A big data-based intelligent factory dynamic collaborative scheduling method and system

Abstract

The application discloses a kind of based on big data's wisdom factory dynamic collaborative scheduling method and system, it is related to wisdom factory management technical field, the method includes: obtaining the state data corresponding to production line, and using state data to establish multidimensional data tensor;Production line collaborative scheduling model is constructed based on multidimensional data tensor, and gradient boosting machine algorithm is used to train production line collaborative scheduling model;Real-time state data collected are input to production line collaborative scheduling model, and the scheduling instruction of each production line is output, and scheduling instruction is transmitted to wisdom factory control end using wireless communication network technology;The execution result of scheduling instruction is monitored using internet of things and monitoring technology, and production line collaborative scheduling model is optimized based on execution result.The application can accurately capture the uncertainty factors in the production environment by collecting and analyzing the state data of each production line, and quickly generate the corresponding scheduling instruction, ensure the consistency and accuracy of decision-making.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart factory management, in particular to a smart factory dynamic collaborative scheduling method and system based on big data. BACKGROUND

[0002] Under the background of the rapid development of intelligentization and big data technology, smart factory, as an important symbol of modern manufacturing informatization, has undergone profound digitalization, networking and intelligentization upgrade through the integration of Internet of Things (IoT), big data analysis, cloud computing and artificial intelligence (AI) and other frontier information technologies. This transformation not only improves the automation level of production, but also enhances the visualization, traceability and predictability of the entire manufacturing process, thereby effectively improving production efficiency, reducing costs, and strengthening product quality and the ability to adapt to market changes.

[0003] In a smart factory, the application of Internet of Things technology is particularly crucial, as it enables real-time feedback for equipment status monitoring and product management, providing management with the possibility of precise control over production activities. At the same time, through the collection and intelligent analysis of massive data, a smart factory can learn and optimize itself, promoting the production process towards a more autonomous direction.

[0004] However, in the actual operation of a smart factory, there are a large number of uncertain factors in the production environment, such as fluctuations in raw material supply and changes in order demand, which pose great challenges to production scheduling. Traditional dynamic collaborative scheduling methods are not up to the task when faced with these complex and variable situations. Such methods usually rely on manual information collection and decision-making, which not only consumes time and effort, but also is susceptible to personal experience and subjective judgment, making it difficult to ensure the consistency and accuracy of decisions. In addition, they often lack sufficient flexibility to quickly respond to changing production conditions, failing to meet the requirements of today's manufacturing industry for high efficiency and low latency.

[0005] To address the problems in the related art, no effective solutions have been proposed so far. SUMMARY

[0006] To address the problems in the related art, the present application proposes a smart factory dynamic collaborative scheduling method and system based on big data to overcome the aforementioned technical problems existing in the prior art.

[0007] To this end, the specific technical solutions adopted by the present application are as follows:

[0008] According to one aspect of the present application, a smart factory dynamic collaborative scheduling method based on big data is provided, which comprises the following steps:

[0009] S1, according to the running state of each production line in the smart factory, obtain the state data corresponding to the production line, and establish a multi-dimensional data tensor using the state data;

[0010] S2, based on the multi-dimensional data tensor, a production line collaborative scheduling model is constructed, and a gradient boosting machine algorithm is used to train the production line collaborative scheduling model, and the final production line collaborative scheduling model is obtained according to the training result;

[0011] S3, input the collected real-time state data into the production line collaborative scheduling model, output the scheduling instruction of each production line, and transmit the scheduling instruction to the smart factory control terminal by using wireless communication network technology;

[0012] S4, the smart factory control terminal receives and executes the scheduling instruction, monitors the execution result of the scheduling instruction by using the Internet of Things and monitoring technology, and optimizes the production line collaborative scheduling model based on the execution result;

[0013] Based on the multi-dimensional data tensor, a production line collaborative scheduling model is constructed, and a gradient boosting machine algorithm is used to train the production line collaborative scheduling model, and the final production line collaborative scheduling model is obtained according to the training result, including the following steps:

[0014] S21, the principal component analysis algorithm is used to process the multi-dimensional data tensor to extract the feature data of each process of the production line, and the feature data is integrated to obtain a feature data set;

[0015] S22, based on the feature data set, a production line collaborative scheduling model is designed and constructed, and the production line collaborative scheduling model takes the feature data set as input and the scheduling instruction of each production line as output;

[0016] S23, use the gradient boosting machine algorithm to train the production line collaborative scheduling model to obtain the final production line collaborative scheduling model.

