Discover real-time opportunities for increased productivity

Abstract

In a real-time opportunity discovery method for improving productivity in a manufacturing process, a processor extracts a set of features from time series data based on uncontrolled variables for the time series data by autoencoding using a neural network. The processor identifies one or more operation modes based on the extracted features, including dimensionality reduction using representation learning from the time series data. The processor identifies neighborhoods of the current operation state based on the extracted features. The processor compares the current operation state with historical operation states based on time series data of the same operation mode. The processor discovers operation opportunities based on a comparison of the current operation state with the historical operation states using the neighborhoods. The processor identifies controlled variables of the same mode associated with the current operation state.

Description

[Technical Field]

[0001] The present disclosure relates generally to the field of machine learning, and more particularly to real-time opportunity discovery using historical data encoding to improve productivity in manufacturing processes. [Background technology]

[0002] Many production processes can be highly complex. For example, in oil sands production, mined ore may go through several stages of extraction, upgrading, and refining. Similar processes can be found in food and steel production. Outflows from upstream processes can become inflows to downstream processes. Each stage can involve numerous components and processes, making the system dynamic. In a typical oil sands process, operations run at full capacity when sufficient feedstock is available from mining and all components are functioning properly. Upgrading operations can be run without a vacuum. When bitumen from processed oil sands is of low quality (e.g., high chloride concentrations), it may be operated in a reduced production mode to avoid deterioration of the coking unit. When feedline maintenance is performed, it may be operated at partial capacity. When bitumen from processed oil sands is of low quality (e.g., high chloride concentrations), it may be operated in a reduced production mode to avoid deterioration of the coking unit. Partial capacity operation may occur when feed lines are undergoing maintenance.

[0003] Traditionally, the majority of bitumen produced is upgraded to synthetic crude oil before being sold to refineries in the market. However, some bitumen is suitable enough to be sent directly to high-conversion refineries capable of processing heavy / sour crude oil. Examples of such diluted bitumen sold directly to refineries include products from on-site facilities and other locations. Petroleum products can be produced from oil sands through three basic steps: i) extraction of bitumen from the oil sands, during which solids and water are removed; ii) upgrading of heavy bitumen to lighter intermediate crude products; and iii) refining of crude oil into final products such as gasoline, lubricants, and diluents. All of these processes involve numerous sequential steps of physical or chemical transformations to convert one substance into another. Achieving multiple goals within such a production system requires an optimal balance of processes. Plant operators must seek out opportunities to improve productivity, for example, to use fewer feedstocks, make additives more affordable, or produce more final products. It is also necessary to focus on specific areas with high business value to provide incremental value at local steps in the manufacturing process. It is also necessary to find opportunities for cost, material, and energy saving models that can help increase additional profits limited to local steps within a relatively short time window of the manufacturing process that encompasses those opportunities within the overall plant operations. Summary of the Invention

[0004] A real-time opportunity discovery method for improving productivity in a manufacturing process addresses certain shortcomings of the prior art and provides additional advantages. Advantageously, a processor extracts a set of features from time series data based on uncontrolled variables for the time series data by autoencoding using a neural network. The processor identifies one or more operation modes based on the extracted features, including dimensional reduction using representation learning from the time series data. The processor identifies a neighborhood of the current operation state based on the extracted features. The processor compares the current operation state with historical operation states based on time series data for the same operation mode. The processor discovers operation opportunities based on a comparison of the current operation state with the historical operation states using the neighborhood. The processor identifies controlled variables of the same mode associated with the current operation state. The processor recommends an action strategy based on the one or more controlled variables, the one or more uncontrolled variables, and the target productivity.

[0005] In one or more embodiments, a computer-implemented method is provided for monitoring time-series data generated from one or more sensors. For example, the time-series data can be data from oil sands operations and production processes. For example, petroleum products can be produced from oil sands through several stages, such as extraction, upgrading, and refining. Advantageously, an optimal balance of processes is provided to achieve multiple goals within such a production system.

[0006] In one or more embodiments, a computer-implemented method is provided for extracting a set of features from time series data based on one or more uncontrolled variables for the time series data by autoencoding using a neural network, e.g., an LSTM autoencoder. Advantageously, the LSTM autoencoder not only uses the set of features for training, but also learns the set of features. The set of features is information about the time series data. The set of features may be measurable individual attributes or characteristics of a phenomenon observed in the time series data. The opportunity discovery module can select a subset of relevant features for creating a predictive model based on uncontrolled variables that define a user's lack of control over the time series data. Advantageously, the opportunity discovery module can reduce the number of resources required to describe the time series data. The opportunity discovery module can construct a combination of uncontrolled variables that describes the time series data with sufficient accuracy using the LSTM autoencoder. Advantageously, an LSTM autoencoder can invoke an auto-encoding process based on time series data to reduce the dimensionality of the sensor tag space into a bounded embedding space.

[0007] In one or more embodiments, a computer-implemented method is provided for identifying one or more operation modes based on extracted features, including dimensionality reduction using representation learning from time series data. Advantageously, dimensionality reduction transforms the time series data from a high-dimensional space to a low-dimensional space such that the low-dimensional representation preserves certain meaningful attributes of the original data, ideally close to its intrinsic dimensionality. Advantageously, a neighborhood of the current operation state can be identified. This neighborhood can be a dynamic mode within the same operation mode and can be found by the Euclidean distance between historical operation states and the current operation state. Advantageously, an automated process is provided rather than relying on rule-based mode detection, which requires a large amount of prior knowledge and memorized principles. The opportunity discovery module can achieve opportunity realization through analysis using unsupervised machine learning. For example, the opportunity discovery module can identify operation opportunities by comparing the current state with similar historical operations within that mode or neighborhood.

[0008] In one or more embodiments, a computer-implemented method is provided for using an opportunity discovery module to compare a current operational state with a historical operational state based on time series data of the same operational mode among one or more operational modes. Advantageously, the opportunity discovery module can identify a particular mode in which the current operational state exists. The opportunity discovery module can project clusters using t-distributed stochastic neighbor embedding (t-SNE) compression to create a graph. The opportunity discovery module can use t-SNE compression to embed high-dimensional points into a lower dimension in a manner that considers similarities between points. In one example, the opportunity discovery module can achieve high bitumen extraction by comparing the current operational state with other operations in the same mode. The opportunity discovery module can analyze episodes with poor performance and discover operational opportunities for improvement.

[0009] In one or more embodiments, a computer-implemented method is provided for discovering operational opportunities based on a comparison of current and historical operational states. Advantageously, operational opportunities can be identified by comparing the current state with similar historical operations within the mode or vicinity. In one example, an operational opportunity can be a set of operational changes inferred from historical operational states to increase the current operational state to higher production within a defined short future period, e.g., within a two-hour window. In another example, an operational opportunity can be a set of operational changes inferred from historical operational states to decrease the current operational state to lower additive or ingredient usage within a defined short future period. Other suitable opportunities can be found.

[0010] In one or more embodiments, a computer-implemented method is provided for identifying controlled variables of the same mode associated with a current operational state. Advantageously, these controlled variables can be used to calculate rewards (or opportunities) from episodes of time-series data. For example, these controlled variables can be production rates and raw material variables that a user can optimize based on the best neighboring episodes found. These controlled variables can be identified from an established neighborhood of similar historical uncontrolled variables to generate possible action strategies associated with the current state based on the time-series data.

[0011] In one or more embodiments, a computer-implemented method for recommending an action strategy based on controlled variables, uncontrolled variables, and a target productivity is provided. Advantageously, a similarity measurement can be defined to identify historical episodes from time-series data having similar operational states based on a comparison of current and historical operational states. Episodes can be created from historical episodes that exhibit higher productivity or throughput. Scores can be generated based on alternative action strategies, and the scores can be used to recommend action strategies based on the scoring for each alternative action strategy.

