Generating a hybrid sensor for compensating for invasive sampling

Abstract

The hybrid sensor can be generated by training a machine learning model, such as a neural network, based on a training data set. The training data set can include a first time series of upstream sensor data (504, 514, 702, 802) having a forward dependence on a target variable (108, 110, 112, 604, 708, 804), a second time series of downstream sensor data (508, 518, 704, 806) having a backward dependence on the target variable (108, 110, 112, 604, 708, 804), and a time series of measured target variable (108, 110, 112, 604, 708, 804) data associated with the target variable (108, 110, 112, 604, 708, 804). The target variable (108, 110, 112, 604, 708, 804) has a lower measurement frequency than the measurement frequencies associated with the upstream sensor data (504, 514, 702, 802) and the downstream sensor data (508, 518, 704, 806). The hybrid sensor can estimate a value of the target variable (108, 110, 112, 604, 708, 804) at a given time, e.g., during which no actual measured target variable (108, 110, 112, 604, 708, 804) value is available.

Description

Background Technology

[0001] This application generally relates to computers and computer applications, and more specifically to the generation of hybrid sensors, for example, in process industries such as engineering, manufacturing, and / or another industrial or industrial process.

[0002] In processes such as petroleum, chemicals, and food, operational health depends on the complete and continuous monitoring of physical variables (e.g., pressure, temperature, and flow rate) and the chemical composition and / or concentrations within the engineered process. Systems enabling such process engineering can monitor these physical variables by deploying sensors within the engineered process. However, such monitoring may require reliance on invasive sampling methods and may involve time delays until the sensor data is known. For example, monitoring of chemical analysis in liquid streams in industrial processes relies on invasive manual sampling in a laboratory and is performed offline. This invasive manual sampling and laboratory analysis can introduce potential problems, such as a reduction in the number of samples that can be actually sampled and a lag in knowing the current chemical composition of the stream. Both of these problems lead to a higher likelihood of missing opportunities to fine-tune engineered processes for high productivity. Summary of the Invention

[0003] Systems and methods for generating hybrid sensors, for example, in industrial processes, can be provided. In one aspect, the system may include a hardware processor. A storage device may be coupled to the hardware processor. The hardware processor may be configured to receive a first time series of upstream sensor data having a forward dependency on a target variable. The hardware processor may also be configured to receive a second time series of downstream sensor data having a backward dependency on the target variable. The hardware processor may further be configured to receive a time series of target variable data of measurements associated with the target variable, the target variable having a measurement frequency lower than the measurement frequency associated with the upstream and downstream sensor data. The hardware processor may also be configured to determine a first time window of lag representing the forward dependency between the upstream sensor data and the target variable. The hardware processor may also be configured to determine a second time window of lag representing the backward dependency between the downstream sensor data and the target variable. The hardware processor can also be configured to train a machine learning model (such as a neural network) based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time.

[0004] In another aspect, a system may include a hardware processor. A storage device may be coupled to the hardware processor. The hardware processor may be configured to receive upstream sensor data of a first time series having a forward dependency on a target variable. The hardware processor may also be configured to receive a second time series of downstream sensor data having a backward dependency on the target variable. The hardware processor may further be configured to receive a time series of measured target variable data associated with the target variable, the target variable having a measurement frequency lower than the measurement frequency associated with the upstream and downstream sensor data. The hardware processor may also be configured to determine a first time window of lag representing the forward dependency between the upstream sensor data and the target variable. The hardware processor may further be configured to determine a second time window of lag representing the backward dependency between the downstream sensor data and the target variable. The hardware processor may also be configured to train a machine learning model (such as a neural network) based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. The hardware processor can be further configured to create a causal data structure that forward links a first time series of upstream sensor data with a time series of measured target variable data, and backward links a second time series of downstream sensor data with a time series of measured target variable data.

[0005] On the other hand, the system may include a hardware processor. A storage device may be coupled to the hardware processor. The hardware processor may be configured to receive upstream sensor data of a first time series having a forward dependency on a target variable. The hardware processor may also be configured to receive a second time series of downstream sensor data having a backward dependency on the target variable. The hardware processor may also be configured to receive a time series of measured target variable data associated with the target variable, the target variable having a measurement frequency lower than the measurement frequency associated with the upstream and downstream sensor data. The hardware processor may also be configured to determine a first time window of lag representing the forward dependency between the upstream sensor data and the target variable. The hardware processor may also be configured to determine a second time window of lag representing the backward dependency between the downstream sensor data and the target variable. The hardware processor may also be configured to train a machine learning model (such as a neural network) based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. The hardware processor can be further configured to create a causal data structure that forward-links a first time series of upstream sensor data with the time series of the measured target variable data, and backward-links a second time series of downstream sensor data with the time series of the measured target variable data. The upstream and downstream sensor data can be selected based on this causal data structure.

[0006] In another aspect, the system may include a hardware processor. A storage device may be coupled to the hardware processor. The hardware processor may be configured to receive upstream sensor data of a first time series having a forward dependency on a target variable. The hardware processor may also be configured to receive a second time series of downstream sensor data having a backward dependency on the target variable. The hardware processor may also be configured to receive a time series of measured target variable data associated with the target variable, the target variable having a measurement frequency lower than the measurement frequency associated with the upstream and downstream sensor data. The hardware processor may also be configured to determine a first time window of lag representing the forward dependency between the upstream sensor data and the target variable. The hardware processor may also be configured to determine a second time window of lag representing the backward dependency between the downstream sensor data and the target variable. The hardware processor may also be configured to train a machine learning model (such as a neural network) based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. The trained neural network can estimate the value of the target variable at a given time. Based on the value of the target variable estimated for a given time, the hardware processor can also control the setpoint in the industrial process.

[0007] In another aspect, a system may include a hardware processor. A storage device may be coupled to the hardware processor. The hardware processor may be configured to receive upstream sensor data of a first time series having a forward dependency on a target variable. The hardware processor may also be configured to receive a second time series of downstream sensor data having a backward dependency on the target variable. The hardware processor may further be configured to receive a time series of measured target variable data associated with the target variable, the target variable having a measurement frequency lower than the measurement frequency associated with the upstream and downstream sensor data. The hardware processor may also be configured to determine a first time window of lag representing the forward dependency between the upstream sensor data and the target variable. The hardware processor may further be configured to determine a second time window of lag representing the backward dependency between the downstream sensor data and the target variable. The hardware processor may also be configured to train a machine learning model (such as a neural network) based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. A trained machine learning model (such as a neural network) can estimate the value of a target variable at a given time. Based on the estimated value of the target variable for a given time, a hardware processor can further control the setpoint in an industrial process. This industrial process may include oil sands processing, and the target variable may include characteristics associated with tailings from the oil sands processing.

[0008] In another aspect, a system may include a hardware processor. A storage device may be coupled to the hardware processor. The hardware processor may be configured to receive upstream sensor data of a first time series having a forward dependency on a target variable. The hardware processor may also be configured to receive a second time series of downstream sensor data having a backward dependency on the target variable. The hardware processor may further be configured to receive a time series of measured target variable data associated with the target variable, the target variable having a measurement frequency lower than the measurement frequency associated with the upstream and downstream sensor data. The hardware processor may also be configured to determine a first time window of lag representing the forward dependency between the upstream sensor data and the target variable. The hardware processor may further be configured to determine a second time window of lag representing the backward dependency between the downstream sensor data and the target variable. The hardware processor may also be configured to train a machine learning model (such as a neural network) based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. Machine learning models such as neural networks can include an aggregation of the following: a first neural network trained to predict the value of a target variable based on a first time series of forward dependencies on upstream sensor data; a second neural network trained to predict the value of a target variable based on a second time series of backward dependencies on downstream sensor data; and a third neural network trained to predict the value of a target variable based on a time series of measured target variable data.

[0009] In another aspect, a system may include a hardware processor. A storage device may be coupled to the hardware processor. The hardware processor may be configured to receive upstream sensor data of a first time series having a forward dependency on a target variable. The hardware processor may also be configured to receive a second time series of downstream sensor data having a backward dependency on the target variable. The hardware processor may further be configured to receive a time series of measured target variable data associated with the target variable, the target variable having a measurement frequency lower than the measurement frequency associated with the upstream and downstream sensor data. The hardware processor may also be configured to determine a first time window of lag representing the forward dependency between the upstream sensor data and the target variable. The hardware processor may further be configured to determine a second time window of lag representing the backward dependency between the downstream sensor data and the target variable. The hardware processor may also be configured to train a machine learning model (such as a neural network) based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. Machine learning models, such as neural networks, can include aggregations of: a first neural network trained to predict the value of a target variable based on a first time series data with forward dependence on upstream sensor data; a second neural network trained to predict the value of the target variable based on a second time series data with backward dependence on downstream sensor data; and a third neural network trained to predict the value of the target variable based on a time series data of measured target variable data. The reporting time of the measured target variable may have a delay from the harvest time of the target variable, and the training dataset including the time series data of the measured target variable may include data of the measured target variable determined at the reporting time of the corresponding harvest time of the measured target variable. Machine learning models (such as neural networks) can estimate the value of the target variable at a given time without delay during runtime.

