Method and system for optimizing flocculant dosage in sewage treatment

Abstract

The invention provides a method for optimizing flocculant dosage in sewage treatment, executed by one or more processors executing a computer program, for determining a most relevant set of feature objects as first input data for a correlation analysis of each of a plurality of feature objects with the object. And collecting the group of first input data as a complete sample set, and decomposing the sample set data into feature data and label data. Sliding window-based time segmentation of data is performed for a plurality of cycles of duplicated tasks, the plurality of cycles including a training cycle and a test cycle. And for the input feature data and label data, determining object class prediction in the input data by using a neural network, and outputting second output data along with prediction of one or more attributes. Because the Pearson correlation coefficient analysis is carried out on the feature object and the denoising processing is carried out on the historical data, the prediction accuracy of the method is improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates generally to predicting a dosage of a flocculant for wastewater treatment, and more particularly to employing one or more neural networks to generate a neural network model for optimizing flocculant dosage in wastewater treatment. BACKGROUND

[0002] The operation process of a mining enterprise producing sand and gravel generally includes raw material mining, multi-stage crushing, screening, sand washing, and finished product processing. In order to ensure the cleanliness of the finished sand and gravel, the crushed sand and gravel needs to be washed, and the wastewater after washing will flow into the wastewater treatment tank through the pipeline. When treating wastewater, the sludge in the wastewater needs to be precipitated to form a state of sludge in the lower layer and clean water in the upper layer. The clean water is used for recycling, and the sludge is subjected to pressure filtration treatment to finally form a mud cake.

[0003] Therefore, wastewater treatment is very critical, which can realize purification and regeneration of wastewater. At present, in order to make the sludge in the wastewater settle quickly and obtain clean water with low turbidity, a flocculant is often added to the wastewater. The available flocculants include organic flocculants (PAM) and inorganic flocculants (PAC), and the appropriate flocculant needs to be reasonably selected according to the water quality of the wastewater.

[0004] In addition, the dosage of the flocculant is a key factor affecting wastewater treatment. Too little flocculant dosage affects the efficiency and quality of wastewater treatment, and too much flocculant causes waste of cost. At present, the common method is to obtain the best flocculant type and flocculant dosage through sampling and experimental methods in the laboratory. However, in fact, because the types of pollutants in wastewater are various, and the wastewater treatment process is also variable and complex, the flocculant dosage determined in the laboratory is not ideal in actual treatment.

[0005] Therefore, considering the key factors that may affect the flocculant dosage, the demand for the effect of obtaining the optimal flocculant dosage on wastewater treatment is increasing. At present, some neural network models are used to predict the dosage of the flocculant on the market, but there are still the following problems: (1) The historical data of the flocculant used in training the neural network model is determined by artificial experience or measured in a single environment in the laboratory, which leads to a large error in the predicted results.

[0006] (2) The factors affecting the dosage of the flocculant are considered too widely or simply, for example, only the turbidity and PH value of the wastewater are considered to predict the dosage of the flocculant, and other key factors are ignored, which leads to inaccurate prediction results; for example, too many factors are considered, too much data is input into the neural network model, data noise is formed, and the training results are inaccurate. SUMMARY

[0007] The present application aims to provide an accurate neural network model for optimizing the dosage of flocculants in sewage treatment, solving the problem of inaccurate prediction of the current neural network model.

[0008] In one or more illustrative examples, a method for optimizing the dosage of flocculants in sewage treatment is performed by one or more processors executing a computer program.

[0009] For each feature object in the plurality of feature objects and the relevance analysis of the object, a most relevant set of feature objects is determined as first input data.

[0010] The above-mentioned set of first input data is collected as a complete sample set, and the sample set data is decomposed into feature data and label data.

[0011] For a plurality of cycles of repeated tasks, perform a sliding window-based time segmentation of data, the plurality of cycles including a training cycle and a test cycle.

[0012] For the input feature data and label data, using a neural network to determine object class prediction in the input data, together with one or more attribute predictions, output second output data.

[0013] In one or more illustrative examples, a system for optimizing the dosage of flocculants in sewage treatment is provided, the system comprising: a memory configured to store first input data including a plurality of feature objects for optimizing sewage treatment and a most relevant set of feature objects determined finally collected; a processor operatively connected to the memory and configured to execute an optimization of flocculant dosage analysis application in sewage treatment, For each feature object in the plurality of feature objects and the relevance analysis of the object, a most relevant set of feature objects is determined as first input data.