[0012] In one or more embodiments, a computer-implemented method is provided for outputting an action strategy to a user. Advantageously, the user interface can present nearby episodes using a time-stamped chart. The estimated gain of the action strategy can be presented.

[0013] In another aspect, a computer program product is provided that includes one or more computer-readable storage media and program instructions collectively stored on the one or more computer-readable storage media. Advantageously, the program instructions extract a set of features from time-series data based on uncontrolled variables for the time-series data by neural network autoencoding. The program instructions identify one or more operation modes based on the extracted features, including dimensionality reduction using representation learning from the time-series data. The program instructions identify neighborhoods for a current operation state based on the extracted features. The program instructions compare the current operation state with historical operation states based on time-series data for the same operation mode. The program instructions discover operation opportunities based on a comparison of the current operation state with the historical operation states using the neighborhoods. The program instructions identify controlled variables of the same mode associated with the current operation state. The program instructions recommend an action strategy based on one or more controlled variables, one or more uncontrolled variables, and a target productivity.

[0014] In a further aspect, a computer system is provided that includes one or more computer processors, one or more computer-readable storage media, and program instructions stored on the one or more computer-readable storage media and executed by at least one of the one or more computer processors. Advantageously, the program instructions extract a set of features from time-series data based on uncontrolled variables for the time-series data by neural network autoencoding. The program instructions identify one or more operation modes based on the extracted features, including dimensionality reduction using representation learning from the time-series data. The program instructions identify neighborhoods for a current operation state based on the extracted features. The program instructions compare the current operation state with historical operation states based on time-series data for the same operation mode. The program instructions discover operation opportunities based on a comparison of the current operation state with the historical operation states using the neighborhoods. The program instructions identify controlled variables of the same mode associated with the current operation state. The program instructions recommend an action strategy based on the one or more controlled variables, the one or more uncontrolled variables, and a target productivity.

[0015] Additional features and advantages are realized through the techniques of the present invention. Other embodiments and aspects of the invention are described in detail herein and are considered a part of the claimed invention. [Brief explanation of the drawings]

[0016] [Figure 1] FIG. 1 is a functional block diagram illustrating an operational opportunity discovery environment, according to one embodiment of the present disclosure. [Figure 2] 2 is a flowchart illustrating the operational steps of an opportunity discovery module in the computing device of FIG. 1 in accordance with one embodiment of the present disclosure. [Figure 3]FIG. 2 is an exemplary functional diagram of an opportunity discovery module within the computing device of FIG. 1 in accordance with one embodiment of the present disclosure. [Figure 4] FIG. 2 is an exemplary architecture diagram of an opportunity discovery module within the computing device of FIG. 1, according to one embodiment of the present disclosure. [Figure 5] FIG. 2 is a block diagram of components of the computing device of FIG. 1 in accordance with one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0017] The present disclosure is directed to a system and method for real-time opportunity discovery for improving productivity in manufacturing processes using historical data encoding.

[0018] Embodiments of the present disclosure recognize the need for plant operators to seek out opportunities to improve productivity, such as using fewer raw materials, using cheaper additives, or producing more of the final product. Embodiments of the present disclosure can focus on specific areas with high business value to provide incremental value at local steps in the manufacturing process. Embodiments of the present disclosure can discover opportunities related to cost, raw material, and energy savings models. These opportunities may have relatively short time windows. Embodiments of the present disclosure can help obtain additional profit gains limited to local steps throughout plant operations. Embodiments of the present disclosure can select a complete set of time series data, such as time series data from sensors, e.g., Internet of Things (IoT) sensors, that have a complete picture of the plant situation. Embodiments of the present disclosure can separate the time series data into controlled and uncontrolled variables. In a production process, embodiments of the present disclosure can dynamically provide recommendations and suggestions for improving productivity by consuming fewer raw materials, using cheaper additives, reducing energy consumption, and increasing product output. Embodiments of the present disclosure can derive more timely and accurate opportunities within a window of just a few hours. Embodiments of the present disclosure can utilize auto-encoding techniques to find operation modes and neighborhoods using features extracted as an embedded space, which can achieve time-series data dimensionality reduction and generate production recommendations through historical similarity analysis within the modes or neighborhoods in the embedded space.

[0019] Embodiments of the present disclosure can apply a long short-term memory (LSTM) autoencoder to extract features or an embedded space related to uncontrolled variables. Embodiments of the present disclosure can identify static operation modes by defining neighborhoods or using Gaussian mixture clustering applied to the embedded space. Embodiments of the present disclosure can identify neighbors of the current operation state. This neighborhood can be a dynamic mode within the same operation mode and can be found by the Euclidean distance between the historical operation state and the current operation state. Embodiments of the present disclosure can identify opportunities by identifying improvement recommendations by focusing on differences in controlled variables. Embodiments of the present disclosure can compare the current operation state with other operations in the same mode. Poor episodes with poor performance provide opportunities for improvement. Embodiments of the present disclosure can limit historical episodes by selecting neighbors of the current operation state. Embodiments of the present disclosure can perform validation of the performance accuracy of a predictive model developed using all controlled and uncontrolled variables across time series data.

[0020] The present disclosure will now be described in detail with reference to the figures, in which: Figure 1 is a functional block diagram illustrating an operational opportunity discovery environment, generally designated 100, in accordance with one embodiment of the present disclosure.

[0021] In the illustrated embodiment, the operational opportunity discovery environment 100 includes a computing device 102, time-series data 104, and a network 108. In one embodiment, the computing device 102 can directly access the time-series data 104. In another embodiment, the time-series data 104 can be accessed through a communication network, such as the network 108. In one or more embodiments, the time-series data 104 can be data captured by one or more sensors. For example, the time-series data 104 can be data from an oil sands operation and production process. In an example oil sands operation and production process, some bitumen may be upgraded to synthetic crude oil before being sold to a refinery. Some bitumen may be suitable enough to be sent to a high-conversion refinery that can process heavy crude oil. For example, petroleum products can be produced from oil sands through several stages, such as extraction, upgrading, and refining. For example, solids and water can be removed during the extraction stage, during which bitumen can be extracted from the oil sands. During the upgrading stage, the bitumen can be upgraded to a lighter intermediate crude product. During the refining stage, crude oil can be refined into finished products such as gasoline, lubricants, and diluents. These processes within these stages may involve numerous sequential steps of physical or chemical transformations to convert one substance into another. Reaching multiple goals within such a production system requires an optimal balance of processes.

[0022] In another example, the time series data 104 may be data from a steelmaking process that produces steel from iron ore and / or scrap. Impurities such as nitrogen, silicon, phosphorus, sulfur, and excess carbon may be removed from the iron supply, and alloying elements such as manganese, nickel, chromium, carbon, and vanadium may be added to produce different grades of steel. In yet another example, the time series data 104 may be data from a production process that processes soybeans into soy sauce with additives. In yet another example, the time series data 104 may be data from any other suitable operations and production processes.

[0023] In one or more embodiments, the time series data 104 may be data that includes, for example, uncontrolled variables 122 and controlled variables 124. To ensure the ability to take action to capture production improvement opportunities, the uncontrolled variables 122 and the controlled variables 124 may be separated. For example, the uncontrolled variables 122 may be time-stamped variables and may be defined as a set of variables from sensors over which a user has little or no control. The uncontrolled variables 122 may be parameters used to define how similar operational and production conditions are. The uncontrolled variables 122 may be derived from episodes of the time series data 104. The controlled variables 124 may be time-stamped variables and may be defined as a set of variables from sensors over which a user has control over actions. In one example, the controlled variables 124 may be used to calculate rewards (or opportunities) from episodes of the time series data 104. The controlled variables 124 may be production rates and raw material variables that a user can optimize based on the best neighboring episodes found. Controlled variables 124 can be identified from an established neighborhood of similar historical uncontrolled variables 122 to generate possible action strategies related to the current state based on the time series data 104 .

