Production task time consumption prediction method and device, equipment and medium

Abstract

The embodiment of the invention provides a production task time consumption prediction method and device, equipment and a medium. The method comprises the steps that a production task time consumption prediction model and historical production data are acquired; the historical production data comprises task data of a plurality of production tasks executed in advance; inputting the historical production data into a production task time consumption prediction model; the production task time consumption prediction model comprises a nonlinear processor and a linear processor; generating feature data of the plurality of task data through a nonlinear processor; the feature data comprises linear feature data and nonlinear feature data; through a linear processor, according to the feature data of the multiple pieces of task data, time consumed for executing the next production task is predicted; through the collaborative architecture of processing nonlinear and linear features in stages, the calculation overhead is remarkably reduced on the premise of guaranteeing the prediction precision, and efficient and reliable technical support is provided for real-time task scheduling in an industrial scene.

Description

Technical Field

[0001] The present invention relates to the technical field of production task management and control, and in particular to a method, device, equipment and medium for predicting the time consumption of a production task. Background Art

[0002] In industrial automation production scenarios, real-time and highly accurate prediction of production task duration is a core challenge in achieving dynamic resource scheduling and production line optimization. Although existing prediction methods such as statistical models, machine learning, and physical models have their applications in different scenarios, they all suffer from a fundamental flaw: they cannot simultaneously meet the requirements of prediction accuracy and real-time performance in dynamic and complex industrial environments. Specifically,

[0003] While high-precision methods can model complex nonlinear relationships, they rely on large-scale labeled data for training and have high computational latency during inference, making them difficult to meet the real-time response requirements for sudden task insertions or sudden changes in device status. Low-latency methods, while computationally efficient, can only handle simple linear relationships and are unable to represent dynamic nonlinear effects such as device load fluctuations and multi-task collaboration conflicts, leading to significant prediction errors.

[0004] This problem directly restricts the agility and reliability of production scheduling. Therefore, there is an urgent need for a prediction method that can achieve both high accuracy and low latency in dynamic industrial environments to support real-time decision-making and promote the upgrading of intelligent manufacturing. Summary of the Invention

[0005] In view of the above problems, embodiments of the present invention are proposed to provide a method, apparatus, device and medium for predicting the time consumption of a production task that overcomes the above problems or at least partially solves the above problems.

[0006] In order to solve the above problems, an embodiment of the present invention discloses a method for predicting the time consumption of a production task, the method comprising:

[0007] Obtaining a production task time-consuming prediction model and historical production data; the historical production data includes task data of multiple production tasks previously executed;

[0008] Inputting the historical production data into the production task time consumption prediction model; the production task time consumption prediction model includes: a nonlinear processor and a linear processor;

[0009] Generate characteristic data of the plurality of task data by the nonlinear processor; the characteristic data includes linear characteristic data and nonlinear characteristic data;

[0010] The linear processor predicts the time required to execute the next production task based on the characteristic data of the plurality of task data.

[0011] Optionally, the nonlinear processor includes a plurality of nonlinear processing modules; the plurality of nonlinear processing modules correspond one-to-one to the plurality of task data respectively; and generating feature data of the plurality of task data by the nonlinear processor includes:

[0012] Determining first-order Volterra series expansion terms and second-order Volterra series expansion terms of corresponding task data through the multiple nonlinear processing modules;

[0013] Using the first-order Volterra series expansion term as the linear characteristic data;

[0014] The second-order Volterra series expansion term is used as the nonlinear characteristic data.

[0015] Optionally, the taking the first-order Volterra series expansion term as the linear feature data includes:

[0016] Obtaining first weights corresponding to the multiple nonlinear processing modules;

[0017] Taking the product of the first-order Volterra series expansion term and the corresponding first weight as the linear feature data;

[0018] The step of using the second-order Volterra series expansion term as the nonlinear characteristic data includes:

[0019] The product of the second-order Volterra series expansion term and the first weight is used as the nonlinear characteristic data.

[0020] Optionally, inputting the historical production data into the production task time-consuming prediction model includes:

[0021] Acquire multiple reference feature data generated by the multiple nonlinear processing modules during the previous prediction process based on the historical production data of the previous prediction process;

[0022] For each nonlinear processing module, the corresponding task data in the current prediction process and the reference feature data of an adjacent nonlinear processing module are used as input data.

[0023] Optionally, predicting the time required to execute the next production task based on the characteristic data of the plurality of task data includes:

[0024] Get the second weight;

[0025] The product of the characteristic data of the plurality of task data and the second weight is used as the time consumption for executing the next production task.

[0026] Optionally, the task data of the production task includes the time taken by the industrial robot to perform the production task, the average load of the industrial robot during the production task, the average temperature of the industrial robot during the production task, and the task complexity of the production task.