[0010] In one aspect, a method for generating hybrid sensors in an industrial process may include receiving a first time series of upstream sensor data having a forward dependency on a target variable. The method may further include receiving a second time series of downstream sensor data having a backward dependency on the target variable. The method may also include receiving a time series of measured target variable data associated with the target variable, the target variable having a measurement frequency lower than that associated with the upstream and downstream sensor data. The method may further include determining a first time window representing a lag in the forward dependency between the upstream sensor data and the target variable. The method may further include determining a second time window representing a lag in the backward dependency between the downstream sensor data and the target variable. The method may further include training a machine learning model (such as a neural network) based on a training dataset that includes at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time.

[0011] In another aspect, a method for generating hybrid sensors in an industrial process may include receiving upstream sensor data of a first time series that has a forward dependency on a target variable. The method may also include receiving a second time series of downstream sensor data that has a backward dependency on the target variable. The method may further include receiving a time series of measured target variable data associated with the target variable, the target variable having a lower measurement frequency than the measurements associated with the upstream and downstream sensor data. The method may also include determining a first time window representing a lag in the forward dependency between the upstream sensor data and the target variable. The method may further include determining a second time window representing a lag in the backward dependency between the downstream sensor data and the target variable. The method may also include training a machine learning model (such as a neural network) based on a training dataset that includes at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. The method may also include creating a causal data structure that forward links a first time series of upstream sensor data with a time series of measured target variable data, and backward links a second time series of downstream sensor data with a time series of measured target variable data.

[0012] In another aspect, a method for generating hybrid sensors in an industrial process may include receiving a first time series of upstream sensor data having a forward dependency on a target variable. The method may also include receiving a second time series of downstream sensor data having a backward dependency on the target variable. The method may further include receiving a time series of measured target variable data associated with the target variable, the target variable having a lower measurement frequency than the measurements associated with the upstream and downstream sensor data. The method may also include determining a first time window representing a lag in the forward dependency between the upstream sensor data and the target variable. The method may further include determining a second time window representing a lag in the backward dependency between the downstream sensor data and the target variable. The method may also include training a machine learning model (such as a neural network) based on a training dataset that includes at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. The method may further include creating a causal data structure that forward-links a first time series of upstream sensor data with the time series of the measured target variable data, and backward-links a second time series of downstream sensor data with the time series of the measured target variable data. The upstream and downstream sensor data can be selected based on this causal data structure.

[0013] In another aspect, a method for generating hybrid sensors in an industrial process may include receiving a first time series of upstream sensor data having a forward dependency on a target variable. The method may also include receiving a second time series of downstream sensor data having a backward dependency on the target variable. The method may further include receiving a time series of measured target variable data associated with the target variable, the target variable having a lower measurement frequency than the measurements associated with the upstream and downstream sensor data. The method may also include determining a first time window representing a lag in the forward dependency between the upstream sensor data and the target variable. The method may also include determining a second time window representing a lag in the backward dependency between the downstream sensor data and the target variable. The method may further include training a machine learning model (such as a neural network) based on a training dataset that includes at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. The trained machine learning model (such as a neural network) can estimate the value of the target variable at a given time. Based on the estimated value of the target variable for a given time, a hardware processor can also control a setpoint in the industrial process.

[0014] In another aspect, a method for generating hybrid sensors in an industrial process may include receiving upstream sensor data of a first time series that has a forward dependency on a target variable. The method may further include receiving a second time series of downstream sensor data that has a backward dependency on the target variable. The method may also include receiving a time series of measured target variable data associated with the target variable, the target variable having a lower measurement frequency than the measurements associated with the upstream and downstream sensor data. The method may further include determining a first time window representing a lag in the forward dependency between the upstream sensor data and the target variable. The method may further include determining a second time window representing a lag in the backward dependency between the downstream sensor data and the target variable. The method may further include training a machine learning model (such as a neural network) based on a training dataset that includes at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. The trained machine learning model (such as a neural network) can estimate the value of the target variable at a given time. Based on the estimated value of the target variable for a given time, a hardware processor can further control a setpoint in the industrial process. The industrial process may include oil sands treatment, and the target variable includes characteristics associated with tailings.

[0015] In another aspect, a method for generating hybrid sensors in an industrial process may include receiving upstream sensor data of a first time series that has a forward dependency on a target variable. The method may also include receiving a second time series of downstream sensor data that has a backward dependency on the target variable. The method may further include receiving a time series of measured target variable data associated with the target variable, the target variable having a lower measurement frequency than the measurements associated with the upstream and downstream sensor data. The method may also include determining a first time window representing a lag in the forward dependency between the upstream sensor data and the target variable. The method may further include determining a second time window representing a lag in the backward dependency between the downstream sensor data and the target variable. The method may also include training a machine learning model (such as a neural network) based on a training dataset that includes at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, to estimate the value of the target variable at a given time. Machine learning models such as neural networks can include an aggregation of the following: a first neural network trained to predict the value of a target variable based on a first time series based on a forward dependency of upstream sensor data; a second neural network trained to predict the value of a target variable based on a second time series based on a backward dependency of downstream sensor data; and a third neural network trained to predict the value of a target variable based on a time series based on measured target variable data.

[0016] In one or more aspects of the systems and / or methods disclosed herein, the first time series of upstream sensor data may be one-dimensional or multi-dimensional time series data. In one or more aspects of the systems and / or methods disclosed herein, the second time series of downstream sensor data may be one-dimensional or multi-dimensional time series data. In one or more aspects of the systems and / or methods disclosed herein, the machine learning model may be a neural network or a neural network model.

[0017] A computer-readable storage medium may also be provided for storing a program of instructions that can be executed by a machine to perform one or more of the methods described herein.

[0018] Other features, structures, and operations of various embodiments are described in detail below with reference to the accompanying drawings. In the drawings, the same reference numerals denote the same or similarly functional elements. Attached Figure Description

[0019] Figure 1 This is a timing diagram showing the operation of the system in the embodiment when implementing a hybrid sensor, and a diagram illustrating different sets of measurement results.

[0020] Figure 2This is a diagram illustrating tailings sampling in an oil sands industry, which serves as an example industrial process in one embodiment.

[0021] Figure 3 This is a diagram illustrating relevant metrics in an example of sand and oil tailings from an embodiment.

[0022] Figure 4 Another diagram illustrating the different frequencies at which different sensor data sets are acquired in the embodiments is shown.

[0023] Figure 5 This is a diagram illustrating the system components in an embodiment.

[0024] Figure 6 A causal processing graph (causal relationship graph) is shown in the embodiment.

[0025] Figure 7 This is another diagram showing the time range associated with the data source used in the embodiment to estimate the target sensor value.

[0026] Figure 8 This is a diagram illustrating the dynamic graph used for forward and backward inference relationships in an embodiment.

[0027] Figure 9 This is a diagram illustrating the processing flow of upstream variables, target variables, and downstream variables when determining one or more hidden sensor values ​​in an embodiment.

[0028] Figure 10 This is a diagram illustrating a neural network architecture in an embodiment that can be trained to estimate a target variable at a given time point.

[0029] Figure 11 This is a flowchart illustrating the method in the embodiment.

[0030] Figure 12 This is another flowchart illustrating the method in the embodiment.

[0031] Figure 13 This is a diagram illustrating components of a system in one embodiment, which in one embodiment can generate a hybrid sensor.

[0032] Figure 14 A schematic diagram of an exemplary computer or processing system that can implement a hybrid sensor system in one embodiment is shown. Specific Implementation

[0033] In one aspect, a system, apparatus, and method for generating hybrid sensors are disclosed. For example, the system, apparatus, and method can generate hybrid sensors to compensate for costly invasive sampling in industrial processes such as manufacturing, engineering, chemical, and other industrial processes by using relevant high-frequency sampling from non-invasive sensors. In one aspect, in one embodiment, the hybrid sensor can provide a technological improvement over existing one or more sensor devices or manual sampling that cannot be used to measure resulting data at relatively high frequencies (e.g., the desired rate of obtaining sensor data).