[0024] In the example of froth production, uncontrolled variables 122 may be environmental variables over which the user may have little or no control, such as ambient temperature, flash, cloud, viscosity, hydrogen availability, coker rate, and straight-run production. Controlled variables 124 may be variables over which the user has control, such as production rates and feedstock variables that the user can optimize or modify. Controlled variables 124 may include, for example, the feedstock for diesel hydrotreating or catalytic hydrotreating, the feed slate for diesel hydrotreating using low vacuum gas oil, low vacuum gas oil, side-draw kerosene, heavy naphtha, and coker kerosene. Diesel hydrotreating or catalytic hydrotreating may primarily reduce undesirable species from the straight-run diesel fraction by selectively reacting them with hydrogen in a high-temperature, moderate-pressure reactor. To successfully produce ultra-low sulfur diesel, organic sulfur species, including substituted dibenzothiophenes and other heat-resistant sulfur species, must be removed. Numerous reactions may occur in parallel on the surface of diesel hydrotreating catalysts, including hydrodesulfurization, hydrodenitrogenation, and aromatic ring saturation / hydrogenation. The feed to a diesel hydrotreating unit may have a nominal distillation range of 300-700°F (approximately 150-370°C). Different process designs and flow schemes can be used for diesel hydrotreating, depending on the process objectives and the characteristics of the feed being processed.

[0025] In various embodiments of the present disclosure, the computing device 102 may be a laptop computer, a tablet computer, a netbook computer, a personal computer (PC), a desktop computer, a mobile phone, a smartphone, a smart watch, a wearable computing device, a personal digital assistant (PDA), or a server. In another embodiment, the computing device 102 represents a computing system that utilizes clustered computers and components to act as a single pool of seamless resources. In other embodiments, the computing device 102 may represent a server computing system that utilizes multiple computers as a server system, such as in a cloud computing environment. In general, the computing device 102 may be any computing device or combination of devices that access the opportunity discovery module 110 and the network 108 and is capable of processing program instructions and executing the opportunity discovery module 110 in accordance with embodiments of the present disclosure. As shown and described in further detail with respect to FIG. 5 , the computing device 102 may include internal and external hardware components.

[0026] Additionally, in the illustrated embodiment, the computing device 102 includes an opportunity discovery module 110. In the illustrated embodiment, the opportunity discovery module 110 is located on the computing device 102. However, in other embodiments, the opportunity discovery module 110 can be external and accessed via a communications network, such as the network 108. This communications network can be, for example, a local area network (LAN), a wide area network (WAN) such as the Internet, or a combination of the two, and can include wired, wireless, fiber optic connections, or other connections known in the art. In general, this communications network can be any combination of connections and protocols that facilitate communication between the computing device 102 and the opportunity discovery module 110 in accordance with the desired embodiment of the present disclosure.

[0027] In the illustrated embodiment, opportunity discovery module 110 includes LSTM autoencoder 112, predictive model 114, variable identification module 116, strategy recommendation module 118, and output module 120. In the illustrated embodiment, LSTM autoencoder 112, predictive model 114, variable identification module 116, strategy recommendation module 118, and output module 120 are located on computing device 102. However, in other embodiments, LSTM autoencoder 112, predictive model 114, variable identification module 116, strategy recommendation module 118, and output module 120 may be external and accessed via a communications network, such as network 108.

[0028] In one or more embodiments, the opportunity discovery module 110 is configured to monitor time series data 104 generated from one or more sensors. For example, the time series data 104 can be data from an oil sands operation and production process. For example, petroleum products can be produced from oil sands through several stages, such as extraction, upgrading, and refining. For example, solids and water can be removed during the extraction stage, during which bitumen can be extracted from the oil sands. During the upgrading stage, bitumen can be upgraded to lighter intermediate crude products. During the refining stage, crude oil can be refined into final products such as gasoline, lubricants, and diluents. These processes within these stages can involve multiple sequential steps of physical or chemical transformations to convert one substance into another. An optimal balance of processes is required to achieve multiple goals within such a production system. In another example, the time series data 104 can be data from a steelmaking process, which produces steel from iron ore and / or scrap. Impurities such as nitrogen, silicon, phosphorus, sulfur, and excess carbon can be removed from the iron supply, and alloying elements such as manganese, nickel, chromium, carbon, and vanadium can be added to produce different grades of steel. In yet another example, the time series data 104 can be data from a production process in which soybeans can be processed into soy sauce with additives. In yet another example, the time series data 104 can be data from any other suitable operations and production processes.

[0029] In one or more embodiments, the opportunity discovery module 110 is configured to extract a set of features from the time series data 104 using a recurrent neural network, such as an LSTM autoencoder 112, based on one or more uncontrolled variables 122 for the time series data 104. The opportunity discovery module 110 can use the set of features for training, as well as learn the set of features using the LSTM autoencoder 112. The set of features is information about the time series data 104. The set of features may be measurable individual attributes or characteristics of a phenomenon observed in the time series data 104. The opportunity discovery module 110 can select a subset of relevant features for creating the predictive model 114 based on the uncontrolled variables 122 that define the time series data 104 beyond the user's control. The opportunity discovery module 110 can reduce the number of resources required to describe the time series data 104. The opportunity discovery module 110 can use an LSTM autoencoder 112 to construct a combination of uncontrolled variables 122 that describes the time series data 104 with sufficient accuracy. The LSTM autoencoder 112 can be an artificial neural network used to learn efficient data encoding in an unsupervised manner. The LSTM autoencoder 112 can be a recurrent neural network with an autoencoder implementation for the time series data 104 using an encoder-decoder LSTM architecture. The LSTM autoencoder 112 can learn a representation (encoding) for the time series data 104, for example, for dimensionality reduction. From the reduced encoding, the LSTM autoencoder 112 can generate a representation of the time series data 104 that is as close as possible to the original input. The LSTM autoencoder 112 can include an encoder and a decoder. The encoder can use raw data (e.g., the time series data 104) as input and generate features or representations as output, while the decoder uses extracted features from the encoder as input and reconstructs the original input raw data as output.Training can be repeated until a stopping criterion is met. The LSTM autoencoder 112 can invoke an auto-encoding process based on the time series data 104 to reduce the dimensionality of the sensor tag space into a limited embedding space. The LSTM autoencoder 112 can invoke a clustering method to generate latent modes. The LSTM autoencoder 112 can invoke a type of artificial neural network used to learn efficient data encoding in an unsupervised manner. The LSTM autoencoder 112 can obtain fixed-size vectors from the time series data 104.