[0017] Further, according to the running state of each production line in the smart factory, the state data corresponding to the production line is obtained, and a multi-dimensional data tensor is established using the state data, including the following steps:

[0018] S11, according to the running state of each production line in the smart factory, determine the dimension of the state data;

[0019] S12, use the pre-set data interface to collect state data from each production line, and pre-process the state data;

[0020] S13, based on the pre-processed state data, integrate the state data of each process of each production line to obtain the preliminary running state information of the corresponding production line;

[0021] S14. Analyze the preliminary operating status information, identify the status indicators that can reflect the operating status of the production line, and combine the identified status indicators according to the determined dimensions to establish a multi-dimensional data matrix of the production line.

[0022] S15. Repeat steps S13 to S14 to obtain the multi-dimensional data matrix corresponding to each production line.

[0023] S16. Summarize the multi-dimensional data matrices corresponding to each production line and establish a multi-dimensional data tensor.

[0024] Furthermore, principal component analysis (PCA) is used to process the multidimensional data tensor to extract feature data from each process of the production line, and the feature data is then integrated to obtain a feature dataset, including the following steps:

[0025] S211. Based on the attributes and structural characteristics of each process in the production line, classify the multidimensional data tensors and construct the corresponding data structure models.

[0026] S212. Analyze the needs of production line collaborative scheduling, select several state data related to scheduling needs in the data structure model, and establish a candidate feature dataset;

[0027] S213. Use principal component analysis algorithm to process the candidate feature dataset and extract feature data of each process in the production line;

[0028] S214. Based on the preset data flow and scheduling rules of each process in the production line, integrate the extracted feature data to obtain a feature dataset.

[0029] Furthermore, the data flow formula for each process in the production line is as follows:

[0030]

[0031] In the formula, n ( t )express t The data flow of each process on the production line at any given moment;

[0032] t Indicates the extraction time of feature data;

[0033] N Indicates the total number of feature data;

[0034] i Indicates the first i One feature data;

[0035] M Represents the candidate feature dataset;

[0036] λrepresenting the amplitude of the oscillation present in the data stream;

[0037] p representing the attributes of the feature data;

[0038] c representing the non-production data in each process of the production line.

[0039] Further, based on the feature data set, a production line collaborative scheduling model is designed and constructed, the production line collaborative scheduling model takes the feature data set as input and the scheduling instructions of each production line as output, including the following steps:

[0040] S221, based on the feature data set, the number of nodes of the input layer is designed to match the dimension of the feature data set;

[0041] S222, the structure of the hidden layer is designed to process the input feature data set and further extract the input features, the structure of the hidden layer includes the number of layers, the number of neurons in each layer and the activation function;

[0042] S223, according to the requirements of the production line collaborative scheduling, the number of nodes of the output layer is designed to correspond to and generate the required number of scheduling instructions of each production line;

[0043] S224, using a deep learning framework, the input layer, the hidden layer and the output layer designed are connected to construct the production line collaborative scheduling model.

[0044] Further, the gradient boosting machine algorithm is used to train the production line collaborative scheduling model to obtain the final production line collaborative scheduling model, including the following steps:

[0045] S231, the feature data set is divided into a training set and a test set, and the parameters of the production line collaborative scheduling model are initialized;

[0046] S232, each training data in the training set is traversed, the gradient of the loss function is calculated using the parameters of the current production line collaborative scheduling model, and the gradient calculation is optimized combined with the binary ant colony algorithm;

[0047] S233, combined with the gradient and the feature data in the feature data set, a new training set is established, the current weak decision tree is trained using the new training set, and the optimal scaling factor is determined;

[0048] S234, according to the output result of the weak decision tree and the optimal scaling factor, the parameters of the production line collaborative scheduling model are updated, and the updated parameters are used as the starting parameters for the next iteration;

[0049] S235, repeat the above steps S232 to S234 until the preset number of iterations is reached;

[0050] S236, integrate the weak decision trees trained in all iterations according to the corresponding scaling factors to obtain a production line collaborative scheduling model after training;

[0051] S237, evaluate the production line collaborative scheduling model after training using a test set to obtain a final production line collaborative scheduling model.