[0034] For example, a system capable of generating hybrid sensors may include computer-implemented components, such as those implemented and / or operating on or coupled to one or more processors or hardware processors. Methods for generating hybrid sensors may be executed and / or operated on one or more such processors or hardware processors. For example, one or more hardware processors may include components such as programmable logic devices, microcontrollers, memory devices, and / or other hardware components that can be configured to perform the corresponding tasks described in this disclosure. Coupled memory devices may be configured to selectively store instructions executable by one or more hardware processors.

[0035] The processor may be a central processing unit (CPU), a graphics processing unit (GPU), a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), another suitable processing component or device, or one or more combinations thereof. The processor may be coupled to a memory device. The memory device may include random access memory (RAM), read-only memory (ROM), or another memory device, and may store data and / or processor instructions for implementing various functions associated with the methods and / or systems described herein. The processor may execute computer instructions stored in memory or received from another computer device or medium.

[0036] As sensors are deployed in an engineering process or another industrial process, the system in this embodiment can continuously monitor physical variables using sensing devices, for example, by high-frequency, non-invasive, online sampling at intervals (e.g., at fixed intervals). Alternatively, the system can receive or acquire continuously monitored physical variables.

[0037] In one embodiment, the system can generate a hybrid sensor that estimates the number of targets falling in the unmeasured space between any two consecutive target variable measurements, which are considered to be far apart (e.g., separated by a relatively long insertion duration). In one embodiment, the system can create a causal graph or data structure to obtain the hybrid sensor, which fuses signals from multiple sources in a manner that utilizes the causal graph or data structure. The hybrid sensor can be implemented for various processing and manufacturing industries, for example, where composition, purity, or any quality-related quantity or variable is the target variable in question. The system can obtain timely and accurate predictions of quantities that, for example, were previously unavailable at the desired frequency. The hybrid sensor can provide finer resolution and earlier availability of target quantities, thereby allowing for better proactive action and / or decision-making. For example, based on the predicted number of unobserved targets, the system can automatically or autonomously control one or more setpoints, for example, automatically or autonomously changing one or more setpoints in an industrial process.

[0038] Figure 1 This is a timing diagram illustrating the operation of the system in the embodiment when implementing a hybrid sensor setup, and a diagram showing different sets of measurement results. In the embodiment, the system can operate with different types or sets of measurement results. Measurement results provide variables or quantities, such as temperature, pressure, the amount of components, and / or others. The upstream measurement result set can be measured by one or more sensors at a relatively high frequency (e.g., every 5 minutes) in online sensing or at another time interval (which may be a fixed interval). For example, one or more sensors measure upstream data in real time or near real time. The time series of upstream measurements is shown as follows. Figure 1 A series of ellipses. Multiple distinct sets of downstream measurements can exist. These downstream measurement sets can be measured by one or more sensors at a relatively high frequency, which is also online sensing, such as every 5 minutes or another time interval that could be fixed. For example, one or more sensors measure downstream data in real-time or near real-time. The time series of downstream measurements is shown as... Figure 1 A series of 'x's. The target measurement results (which can be measured by one or more sensors or determined based on analysis) corresponding to the target variable or quantity are only available at a slower or lower frequency (e.g., every 12 hours) than the upstream and downstream measurement result sets. The time series of the target measurement results over time is shown as... Figure 1 A series of rhombuses in the middle.

[0039] In an embodiment, a causal graph-driven model-based hybrid sensor can infer an unobserved target variable 110 at any time “t” 102, wherein the system operates on evidence (measured sensor data) from three sources: upstream finely measured covariates 104, e.g., sensor measurements from a time window prior to time “t” 102 (e.g., from an instantaneous left-hand time interval prior to time “t” 102); downstream finely measured covariates 106, e.g., sensor measurements from a time window after time “t” 102 (e.g., from an instantaneous right-hand time interval after time “t” 102); and the most recently available measurement for the target variable 108. The upstream measurements measure upstream materials or conditions, such as raw materials input into an industrial process; the downstream measurements measure downstream materials or conditions, such as byproducts, intermediate products, production, and / or operating conditions generated in the process. In one embodiment, a causal graph-driven model-based hybrid sensor can be implemented as a neural network architecture, wherein the neural network is trained or run to predict, for example, an unobserved target variable 110 between two observed target variables 108 and 112.

[0040] In short, an artificial neural network (ANN) or neural network (NN) is a machine learning model that can be trained to predict or classify input data. An artificial neural network can comprise a series of interconnected layers of neurons such that the output signals of neurons in one layer are weighted and passed to neurons in the next layer. A neuron Ni in a given layer can be connected to one or more neurons Nj in the next layer, and different weights wij can be associated with each neuron-neuron connection Ni-Nj to weight the signal transmitted from Ni to Nj. Neuron Nj generates its output signal based on its accumulated input, and the weighted signal can propagate from the input to the output neuron layer in successive layers of the network. The artificial neural network machine learning model can undergo a training phase in which the set of weights associated with the corresponding neuron layers is determined. In an iterative training scheme, the network is exposed to a set of training data, where the weights are repeatedly updated as the network “learns” the training data. The resulting trained model with weights defined via the training operation can be applied to perform tasks based on new data.

[0041] For example, an industrial process may include one or more sensors that measure one or more corresponding variables or quantities, which are set or input. An example of an industrial method could be an oil sands process, which may include extracting bitumen from oil sands deposits, which can result in diluted bitumen; bitumen lifting, which can result in synthetic crude oil; and crude oil refining, which can produce final products such as gasoline, diesel, lubricants, and bitumen. In such industrial processes, measurements or sampling are performed to monitor the process and the quality of the intermediate and / or final products produced. However, obtaining some of these measurements can be invasive and may not be available at desired frequencies. On the other hand, samples requiring laboratory treatment can lead to delays in obtaining measurement results, for example, depending on the rate of laboratory treatment. In embodiments, hybrid sensors can, for example, measure or predict such unobserved measurements in near real-time and can allow for faster responses to correcting potential problems detected in the process and / or improve the quality of intermediate and / or final products produced in the process.

[0042] Figure 2 This is a diagram illustrating tailings sampling in an oil sands industry as an example industrial process in one embodiment. Tailings sampling analysis can be performed twice daily. Tailings are a mixture of water, sand, clay, and residual bitumen, and are a byproduct of the hot water treatment process 204 used to separate bitumen from sand and clay. Tailings are stored in a large-scale engineering dam and dam system called tailings pond 202. Tailings pond 202 is a settling tank that enables the separation and continuous recycling of treated water. Water is continuously recycled from the tailings pond back to the extraction process, thereby reducing the use of freshwater from rivers and other sources. Oil sands producers recycle 78%–86% of their water. Tailings can have environmental and product outcome impacts. The tailings composition is associated with high-frequency, (online) non-invasive sensors 206 (e.g., measured every 5 minutes) that correspond to upstream variables such as feedstock and downstream measurements or sensor values ​​208 corresponding to downstream materials produced by hot water treatment (after primary treatment). The hybrid sensor in the embodiment can measure tailings, for example, between observed samples (e.g., between sample analyses), and can provide continuous measurement results of tailings 210.

[0043] Figure 3This is a diagram illustrating relevant metrics in an example of asphalt tailings from an embodiment. In the primary asphalt treatment method 304, the raw material 302 is treated with hot water. Various sensors measure upstream sensor data or metrics, such as raw material content measurements and inflow and process variables 310. The hot water treatment method produces tailings (byproducts) 306. The remaining portion of the asphalt mixture (excluding tailings) is treated with asphalt secondary treatment (e.g., centrifugation 308). After hot water treatment 304, various sensors also measure downstream sensor data or metrics, such as post-treatment content measurements and effluent variables 312. In this example, the target variable 314 could be a measurement associated with a tailings sample, such as the content of a tailings sample produced as a byproduct (or intermediate variable) of the thermal treatment process. Such measurements are typically performed at a lower frequency than upstream sensor metrics 310 and downstream sensor metrics 312, and may incur costly invasive sampling, such as offline sampling, for example, laboratory analysis. Both upstream sensor metric 310 and downstream sensor metric 312 are high-frequency sampling non-invasive sensor metrics, for example, sampling at a higher frequency than the target variable sampling.

