is normalized, and the core parameters associated with the key quality indicators of offline detection are screened out through feature importance analysis, and a core parameter set is constructed. Based on the core parameter set and the corresponding offline detection key quality index data, an ensemble learning algorithm is used to train a soft measurement model.

5. The method for quality control of emulsified asphalt according to claim 4, characterized in that, In step 2, the data in the model training sample library is normalized, and core parameters associated with key quality indicators of offline detection are selected through feature importance analysis, constructing a core parameter set as follows: The historical production process parameters and historical raw material characteristic parameters in the model training sample library are normalized. Feature importance analysis was performed using the random forest algorithm. Normalized historical production process parameters and historical raw material characteristic parameters are used as input features of the random forest algorithm, and the corresponding offline detection key quality index data are used as output labels to train the random forest model. The importance score of the input features of each random forest algorithm to the output label is obtained through the random forest model. The importance score is quantified by the information gain of the input features in the decision tree splitting process of the random forest algorithm. Input features with importance scores higher than the feature importance threshold are selected as core parameters, and all core parameters are obtained to form a core parameter set; The importance scores of the input features of each random forest algorithm to the output label are obtained by using the random forest model, forming a mapping table of historical production process parameters, historical raw material characteristic parameters and corresponding offline detection key quality indicator data.

6. The method for quality control of emulsified asphalt according to claim 5, characterized in that, In step 2, the feature importance threshold is: obtain the importance scores of the input features of all random forest algorithms to the output labels to form an importance score set; obtain the average value of the importance scores in the importance score set; The average importance score in the importance score set is used as the feature importance threshold.

7. The method for quality control of emulsified asphalt according to claim 5, characterized in that, In step 2, based on the core parameter set and the corresponding offline detection key quality index data, an ensemble learning algorithm is used to train the soft measurement model, as follows: The core parameter set is used as the input feature of the ensemble learning algorithm, and the offline detection key quality index data that corresponds one-to-one with the core parameter set is used as the output label of the ensemble learning algorithm. Based on the input features and output labels of the ensemble learning algorithm, the ensemble learning algorithm is used for model training. The mapping relationship between core parameters and key quality indicators is learned through the ensemble learning algorithm to obtain the initial model. Cross-validation was used to optimize the parameters of the initial model, and outlier parameters related to overfitting and underfitting were removed to obtain the optimal combination of model parameters. The optimal combination of model parameters was then substituted into the initial model to obtain the optimized soft sensor model. The soft measurement model takes core parameters as input and outputs predicted values ​​of key quality indicators for offline detection corresponding to the core parameters.

8. The method for quality control of emulsified asphalt according to claim 7, characterized in that, In step 4, if the predicted values ​​of key quality indicators from offline detection are determined to be out of tolerance, a targeted correction instruction is generated based on a multi-dimensional real-time parameter set, as follows: When the predicted value of the key quality indicator data of offline detection is determined to be out of tolerance, the target parameter associated with the key quality indicator data of offline detection is extracted from the multi-dimensional real-time parameter set. Based on historical production process parameters, historical raw material characteristic parameters, and corresponding offline key quality indicator data, an association rule mining algorithm is used to extract the association relationship between historical production process parameters, historical raw material characteristic parameters, and offline key quality indicator data to form parameter-quality association rules. All parameter-quality association rules are obtained to form a parameter-quality association rule library. Based on the parameter-quality correlation rule base, a quantitative correlation relationship is generated between the deviation of the target parameter and the key quality indicators of offline detection of out-of-tolerance, as follows: The first step is to retrieve the parameter-quality association rules corresponding to the offline detection key quality indicator data and target parameters that are currently out of tolerance from the parameter-quality association rule library; The second step is to obtain the mean and standard deviation of the target parameters for qualified batches corresponding to the target parameters based on the historical qualified batch data of emulsified asphalt production. Obtain the current target parameter value, and then calculate the deviation quantification value by dividing (current target parameter value - mean target parameter value of qualified batches) by the standard deviation of target parameters of qualified batches. The third step is to take the deviation range of the offline key quality indicator data as the value of the predicted value of the offline key quality indicator data if it exceeds the qualified range of the offline key quality indicator data. Based on the retrieved parameter-quality correlation rules, a mapping relationship is established between the quantified deviation value and the out-of-tolerance range of key quality indicator data in offline detection. Based on the mapping relationship between the quantified deviation value and the out-of-tolerance range of key quality indicator data in offline detection, the adjustment direction and adjustment range of the target parameters are determined, and preliminary correction instructions are generated.

9. The method for quality control of emulsified asphalt according to claim 8, characterized in that, In step 4, when the predicted value of a key quality indicator is determined to be out of tolerance, the target parameters associated with the out-of-tolerance indicator are extracted from the multi-dimensional real-time parameter set, as follows: When the predicted value of a certain offline key quality indicator data exceeds the error, the association mapping table between historical production process parameters, historical raw material characteristic parameters and the corresponding offline key quality indicator data is called. Historical production process parameters and / or historical raw material characteristic parameters with importance scores higher than the feature importance threshold corresponding to the offline key quality indicator data are selected. Then, the corresponding production process parameters and / or raw material characteristic parameters are extracted from the multi-dimensional real-time parameter set as target parameters associated with the offline key quality indicator data.

10. A quality control system for emulsified asphalt, used with any one of the emulsified asphalt quality control methods according to claims 1 to 9, characterized in that, include: Data acquisition module: During the production of emulsified asphalt, online acquisition of production process parameters and raw material characteristic parameters forms a multi-dimensional real-time parameter set; Key quality indicator data prediction module: acquire qualified and unqualified batch data of emulsified asphalt in historical production, and build a model training sample library; based on the model training sample library, use an ensemble learning algorithm to train a soft measurement model, which is used to output the predicted values ​​of key quality indicator data for offline detection; Out-of-tolerance state judgment module: Obtain the acceptable range of offline detection key quality indicator data. If the predicted value of the offline detection key quality indicator data exceeds the acceptable range of the offline detection key quality indicator data, it is determined that the predicted value of the offline detection key quality indicator data is in an out-of-tolerance state. Correction instruction generation module: If the predicted value of the key quality indicator data of offline detection is determined to be out of tolerance, a targeted correction instruction is generated based on a multi-dimensional real-time parameter set.

Images