[0027] Optionally, the production task time consumption prediction model is trained according to the following steps:

[0028] Get training data and validation data;

[0029] Inputting the training data into the production task time-consuming prediction model;

[0030] Generate feature data of the training data by the nonlinear processor; the feature data includes linear feature data and nonlinear feature data;

[0031] Predicting the time consumption of a training production task according to the characteristic data of the plurality of training data by the linear processor;

[0032] Determining the time consumption of the prediction training production task and the error value of the verification data;

[0033] When the error value is greater than a preset threshold, the linear parameters and nonlinear parameters of the production task time consumption prediction model are updated according to the preset stochastic gradient descent algorithm, and the above training process is repeated until the preset training conditions are met.

[0034] On the other hand, an embodiment of the present invention further discloses a time-consuming prediction device for a production task, the device comprising:

[0035] A prediction preparation module is used to obtain a production task time-consuming prediction model and historical production data; the historical production data includes task data of multiple production tasks previously executed;

[0036] A data input module is used to input the historical production data into the production task time consumption prediction model; the production task time consumption prediction model includes: a nonlinear processor and a linear processor;

[0037] A feature extraction module, configured to generate feature data of the plurality of task data through the nonlinear processor; the feature data includes linear feature data and nonlinear feature data;

[0038] The time consumption prediction module is used to predict the time consumption of executing the next production task according to the characteristic data of the plurality of task data through the linear processor.

[0039] Optionally, the nonlinear processor includes a plurality of nonlinear processing modules; the plurality of nonlinear processing modules correspond one-to-one to the plurality of task data respectively; the feature extraction module includes:

[0040] A Volterra expansion submodule, configured to determine a first-order Volterra series expansion term and a second-order Volterra series expansion term of the corresponding task data through the plurality of nonlinear processing modules;

[0041] a linear feature determination submodule, configured to use the first-order Volterra series expansion term as the linear feature data;

[0042] The nonlinear feature determination submodule is used to use the second-order Volterra series expansion term as the nonlinear feature data.

[0043] Optionally, the linear feature determination submodule includes:

[0044] A first weight obtaining unit, configured to obtain first weights corresponding to the plurality of nonlinear processing modules;

[0045] a weighted linear feature determining unit, configured to use the product of the first-order Volterra series expansion term and the corresponding first weight as the linear feature data;

[0046] The nonlinear feature determination submodule includes:

[0047] The weight line nonlinear characteristic determination unit is used to use the product of the second-order Volterra series expansion term and the first weight as the nonlinear characteristic data.

[0048] Optionally, the data input module includes:

[0049] A reference feature submodule, configured to obtain a plurality of reference feature data generated by the plurality of nonlinear processing modules in a previous prediction process with respect to historical production data of the previous prediction process;

[0050] The input data determination submodule is used to use the corresponding task data in the current prediction process and the reference feature data of an adjacent nonlinear processing module as input data for each nonlinear processing module.

[0051] Optionally, the time consumption prediction module includes:

[0052] A second weight acquisition submodule, used to acquire a second weight;

[0053] The weighted time consumption prediction submodule is used to use the product of the characteristic data of the plurality of task data and the second weight as the time consumption for executing the next production task.

[0054] Optionally, the task data of the production task includes the time taken by the industrial robot to perform the production task, the average load of the industrial robot during the production task, the average temperature of the industrial robot during the production task, and the task complexity of the production task.

[0055] Optionally, the production task time consumption prediction model is trained by the following modules:

[0056] A training data acquisition module is used to obtain training data and verification data;

[0057] A training data input module, used for inputting the training data into the production task time consumption prediction model;

[0058] A training feature acquisition module, configured to generate feature data of the training data through the nonlinear processor; the feature data includes linear feature data and nonlinear feature data;

[0059] A training result acquisition module, configured to predict the time consumption of a training production task based on the feature data of the plurality of training data through the linear processor;

[0060] An error value determination module, used to determine the error value of the predicted training production task time and the verification data;

[0061] The training module is used to update the linear parameters and nonlinear parameters of the production task time consumption prediction model according to the preset stochastic gradient descent algorithm when the error value is greater than the preset threshold, and repeat the above training process until the preset training conditions are met.

[0062] Accordingly, an embodiment of the present invention discloses an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, the various steps of the embodiment of the time-consuming prediction method for the above-mentioned production task are implemented.

[0063] Accordingly, an embodiment of the present invention discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various steps of the embodiment of the time consumption prediction method for the above-mentioned production task are implemented.