[0044] Figure 4 Another diagram illustrating the different frequencies at which different sensor data sets are acquired in the embodiments is shown. Upstream sensor metrics 402 may be acquired at fixed time intervals, such as at a first frequency. Downstream sensor metrics 404 may be acquired at another fixed time interval (e.g., at a second frequency). The target variable 406 for which it is desired to estimate its unobserved or unmeasured values ​​may be a variable or quantity that is measured at a frequency lower than the first and second frequencies.

[0045] In embodiments, the system and / or method combine actually monitored lower frequency values ​​with estimated higher frequency values ​​to generate a hybrid sensor. The hybrid sensor allows for the inference and storage of values ​​for variables with invasive, low-frequency measurements; otherwise, at a much higher rate comparable to large-scale data analysis of high-frequency, non-invasive measurements. The comparable and consistent availability of measurements at fine frequencies enables better data-driven decision-making and control for processes with optimal recovery and yield at acceptable quality. In embodiments, the hybrid sensor can allow for the detection of potential environmental breaches, for example, before a potential disaster or catastrophe occurs due to cumulative effects. On the other hand, the hybrid sensor can allow for the earlier identification of any substantial loss of expensive or economical raw materials.

[0046] Figure 5This is a diagram illustrating system components in an embodiment. The illustrated components include computer-implemented components, such as one or more processors or hardware processors that execute computer processes and / or threads as described above. In one embodiment, the system may utilize backward inference using downstream (e.g., fine) measurement results and forward inference using upstream (e.g., fine) measurement results and past self (e.g., coarse) measurement results when identifying one or more hidden values ​​for a hybrid sensor, respectively. For example, storage device 502 may store historical data including upstream measurement result (or metric) data (such as sensor measurement result data), downstream measurement result (or metric) data (such as sensor measurement result data), and hybrid sensor data. As described above, upstream measurement result data may include data associated with input materials and / or conditions in a process (e.g., an industrial, engineering, or chemical process). Downstream measurement result data may include data associated with the effects on the process caused by the treatment of input materials and / or conditions. Hybrid sensor data may include data associated with variables or properties of the process for which the hybrid sensor estimates unobserved values ​​at points in time (e.g., between times of actual measurement of the metric). Hybrid sensor data may include historical actual measurement results of variables or attributes, and may also include previously estimated values. Upstream and downstream data can be acquired or collected by sensors performing high-frequency sampling (e.g., at frequencies higher than those of hybrid sensor data) in real time during industrial processes (such as physical or chemical processes). The acquired sensor results may be stored back into a storage repository (e.g., storage device 502) for example, instantaneously, in real time, or near real time.

[0047] The computer processor may receive or acquire (directly or indirectly) upstream data 504, downstream data 508, and hybrid sensor data 506, for example, from storage device 502 storing a historical database. The computer processor may also receive estimated time points. In an embodiment, a time window before the estimated time point and a time window after the estimated time point are determined. In subsequent processing, upstream data measured within the time window before the estimated time point can be used, and downstream data measured within the time window after the estimated time point can be used. For simplicity, the time window before the estimated time point is also referred to as a first time window. For simplicity, the time window after the estimated time point is also referred to as a second time window. In an embodiment, the received data may be data within those time windows.

[0048] At 510, the computer processor may establish an in-memory data structure (e.g., a causal graph or causal processing graph) representing causality or relationship between measurement outcomes (e.g., upstream and downstream measurements). The causal graph or causal processing graph (also referred to as a qualitative causal graph) may include specific upstream and downstream variable names and / or identifiers: this information may be received as input. In an embodiment, the box shown at 510 includes a data structure that captures the qualitative information in the form of a corresponding graph data structure, which is also associated with corresponding time-series data or sensor measurements for each of these variables (i.e., the quality variables in question that are difficult to sense / measure, whose values ​​are estimated using embodiments of the methods disclosed herein by utilizing the qualitative graph, with a corresponding computational data structure containing time-series measurements).

[0049] In an embodiment, the computer processor can perform feature selection 512, identifying upstream and downstream sensors that have a strong relationship with the target variable. Other feature selection methods can be used, which can use statistical data to evaluate the relationship between each input variable and the target variable, and select those input variables with the strongest relationship to the target variable. The selected features can include data that can be determined to be relevant to identifying or estimating data from the mixed sensors. Examples of features may include material content and operating conditions, such as temperature, pressure, and other operating conditions. For example, using... Figure 2 and Figure 3 In the example shown, features may include raw material content measurements, inflow and process variables, post-processing content measurements, and outflow variables. Hybrid sensor data may also be selected as features, such as tailings sampling data. In this embodiment, the feature selection at 512 calculates parameters “tau_1” (also referred to as tau_1) and “tau_2” (also referred to as tau_2), which are hysteresis and leading parameters corresponding to the graphical data structure. For example, these parameters are calculated based on time-series measurement data held or stored in the graphical data structure.

[0050] In one embodiment, the computer processor constructs and uses an in-memory causal processing graph data structure (e.g., an internal computer data structure for computer processing) and identifies the selected upstream and downstream sensors based on the causal processing graph data structure.

[0051] In one embodiment, a computer processor can estimate the hidden value of a hybrid sensor by forward inference or estimation using upstream data 514, self-inference or estimation using hybrid sensor data 516, and forward inference or estimation using downstream data 518. The computer processor aggregates the estimated hidden values ​​to a final sensor value at 520. In an embodiment, the components shown at 514, 516, 518, and 520 may include neural networks or aggregations of neural networks. The final value may be used, for example, for application 522. Application 522 may include providing a hybrid sensor, providing an environmental hazard emission warning, a productivity warning, or another application. In one aspect, the hybrid sensor may be connected to a physical system and activate a controller to change one or more of the inputs (e.g., the material content and / or operating conditions used in the physical system). For example, a valve or conduit that controls the amount of material content added to the physical system may be opened or closed to control the quantity. As another example, a setpoint may be controlled to set operating conditions such as temperature, pressure, and / or another condition. In an embodiment, the hybrid sensor value thus determined may be used to automatically control the physical system in real time or near real time, for example, by having the computer processor communicate directly with the controller of the physical system. The determined hybrid sensor values ​​can also be stored, for example, in real-time or near real-time, in storage device 502. The stored hybrid sensor values ​​can then be made available to application 522.

[0052] Figure 6 A causal processing graph (causality graph) is shown in the embodiment. Upstream process variables (upstream sensors or features) 602 that influence the target variable 604 are linked in a data structure. The target variable 604 is also linked to downstream process variables 606 that are influenced by the target component. In this embodiment, upstream process variables 602 that influence downstream process variables 606 may also be linked.

[0053] In this embodiment, causality is achieved from an industrial process sequence, such as a physical and chemical process sequence. tau_1, tau_2, and tau_3 are values ​​greater than 0, representing time lags for the effect. One or more of these lags can be used as time series windows for analysis, for example, to identify or estimate unobserved hybrid sensors. For example, the time period from time point t-tau_1 to time point t (the time when unobserved hybrid sensors are to be estimated), during which the effect of upstream process variables on the target variable can be observed through the relationship between upstream process variables and the target variable (also referred to as the first time window above).

[0054] Downstream variable 606 can be observed during the time period from time point t to time point t+tau_2 (also referred to as the second time window). In the embodiment, tau-3 is greater than tau-1, tau-3 is greater than tau-2, and tau-3 can be equivalent to tau-1+tau2.

[0055] Figure 7 This is another diagram illustrating the time range associated with the data source used in the embodiment to estimate the target sensor value. The data source includes upstream data (x_u) 702 acquired or monitored during the time period t-tau_1 to t, downstream data (x_d) 704 acquired or monitored during the time period t to t+tau_2, and past monitored target values ​​(y(t)) 706. The target variable 708, e.g., the hybrid sensor value, is estimated at time t. The system (e.g., the system's computer processor) can generate near real-time hybrid sensor values. In this respect, the maximum delay for the sensor value determined at time point t is tau_2. The value of tau_2 is small, for example, 2 minutes. In such an example, the time interval between downstream data samples may be less than tau_2.

[0056] Figure 8 This is a diagram illustrating another dynamic graph used for forward and backward inference relationships in an embodiment. The time series 802 of the upstream data has a forward inference relationship with the time series 804 of the target variable and the time series 806 of the downstream data (e.g., shown in 808 and 810), from the previous time step to the next time step. In the time series 804 of the target variable, the target variable has a relationship with the next target variable in the next time step, for example, as shown in 812. The time series 806 of the downstream data has a backward inference relationship with the time series of the target variable; for example, the downstream data in the next time step has a relationship with the target variable in the previous time step, for example, as shown in 814.