[0014] For a plurality of cycles of repeated tasks, perform a sliding window-based time segmentation of data, the plurality of cycles including a training cycle and a test cycle.

[0015] For the input feature data and label data, using a neural network to determine object class prediction in the input data, together with one or more attribute predictions, output second output data.

[0016] In one or more illustrative examples, a non-transitory computer readable medium includes an analysis application instruction for optimizing the dosage of flocculants in sewage treatment, the instruction when executed by one or more processors causes the one or more processors to: For each feature object in the plurality of feature objects, determine a most relevant set of feature objects as a first input data based on relevance analysis of the object.

[0017] Perform a sliding window based time segmentation of the data for a plurality of cycles of a repetitive task, the plurality of cycles including a training cycle and a testing cycle.

[0018] For the input feature data and label data, determine object class predictions in the input data along with one or more attribute predictions using a neural network, outputting a second output data Compared with the prior art, the present application has the following technical effects: 1. Among all possible factors affecting the dosage of flocculant, the Pearson correlation coefficient is applied for screening to obtain input feature objects for a neural network model, including suspended matter concentration, suspended matter particle size, suspended matter surface charge, dosage of flocculant, water temperature and PH value of the wastewater to be treated. Therefore, the dimension of the neural network training data is reduced, and the training efficiency and accuracy are improved.

[0019] 2. The historical data is trained by a first neural network to eliminate outliers in the historical data, update the output of the effective historical data, and then use the data output by the first neural network model as the input of a second neural network model to finally predict the dosage of the flocculant, thereby improving the accuracy of the model prediction. BRIEF DESCRIPTION OF DRAWINGS

[0020] The disclosure of the present application will become more readily apparent from the following description of the drawings. As will be readily appreciated by one skilled in the art, these drawings are merely to illustrate the present application and are not intended to limit the scope of the protection afforded the present application. In which: Figure 1 shows a feature data and label data structure according to an embodiment of the present application; Figure 2 shows a long short-term memory (LSTM) RNN neural network model according to an embodiment of the present application; Figure 3 shows an example process of denoising the first input data according to an embodiment of the present application; Figure 4 shows an example process of predicting the first input data according to an embodiment of the present application; Figure 5 shows a method for optimizing the dosage of flocculant in wastewater treatment according to an embodiment of the present application; Figure 6 shows a system for optimizing the dosage of flocculant in wastewater treatment according to an embodiment of the present application; DETAILED DESCRIPTION

[0021] Various embodiments will be described more fully below with reference to the accompanying drawings, which form a part of the specification and illustrate specific examples of practical embodiments by way of illustration. However, this specification may be embodied in many different forms and should not be construed as limiting itself to the embodiments set forth herein; rather, these embodiments are provided so that this specification will be exhaustive and complete, and will fully convey the scope of the invention to those skilled in the art. Among other things, this specification may be embodied as a method or apparatus. Therefore, any of the various embodiments herein may take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Therefore, the following description should not be considered limiting.

[0022] Throughout the specification and claims, unless the context clearly indicates otherwise, the following terms shall have the meaning explicitly associated herein: The phrase "in one embodiment" as used herein does not necessarily refer to the same embodiment, although it may. Therefore, as described below, various embodiments of the invention can be readily combined without departing from the scope or spirit of the invention.

[0023] As used herein, unless the context clearly indicates otherwise, the term “or” is an inclusive “or” operator and is equivalent to the term “and / or”.

[0024] Unless the context clearly indicates otherwise, the term "based on" is not exclusive and allows for the use of additional factors not described.

[0025] Throughout the following discussion, extensive reference will be made to servers, service-side components, interfaces, engines, modules, clients, peers, ports, platforms, or other systems formed by computing devices. It should be understood that the use of such terms should be considered as referring to one or more computing devices having at least one processor (e.g., ASIC, FPGA, DSP, x86, ARM, ColdFire, GPU, multi-core processor, etc.) configured to execute software instructions stored on a computer-readable tangible, non-transitory medium (e.g., hard disk drive, solid-state drive, RAM, flash memory, ROM, etc.). For example, a server may include one or more computers operating as a web server, database server, or other type of computer server in a manner that fulfills the described roles, responsibilities, or functions. It should be further understood that the disclosed computer-based algorithms, processes, methods, or other types of instruction sets may be embodied as a computer program product comprising a non-transitory, tangible, computer-readable medium storing instructions that cause a processor to perform the disclosed steps. Various servers, systems, databases, or interfaces may exchange data using standardized protocols or algorithms based on HTTP, HTTPS, AES, public-key-private-key exchange, web service APIs, known financial transaction protocols, or other electronic information exchange methods. Data exchange can be conducted through packet-switched networks, circuit-switched networks, the Internet, LANs, WANs, VPNs, or other types of networks.