[0064] The embodiments of the present invention include the following advantages: through a nonlinear processor, feature data of multiple task data is generated; the feature data includes linear feature data and nonlinear feature data, which can model implicit dynamic nonlinear relationships from complex production data, making up for the lack of accuracy of traditional linear models; through the linear processor, low-complexity linear operations are directly performed on the extracted features, avoiding the high computational delay of traditional deep learning models, meeting the real-time prediction requirements, and realizing the coordinated optimization of prediction accuracy and real-time performance in dynamic industrial scenarios. At the same time, the model is based on historical data-driven training, can adapt to changes in the production environment, and get rid of dependence on fixed physical models. Ultimately, through the collaborative architecture of processing nonlinear and linear features in stages, the computational overhead is significantly reduced while ensuring prediction accuracy, providing efficient and reliable technical support for real-time task scheduling in industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a flowchart of the steps of an embodiment of a method for predicting the time consumption of a production task according to the present invention;

[0066] Figure 2 This is a structural diagram of a production task time consumption prediction model according to an embodiment of a production task time consumption prediction method of the present invention;

[0067] Figure 3 is a structural diagram of a nonlinear processing module of a nonlinear processor in an embodiment of the present invention;

[0068] Figure 4 This is a flow chart of predicting production tasks according to an embodiment of the present invention;

[0069] Figure 5 It is a structural block diagram of an embodiment of a time-consuming prediction device for a production task of the present invention. DETAILED DESCRIPTION

[0070] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0071] One of the core concepts of the embodiment of the present invention is to use a custom nonlinear processor to process data first, extract linear and nonlinear features of the data, and then use the linear processor to further integrate the features to predict the time consumption of production tasks.

[0072] Reference Figure 1 , shows a flowchart of an embodiment of a method for predicting the time consumption of a production task according to the present invention, which may specifically include the following steps:

[0073] Step 101: Obtain a production task time consumption prediction model and historical production data; the historical production data includes task data of multiple production tasks previously executed;

[0074] In one embodiment, the task data of the production task includes the time taken by the industrial robot to perform the production task, the average load of the industrial robot during the production task, the average temperature of the industrial robot during the production task, and the task complexity of the production task.

[0075] Before predicting the time consumption of a production task, the required data must be collected in advance. The type of data to be prepared needs to be flexibly adjusted according to the different production tasks. Taking the production time consumption prediction of an industrial palletizing robot as an example, the types and methods of collecting historical production data can be as follows:

[0076] Industrial palletizing robots typically perform a specific production task on a production line, often in a cyclical manner. The data required for collection may include historical task durations and associated task information. In the context of an industrial palletizing robot's production tasks, this associated information may include the robot's average load, average temperature, and task complexity.

[0077] The time it takes for an industrial palletizing robot to perform a production task refers to the historical data used to predict the execution time of previous production tasks of the industrial palletizing robot. The execution time of each production task of the industrial palletizing robot is the period from the end of the last task to the end of this task. In order to accurately collect and measure the execution time of each production task, the voltage signal of the industrial robot's emergency stop switch can be connected to the input IO (Input / Output) of the PLC (Programmable Logic Controller). The emergency stop switch is an important component of the industrial robot safety system, and the voltage signal of the emergency stop switch can be used to determine the working status of the robot. When the emergency stop switch is triggered, the power supply of the robot will be cut off, thereby stopping all operations. At this time, the voltage signal is a low level, that is, a falling edge. The specific collection method can be: when the falling edge of the voltage is detected, record the current time t start When the second falling edge is detected, the current time t is recorded. end . Calculate the difference between the two timestamps and get the execution time of the most recent task t(0) = t end -t start, and save it. Processing of historical production task time data: record the most recent task completion time as t(0), the previous task completion time as t(1), the previous nth task completion time as t(n), the next task completion time as t(-1), and the next nth task completion time as t(-n). Each time a new task time t is collected new , will update the index of the previously collected time, that is, t new is replaced by t(0), t(0) is replaced by t(1), and so on.

[0078] The average load of an industrial palletizing robot during a production task refers to the load the robot bears during the task. The average temperature of an industrial palletizing robot during a production task refers to the robot's operating temperature during the task. The task complexity of a production task refers to its complexity and can be expressed as a value between 0 and 1, with 0 representing extremely simple and 1 representing extremely complex. The average load and average temperature can be expressed as the arithmetic mean of the load and temperature of the industrial robot during the task.

[0079] In one example, the production task time consumption prediction model may be a prediction model based on Volterra technology (PCAPM-SOV model):

[0080] Reference Figure 2 , Figure 2 A structural diagram showing a production task time consumption prediction model according to an embodiment of a production task time consumption prediction method of the present invention is shown:

[0081] The PCAPM-SOV model has a three-layer structure: the input layer, the intermediate layer, and the output layer. The input layer consists of the input queue of each module, and the intermediate layer is composed of multiple Volterra modules and linear processor modules. The intermediate layer consists of two connected parts: a nonlinear processor and a linear part. The nonlinear processor adopts a pipelined structure, containing multiple Volterra modules, and implements a nonlinear mapping from the input layer to the intermediate layer. The linear part is a traditional linear processor, which performs a linear mapping from the intermediate space to the output space. The nonlinear processor is connected to the linear part, and the input of the linear part consists of the output of the nonlinear processor, thus forming a complete network structure. Therefore, this combined structure that contains both linear and nonlinear elements can extract more relationships contained in the input queue, thereby more effectively fitting the production system and achieving more accurate predictions.