[0057] Figure 9This is a diagram illustrating the processing flow of upstream variables, target variables, and downstream variables when determining one or more hidden sensor values ​​in an embodiment. In this embodiment, the process flow uses a dynamic diagram. Upstream variables or data or sensor data 902 include one or more upstream variables or features, such as U1…Uk, where k indicates the number of different upstream variables. Although not shown, time series for each of the different upstream variables (e.g., from t to t-tau_1) can be used. In one embodiment, target variable 904 is a variable, such as y, and its value is estimated at time ty(t). Downstream variables or data or sensor data 906 may include one or more downstream variables or features, such as D1…D1, where l (the letter l) indicates the number of different downstream variables. Although not shown, time series for each of the different upstream variables (e.g., from t to t+tau_2) can be used.

[0058] Figure 10 This is a diagram illustrating a neural network architecture in one embodiment, which can be trained to estimate a target variable at a given time point. For example, the trained neural network can be used as a hybrid sensor.

[0059] Neural network (f) U )1002 Perform upstream estimation. Using the ground truth data from the upstream data and the target variable influenced by the upstream data, the neural network (f U The neural network 1002 is trained to estimate a target variable from a given feature value (e.g., a time series) of upstream data at any given time point. In an embodiment, the neural network 1002 computes intermediate features, i.e., values ​​output by the neural network. These intermediate features (or values ​​generated by the neural network) capture the separable effects of lagged values ​​of individual upstream variables on the target. The input to the neural network 1002 is a historical window for each upstream variable (XU1, XU2, ...). For each of these upstream variables, e.g., XU_1, the input to the neural network 1002 could be a window from (t-tau)... Dmax -tau U1 ) begins and in (t-tau Dmax The window of time series data ending at (t-tau) Dmax -tau U1 ) begins and in (t-tau Dmax The consecutive values ​​on the window that end.

[0060] The input to the neural network can be XU1(t-tau) Dmax -tau U1 )...XUk(t-tau Dmax -tau UkIn this embodiment, the input to the neural network is a history window of a predetermined length (e.g., W ≥ 1) for each upstream variable (XU1, XU2, ...). For each of these upstream variables, such as XU_k, the input may be a history window from (t - tau) Dmax -tau U1 ) begins and in (t-tau Dmax The window of time series data ending at (t-tau) Dmax -tau U1 The continuous values ​​on the window that end. In the embodiment, each training point corresponds to a time point in which the target variable has been measured; for example, if t* is a time point in which the target has been measured in the training data, then (t-tau) Dmax = t*), and all different input windows relative to the above training time point t* can be implied by the above description. In an embodiment, this process selects the training point t*, such that the process also has measurements for all downstream variables, which are measurements taken at times after t*, i.e., at (tau) Dmax Measurements at +t*). After this training, if the current time is t, the program is always configured to measure at time (t-tau). Dmax The target variable is estimated at (). The output of neural network 1002 is the estimated target variable.

[0061] Here, "t" is used to represent the current time when the system and / or method of one embodiment disclosed herein are used in practice. Relative to real-time "t", the system can operate at time (t-tau). Dmax This generates estimates of the target variable. These estimates are relative to a point in time, i.e., relative to time (t-tau). Dmax The system is guaranteed to have all downstream variables at its corresponding “advance” time, because the most finite such “advance” time is when its “advance” time is tau. Dmax The downstream variables. Since the system is relative to time (t-tau) Dmax The system is in real-time "t", so it can access all measurements of downstream variables.

[0062] tau Dmax It refers to the time window of maximum delay between the point in time when the target variable is being estimated and tau_2 (the end of the sampling time point for downstream data).

[0063] "Ahead of time" relative to any other time, such as "tau" or \tau, refers to a time to the right (later, or in the future) relative to time \tau.

[0064] "Lag" relative to any time (e.g., \tau) refers to the time relative to \tau that is to the left (before or in the past).

[0065] \tau_{U_k}: All of these are "lags" because the upstream variables with influence / causation are lagged values ​​from the past. "U" refers to upstream. U1 to Uk refer to k distinct upstream data. For example, one could use k upstream data, each with a time series value.

[0066] \tau_{D_k}: All of these are "early" because the downstream variables that influence / cause-related relationships are later / earlier values ​​from the relatively future. "D" refers to downstream.

[0067] Here, "t" is used to represent the current time, for example, when using the systems and / or methods disclosed herein in practice. The system can be used at time (t-tau) relative to real-time "t". Dmax This generates estimates of the target variable relative to time (t-tau). Dmax The system uses the lagged values ​​of upstream variables. For the upstream variable U_1, this would mean that variable U_1 is relative to the estimated time point (i.e., time (t-tau)). Dmax The lag value of )) has a lag of \tau_{U_1}: therefore, the timestamp t-tau Dmax -tau U1 As described above, for each of these upstream variables (e.g., XU_1), the input can be a window of continuously measured time series data, ranging from (t-tau) Dmax -tau U1 ) begins and in (t-tau Dmax The end of (t-tauDmax) refers to the continuous values ​​within the window that end at (t-tauDmax).

[0068] Neural network (f) T )1004 performs self-estimation. By using the true data of the target variable and its impact on the next target variable, the neural network (f T The neural network 1004 is trained to estimate or predict the next value of the target variable given a value of the target variable. The input to the neural network 1004 can be y(t-tau). Dmax-tau_{y}). "tau_{y}" represents the timestamp of the most recent measurement of the target variable "y" in the past, for a coarse measurement. It is also a window of consecutive past measurements of length W (>=1), ending at timestamp \tau_{y}. When W=1, it truly represents the most recent measurement. When W=2, it represents the two most recent measurements from the past, and so on. W is a modeling parameter. W can be pre-selected or pre-configured as a configuration parameter. The output of neural network 1004 is the estimated target variable.

[0069] Neural network (f) D )1006 performs downstream estimation. Using real data from the downstream data and the target variable affecting said downstream data (backward inference), a neural network (f) can be trained. D The neural network 1006 is used to predict or estimate a target variable given future downstream data. The input to the neural network 1006 can be x. D1 (t-tau Dmax -tau D1 ), ..., x D1 (t-tau Dmax -tau D1 In D1...D1, l (letter l) refers to the number of different downstream variables l...l (letter l).

[0070] Neural network (g) 1008 aggregates neural networks 1002, 1004, and 1006 to output an aggregated estimate of the target variable. For example, neural network (g) can be trained based on real data at 1002, 1004, and 1006 to output an aggregated target value. For instance, neural network (g) 1008 can estimate y(t-tau) Dmax ).

[0071] y(t-tau Dmax )=g(f U (x U1 (t-tau Dmax -Tau U1 ), ..., x Uk (t-tau Dmax -tau Uk )), f T (y(t-tau Dmax -Tau y ),

[0072] f D (x D1 (t-tau Dmax -Tau D1 ), ..., x D1 (t-tauDmax -tau D1 ))

[0073] Tau Dmax =max i tau Di For example, Tau Dmax It is the end time point of the time series window for downstream data. The neural network (g)1008 is used, for example, as a hybrid sensor.

[0074] Here, "t" refers to, for example, the current real-time "t" when it relates to a system used in real-time. Relative to real-time "t", the system can be used in time (t-tau). Dmax This generates an estimate of the target variable.

[0075] Figure 11 This is a flowchart illustrating the method in the embodiments. A processor or hardware processor can execute the method in the embodiments. At 1102, the method may include estimating the upstream lag window length (e.g., tau_1) related to the forward dependency of the target variable measured at the time the target variable was acquired. In the embodiments, known standard time series statistical techniques are used to estimate lags and advances. For example, for different values ​​of lag (e.g., 1, 2, 3, 4, ...), the cross-correlation between the target and each upstream variable is calculated, and the lag corresponding to the highest cross-correlation between the upstream variable and the target is selected as the "lag" parameter. In another approach, the system may leave this "lag" parameter as a model hyperparameter over a lag range (e.g., [1, 2, 3, ..., L]) and select the best lag value for each upstream variable using a validation set and hyperparameter tuning. "Lagged relative to any time, e.g., tau" refers to time to the left (before, or in the past) relative to time tau. tau_{U_k}: All these values ​​are "lags" because the upstream variables with influence / causation are lagged values ​​from the past.