[0026] As used in the description herein and throughout the following claims, when a system, engine, server, device, module or other computing element is described as being configured to perform or implement functions on data in memory, the meaning of "configured to" or "programmed to" is defined as one or more processors or cores of the computing element being programmed by a set of software instructions stored in the memory of the computing element to perform that set of functions on target data or data objects stored in memory.

[0027] It should also be understood that the following instructions are not intended as a broad overview, and therefore, concepts may be simplified for clarity and brevity.

[0028] Factors affecting flocculation effectiveness mainly include water quality characteristics, type and dosage of flocculant, stirring conditions and flocculation time, temperature and pH of the wastewater to be treated, and environmental factors. Water quality characteristics mainly include the concentration and particle size of suspended solids in the wastewater, as well as the surface charge of the suspended solids. The types of flocculants that can be selected are mainly organic flocculants (PAM) and inorganic flocculants (PAC). The dosage of the flocculant is the label data that needs to be predicted in this application, such as... Figure 1As shown. Proper stirring promotes sufficient contact and mixing between the flocculant and suspended solids, which is beneficial for forming large and dense flocs. Stirring too fast or too slow is detrimental to the flocculation effect. The flocculation time refers to the time required for wastewater to reach the turbidity standard after the addition of flocculant. It should be noted that flocculation time is an indicator of flocculant dosage; a suitable flocculant dosage requires less flocculation time, while an unreasonable dosage requires more. The environmental factors mentioned include temperature, humidity, and light.

[0029] All these factors affecting flocculant dosage are the feature objects of interest in this application. Before training the neural network model, it is necessary to perform correlation analysis between each feature object and the object to eliminate influencing factors with low correlation.

[0030] The Pearson correlation coefficient is a statistic used to measure the strength and direction of the linear relationship between two variables. Its value range is between -1 and 1, where: (1) when the Pearson correlation coefficient is 1, it means that the two variables are perfectly positively correlated; (2) when the Pearson correlation coefficient is -1, it means that the two variables are perfectly negatively correlated; (3) when the Pearson correlation coefficient is 0, it means that the two variables have no linear relationship.

[0031] The formula for calculating the Pearson correlation coefficient is as follows:

[0032] in, and These are the i-th observations of the two variables; and is the sample mean of the two variables; n is the number of observations.

[0033] Based on the Pearson correlation coefficient, a set of characteristic parameters related to flocculation effectiveness was calculated, including the suspended solids concentration, suspended solids particle size, suspended solids surface charge, type of flocculant, flocculant dosage, wastewater temperature and pH, stirring speed, and reaction time. Among these characteristic parameters, the suspended solids concentration, suspended solids particle size, suspended solids surface charge, flocculant dosage, wastewater temperature, and pH are variables that vary with each wastewater treatment. However, the type of flocculant, stirring speed, and reaction time were experimentally determined before wastewater treatment.

[0034] Table 1

[0035] Table 1 compares three different brands of polyacrylamide (PAM) flocculants with three other different brands of polyaluminum chloride (PAC) flocculants, thus eliminating the potential influence of brand. The flocculation effect shows that polyacrylamide (PAM) is superior to polyaluminum chloride (PAC) for wastewater treatment in mining enterprises; therefore, polyacrylamide (PAM) is the final flocculant chosen.

[0036] Table 2

[0037] Table 2 shows the flocculation effect of polyacrylamide (PAM) flocculants at different stirring speeds. The stirring speed is divided into 5 equal parts from 0-25 r / min. It can be seen from the flocculation effect that the stirring speed of 10-15 r / min is the best for wastewater treatment in mining enterprises. Therefore, 10-15 r / min was finally determined as the optimal stirring speed.