[0082] The Volterra series is a mathematical tool commonly used in the modeling and prediction of nonlinear systems. As a modeling tool for nonlinear systems, the Volterra series can effectively handle complex nonlinear relationships. By expressing a nonlinear system as a superposition of a series of convolution integrals, the Volterra series can capture the dynamic characteristics of the system and provide more accurate prediction results. In addition, the Volterra series has high flexibility and identifiability. A truncated p-order Volterra series expansion is as follows:

[0083]

[0084] Among them, x(n) and y(n) are the input and output of the system respectively, h i represents the i-th-order Volterra kernel coefficient, and N is the memory length. In practical engineering, second-order Volterra series (SOV) or third-order Volterra series (TOV) are often used to reduce the computational burden. This production task time prediction model uses the following second-order Volterra series expansion to model nonlinear systems:

[0085]

[0086] Among them, h o , h1 and h2 are constant, first-order and second-order Volterra kernels respectively, and N represents the memory length of the Volterra kernel. It can be written in matrix form as follows: y(n) = H T (n)X(n)

[0087] Among them, the first-order Volterra kernel can characterize the linear characteristics of the data, and the second-order Volterra kernel can characterize the nonlinear characteristics of the data; T refers to the transpose operation, and the Volterra kernel vector has (N+2)(N+1) / 2 kernel coefficients. The corresponding input vector also consists of the same number of input values, that is,

[0088] H(n)=[h0,h1(0),…h1(N-1),h2(0,0),…,h2(N-1,N-1)], X(n)=[1,x(n),…,x(n-N+1),x 2 (n),x(n)x(n-1),…,x 2 (n-N+1)]

[0089] The model generates an output from a set of inputs: a predicted time for the next production task. Each module in the PCAPM-SOV model receives an input queue containing the necessary information for predicting future production task times. In the nonlinear processor, each module's input queue contains p+3 elements: p historical production task time data and three production task-related information data for industrial palletizing robots.

[0090] For example, when p=1, the internal structure of the module in the first nonlinear processor is as follows: Figure 3 shown

[0091] Figure 3 A structural diagram of a nonlinear processing module of a nonlinear processor in an embodiment of the present invention is shown.

[0092] Its input consists of three parts: constant 1, input matrix T i And internal feedback r2. Among them, 1 is the constant term input fixed by the second-order Volterra series, T1 = [t(0), x1, x2, x3], because T1 = [t(0)] (when p = 1), and r2 is the output result z2 of the second module last time.

[0093] Therefore, the input X1 after the second-order Volterra expansion becomes

[0094]

[0095] The number of elements in X1 is 21. Among them, t(0), x1, x2, x3, and r2 are the first-order terms after expansion, t(0)·x1, t(0)·x2, t(0)·x3, t(0)·r2, x1·x2, x1·x3, x1·r2, x2·x3, x2·r2, x3·r2, t 2 (0), is the second-order term after expansion. Therefore, the length of the nonlinear weight vector H1 of module 1 is also 21, which is recorded as

[0096] H1=[h1,h2,…,h 21 ] Therefore, the output z1 of module 1 is:

[0097]

[0098] The complete calculation process of the PCAPM-SOV model is as follows: the input matrix is ​​first calculated by the nonlinear processor, and the output of the nonlinear processor is used as the input of the linear part and calculated by the linear processor to obtain the final output. Therefore, the specific calculation process is divided into the following two parts:

[0099] (1) Nonlinear processor

[0100] Without loss of generality, for the i-th module, its input queue T i There are p+3 elements in T, denoted as: i =[T i,a ,T i,b ]

[0101] =[t((i-1)p),t((i-1)p+1),…,t(ip-1),x1,x2,x3] Each module also has an internal feedback value. The internal feedback value of the i-th module is r i for

[0102]

[0103] The internal feedback of the mth module is the last output z of the module m For other modules, the internal feedback is the output of the next module in the previous operation. For example, the internal feedback of the first module is the output result z2 of the second module in the previous operation. i With r i Input into the module, after the second-order Volterra

[0104] After expansion, it becomes the expanded input X i ,Right now

[0105]

[0106] In a nonlinear processor, each module has its weight vector. The weight vector H of the i-th module is i Defined as: H i =[h i1 ,h i2 ,…,h iL ] Among them, L and vector X i and H i The length is the same as that of , and the length can be determined by the formula given below:

[0107]

[0108] Then, the output z of the i-th module in the PCAPM-SOV nonlinear processor is i for:

[0109]

[0110] (2) Linear part:

[0111] The linear part of PCAPM-SOV is a linear processor whose input is composed of the outputs of all modules in the nonlinear processor, namely

[0112] Z=[z1,z2,…,z m ]

[0113] The weight vector W of the linear processor is recorded as:

[0114] W=[w1,w2,…,w m ]

[0115] Therefore, the output of the linear part, that is, the output y of the model is

[0116] y=ZW T

[0117] In one embodiment, the production task time consumption prediction model can be trained by the following steps:

[0118] Get training data and validation data;

[0119] Training data refers to the sample set used for model learning, which contains historical production task time and related information of production tasks; verification data refers to an independent data set used to evaluate the generalization ability of the model.