[0030] In one or more embodiments, the opportunity discovery module 110 is configured to identify one or more operation modes based on extracted features, including dimensionality reduction using representation learning from the time series data 104. Dimensionality reduction may involve transforming the time series data 104 from a high-dimensional space to a low-dimensional space such that the low-dimensional representation preserves certain meaningful attributes of the original data, ideally close to its intrinsic dimensionality. The opportunity discovery module 110 may identify a neighborhood of the current operation state. This neighborhood may be a dynamic mode within the same operation mode and may be found by the Euclidean distance between the historical operation state and the current operation state. The opportunity discovery module 110 may define this neighborhood or identify static operation modes using Gaussian mixture clustering applied to the embedded space. The mode may include a plant operation situation. A hard category type of mode may relate to a specific operation configuration of an entire production process, such as manufacturing pipe configurations or unit operation situations. A soft category type of mode may include production levels of local or global operations. The sands oil industry has a complex process for converting sands oil into synthetic crude oil. Multiple stages, including primary extraction, secondary extraction, and upgrading, are involved in completing synthetic crude oil production. Each stage involves multiple components and processes, making the system dynamic. Some of the modes involved may be explicitly known by field engineers. Other modes may be hidden and can be identified by advanced analytical models. The opportunity discovery module 110 can automatically detect operation modes by learning from historical sensor data and other production data, achieving compact feedback of the mode form. The opportunity discovery module 110 can determine operation mode algorithms and models.The opportunity discovery module 110 can use operational modes as benchmarks when similar production conditions existed in the past or when new modes may be identified. These operational modes can be used to recommend better control actions or automatically change control parameters. If a new mode is identified, it can be stored to expand the system's memory and knowledge. The opportunity discovery module 110 can calculate historical benchmarking of the detected mode against historical plant data when the plant was in the same mode. The opportunity discovery module 110 can use the detected mode as a basis for identifying similar periods from history. For example, periods from history can be identified when the same mode was active. The opportunity discovery module 110 can further identify key differences between current and historical variable values ​​using factor analysis on a set of process variables that jointly define the detected mode. The opportunity discovery module 110 can calculate and display possible low-level operational / production root causes to field engineers. The opportunity discovery module 110 can provide analytical techniques for identifying or classifying modes throughout complex manufacturing processes. The opportunity discovery module 110 can provide detailed or general multivariate mode identification, and as a side effect, can also achieve segmentation of unsegmented data into subsets. Rather than relying on rule-based mode detection, which requires a lot of prior knowledge and memorized principles, the opportunity discovery module 110 can provide an automated process that offers a simpler, more human-involved approach to mode generation. The opportunity discovery module 110 can achieve opportunity realization through analysis using unsupervised machine learning. For example, the opportunity discovery module 110 can identify operational opportunities by comparing the current state with similar historical operations within the mode or its vicinity.

[0031] In one or more embodiments, the opportunity discovery module 110 is configured to compare a current operational state with a historical operational state based on time-series data of the same operational mode among one or more operational modes. The opportunity discovery module 110 can identify the specific mode in which the current operational state exists. The opportunity discovery module 110 can project clusters to create a graph using t-SNE (t-SNE) compression. t-SNE is a dimensionality reduction algorithm suitable for visualizing high-dimensional data, such as time-series data 104. The opportunity discovery module 110 can use t-SNE compression to embed high-dimensional points into a lower dimension in a manner that considers similarities between points. In one example, the opportunity discovery module 110 can achieve high bitumen extraction by comparing the current operational state with other operations in the same mode. The opportunity discovery module 110 can analyze episodes with poor performance and discover operational opportunities for improvement. The opportunity discovery module 110 can focus on those operational episodes in history that have reached high froth production and can extract key operational actions from these episodes to benefit current operations.

[0032] In one or more embodiments, the opportunity discovery module 110 is configured to discover operational opportunities based on a comparison of current and historical operational states. Operational opportunities can be identified by comparing the current state to similar historical operations within the mode or vicinity. In one example, an operational opportunity can be a set of operational changes inferred from historical operational states to increase the current operational state to higher production within a defined short future period, such as within a two-hour window. In another example, an operational opportunity can be a set of operational changes inferred from historical operational states to decrease the current operational state to lower additive or ingredient usage within a defined short future period. Other suitable opportunities can be found.

[0033] In one or more embodiments, the opportunity discovery module 110 is configured to identify controlled variables 124 of the same mode associated with the current operational state. The controlled variables 124 may define actions that a user can control. The controlled variables 124 may be time-stamped variables and may be defined as a set of variables from sensors for which a user can control actions. In one example, the controlled variables 124 may be used to calculate rewards (or opportunities) from episodes of the time-series data 104. For example, the controlled variables 124 may be production rate and feedstock variables that a user can optimize based on the best neighbor episode found. The controlled variables 124 may be identified from an established neighborhood of similar historical uncontrolled variables 122 to create possible action strategies associated with the current state based on the time-series data 104. In a froth production example, the controlled variables 124 may include, for example, feedstock for diesel hydrotreating or catalytic hydrotreating, a feed slate for diesel hydrotreating using low vacuum gas oil, low vacuum gas oil, side-draw kerosene, heavy naphtha, and coker kerosene. Diesel hydrotreating, or catalytic hydrotreating, may primarily involve the reduction of undesirable species from straight-run diesel fractions by selectively reacting them with hydrogen in reactors at high temperatures and moderate pressures. Successful production of ultra-low sulfur diesel requires the removal of organic sulfur species, including substituted dibenzothiophenes and other heat-resistant sulfur species. Numerous reactions may occur in parallel on the surface of diesel hydrotreating catalysts, including hydrodesulfurization, hydrodenitrogenation, and aromatic ring saturation / hydrogenation.

[0034] In one or more embodiments, the opportunity discovery module 110 is configured to recommend an action strategy based on the controlled variables 124, the uncontrolled variables 122, and the target productivity. The opportunity discovery module 110 can define a similarity measure to identify historical episodes from the time-series data 104 having similar operational states based on a comparison of the current state with the historical operational states. The opportunity discovery module 110 can generate episodes from the historical episodes that exhibit higher productivity or throughput. The opportunity discovery module 110 can generate scores based on the alternative action strategies and can recommend action strategies based on the scoring for each alternative action strategy.

[0035] In one or more embodiments, the opportunity discovery module 110 is configured to output an action strategy to a user. The opportunity discovery module 110 may provide a user interface for interfacing with the user. The opportunity discovery module 110 may provide other suitable output means with the user. The opportunity discovery module 110 may provide an indicator or alert to the user regarding the discovered operation opportunity. The opportunity discovery module 110 may display the operation modes as a graph using a t-SNE method. The opportunity discovery module 110 may display nearby episodes using a time-stamped chart. The opportunity discovery module 110 may display an estimated gain of the action strategy.

[0036] In one or more embodiments, the LSTM autoencoder 112 is configured to extract a set of features from the time series data 104 based on uncontrolled variables 122 for the time series data 104. In one example, the LSTM autoencoder 112 may be an artificial neural network used to learn efficient data encoding in an unsupervised manner. The LSTM autoencoder 112 may be capable of automatically extracting the influence of past events. The LSTM autoencoder 112 may be a recurrent neural network with an autoencoder implementation for the time series data 104 using an encoder-decoder LSTM architecture. The LSTM autoencoder 112 may learn a representation (encoding) for the time series data 104, for example, for dimensionality reduction. The LSTM autoencoder 112 may generate a representation from the reduced encoding that is as close as possible to the original input of the time series data 104. The LSTM autoencoder 112 may include an encoder and a decoder. The encoder can use raw data (e.g., time series data 104) as input and generate features or representations as output. The decoder can use extracted features from the encoder as input and reconstruct the original input raw data as output. Training of the LSTM autoencoder 112 can be repeated until a stopping criterion is met. The LSTM autoencoder 112 can invoke an auto-encoding process using the time series data 104 to reduce the dimensionality of the sensor tag space to a bounded embedding space. The LSTM autoencoder 112 can invoke a clustering method to generate latent modes. The LSTM autoencoder 112 can invoke a type of artificial neural network used to learn efficient data encoding in an unsupervised manner. The LSTM autoencoder 112 can obtain fixed-size vectors from the time series data 104.The LSTM autoencoder 112 can not only use this set of features for training, but also train the set of features itself. The set of features can be information about the time series data 104. The set of features can be measurable individual attributes or characteristics of phenomena observed from the time series data 104. The LSTM autoencoder 112 can select a subset of relevant features for creating the predictive model 114 based on uncontrolled variables 122 that define the time series data 104 beyond the user's control. The LSTM autoencoder 112 can reduce the number of resources required to describe the time series data 104. The LSTM autoencoder 112 can construct combinations of uncontrolled variables 122 that describe the time series data 104 with sufficient accuracy.