[0076] In 1104, the method may include estimating the downstream lag window length (e.g., tau_2) related to the backward dependency of the target variable measured at the time the target variable was acquired. In an embodiment, known standard time series statistical techniques are used to estimate the lags and advances. For example, for different values ​​of lag (e.g., 1, 2, 3, 4, ...), the cross-correlation between the target and each downstream variable is calculated, and the advance corresponding to the highest cross-correlation between the downstream variable and the target is selected as the "advance" parameter. In another approach, the system may leave this "advance" parameter as a model hyperparameter over a lag range (e.g., [1, 2, 3, ..., L]) and select the best advance value for each downstream variable using a validation set and hyperparameter tuning. "Advance" relative to any time (e.g., tau) refers to a time to the right (later, or in the future) relative to time tau. tau_{D_k}: All these values ​​are "advance" because the downstream variable with influence / causation is a later / earlier value from the relatively future.

[0077] At 1106, the method may include integrating measurements of an upstream variable from the immediately preceding time window (from the target variable acquisition time), the duration (length) of which is tau_1 from 1102. The upstream variable includes measurements with a finer or higher frequency than those used to generate the target variable estimated by the mixed sensors. This time window is also referred to as the first time window. In embodiments, various time series feature derivations may also be performed on the first time window. Known techniques can be used to derive time series features from time series windows with one or more consecutive measurements.

[0078] At 1108, the method may include integrating measurements of downstream variables from a subsequent or immediately following time window (the time from which the target variable was acquired), the duration (length) of which is tau_2 from 1104. Downstream variables include measurements with a finer or higher frequency than those generated by generating the mixed sensor data to estimate the target variable. This time window is also referred to as a second time window. In embodiments, various time series feature derivations may be performed within the second time window. Known techniques can be used to derive time series features from time series windows with one or more consecutive measurements.

[0079] At 1110, the method may include using supervised and / or real data available in historical data of measurements (e.g., laboratory measurements) to learn (train) a mapping from “input” to “output” (target variable) of a combination of measurement results across features from 1106 and 1108, where the “output” (target variable) is a hybrid sensor for variables measured less frequently (e.g., laboratory measurement variables, which may have delays). In an example, where there is a time delay from acquiring and obtaining measurement result data, for example due to laboratory processing time, these “output” variables may be temporally aligned with corresponding historical acquisition periods. In one embodiment, learning includes training one or more neural networks, such as... Figure 10 As shown.

[0080] For historical learning in this embodiment, the processing at 1106 and 1008 considers upstream and downstream windows of size relative to all historical acquisition periods, generated by the processing at 1102 and 1104, respectively, for all historical acquisition periods, there are corresponding target variables, such as aligned reported laboratory data of the variable of interest (e.g., the target variable for which hybrid sensors are being generated).

[0081] In 1112, the learned mapping or neural network can run at any point in time (e.g., "t_any") by constructing upstream and downstream windows of lengths tau_1 and tau_2 relative to the time point (t_any-tau_2) to obtain the estimate of the hybrid sensor at the time point (t_any-tau_2). While this example illustrates a simple scenario with one upstream variable and one downstream variable, the method can be operated with multiple upstream and / or multiple downstream variables. In embodiments, the various "lag" and "lead" values ​​can be different for each specific lag (upstream) or lead (downstream) variable. The values ​​used to train the neural network and the values ​​derived through the neural network can be the same values ​​for these "lag" and "lead" values, respectively, corresponding to the different upstream and downstream variables.

[0082] This method can provide a more refined estimate of the frequency of a slowly measured variable of interest (target variable). For example, at any point in continuous time (e.g., t_any), the method estimates the target variable of the previous time step, i.e., (t-tau_2).

[0083] Return to reference Figure 4For example, in an example scenario where there is a delay between the actual sampling time and the time of determining the measurement for the target variable (of interest, for which a hybrid sensor is generated), the following terms may be applied: “the time period of acquisition” refers to the time when the sample is drawn 408 (e.g., and sent for analysis, such as to the laboratory to determine the measurement); “the time period of reporting” refers to the time when a laboratory report (or another analytical report) 410 is available for the corresponding acquired sample after a laboratory-related delay (or another analytical delay).

[0084] To create training data that includes input-output data, the method can align or shift lab-reported values ​​(or measurements of target variables with a delay between the collected and reported time periods) to the corresponding "collected time." In embodiments, the method also partitions the time range of the target metric (e.g., measured using invasive sampling) based on the time from the collected time period to different time periods. The training input-output data includes, for example, time series of upstream data measured in a first time window and time series of downstream data measured in a second time window (as input), and a measured target variable having a collection time after the first time window and before the second time window, wherein the first time window, the collection time, and the second time window are temporally continuous. An aggregated neural network trained on the training data (e.g., Figure 10 (As shown in the illustration) to provide a hybrid sensor that can measure a target variable upon request or at a given time. In an embodiment, different variables may have different measurement rates.

[0085] For example, for a given time t, the hybrid sensor measurement at time t is calculated based on: the upstream sensor value measured within the time window [t-tau_1, t]; the downstream value measured within the time window [t, t+tau_2]; and the measured target variable metric (e.g., laboratory results with a target metric of “report time”). The time used in training associated with the measured target variable metric is an alignment or shift time shifted to the corresponding acquisition time (here, the acquisition time is less than t). In embodiments, different variables may have different measurement rates. Here, “t” is used to represent the current time, e.g., when using or implementing the system disclosed herein in practice. The system can be at a time (t-tau_1, t+tau_2) relative to real-time “t”. Dmax This generates an estimate of the target variable. Relative to a point in time, i.e., relative to time (t-tau). Dmax The system is guaranteed to have all downstream variables at its corresponding “advance” time, because the most finite such “advance” time will be the time when the “advance” time is the downstream variable of tauDmax. Since relative to time (t-tau)Dmax In real time "t", the system accesses all measurements of downstream variables.

[0086] In an embodiment, the method can use a causal graph to pose the estimate of a target variable (e.g., a laboratory variable) using a bidirectional approach (e.g., using both forward effects from upstream variables and backward effects from downstream variables) over a near-continuous time period. Previously measured target variables can also be used to estimate the next target variable. This estimation can be based on these three sets of measurements, for example, using a combined estimate. Alternatively, the estimation can be performed based on those three sets of measurements, for example, using a combination of individual measurements estimated for the measurement sets.

[0087] In an embodiment, the method can match low-rate monitoring of one sensor with high-rate monitoring of another. For example, the method can fill gaps in low-rate monitoring with values ​​estimated using a causal relationship plot and the true values.

[0088] In embodiments, the method can generate hybrid sensors to address problems or difficulties associated with slow-monitoring rate sensors. For example, the method generates hybrid sensors in situations where high-frequency and automated monitoring is difficult or impossible (e.g., in tailings monitoring), creating hybrid sensors for those sensors that are monitored less frequently. In one aspect, the method uses slow-rate measurements and causal graphs to find values ​​not directly measured by the sensors.

[0089] In embodiments, a method, system, and apparatus may be provided that generates hybrid sensors for a target variable or metric in an industrial process, such as a manufacturing or chemical production process. This target variable or metric may include expensive, invasive measurements with large sampling intervals (low-frequency measurements) and delays due to analytical (e.g., laboratory) work. In one embodiment, the method, for example, creates a qualitative causal graph or data structure that illustrates what the links are between upstream and downstream causal edges. The method also receives sensor data from the measured variable in the industrial process at their available sampling and measurement rates. The method also estimates the target variable at the same or similar high rate as the set of upstream and downstream variables. The method may further create hybrid sensors using recent laboratory measurements and estimates of the target variable. The method may send alarm or signaling messages to the industrial process based on the results of the generated hybrid sensors, such as, but not limited to, hazardous emission alarms, productivity-related process warnings, and / or other warnings. The method may also invoke or control setpoint changes or actuate physical actuators in the industrial process to correct erroneous conditions determined based on the hybrid sensor values.

[0090] In an embodiment, the method may create a qualitative causal relationship graph or data structure, for example, by: determining a set of candidate upstream variables with a higher sampling and measurement rate relative to the target variable based on the causal relationship between the candidate upstream variable set and the target variable in the process; and determining a set of candidate downstream variables with a higher sampling and measurement rate relative to the target variable based on the causal relationship between the downstream variable sets in the process and the target variable.