[0038] Table 3

[0039] Table 3 shows the flocculation effect of polyacrylamide (PAM) flocculants at different reaction times, namely 10 min, 30 min, 60 min, 90 min, and 120 min. It can be seen from the flocculation effect that the flocculation effect is best when the reaction time is 30-60 min for wastewater treatment in mining enterprises. Therefore, 30-60 min was finally determined as the optimal reaction time.

[0040] Based on the above analysis, the characteristic objects for neural network training are determined to include the suspended solids concentration, suspended solids particle size, suspended solids surface charge, flocculant dosage, and water temperature and pH value of the wastewater to be treated. The suspended solids concentration and particle size of the wastewater to be treated are determined by the membrane filtration method. Specifically, a certain number of water samples are filtered through a membrane, and then the membrane is observed under a microscope. If there are suspended solids on the membrane, the size and shape of the suspended solids can be observed under a microscope, thereby calculating the concentration and particle size of the suspended solids. The suspended solids surface charge is measured by the infrared induction test method. The specific test method is as follows: (1) First, align the instrument with the object to be tested. The infrared induction test head is about 25 mm away from the object to be tested. When the three points converge into a circle within the aperture, the reading can be accurately taken; (2) The instrument can display the current static charge value in KV, which can be converted to V, and can also show whether it is a positive charge or a negative charge; (3) If the object to be tested is relatively large, multiple measurements can be taken and the average value can be taken.

[0041] Historical data on the concentration of suspended solids, particle size, surface charge of suspended solids, flocculant dosage, water temperature, and pH of the wastewater to be treated were collected as the first input data 301, and also served as training and testing data for the neural network. In reality, much of the historical data consists of flocculant historical data measured under single laboratory conditions based on human experience. Some of this data constitutes outliers mixed into the training set, and the model may learn these abnormal patterns, leading to larger errors in the prediction results.

[0042] Therefore, removing data noise before inputting the first input data 301 into the first neural network model 302 for prediction becomes particularly important. To this end, this application introduces a Long Short-Term Memory (LSTM) RNN 200 to perform noise preprocessing on the first input data 301, outputting the first output data 303. The structure of the LSTM is as follows: Figure 2 As shown. It should be noted that the Long Short-Term Memory (LSTM) RNN 200 model used in this application is only one example, not a limitation. Other training models can also be used, such as random forests, which reduce the impact of noise through voting by multiple decision trees; gradient boosting trees, which automatically handle noise and support missing values ​​and distributed computation; adaptive weighting mechanisms, which dynamically reduce the weight of noisy samples; and Transformer architecture, which focuses on key information through self-attention mechanisms.

[0043] Long Short-Term Memory (LSTM) RNNs, compared to ordinary recurrent neural network (RNN) models, introduce memory units to store important information, and a series of gate control mechanisms (input gate, forget gate, output gate) to control the storage, forgetting, and output of information. This design allows LSTMs to remember key information in a data sequence while ignoring invalid data, resulting in relatively accurate predictions. The forget gate receives the current input and the hidden state from the previous time step, and generates a vector between 0 and 1 using the sigmoid function, representing the proportion of forgetting. This result is multiplied by the state of the memory unit, thus achieving selective forgetting. The calculation formula is as follows:

[0044] in, It is the output of the forget gate. It is the weight matrix of the forget gate. It is the hidden state of the previous time step. This is the input for the current time step. It is the bias vector of the forget gate. It is the sigmoid activation function, which restricts the value to between 0 and 1.

[0045] To improve model training performance, the first input data (301) of the feature objects typically needs to be normalized or standardized. This eliminates the influence of differences in units and scales between features, treating each feature dimension equally. Next, a sliding window of samples needs to be constructed, including feature data and label data. The feature data is used to calculate the label data, which represents the prediction target. After completing the above data processing, the LSTM model is built and trained. The input is the feature data from the past N time steps, and the output is the predicted value for the current time step. After training, the model can be used to predict the value at each time step, and then the residual between the predicted and actual values, i.e., the error, is calculated. For normal data, the error should be small, while the error at outliers will be relatively large.