[0037] In one or more embodiments, the predictive model 114 is configured to discover operational opportunities based on the time series data 104. Operational opportunities can be identified by comparing the current state to similar historical operations within the mode or vicinity. In one example, an operational opportunity can be a set of operational changes inferred from historical operational states to increase the current operational state to higher production within a defined short-term future timeframe. In another example, an operational opportunity can be a set of operational changes inferred from historical operational states to decrease the current operational state to lower additive or ingredient usage within a defined short-term future timeframe. Other suitable opportunities can be found.

[0038] In one or more embodiments, the predictive model 114 is configured to identify one or more operational modes based on extracted features, including dimensionality reduction using representation learning, from the time-series data 104. The predictive model 114 can identify a neighborhood of the current operational state. This neighborhood can be a dynamic mode within the same operational mode and can be found by the Euclidean distance between the historical and current operational states. The predictive model 114 can define this neighborhood or identify a static operational mode using Gaussian mixture clustering applied to the embedded space. The modes can include plant operation conditions. The sands oil industry involves a complex process for converting sands oil to synthetic crude oil. Multiple stages, including primary extraction, secondary extraction, and upgrading, are involved in completing the synthetic crude oil production. Each stage involves multiple components and processes. Some of the modes may be explicitly known by field engineers. Other modes may be hidden and can be identified by advanced analytical models. The predictive model 114 can automatically detect operation modes by learning from historical sensor data and other production data, achieving compact feedback in the form of modes. The predictive model 114 can use these operation modes as benchmarks when similar production conditions existed in the past or when new modes may be identified. The predictive model 114 can use these operation modes to recommend better control actions or automatically change control parameters. The predictive model 114 can calculate historical benchmarking of the detected mode against historical plant data when the plant was in the same mode. The predictive model 114 can use the detected mode as a basis for identifying similar periods from history. For example, periods from history can be identified when the same mode was active.The predictive model 114 can further identify key differences between current and historical variable values ​​using factor analysis on a set of process variables that jointly define the detected mode. The predictive model 114 can calculate and display possible root causes of low-level operations / production to field engineers. The predictive model 114 can provide an automated process that provides a simpler, more human-involved approach to generating modes. The predictive model 114 can achieve opportunity realization through analysis using unsupervised machine learning. For example, the predictive model 114 can identify operational opportunities by comparing the current state with similar historical operations within that mode or its vicinity. The predictive model 114 can identify the specific mode in which the current operational state exists. The predictive model 114 can project clusters using t-SNE compression to create a graph. t-SNE can be a dimensionality reduction algorithm suitable for visualizing high-dimensional data, such as time series data 104. The predictive model 114 can use t-SNE compression to embed high-dimensional points into lower dimensions in a manner that considers similarities between points. In one example, the predictive model 114 achieves high bitumen extraction by comparing the current operational state with other operations in the same mode. The predictive model 114 can analyze episodes with poor performance and discover operational opportunities for improvement. The predictive model 114 can focus on those operational episodes in history that reached high froth production and extract key operational actions from these episodes to benefit current operations.

[0039] In one or more embodiments, the variable identification module 116 is configured to identify uncontrolled variables 122 and controlled variables 124 from the time series data 104. Based on the time series data 104, the uncontrolled variables 122 can be separated from the controlled variables 124 to ensure the ability to take action to capture production improvement opportunities. For example, the uncontrolled variables 122 can be time-stamped variables and can be defined as a set of variables from sensors over which a user has little or no control. The uncontrolled variables 122 can be parameters used to define how similar operational and production conditions are. The uncontrolled variables 122 can be derived from episodes of the time series data 104. The controlled variables 124 can be time-stamped variables and can be defined as a set of variables from sensors over which a user has control. In one example, the controlled variables 124 can be used to calculate rewards (or opportunities) from episodes of the time series data. The controlled variables 124 can be production rate and raw material variables that a user can optimize based on the best neighboring episodes found. The controlled variables 124 may be identified from an established neighborhood of similar historical uncontrolled variables 122 to generate possible action strategies associated with the current state based on the time series data 104. In one or more embodiments, the variable identification module 116 is configured to identify one or more controlled variables 124 of the same mode associated with the current operating state.

[0040] In one or more embodiments, the strategy recommendation module 118 is configured to recommend an action strategy based on the controlled variables 124, the uncontrolled variables 122, and the target productivity. The strategy recommendation module 118 can define a similarity measure to identify historical episodes from the time-series data 104 having similar operational states based on a comparison of the current state with the historical operational states. The strategy recommendation module 118 can generate episodes from the historical episodes that exhibit higher productivity or throughput. The strategy recommendation module 118 can generate scores based on the alternative action strategies and recommend action strategies based on the scoring for each alternative action strategy.

[0041] In one or more embodiments, the output module 120 is configured to output the action strategy to a user. The output module 120 may provide a user interface for interfacing with the user. The output module 120 may provide other suitable output means for communicating with the user. The output module 120 may provide an alert to the user regarding discovered operational opportunities. The output module 120 may display their operational modes using t-SNE methods. The output module 120 may display nearby episodes using a time-stamped chart. The output module 120 may display the estimated gain of the action strategy.

[0042] FIG. 2 is a flowchart 200 illustrating the operational steps of the opportunity discovery module 110 according to one embodiment of the present disclosure.

[0043] The opportunity discovery module 110 operates to monitor time series data 104 generated from one or more sensors. The opportunity discovery module 110 further operates to extract a set of features from the time series data 104 based on uncontrolled variables for the time series data 104 by autoencoding using a neural network, e.g., an LSTM autoencoder 112. The opportunity discovery module 110 operates to identify one or more operation modes based on the extracted features, including dimensionality reduction using representation learning from the time series data 104. The opportunity discovery module 110 operates to compare a current operation state with a historical operation state based on the time series data 104 for the same operation mode. The opportunity discovery module 110 operates to discover operation opportunities based on the comparison between the current operation state and the historical operation state. The opportunity discovery module 110 operates to identify controlled variables 124 for the same mode associated with the current operation state. The opportunity discovery module 110 operates to recommend an action strategy based on the controlled variables 124, the uncontrolled variables 122, and the target productivity. The opportunity discovery module 110 operates to output an action strategy to the user.

[0044] In step 202, the opportunity discovery module 110 monitors time series data 104 generated from one or more sensors. For example, the time series data 104 can be data from an oil sands operation and production process. For example, petroleum products can be produced from oil sands through several stages, such as extraction, upgrading, and refining. For example, solids and water can be removed during the extraction stage, during which bitumen can be extracted from oil sands. During the upgrading stage, bitumen can be upgraded to lighter intermediate crude products. During the refining stage, crude oil can be refined into final products such as gasoline, lubricants, and diluents. These processes within these stages may involve multiple sequential steps of physical or chemical transformations to convert one substance into another. An optimal balance of processes is required to achieve multiple goals within such a production system. In another example, the time series data 104 can be data from a steelmaking process, which produces steel from iron ore and / or scrap. Impurities such as nitrogen, silicon, phosphorus, sulfur, and excess carbon can be removed from a supply of iron, and alloying elements such as manganese, nickel, chromium, carbon, and vanadium can be added to produce different grades of steel. In yet another example, the time series data can be data from a production process where soybeans can be processed into soy sauce with additives. In yet another example, the time series data can be data from any other suitable operations and production processes.