[0120] Inputting the training data into the production task time-consuming prediction model;

[0121] Generate feature data of the training data by the nonlinear processor; the feature data includes linear feature data and nonlinear feature data;

[0122] Predicting the time consumption of a training production task according to the characteristic data of the plurality of training data by the linear processor;

[0123] Determining the time consumption of the prediction training production task and the error value of the verification data;

[0124] The error value can be determined by using methods such as mean square error (MSE) or mean absolute error (MAE).

[0125] When the error value is greater than a preset threshold, the linear parameters and nonlinear parameters of the production task time consumption prediction model are updated according to the preset stochastic gradient descent algorithm, and the above training process is repeated until the preset training conditions are met.

[0126] The preset training condition may be when the error value is less than a preset threshold or the number of training times reaches a preset threshold.

[0127] The above steps are explained in detail using the training process of the production task time prediction model of an industrial palletizing robot as an example:

[0128] The input and calculation processes of the model have been described in detail in the above introduction to the model, so we will not go into details here.

[0129] In one example, the linear parameters and nonlinear parameters of the production task time consumption prediction model may be weight vectors of a linear processor and weight vectors of a nonlinear processor. The process of updating the linear parameters and nonlinear parameters of the production task time consumption prediction model described above may be as follows:

[0130] When the next production task time t(-1) is collected, the model's prediction error e can be calculated using the next production task time. By defining a cost function, the NLMS (Normalized Least Mean Squares) adaptive algorithm is used to adjust parameters in the direction that minimizes the cost function. NLMS is an adaptive filtering algorithm used to adjust parameters in real time in dynamic systems to minimize the mean square value of the output error. Its core concept is to balance convergence speed and stability by normalizing the step size.

[0131] The parameters that need to be updated in the model are the nonlinear weights (H1, H2, ..., H m ) and linear weights (W). Since the nonlinear processor is coupled with the linear processor, the weights of the nonlinear part need to be updated first, and then the weights of the linear part are updated. The adaptive update algorithms of the two parts are as follows:

[0132] (1) Nonlinear part

[0133] Define the cost function J as:

[0134] J=e 2

[0135] Where is the difference between the next observation value t(-1) of e and the predicted value y, that is,

[0136] e=t(-1)-y

[0137] Then, according to the stochastic gradient estimation algorithm, the derivative of J with respect to the weight vector is obtained

[0138]

[0139] Then the weight vector H i The adaptive update equation is:

[0140]

[0141] Among them, η h is the learning rate of the nonlinear part. Therefore, the update formula based on the NLMS algorithm finally becomes:

[0142]

[0143] in represents the two-norm

[0144] (2) Linear part

[0145] Similarly, according to the gradient algorithm, the derivative of J with respect to the weight vector W is obtained:

[0146]

[0147] Then the adaptive update equation of the weight vector is as follows:

[0148]

[0149] where η w Is the learning rate parameter of the linear part. Therefore, the update formula based on the NLMS algorithm becomes:

[0150]

[0151] Step 102: Input the historical production data into the production task time consumption prediction model; the production task time consumption prediction model includes: a nonlinear processor and a linear processor;

[0152] In one embodiment, step 102 may include the following sub-steps:

[0153] Sub-step S11, obtaining a plurality of reference feature data generated by the plurality of nonlinear processing modules in the previous prediction process based on the historical production data of the previous prediction process;

[0154] Sub-step S12: for each nonlinear processing module, the corresponding task data in the current prediction process and the reference feature data of an adjacent nonlinear processing module are used as input data.

[0155] This part is the feedback value part in the above content. Since it has been explained in detail above, it will not be repeated here.

[0156] Step 103: generating feature data of the plurality of task data by the nonlinear processor; the feature data includes linear feature data and nonlinear feature data;

[0157] In one embodiment, the nonlinear processor includes a plurality of nonlinear processing modules; the plurality of nonlinear processing modules correspond one-to-one to the plurality of task data respectively; step 103 may include the following sub-steps:

[0158] Sub-step S21, determining the first-order Volterra series expansion term and the second-order Volterra series expansion term of the corresponding task data through the multiple nonlinear processing modules;

[0159] Sub-step S22, using the first-order Volterra series expansion term as the linear feature data;

[0160] Sub-step S23: using the second-order Volterra series expansion term as the nonlinear characteristic data.

[0161] The Volterra series can be combined with the NLMS algorithm for online updates to adapt to changes in the production environment in real time.