[0046] Therefore, in order to remove outliers from the first input data 301, a loss function 303 needs to be introduced as an evaluation metric for the first neural network model 302 to measure the difference or error between the output of the first neural network model 302 and the true label. The loss function 303 is typically a non-negative real number function, expressed as... Where Y is the actual value, f(X) is the model's predicted value, and X is the input data. The smaller the value of the loss function 303, the closer the model's prediction is to the actual value, and the better the model's performance. For regression tasks, the mean squared error (MSE) is often used as the loss function 303, calculated as follows:

[0047] Where n is the number of samples, It is the actual value of the i-th sample. It is the predicted value of the i-th sample.

[0048] To filter MSE (Mean Error Estimate), a threshold of 304 needs to be set. Points exceeding this threshold are considered outliers and deleted, while points within the threshold are considered valid data and retained. This can be achieved by using statistical methods to calculate the mean error (MME). ) and standard deviation ( ), set as mean plus or minus three standard deviations ( ).

[0049] To more accurately filter the first input data 301, a dynamic threshold algorithm 304 is introduced, which dynamically adjusts the threshold according to the historical residual distribution to adapt to data changes. Specifically, (1) error calculation: recording the residual (error) between the predicted value and the true value. (2) Dynamic threshold 304, the threshold is dynamically adjusted according to the statistical characteristics of historical errors (such as mean, standard deviation). value ,in and The mean and standard deviation of the error are given, and k is the adjustment coefficient; (3) Anomaly detection: If the current prediction error is... If the value exceeds the threshold range, it is considered abnormal.

[0050] If necessary, the first output data 305 can be used as the output of the first neural network 303 to output new first output data 305 until the first output data 305 meets the requirements. Figure 3 As shown.

[0051] After removing outlier data, the first output data 305 is output. This first output data 305 is then used as the input data for the second neural network model 401, with training and testing data divided in a 7:3 ratio. First, it passes through an LSTM layer. At this point, the first output data 305 needs to be transformed into a three-dimensional tensor, in the form of [batch_size, seq_len, input_size], where batch_size represents the number of samples in each batch; seq_len represents the sequence length of the samples, corresponding to time_step; and input_size represents the number of features in each sample. In this application, batch_size is the number of samples in the first output data 301, i.e., the number of data such as suspended solids concentration, suspended solids particle size, suspended solids surface charge, flocculant dosage, wastewater temperature and pH value. seq_len represents the suspended solids concentration, suspended solids particle size, suspended solids surface charge, flocculant dosage, wastewater temperature and pH value, where the flocculant dosage is label data and the others are feature data.

[0052] The LSTM layer output is out = [batch_size, seq_len, hidden_size], where hidden_size is the output data of the intermediate hidden layers, and out stores the predicted values ​​of the model at each time step along the seq_len dimension. Only the last time step contains the final predicted value we need, i.e., the flocculant dosage. The output from the previous step is fed into the Dense fully connected layer to obtain [batch_size, output_size]. Here, output_size is defined according to the task requirements. For example, if we need to predict the value of a single element at one future time step, then output_size = 1; if we need to predict the value of a single element at ten future time steps, then output_size = 10.

[0053] This application also provides a system 600 for optimizing flocculant dosage in wastewater treatment, including a memory configured to store data collected for optimizing multiple feature objects in wastewater treatment and a final determined most relevant set of feature objects; and a processor operatively connected to the memory and configured to perform an application for optimizing flocculant dosage in wastewater treatment, to perform a correlation analysis between each of the multiple feature objects and the object, determining the most relevant set of feature objects as first input data. The first set of input data is collected as a complete sample set, and the sample set data is decomposed into feature data and label data. A sliding window-based time segmentation of the data is performed for multiple cycles of a repetitive task, the multiple cycles including training cycles and testing cycles. For the input feature data and label data, a neural network is used to determine object class predictions in the input data, along with one or more attribute predictions, and output second output data.

[0054] A system 600 for optimizing flocculant dosage in wastewater treatment specifically includes a user module 601, which enables user login and registration, and administrator authorization. The table fields include user ID, username, login password, and permission level. A training module 602 trains a Long Short-Term Memory (LSTM) RNN 200 based on the first input data 301 to obtain the optimal LSTM RNN 200 model. The table fields include suspended solids concentration, suspended solids particle size, suspended solids surface charge, flocculant dosage, wastewater temperature, and pH value. A prediction module 603 predicts the second output data, i.e., the flocculant dosage for the wastewater to be treated, based on the optimal LSTM RNN 200 model.