[0045] In step 204, the opportunity discovery module 110 extracts a set of features from the time series data 104 by autoencoding using a neural network, such as an LSTM autoencoder 112, based on one or more uncontrolled variables 122 for the time series data 104. The opportunity discovery module 110 can use the set of features for training, or can learn the set of features using the LSTM autoencoder 112. The set of features can be information about the time series data 104. The set of features can be measurable individual attributes or characteristics of phenomena observed in the time series data 104. The opportunity discovery module 110 can select a subset of relevant features for creating the predictive model 114 based on the uncontrolled variables 122 that define the time series data 104 beyond the user's control. The opportunity discovery module 110 can reduce the number of resources required to describe the time series data 104. The opportunity discovery module 110 can use the LSTM autoencoder 112 to construct a combination of uncontrolled variables 122 that describes the time series data 104 with sufficient accuracy. The LSTM autoencoder 112 can be an artificial neural network used to learn efficient data encoding in an unsupervised manner. The LSTM autoencoder 112 can be a recurrent neural network with an autoencoder implementation for the time series data 104 using an encoder-decoder LSTM architecture. The LSTM autoencoder 112 can learn a representation (encoding) for the time series data 104, for example, for dimensionality reduction. From the reduced encoding, the LSTM autoencoder 112 can generate a representation of the time series data 104 that is as close as possible to the original input. The LSTM autoencoder 112 can include an encoder and a decoder.The encoder can use raw data (e.g., time series data 104) as input and generate features or representations as output. The decoder can use extracted features from the encoder as input and reconstruct the original raw input data as output. Training can be repeated until a stopping criterion is met. The LSTM autoencoder 112 can invoke an auto-encoding process using the time series data 104 to reduce the dimensionality of the sensor tag space to a limited embedding space. The LSTM autoencoder 112 can invoke a clustering method to generate latent modes. The LSTM autoencoder 112 can invoke a type of artificial neural network used to learn efficient data encoding in an unsupervised manner. The LSTM autoencoder 112 can obtain fixed-size vectors from the time series data 104.

[0046] In step 206, the opportunity discovery module 110 identifies one or more operation modes based on extracted features, including dimensionality reduction using representation learning from the time-series data 104. The opportunity discovery module 110 can identify neighbors of the current operation state. This neighbor can be a dynamic mode within the same operation mode and can be found by the Euclidean distance between the historical operation state and the current operation state. The opportunity discovery module 110 can define this neighborhood or identify static operation modes using Gaussian mixture clustering applied to the embedded space. Modes may include plant operation situations. Mode hard category types may relate to specific operation configurations of the entire production process, such as manufacturing pipe configurations or unit operation states. Mode soft category types may include production levels of local or global operations. The sands oil industry involves a complex process of converting sands oil to synthetic crude oil. To complete synthetic crude oil production, multiple stages are involved, including primary extraction, secondary extraction, and upgrading. Each stage involves multiple components and processes, making the system dynamic. Some of the included modes may be explicitly known by field engineers. Other modes may be hidden and can be identified by advanced analytical models. The opportunity discovery module 110 can automatically detect operation modes by learning from historical sensor data and other production data, achieving compact feedback in the form of modes. The opportunity discovery module 110 can determine operation mode algorithms and models. The opportunity discovery module 110 can use these operation modes as benchmarks when similar production conditions existed in the past or when new modes may be identified. These operation modes can be used to recommend better control actions or automatically change control parameters.If a new mode is identified, it can be saved to expand the system's memory and knowledge. The opportunity discovery module 110 can calculate historical benchmarking of the detected mode against historical plant data when the plant was in the same mode. The opportunity discovery module 110 can use the detected mode as a basis for identifying similar periods from history. For example, periods from history can be identified when the same mode is active. The opportunity discovery module 110 can further identify key differences between current and historical variable values ​​using factor analysis on a set of process variables that jointly define the detected mode. The opportunity discovery module 110 can calculate and display possible low-level operational / production root causes to field engineers. The opportunity discovery module 110 can provide analytical techniques for identifying or classifying modes throughout a complex manufacturing process. The opportunity discovery module 110 can provide detailed or general multivariate mode identification, and as a side effect, can also achieve partitioning of unpartitioned data into subsets. Rather than relying on rule-based mode detection, which requires prior knowledge and memorized principles, the opportunity discovery module 110 can provide an automated process that offers a simpler, more human-involved approach to generating modes. The opportunity discovery module 110 can achieve opportunity realization through analysis using unsupervised machine learning. For example, the opportunity discovery module 110 can identify operational opportunities by comparing the current state with similar historical operations within the mode or neighborhood.

[0047] In step 208, the opportunity discovery module 110 compares the current operational state with historical operational states based on time series data of the same operational mode among one or more operational modes. The opportunity discovery module 110 can identify the specific mode in which the current operational state exists. The opportunity discovery module 110 can project clusters using t-SNE compression to create a graph. In one example, t-SNE is a dimensionality reduction algorithm suitable for visualizing high-dimensional data, such as time series data 104. The opportunity discovery module 110 can use t-SNE compression to embed high-dimensional points into a lower dimension in a manner that considers similarities between points. In one example, the opportunity discovery module 110 can achieve high bitumen extraction by comparing the current operational state with other operations in the same mode. The opportunity discovery module 110 can analyze episodes with poor performance and discover operational opportunities for improvement. The opportunity discovery module 110 can focus on those operational episodes in history that have reached high froth production and can extract key operational actions from these episodes to benefit current operations.

[0048] In step 210, the opportunity discovery module 110 discovers operational opportunities based on a comparison of the current operational state with the historical operational state. The operational opportunities can be identified by comparing the current state with similar historical operations within the mode or vicinity. In one example, the operational opportunity can be a set of operational changes inferred from the historical operational state to increase the current operational state to higher production within a specified short future timeframe. In another example, the operational opportunity can be a set of operational changes inferred from the historical operational state to decrease the current operational state to lower additive or ingredient usage within a specified short future timeframe. Other suitable opportunities can be found.

[0049] In step 212, the opportunity discovery module 110 identifies one or more controlled variables 124 of the same mode associated with the current operational state. The controlled variables 124 may define actions that the user can control. The controlled variables 124 may be time-stamped variables and may be defined as a set of variables from sensors for which the user can control actions. In one example, the controlled variables 124 may be used to calculate rewards (or opportunities) from episodes of the time-series data 104. For example, the controlled variables 124 may be production rate and feedstock variables that the user can optimize based on the best-neighbor episodes found. The controlled variables 124 may be identified from an established neighborhood of similar historical uncontrolled variables 122 to create possible action strategies associated with the current state based on the time-series data 104. In the froth production example, the controlled variables 124 may include, for example, the feedstock for diesel hydrotreating or catalytic hydrotreating, the feed slate for diesel hydrotreating using low vacuum gas oil, low vacuum gas oil, side-draw kerosene, heavy naphtha, and coker kerosene.

[0050] In step 214, the opportunity discovery module 110 recommends an action strategy based on the controlled variables 124, the uncontrolled variables 122, and the target productivity. The opportunity discovery module 110 can define a similarity measure to identify historical episodes from the time series data 104 having similar operational states based on a comparison of the current state with the historical operational states. The opportunity discovery module 110 can create episodes from historical episodes that exhibit higher productivity or throughput. The opportunity discovery module 110 can generate scores based on the alternative action strategies and can recommend an action strategy based on the scoring for each alternative action strategy.

[0051] In step 216, the opportunity discovery module 110 outputs the action strategy to the user. The opportunity discovery module 110 may provide a user interface for interfacing with the user. The opportunity discovery module 110 may provide other suitable output means for communicating with the user. The opportunity discovery module 110 may display signals regarding the discovered operational opportunities. The opportunity discovery module 110 may display operation modes using t-SNE methods. The opportunity discovery module 110 may display nearby episodes using time-stamped charts. The opportunity discovery module 110 may display an estimated gain of the action strategy.

[0052] FIG. 3 illustrates an example functional diagram of opportunity discovery module 110 in accordance with one or more embodiments of the present disclosure.