[0091] In embodiments, a combined estimate of the three individual estimates can provide a more frequent estimate of the target variable. The combined estimate includes bidirectional estimation. For example, the method uses forward estimation to predict the target variable using a candidate upstream set of the variable within a time window dedicated to that estimation. The method also uses backward estimation to perform backward time inference of the target variable using a candidate downstream set of metrics within a time window dedicated to that estimation. The method further uses historical values ​​of the target variable's actual measurements. The method can create a hybrid sensor by aggregating the above three estimates using aggregated estimation, and, for example, the target variable (hybrid sensor) value at a given time can be determined or estimated using weights and attention to each of the three estimates.

[0092] In the embodiments, forward and backward estimation can be accomplished using physical or chemical principles to develop empirical estimation models and / or using deep learning models based on data science methods.

[0093] In an embodiment, time windows can be created for forward and backward estimation by using physical dwell time and physical processing time to calculate time windows and lags for causal relationships between multiple variables, and / or by using time series correlation to analyze time windows using a data-driven approach.

[0094] Figure 12This is another flowchart illustrating the method in the embodiment. At 1202, the method includes receiving a first time series of upstream sensor data that has a forward dependency on a target variable. The first time series of upstream sensor data may be one-dimensional or multi-dimensional time series data. At 1204, the method includes receiving a second time series of downstream sensor data that has a backward dependency on the target variable. The second time series of downstream sensor data may be one-dimensional or multi-dimensional time series data. At 1206, the method includes receiving a time series of target variable data of measurements associated with the target variable. The target variable has a measurement frequency lower than the measurement frequency associated with the upstream and downstream sensor data. At 1208, the method includes determining a first time window of lag representing the forward dependency between the upstream sensor data and the target variable. At 1210, the method includes determining a second time window of lag representing the backward dependency between the downstream sensor data and the target variable. In 1212, the method includes training a machine learning model (such as a neural network) based on a training dataset that includes at least a first time series of upstream sensor data in a first time window, a second time series of downstream sensor data in a second time window, and a time series of measured target variable data, to estimate the value of the target variable at a given time. The training dataset may include multiple such time series datasets.

[0095] The method may further include creating a causal data structure that forward-links a first time series of upstream sensor data with the time series of the measured target variable data, and backward-links a second time series of downstream sensor data with the time series of the measured target variable data. The upstream and downstream sensor data can be selected based on this causal data structure.

[0096] In one embodiment, a trained machine learning model (such as a neural network) estimates the value of a target variable at a given time. In another embodiment, based on the estimated value of the target variable for a given time, the hardware processor also controls a setpoint in an industrial process. For example, the industrial process includes oil sands processing, and the target variable includes characteristics associated with tailings.

[0097] In an embodiment, the machine learning model, such as a neural network, includes an aggregation of a first neural network, a second neural network, and a third neural network. The first neural network is trained to predict the value of a target variable based on a first time series of forward dependencies of upstream sensor data, the second neural network is trained to predict the value of the target variable based on a second time series of backward dependencies of downstream sensor data, and the third neural network is trained to predict the value of the target variable based on a time series of measured target variable data.

[0098] In one embodiment, the reporting time of the measured target variable has a delay from the acquisition time of the target variable. The training dataset, which includes time-series data of the measured target variable, may include data of the measured target variable determined at the reporting time corresponding to the acquisition time of the measured target variable. In another embodiment, a machine learning model, such as a neural network, estimates the value of the target variable at a given time without delay during runtime.

[0099] Figure 13This diagram illustrates components of a system in one embodiment that can generate a hybrid sensor. One or more hardware processors 1302, such as a central processing unit (CPU), graphics processing unit (GPU) and / or field-programmable gate array (FPGA), application-specific integrated circuit (ASIC), and / or another processor, may be coupled to a memory device 1304 and generate the hybrid sensor. The memory device 1304 may include random access memory (RAM), read-only memory (ROM), or another memory device and may store data and / or processor instructions for implementing various functions associated with the methods and / or systems described herein. One or more processors 1302 may execute computer instructions stored in the memory 1304 or received from another computer device or medium. The memory device 1304 may, for example, store instructions and / or data for the functions of one or more hardware processors 1302 and may include an operating system and other programs containing instructions and / or data. One or more hardware processors 1302 can receive inputs, such as a first time series of upstream sensor data with a forward dependency on a target variable, a second time series of downstream sensor data with a backward dependency on the target variable, and a time series of target variable data of measurements associated with the target variable. The first time series of upstream sensor data can be one-dimensional or multi-dimensional time series data. The second time series of downstream sensor data can also be one-dimensional or multi-dimensional time series data. In one aspect, one or more hardware processors 1302 can also determine a first time window representing the forward dependency and a second time window representing the backward dependency between the downstream sensor data and the target variable. One or more hardware processors 1302 can train a machine learning model, such as a neural network, based on the training dataset described above. The hardware processors 1302 can run the trained machine learning model (such as a neural network) to estimate the value of the target variable at a given time. Based on the estimated value of the target variable for a given time, the hardware processors 1302 can further control the setpoint in the industrial process. In this embodiment, training data may be stored in storage device 1306 or received from a remote device via network interface 1308, and may be temporarily loaded into memory device 1304 for training machine learning models, such as neural networks. One or more hardware processors 1302 may be coupled to interface devices, such as network interface 1308 for communicating with remote systems, for example, via a network, and input / output interface 1310 for communicating with input and / or output devices such as a keyboard, mouse, display, and / or others.

[0100] Figure 14A schematic diagram of an example computer or processing system that can implement a hybrid sensor generation system in one embodiment is shown. The computer system is merely one example of a suitable processing system and is not intended to impose any limitation on the scope of use or functionality of the embodiments of the methods described herein. The processing system shown can operate with many other general-purpose or special-purpose computing system environments or configurations. Suitable for use with Figure 14 Examples of well-known computing systems, environments, and / or configurations used in conjunction with the processing systems shown may include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments that include any of the aforementioned systems or devices.

[0101] A computer system can be described within the general context of computer system executable instructions (such as program modules) executed by the computer system. Generally, program modules can include routines, programs, objects, components, logic, data structures, etc., that perform specific tasks or implement specific abstract data types. Computer systems can be implemented in distributed cloud computing environments, where tasks are performed by remote processing devices linked via communication networks. In distributed cloud computing environments, program modules can reside in local and remote computer system storage media, including memory storage devices.

[0102] The components of the computer system may include, but are not limited to, one or more processors or processing units 12, system memory 16, and a bus 14 that couples the various system components, including system memory 16, to processor 12. Processor 12 may include modules 30 that perform the methods described herein. Modules 30 may be programmed into an integrated circuit of processor 12, or loaded from memory 16, storage device 18, or network 24, or a combination thereof.

[0103] Bus 14 can represent one or more of several types of bus architectures, including memory buses or memory controllers, peripheral buses, accelerated graphics ports, and processor or local buses using any of the various bus architectures. By way of example and not limitation, such architectures include Industry Standard Architecture (ISA) buses, Micro Channel Architecture (MCA) buses, Enhanced ISA (EISA) buses, Video Electronics Standards Association (VESA) local buses, and Peripheral Component Interconnect (PCI) buses.

[0104] Computer systems can include a variety of computer-readable media. Such media can be any available media that can be accessed by a computer system, and can include volatile and non-volatile media, removable and non-removable media.

[0105] System memory 16 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory or others. The computer system may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 18 may be provided for reading from and writing to non-removable, non-volatile magnetic media (e.g., a "hard disk drive"). Although not shown, disk drives for reading from or writing to removable non-volatile disks (e.g., "floppy disks") and optical disk drives for reading from or writing to removable non-volatile optical disks (such as CD-ROMs, DVD-ROMs, or other optical media) may be provided. In such cases, each may be connected to bus 14 via one or more data media interfaces.

[0106] The computer system may also communicate with one or more external devices 26 (e.g., keyboard, pointing device, display 28, etc.); and / or any device that enables the computer system to communicate with one or more other computing devices (e.g., network card, modem, etc.). Such communication may occur via input / output (I / O) interface 20.

[0107] Furthermore, the computer system can communicate with one or more networks 24 (such as a local area network (LAN), a general wide area network (WAN), and / or a public network (e.g., the Internet)) via network adapter 22. As shown, network adapter 22 communicates with other components of the computer system via bus 14. It should be understood that, although not shown, other hardware and / or software components may be used in conjunction with the computer system. Examples include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archiving storage systems.

[0108] This invention can be a system, method, and / or computer program product with any possible level of technical detail integration. The computer program product may include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to execute aspects of the invention.