[0162] In one embodiment, sub-step S22 may include the following sub-steps:

[0163] Sub-step S221, obtaining first weights corresponding to the multiple nonlinear processing modules;

[0164] Sub-step S222, taking the product of the first-order Volterra series expansion term and the corresponding first weight as the linear feature data;

[0165] Sub-step S23 may include the following sub-steps:

[0166] Sub-step S231, taking the product of the second-order Volterra series expansion term and the first weight as the nonlinear characteristic data

[0167] Using the first weight to assign importance to different task data enhances the model's adaptability to diverse production scenarios. The introduction of weights also automatically suppresses the contribution of low-signal-to-noise data, improving model robustness. Multiplication operations are hardware-accelerated through matrix operations, meeting real-time prediction requirements.

[0168] Step 104: using the linear processor to predict the time required to execute the next production task based on the characteristic data of the plurality of task data.

[0169] In one embodiment, step 104 may include the following sub-steps:

[0170] Sub-step S31, obtaining a second weight;

[0171] Sub-step S32: taking the product of the characteristic data of the plurality of task data and the second weight as the time consumption for executing the next production task.

[0172] The second weight integrates the linear and nonlinear characteristics of all modules to achieve global optimal prediction across tasks. The regularization constraint of the second weight prevents overfitting, improves the model's migration capabilities across multiple factories and product lines, and assists in production scheduling decisions.

[0173] In one example, a method for predicting the time consumption of a production task of the present invention may be implemented as follows:

[0174] Suppose an industrial palletizing robot is performing production operations in a manufacturing plant. The robot is installed in a fixed position, does not stop, and executes a fixed process in a loop.

[0175] Deploy PLC to collect task data of the industrial palletizing robot's production tasks;

[0176] Build a prediction model on a computer. Use PyCharm or Matlab software to build the PCAPM-SOV model. Set the model parameters as follows: the number of modules, m, is set to 20; the number of external inputs per module, p, is set to 1; the initial values ​​of the weight vectors for the nonlinear and linear parts are set to random values ​​in the interval [0, 1] that follow a Gaussian distribution; and the adaptive iterative update algorithm uses NLMS.

[0177] When the computer receives data from the PLC, it saves the data and uses the latest 20 sets of data. For example, when 20 sets of data are received ([t(1), x1, x2, x3], …, [t(20), x1, x2, x3]), these 20 sets of data are input into PCAMP-SOV, where the external input of the first module of PCAPM-SOV is t(1), x1, x2, x3, the external input of the second module is t(2), x1, x2, x3, and the external input of the last module is t(20), x1, x2, x3. This is the first time the model is run, so the internal feedback signal of each module is 0. The output of the model is calculated by deducing the calculation formula of the model output in this application. This output is the predicted value of the next production task time, that is, the predicted value of t(21)

[0178] When t(21) is received, the accuracy of the prediction for t(21) in the previous step can be verified. Since the weight vector in the model has not been fully trained, the early prediction effect is not good, and the weight vector of the model needs to be updated using the NLMS algorithm. After the weight vector is updated, [t(2), x1, x2, x3], …, [t(21), x1, x2, x3] is used as the input of the model to calculate the model output. This output is the predicted value of the next production task time, that is, the predicted value of t(22).

[0179] Similarly, every time data is received from the PLC, the weight vector is updated. After the update is completed, the predicted value of the next production task is calculated.

[0180] Once the model's predictions reach a certain level, they can be used to guide production and practical applications. For example, the time required to complete a production task can be used to predict the number of production tasks that the industrial palletizing robot can complete in the next day, month, or year.

[0181] Reference Figure 4 , shows a flowchart of a production task prediction according to an embodiment of the present invention:

[0182] By running the trained PCAPM-SOV model and replacing the next observation value t(-1) with the predicted value y, task association information data T is generated at the same time. i,b It can be collected and calculated again when the task is executed. Therefore, the input queue T can be updated i , and then the model loops until the termination condition is met, outputting N. The core idea is to use the model's output, the predicted value y for the next production task, using a recursive method to iteratively predict the time required for the next n production tasks, ultimately calculating the number of production tasks completed the following day. Similarly, the loop termination condition can be changed to any number of days D, and the loop judgment condition can be changed to H >= 24*D, thus predicting the number of production tasks completed for the next D days.

[0183] When predicting the number of production tasks that the industrial palletizing robot can complete in the next day, month, or year, there is no need to update the model's weight parameters each time the model output is calculated. This is because without observations (data sent by the PLC), the NLMS algorithm cannot be used to update the weight vector. For example, the model runs 36,000 times to reach convergence, with good performance. At this point, it is used to predict the number of production tasks for the next day. x1, x2, x3 are used as model inputs to calculate the model output At this point, the industrial robot is still running and has not completed its 36001st production task, and the PLC has not yet sent t(36002), x1, x2, x3. This is because the computer's calculation time is less than the time it takes for the industrial robot to run once.