[0053] In the example of FIG. 3 , the opportunity discovery module 110 can achieve opportunity realization through analysis using unsupervised machine learning. Operational opportunities can be identified by comparing the current state (or scenario) with similar historical episodes within a mode or neighborhood. The opportunity discovery module 110 can monitor operations as time series data 104 from sensors. The opportunity discovery module 110 can apply an LSTM autoencoder 112 to extract features of the time series data 104 into an embedded space. The LSTM autoencoder 112 can achieve time series representation learning, including dimensionality reduction. The LSTM autoencoder 112 can learn a representation (encoding) for the time series data 104, for example, for dimensionality reduction. From the reduced encoding, the LSTM autoencoder 112 can generate a representation of the time series data 104 that is as close as possible to the original input. In the illustrated embodiment, the LSTM autoencoder 112 includes an encoder 320 and a decoder 322. The encoder 320 can take the original input 324 (e.g., the time series data 104) as input and generate features or representations as output. The decoder 322 can use the extracted features from the encoder 320 as input and reconstruct the original input 324 into a reconstructed input 326 as output. The opportunity discovery module 110 can identify one or more static operation modes 302 (e.g., modes 304) using Gaussian mixture clustering applied to the embedded space. The opportunity discovery module 110 can project the clusters using t-SNE compression to create a graph of the modes 302. The modes 302 can be used as benchmarks from which the system can determine whether similar production conditions existed in the past. The modes 302 can be used to recommend better control actions or automatically change control parameters. The opportunity discovery module 110 can use the embedded space to identify dynamic modes (e.g., neighborhoods 308) for the current operation state 306.The opportunity discovery module 110 can discover operational opportunities, for example, depending on the operation mode 304 or neighborhood 308. In one example, the opportunity discovery module 110 can qualify historical episodes by selecting neighborhoods 308 of the current operational state 306 using the embedding space. The neighborhood 308 can be found by the Euclidean distance between the historical episode and the current state 306, as defined in the feature space. Block 310 illustrates opportunity realization by comparing the current operational state with other operations in the same mode, e.g., mode 304. Operational opportunities are indicated in an opportunity window 312 because poor episodes with poor performance may indicate opportunities for improvement.

[0054] FIG. 4 illustrates an example architecture diagram of the opportunity discovery module 110 in accordance with one or more embodiments of the present disclosure.

[0055] In block 402, a set of features may be selected to create a predictive model 114 based on historical data 404, e.g., time-series data 104. The predictive model 114 may identify potential production gains by using a smaller amount of raw material or additive without reducing production, or by increasing production of an intermediate or final product using the same amount of raw material or additive. The predictive model 114 may predict the potential for using a smaller amount of raw material or additive in production, or for increasing productivity using the same amount of raw material or additive. The predictive model 114 may invoke an opportunity identification algorithm to identify alternative operations to capture the opportunity. The historical data 404 may be data captured by one or more sensors. In one example, the historical data 404 may be data from an oil sands operation and production process. In another example, the historical data 404 may be data from a steelmaking process that produces steel from iron ore, scrap, or both. In yet another example, the historical data 404 may be data from any other suitable operation and production process. The controlled variables 124 and the uncontrolled variables 122 can be used to validate the predictive model 114. The predictive model 114 can use the uncontrolled variables 122 (e.g., environmental variables over which the user has little or no control) to retrieve episodes from the historical data 404 through neighborhood selection 406. The predictive model 114 can use the controlled variables 124 (e.g., effect variables over which the user can control or change) to calculate rewards (or opportunities) from the episodes of the historical data 404 through cluster generation 408. Cluster generation 408 can group a set of objects from the historical data 404 such that objects in the same group (referred to as a cluster) are more similar (in some sense) to each other than to objects in other groups (clusters).

[0056] The alternative control strategies 410 can be action strategies for selecting appropriate actions and changing the environment to realize the final opportunity. The alternative control strategies 410 can enable a user or plant manager to receive real-time support for decision-making to increase productivity. For example, the predictive model 114 can provide scoring 412 and sorting 414 for a recommendation list 416 to the user. In one example, a user interface on a handheld device can be provided to the user. The recommendation list 416 of potential target quantities can be selected from, for example, raw materials, intermediate and final products, or high-value additives.

[0057] 5 illustrates a block diagram 500 of components of a computing device 102 in accordance with an exemplary embodiment of the present disclosure. It should be understood that FIG. 5 illustrates only one implementation and is not intended to imply limitations with respect to the environments in which different embodiments may be implemented. Many modifications to the depicted environment may be implemented.

[0058] The computing device 102 may include a communications fabric 502 that provides communications between a cache 516, memory 506, persistent storage 508, communications unit 510, and input / output (I / O) interface 512. The communications fabric 502 may be implemented to have any architecture designed to communicate data and / or control information between processors (such as microprocessors, communications processors, and network processors), system memory, peripherals, and any other hardware components in the system. For example, the communications fabric 502 may be implemented by one or more buses or crossbar switches.

[0059] Memory 506 and persistent storage 508 are computer-readable storage media. In this embodiment, memory 506 includes random access memory (RAM). Generally, memory 506 may include any suitable computer-readable storage medium, volatile or non-volatile. Cache 516 is a high-speed memory that enhances the performance of computer processor 504 by retaining recently and nearly recently accessed data from memory 506.

[0060] The opportunity discovery module 110 may be stored in persistent storage 508 and memory 506 for execution by one or more of the respective computer processors 504 via cache 516. In one embodiment, persistent storage 508 includes a magnetic hard disk drive. Instead of or in addition to a magnetic hard disk drive, persistent storage 508 may include a solid-state hard drive, a semiconductor memory device, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, or any other computer-readable storage medium capable of storing program instructions or digital information.

[0061] The media used by persistent storage 508 may also be removable. For example, a removable hard drive may be used for persistent storage 508. Other examples include optical and magnetic disks, thumb drives, and smart cards inserted into a drive for transfer to another computer-readable storage medium that is also part of persistent storage 508.

[0062] In these examples, the communications unit 510 provides for communication with other data processing systems or devices. In these examples, the communications unit 510 includes one or more network interface cards. The communications unit 510 may provide communication through the use of either or both physical and wireless communications links. The opportunity discovery module 110 may be downloaded to the persistent storage 508 through the communications unit 510.

[0063] The I / O interface 512 allows for the input and output of data to and from other devices that may be connected to the computing device 102. For example, the I / O interface 512 may provide a connection to external devices 518, such as a keyboard, keypad, touch screen, or some other suitable input device, or a combination thereof. The external devices 518 may further include portable computer-readable storage media, such as thumb drives, portable optical or magnetic disks, and memory cards. Software and data used to implement embodiments of the present invention, such as the opportunity discovery module 110, may be stored on such portable computer-readable storage media and loaded into the persistent storage 508 via the I / O interface 512. The I / O interface 512 also connects to a display 520.

[0064] Display 520 provides a mechanism for displaying data to a user and may be, for example, a computer monitor.

[0065] The programs described herein are identified based on the application in which they are implemented in particular embodiments of the invention. However, it should be understood that any particular program names described herein are used merely as a matter of convenience and thus should not limit the invention to use with the particular application identified or implied by such names or any particular application identified and implied by such names.

[0066] The present invention may be a system, method, or computer program product, or combination thereof, at any possible level of technical detail of integration. The computer program product may include a computer-readable storage medium having computer-readable program instructions thereon for causing a processor to carry out aspects of the present invention.

[0067] The computer-readable storage medium may be any tangible device capable of holding and storing instructions for use by an instruction execution device. The computer-readable storage medium may be, for example, but not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. A non-exhaustive list of more specific examples of computer-readable storage media includes portable computer diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disk read-only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, mechanically encoded devices such as punch cards or raised structures in grooves having instructions recorded thereon, and any suitable combination thereof. As used herein, a computer-readable storage medium should not be construed as being a transitory signal itself, such as an electric wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse traveling in a fiber optic cable), or an electrical signal transmitted over an electrical wire.