[0109] Computer-readable storage media can be tangible means for retaining and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of computer-readable storage media includes: portable computer disks, 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 universal disk (DVD), memory sticks, floppy disks, mechanical encoding devices such as punch cards or protrusions in slots having instructions recorded thereon, and any suitable combination of the foregoing. As used herein, computer-readable storage media should not be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses passing through fiber optic cables), or electrical signals transmitted through wires.

[0110] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to a suitable computing / processing device via a network (e.g., the Internet, a local area network, a wide area network, and / or a wireless network), or to an external computer or external storage device. The network may include copper cables, optical fibers, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to a computer-readable storage medium within the suitable computing / processing device.

[0111] Computer-readable program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​(such as Smalltalk, C++, etc.) and procedural programming languages ​​(such as the "C" programming language or similar programming languages). The computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as a standalone software package, partially on a user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, 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 may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs) may execute computer-readable program instructions by utilizing state information from the computer-readable program instructions to personalize the electronic circuitry in order to perform aspects of this invention.

[0112] The present invention will now be described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should 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.

[0113] These computer-readable program instructions may be provided to a processor of a computer or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / actions specified in one or more blocks of a flowchart and / or block diagram. These computer-readable program instructions may also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner, such that the computer-readable storage medium storing the instructions includes an article of manufacture containing instructions that implement aspects of the functions / actions specified in one or more blocks of a flowchart and / or block diagram.

[0114] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus, or other device to produce computer-implemented processing, such that the instructions executed on the computer, other programmable apparatus, or other device perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0115] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, segment, or portion of instructions, including one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the figures. For example, two blocks shown consecutively may actually be completed as a single step, executed simultaneously, substantially simultaneously, or with partial or complete temporal overlap, or the blocks may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions.

[0116] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, unless the context explicitly indicates otherwise, the singular forms “a,” “an,” and “the” are intended to also include the plural forms. As used herein, the term “or” is an inclusive operator and may mean “and / or” unless the context explicitly or explicitly indicates otherwise. It should also be understood that, when used herein, the terms “comprise,” “comprises,” “comprising,” “includes,” “including,” and / or “having” may specify the presence of said features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or combinations thereof. As used herein, the phrase “in an embodiment” does not necessarily refer to the same embodiment, although it may refer to the same embodiment. As used herein, the phrase “in one embodiment” does not necessarily refer to the same embodiment, although it may refer to the same embodiment. As used herein, the phrase “in another embodiment” does not necessarily refer to different embodiments, although it may refer to different embodiments. Furthermore, the embodiments and / or the components of the embodiments can be freely combined with each other, unless they are mutually exclusive.

[0117] All the means or steps plus functional elements (if any) in the following claims are intended to include any structure, material, action, and equivalent for performing the function in combination with other claimed elements as specifically claimed. The description of the invention has been presented for purposes of illustration and description, but is not intended to be exhaustive or limited to the forms disclosed. Many modifications and variations will be apparent to those skilled in the art without departing from the scope of the invention. Embodiments were chosen and described in order to best explain the principles and practical application of the invention, and to enable others skilled in the art to understand various embodiments of the invention with various modifications suitable for the intended particular use.

Claims

1. A system for generating hybrid sensors in an industrial process, comprising: Hardware processor; as well as A storage device coupled to the hardware processor; The hardware processor is configured to at least: Receive the first time series of upstream sensor data that has a forward dependency on the target variable; Receive a second time series of downstream sensor data that has a backward dependency on the target variable; Receive time series of target variable data of measurements associated with the target variable, the target variable having a lower measurement frequency than the measurement frequency associated with the upstream sensor data and the downstream sensor data; Determine a first time window representing the forward dependency between the upstream sensor data and the target variable; Determine a second time window that represents the backward dependency between the downstream sensor data and the target variable; as well as A machine learning model is trained based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, in order to estimate the value of the target variable at a given time.

2. The system of claim 1, wherein the hardware processor is further configured to create a causal data structure that forward links the first time series of the upstream sensor data with the time series of the measured target variable data, and backward links the second time series of the downstream sensor data with the time series of the measured target variable data.

3. The system of claim 2, wherein the upstream sensor data and the downstream sensor data are selected based on the causal relationship data structure.

4. The system of claim 1, wherein the trained machine learning model estimates the value of the target variable at the given time, wherein the hardware processor further controls a setpoint in the industrial process based on the value of the target variable estimated for the given time.

5. The system of claim 4, wherein the industrial process includes oil sands treatment, and the target variable includes characteristics associated with tailings.

6. The system of claim 1, wherein the machine learning model comprises an aggregation of the following: a first neural network trained to predict the value of the target variable based on the forward dependency of the first time series of the upstream sensor data; A second neural network is trained to predict the value of the target variable based on the backward dependency of the second time series of the downstream sensor data; And a third neural network, which is trained to predict the value of the target variable based on the time series of measured target variable data.

7. The system of claim 6, wherein the reporting time of the measured target variable has a delay from the acquisition time of the target variable, and the training dataset comprising the time series of the measured target variable data includes the data of the measured target variable whose reporting time is determined to be the corresponding acquisition time of the measured target variable, wherein the machine learning model estimates the value of the target variable at the given time without delay.

8. The system of claim 1, wherein the first time series of the upstream sensor data includes multidimensional time series data, and the second time series of the downstream sensor data includes multidimensional time series data.

9. The system according to claim 1, wherein the machine learning model includes a neural network model.

10. A computer program product comprising program instructions, the program instructions being executable by a device to cause the device to: Receive the first time series of upstream sensor data that has a forward dependency on the target variable; Receive a second time series of downstream sensor data that has a backward dependency on the target variable; Receive time series of target variable data of measurements associated with the target variable, the target variable having a lower measurement frequency than the measurement frequency associated with the upstream sensor data and the downstream sensor data; Determine a first time window representing the forward dependency between the upstream sensor data and the target variable; Determine a second time window that represents the backward dependency between the downstream sensor data and the target variable; as well as A machine learning model is trained based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, in order to estimate the value of the target variable at a given time.

11. The computer program product of claim 10, further comprising causing the device to create a causal data structure that forward links the first time series of the upstream sensor data to the time series of the measured target variable data in time, and backward links the second time series of the downstream sensor data to the time series of the measured target variable data in time.

12. The computer program product of claim 11, wherein the upstream sensor data and the downstream sensor data are selected based on the causal relationship data structure.

13. The computer program product of claim 10, wherein the trained machine learning model estimates the value of the target variable at the given time, wherein the device further controls a setpoint in an industrial process based on the value of the target variable estimated for the given time.

14. The computer program product of claim 13, wherein the industrial process includes oil sands treatment, and the target variable includes characteristics associated with tailings.

15. The computer program product of claim 10, wherein the machine learning model comprises an aggregation of the following: a first neural network trained to predict the value of the target variable based on the forward dependency of the first time series of the upstream sensor data; A second neural network is trained to predict the value of the target variable based on the backward dependency of the second time series of the downstream sensor data; And a third neural network, which is trained to predict the value of the target variable based on the time series of measured target variable data.

16. The computer program product of claim 15, wherein the reporting time of the measured target variable has a delay from the acquisition time of the target variable, and the training dataset comprising the time series of the measured target variable data includes data of the measured target variable whose reporting time is determined to be the corresponding acquisition time of the measured target variable, wherein the machine learning model estimates the value of the target variable at the given time without delay.

17. The computer program product of claim 10, wherein the first time series of the upstream sensor data includes multidimensional time series data, and the second time series of the downstream sensor data includes multidimensional time series data.

18. The computer program product of claim 10, wherein the machine learning model comprises a neural network model.

19. A method for generating a hybrid sensor in an industrial process, comprising: Receive the first time series of upstream sensor data that has a forward dependency on the target variable; Receive a second time series of downstream sensor data that has a backward dependency on the target variable; Receive time series of target variable data of measurements associated with the target variable, the target variable having a lower measurement frequency than the measurement frequency associated with the upstream sensor data and the downstream sensor data; Determine a first time window representing the forward dependency between the upstream sensor data and the target variable; Determine a second time window that represents the backward dependency between the downstream sensor data and the target variable; as well as A machine learning model is trained based on a training dataset, the training dataset including at least the first time series of upstream sensor data in the first time window, the second time series of downstream sensor data in the second time window, and the time series of measured target variable data, in order to estimate the value of the target variable at a given time.

20. The method of claim 19, wherein the first time series of the upstream sensor data comprises multidimensional time series data, and the second time series of the downstream sensor data comprises multidimensional time series data, and wherein the machine learning model comprises a neural network model.

Images