[0184] The nonlinear processor is used to generate feature data for the multiple task data; the feature data includes linear feature data and nonlinear feature data, which can model implicit dynamic nonlinear relationships from complex production data, making up for the lack of accuracy of traditional linear models; the linear processor directly performs low-complexity linear operations on the extracted features, avoiding the high computational delay of traditional deep learning models, meeting real-time prediction requirements, and achieving collaborative optimization of prediction accuracy and real-time performance in dynamic industrial scenarios. At the same time, the model is based on historical data-driven training, can adapt to changes in the production environment, and gets rid of dependence on fixed physical models. Ultimately, through the collaborative architecture of processing nonlinear and linear features in stages, the computational overhead is significantly reduced while ensuring prediction accuracy, providing efficient and reliable technical support for real-time task scheduling in industrial scenarios.

[0185] It should be noted that for the sake of simplicity, the method embodiments are described as a series of actions. However, those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0186] Reference Figure 5 , shows a structural block diagram of an embodiment of a production task time consumption prediction device of the present invention, which may specifically include the following modules:

[0187] The prediction preparation module 201 is used to obtain a production task time consumption prediction model and historical production data; the historical production data includes task data of multiple production tasks previously executed;

[0188] The data input module 202 is used to input the historical production data into the production task time consumption prediction model; the production task time consumption prediction model includes: a nonlinear processor and a linear processor;

[0189] A feature extraction module 203 is configured to generate feature data of the plurality of task data through the nonlinear processor; the feature data includes linear feature data and nonlinear feature data;

[0190] The time consumption prediction module 204 is configured to predict the time consumption of executing the next production task according to the characteristic data of the plurality of task data through the linear processor.

[0191] In one embodiment, the nonlinear processor includes a plurality of nonlinear processing modules; the plurality of nonlinear processing modules correspond one-to-one to the plurality of task data; the feature extraction module includes:

[0192] A Volterra expansion submodule, configured to determine a first-order Volterra series expansion term and a second-order Volterra series expansion term of the corresponding task data through the plurality of nonlinear processing modules;

[0193] a linear feature determination submodule, configured to use the first-order Volterra series expansion term as the linear feature data;

[0194] The nonlinear feature determination submodule is used to use the second-order Volterra series expansion term as the nonlinear feature data.

[0195] In one embodiment, the linear feature determination submodule includes:

[0196] A first weight obtaining unit, configured to obtain first weights corresponding to the plurality of nonlinear processing modules;

[0197] a weighted linear feature determining unit, configured to use the product of the first-order Volterra series expansion term and the corresponding first weight as the linear feature data;

[0198] The nonlinear feature determination submodule includes:

[0199] The weight line nonlinear characteristic determination unit is used to use the product of the second-order Volterra series expansion term and the first weight as the nonlinear characteristic data.

[0200] In one embodiment, the data input module includes:

[0201] A reference feature submodule, configured to obtain a plurality of reference feature data generated by the plurality of nonlinear processing modules in a previous prediction process with respect to historical production data of the previous prediction process;

[0202] The input data determination submodule is used to use the corresponding task data in the current prediction process and the reference feature data of an adjacent nonlinear processing module as input data for each nonlinear processing module.

[0203] In one embodiment, the time consumption prediction module includes:

[0204] A second weight acquisition submodule, used to acquire a second weight;

[0205] The weighted time consumption prediction submodule is used to use the product of the characteristic data of the plurality of task data and the second weight as the time consumption for executing the next production task.

[0206] In one embodiment, the task data of the production task includes the time taken by the industrial robot to perform the production task, the average load of the industrial robot during the production task, the average temperature of the industrial robot during the production task, and the task complexity of the production task.

[0207] In one embodiment, the production task time consumption prediction model is trained according to the following steps:

[0208] A training data acquisition module is used to obtain training data and verification data;

[0209] A training data input module, used for inputting the training data into the production task time consumption prediction model;

[0210] A training feature acquisition module, configured to generate feature data of the training data through the nonlinear processor; the feature data includes linear feature data and nonlinear feature data;

[0211] A training result acquisition module, configured to predict the time consumption of a training production task based on the feature data of the plurality of training data through the linear processor;

[0212] An error value determination module, used to determine the error value of the predicted training production task time and the verification data;

[0213] The training module is used to update the linear parameters and nonlinear parameters of the production task time consumption prediction model according to the preset stochastic gradient descent algorithm when the error value is greater than the preset threshold, and repeat the above training process until the preset training conditions are met.

[0214] The nonlinear processor is used to generate feature data for the multiple task data; the feature data includes linear feature data and nonlinear feature data, which can model implicit dynamic nonlinear relationships from complex production data, making up for the lack of accuracy of traditional linear models; the linear processor directly performs low-complexity linear operations on the extracted features, avoiding the high computational delay of traditional deep learning models, meeting real-time prediction requirements, and achieving collaborative optimization of prediction accuracy and real-time performance in dynamic industrial scenarios. At the same time, the model is based on historical data-driven training, can adapt to changes in the production environment, and gets rid of dependence on fixed physical models. Ultimately, through the collaborative architecture of processing nonlinear and linear features in stages, the computational overhead is significantly reduced while ensuring prediction accuracy, providing efficient and reliable technical support for real-time task scheduling in industrial scenarios.