[0055] Database connection: In the backend server code, Python's MySQL library PyMySQL is used to connect to the MySQL database. Configuration information for the database connection is provided, such as hostname, port name, database name, username, and password.

[0056] The foregoing description should be understood as illustrative and exemplary in every respect, not restrictive, and the scope of the invention disclosed herein should not be determined based on the description, but rather on the claims as interpreted in their full breadth permitted by patent law. It should be understood that the embodiments shown and described herein are merely illustrative of the principles of the invention, and various modifications can be made by those skilled in the art without departing from the scope and spirit of the invention. Various other combinations of features can be implemented by those skilled in the art without departing from the scope and spirit of the invention.

Claims

1. A method for optimizing flocculant dosage in wastewater treatment, executed by one or more processors executing a computer program, the method comprising: For each feature object among multiple feature objects, an association analysis is performed between the feature object and the object, and the most relevant set of feature objects is determined as the first input data; Collect the above set of first input data as a complete sample set, and decompose the sample set data into feature data and label data; A sliding window-based time segmentation for multiple epochs of execution data for repetitive tasks, wherein the multiple epochs include training epochs and testing epochs; Given the input feature data and label data, a neural network is used to determine the object class prediction in the input data, along with one or more attribute predictions, and outputs the second output data.

2. The method according to claim 1, characterized in that, The Pearson correlation coefficient is used to analyze the association between each feature object and the object among multiple feature objects, and the most relevant set of feature objects is determined as the first input data.

3. The method according to claim 2, characterized in that, The formula for calculating the Pearson correlation coefficient is as follows: ； in, and These are the i-th observations of the two variables; and is the sample mean of the two variables; n is the number of observations.

4. The method according to any one of claims 1 or 2, characterized in that, The set of characteristic objects includes the suspended solids concentration, suspended solids particle size, suspended solids surface charge, type of flocculant, flocculant dosage, water temperature and pH value of the wastewater to be treated, stirring speed and flocculation time.

5. The method according to claim 4, characterized in that, The suspended solids concentration, suspended solids particle size, suspended solids surface charge, flocculant dosage, wastewater temperature, and pH value of the wastewater to be treated are variables, serving as input data for neural network prediction. The type of flocculant, The stirring speed and flocculation time are quantitative and determined experimentally before neural network prediction.

6. The method according to claim 5, characterized in that, The concentration and particle size of the suspended solids in the wastewater to be treated were determined by the membrane filtration method, and the surface charge of the suspended solids was measured by the infrared induction test method.

7. The method according to claim 1, characterized in that, The first input data is used as the input to the first neural network to output the first output data. The first output data is then used as the input to the second neural network to predict and output the second output data.

8. A system for optimizing flocculant dosage in wastewater treatment, characterized in that, The system includes: The memory is configured to store first input data, including data collected for multiple feature objects used to optimize wastewater treatment and the most relevant set of feature objects determined at the end; The processor, operatively connected to memory, is configured to perform applications that optimize flocculant dosage analysis in wastewater treatment. For each feature object among multiple feature objects, an association analysis is performed between the feature object and the object, and the most relevant set of feature objects is determined as the first input data; A sliding window-based time segmentation for multiple epochs of execution data for repetitive tasks, wherein the multiple epochs include training epochs and testing epochs; Given the input feature data and label data, a neural network is used to determine the object class prediction in the input data, along with one or more attribute predictions, and outputs the second output data.

9. The system according to claim 7, characterized in that, The system also includes: The user module enables user login, registration, and administrator authorization. The training module trains the Long Short-Term Memory (LSTM) RNN 200 based on historical data to obtain the optimal Long Short-Term Memory (LSTM) RNN 200 model. The prediction module predicts the flocculant dosage for the wastewater to be treated based on the optimal Long Short-Term Memory (LSTM) RNN 200 model.

10. A non-transitory computer-readable medium comprising analytical application instructions for optimizing flocculant dosage in wastewater treatment, the instructions, when executed by one or more processors, causing the one or more processors to: For each feature object among multiple feature objects, an association analysis is performed between the feature object and the object, and the most relevant set of feature objects is determined as the first input data; A sliding window-based time segmentation for multiple epochs of execution data for repetitive tasks, wherein the multiple epochs include training epochs and testing epochs; Given the input feature data and label data, a neural network is used to determine the object class prediction in the input data, along with one or more attribute predictions, and outputs the second output data.

Images