[0068] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to a corresponding respective computing / processing device, or can be downloaded to an external computer or external storage device over a network, such as the Internet, a local area network, a wide area network, or a wireless network, or a combination thereof. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, or edge servers, or a combination thereof. A network adapter card or network interface within each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to a computer-readable storage medium within the respective computing / processing device for storage.

[0069] The computer-readable program instructions for carrying out the operations of the present invention may be assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, configuration data for an integrated circuit, or may be source or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Python®, C++, and procedural programming languages ​​such as the "C" programming language or the like. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (e.g., via the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry to carry out aspects of the present invention.

[0070] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0071] These computer-readable program instructions may be provided to a processor of a computer or other programmable data processing apparatus forming a machine, such that the instructions, executed by the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the blocks of these flowchart and / or block diagram illustrations. The computer-readable program instructions may further be stored on a computer-readable storage medium that can direct a computer, programmable data processing apparatus, or other device, or combination thereof, to function in a particular manner, such that the computer-readable storage medium having instructions stored therein comprises an article of manufacture containing instructions that implement aspects of the functions / acts specified in the blocks of these flowchart and / or block diagram illustrations.

[0072] These computer-readable program instructions can then be loaded onto a computer, other programmable data processing device, or other device to cause the computer, other programmable device, or other device to perform a series of operational steps to produce a computer-implemented process, such that the instructions, executed on the computer, other programmable device, or other device, perform the functions / acts specified in the blocks of these flowchart and / or block diagram illustrations.

[0073] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowcharts or block diagrams may represent a module, segment, or portion of instructions, including one or more executable instructions for implementing specified logical functions. In some alternative implementations, the functions shown in the blocks may be performed in an order different from that shown in the figures. For example, two blocks shown in succession may actually be performed as a single step, or may be performed simultaneously, substantially simultaneously, or with partial or complete overlap in time, or the blocks may sometimes be performed in the reverse order, depending on the functionality involved. It should also be noted that each block of the block diagrams and / or flowchart diagrams, and combinations of blocks in the block diagrams and / or flowchart diagrams, may be implemented by a dedicated hardware-based system that performs the specified functions or operations or implements a combination of dedicated hardware and computer instructions.

[0074] The description of various embodiments of the present invention has been provided for illustrative purposes, and the foregoing description is not intended to be exhaustive or to be limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art that do not depart from the scope and spirit of the invention. The terms used herein have been selected to best explain the principles, practical applications, or technical improvements of the embodiments over commercially available technology, or to enable those skilled in the art to understand the embodiments disclosed herein.

[0075] Although specific embodiments of the present invention have been described, those skilled in the art will recognize that there are other embodiments that are equivalent to the described embodiments. It is therefore to be understood that the present invention is not limited by the specific embodiments shown, but rather is limited only by the scope of the appended claims.

Claims

1. 1. A computer-implemented method comprising: extracting, by one or more processors, a set of features from time series data by neural network-based auto-encoding based on one or more uncontrolled variables for the time series data, the one or more uncontrolled variables defining a range of time series data that is not under user control; identifying, by one or more processors, one or more operation modes based on the features extracted by an extraction process including dimensionality reduction using representation learning from the time series data; identifying, by one or more processors, a neighborhood of the current operation state in a feature space based on the extracted features, the neighborhood being a dynamic mode within the same operation mode; comparing, by one or more processors, the current operational state with historical operational states within or near the same operational mode based on the time series data for the same operational mode of the one or more operational modes; discovering, by one or more processors, operation opportunities based on the comparison of the current operation state and the historical operation state; identifying, by one or more processors, one or more controlled variables of the same operation mode associated with the current operation state in which the operation opportunity was discovered from the neighborhood established for the one or more uncontrolled variables in the feature space, wherein the one or more controlled variables define actions that the user can control; and A computer-implemented method comprising: recommending, by one or more processors, an action strategy based on the one or more controlled variables, the one or more uncontrolled variables, and a target productivity.

2. monitoring, by one or more processors, the time series data generated from one or more sensors; The computer-implemented method of claim 1 further comprising:

3. The computer-implemented method of claim 1 or 2, wherein the neural network is a long short-term memory autoencoder.

4. 4. The computer-implemented method of claim 1, wherein discovering the operation opportunity is based on the comparison of the current operation state and the historical operation state using the same operation mode.

5. 5. The computer-implemented method of claim 1, wherein the operational opportunities are selected from the group consisting of a set of operational changes inferred from the historical operational states to increase the current operational state to higher production within a specified short time in the future, and a set of operational changes inferred from the historical operational states to decrease the current operational state to lower additive or raw material usage within a specified short time in the future.

6. determining, by one or more processors, a similarity measure for identifying historical episodes from the time series data having similar operational states based on the comparison between the current operational state and the historical operational states; and creating, by one or more processors, an episode from said historical episodes that exhibits a selection from a group consisting of higher productivity and throughput. The computer-implemented method of any one of claims 1 to 5, further comprising:

7. outputting the action strategy, wherein the outputting comprises: providing the user with an alert regarding the operational opportunity; representing the one or more modes of operation using a t-distributed stochastic neighbor embedding method; displaying the nearby episodes in the feature space using a time-stamped chart; and Displaying the estimated payoff of said action strategy. The computer-implemented method of any one of claims 1 to 6, comprising:

8. A computer program executed by a computer, A computer program causing the computer to execute the computer-implemented method according to any one of claims 1 to 7.

9. 1. A computer system comprising: one or more computer processors; one or more computer-readable storage media; and program instructions stored on the one or more computer-readable storage media, the program instructions being executed by at least one computer processor of the one or more computer processors, the program instructions comprising: program instructions for extracting a set of features from time series data by neural network-based auto-encoding based on one or more uncontrolled variables for the time series data, the one or more uncontrolled variables defining a time series data that is not under user control; and program instructions for identifying one or more operation modes based on the features extracted by an extraction process including dimensionality reduction using representation learning from the time-series data; program instructions for identifying a neighborhood of a current operation state in a feature space based on the extracted features, the neighborhood being a dynamic mode within the same operation mode; program instructions for comparing the current operational state with historical operational states within or near the same operational mode based on the time series data for the same operational mode of the one or more operational modes; program instructions for discovering operation opportunities based on the comparison of the current operation state and the historical operation state; program instructions for identifying one or more control variables of the same operation mode associated with the current operation state in which the operation opportunity was discovered from the neighborhood established for the one or more non-control variables in the feature space, the one or more control variables defining actions that the user can control; program instructions for recommending an action strategy based on the one or more controlled variables, the one or more uncontrolled variables, and a target productivity; 1. A computer system comprising:

10. program instructions stored on the one or more computer-readable storage media for monitoring the time series data generated from one or more sensors; 10. The computer system of claim 9, further comprising:

11. 11. The computer system of claim 9 or 10, wherein the neural network is a long short-term memory autoencoder.

12. 12. The computer system of claim 9, wherein the program instructions for finding the operation opportunity are based on the comparison of the current operation state and the historical operation state using the same operation mode.

13. 13. The computer system of claim 9, wherein the operational opportunities are selected from the group consisting of: a set of operational changes inferred from the historical operational states to increase the current operational state to higher production within a specified short future time frame; and a set of operational changes inferred from the historical operational states to decrease the current operational state to lower additive or raw material usage within a specified short future time frame.

14. program instructions stored on the one or more computer-readable storage media, the program instructions for outputting the action strategy; and wherein the program instructions for outputting the program instructions for providing an alert to the user regarding the operational opportunity; program instructions for representing the one or more modes of operation using a t-distributed stochastic neighbor embedding method; program instructions for displaying the nearby episodes in the feature space using a time-stamped chart; and program instructions for displaying an estimated gain of the action strategy, wherein the proximity is based on a Euclidean distance between a historical operation state and a current operation state.

Images