[0215] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0216] An embodiment of the present invention also provides an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, the various processes of the embodiment of the time-consuming prediction method for the above-mentioned production task are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0217] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the embodiment of the time-consuming prediction method for the above-mentioned production task are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0218] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0219] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus, or computer program products. Thus, embodiments of the present invention may take the form of a fully hardware embodiment, a fully software embodiment, or an embodiment combining software and hardware. Furthermore, embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0220] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0221] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0222] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0223] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they become aware of the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0224] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or terminal device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or terminal device that includes the element.

[0225] The above is a detailed introduction to the time-consuming prediction method, device, equipment and medium for a production task provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for predicting the time consumption of a production task, characterized in that: The method comprises: Obtaining a production task time-consuming prediction model and historical production data; the historical production data includes task data of multiple production tasks previously executed; Inputting the historical production data into the production task time consumption prediction model; the production task time consumption prediction model includes: a nonlinear processor and a linear processor; Generate characteristic data of the plurality of task data by the nonlinear processor; the characteristic data includes linear characteristic data and nonlinear characteristic data; The linear processor predicts the time required to execute the next production task based on the characteristic data of the plurality of task data.

2. The method for predicting the time consumption of a production task according to claim 1, characterized in that: The nonlinear processor includes a plurality of nonlinear processing modules; The plurality of nonlinear processing modules correspond to the plurality of task data one by one respectively; the generating of feature data of the plurality of task data by the nonlinear processor comprises: Determining first-order Volterra series expansion terms and second-order Volterra series expansion terms of corresponding task data through the multiple nonlinear processing modules; Using the first-order Volterra series expansion term as the linear characteristic data; The second-order Volterra series expansion term is used as the nonlinear characteristic data.

3. The method for predicting the time consumption of a production task according to claim 2, characterized in that: The step of using the first-order Volterra series expansion term as the linear characteristic data includes: Obtaining first weights corresponding to the multiple nonlinear processing modules; Taking the product of the first-order Volterra series expansion term and the corresponding first weight as the linear feature data; The step of using the second-order Volterra series expansion term as the nonlinear characteristic data includes: The product of the second-order Volterra series expansion term and the first weight is used as the nonlinear characteristic data.

4. The method for predicting the time consumption of a production task according to claim 2, wherein: The inputting the historical production data into the production task time-consuming prediction model includes: Acquire multiple reference feature data generated by the multiple nonlinear processing modules during the previous prediction process based on the historical production data of the previous prediction process; For each nonlinear processing module, the corresponding task data in the current prediction process and the reference feature data of an adjacent nonlinear processing module are used as input data.

5. The method for predicting the time consumption of a production task according to claim 1, wherein: The predicting the time required to execute the next production task based on the characteristic data of the plurality of task data includes: Get the second weight; The product of the characteristic data of the plurality of task data and the second weight is used as the time consumption for executing the next production task.

6. The method for predicting the time consumption of a production task according to claim 1, characterized in that: The task data of the production task includes the time consumed by the industrial robot to perform the production task, the average load of the industrial robot during the production task, the average temperature of the industrial robot during the production task, and the task complexity of the production task.

7. The method for predicting the time consumption of a production task according to claim 1, characterized in that: The production task time consumption prediction model is trained according to the following steps: Get training data and validation data; Inputting the training data into the production task time-consuming prediction model; Generate feature data of the training data by the nonlinear processor; the feature data includes linear feature data and nonlinear feature data; Predicting the time consumption of a training production task according to the characteristic data of the plurality of training data by the linear processor; Determining the time consumption of the prediction training production task and the error value of the verification data; When the error value is greater than a preset threshold, the linear parameters and nonlinear parameters of the production task time consumption prediction model are updated according to the preset stochastic gradient descent algorithm, and the above training process is repeated until the preset training conditions are met.

8. A time-consuming prediction device for a production task, characterized in that: The device comprises: A prediction preparation module is used to obtain a production task time-consuming prediction model and historical production data; the historical production data includes task data of multiple production tasks previously executed; A data input module is used to input the historical production data into the production task time consumption prediction model; the production task time consumption prediction model includes: a nonlinear processor and a linear processor; A feature extraction module, configured to generate feature data of the plurality of task data through the nonlinear processor; the feature data includes linear feature data and nonlinear feature data; The time consumption prediction module is used to predict the time consumption of executing the next production task according to the characteristic data of the plurality of task data through the linear processor.

9. An electronic device, characterized in that: include: A processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein when the computer program is executed by the processor, the steps of the method for predicting the time consumption of a production task as claimed in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for predicting the time consumption of a production task according to any one of claims 1 to 7 are implemented.

Images