[0052] Further, traverse each training data in the training set, calculate the gradient of the loss function using the parameters of the current production line collaborative scheduling model, and combine the binary ant colony algorithm to optimize the gradient calculation, including the following steps:

[0053] S2321, traverse each training data in the training set, calculate the gradient value of the loss function to the current parameters after the previous iteration according to each training data through the parameters of the current production line collaborative scheduling model;

[0054] S2322, based on the calculated gradient value, define a discrete search space for the binary ant colony algorithm, and randomly generate parameter solutions as the initial population in the discrete search space using binary coding rules;

[0055] S2323, evaluate the fitness of each parameter solution in the initial population using the loss function, and select the parameter solution according to the fitness value function;

[0056] S2324, combine the selected parameter solution with the roulette selection algorithm for cross processing to obtain a new parameter solution, and perform inverse mutation according to the fitness function to further explore the discrete search space;

[0057] S2325, assign an initial pheromone value to each parameter solution path, and place several ants in the discrete search space to select the next moving direction according to the current pheromone distribution and gradient value;

[0058] S2326, update the pheromone distribution of the discrete search space according to the moving path of the ant and the fitness of the parameter solution;

[0059] S2327, repeat steps S2323 to S2326 until a preset number of iterations is reached to obtain the optimal parameter solution.

[0060] Further, the update formula of the pheromone distribution is:

[0061]

[0062] In the formula, τ ij represents the pheromone concentration on the path i to the path j ;

[0063] t represents a time;

[0064] n represents a natural number;

[0065] ρ represents pheromone evaporation coefficient;

[0066] represents the information amount increment of the path i to the path j in this cycle.

[0067] Further, the collected real-time state data is input into a production line collaborative scheduling model, scheduling instructions of each production line are output, and the scheduling instructions are transmitted to the control end of the smart factory by using wireless communication network technology, including the following steps:

[0068] S31, collecting real-time state data from each production line through a pre-set data interface, and pre-processing the real-time state data;

[0069] S32, extracting feature data from the pre-processed real-time state data, establishing a real-time feature data set, and inputting the real-time feature data set into the production line collaborative scheduling model;

[0070] S33, the production line collaborative scheduling model outputs scheduling instructions of each production line according to the input feature data set;

[0071] S34, using wireless communication network technology, transmitting the generated scheduling instructions to the control end of the smart factory.

[0072] According to another aspect of the present application, a dynamic collaborative scheduling system of a smart factory based on big data is also provided, which comprises a data collection and preprocessing module, a model construction and training module, a scheduling instruction generation and transmission module, and an execution monitoring and model optimization module;

[0073] The data collection and preprocessing module, the model construction and training module, the scheduling instruction generation and transmission module, and the execution monitoring and model optimization module are sequentially connected;

[0074] The data collection and preprocessing module is used to obtain state data corresponding to the production line according to the running state of each production line in the smart factory, and to establish a multi-dimensional data tensor by using the state data;

[0075] The model construction and training module is used to construct a production line collaborative scheduling model based on the multi-dimensional data tensor, and to train the production line collaborative scheduling model by using a gradient boosting machine algorithm, so as to obtain a final production line collaborative scheduling model according to the training result;

[0076] The scheduling instruction generation and transmission module is configured to input the collected real-time state data into the production line collaborative scheduling model, output scheduling instructions of each production line, and transmit the scheduling instructions to the intelligent factory control terminal by using wireless communication network technology.

[0077] The execution monitoring and model optimization module is configured to receive and execute the scheduling instructions by the intelligent factory control terminal, monitor the execution results of the scheduling instructions by using Internet of Things and monitoring technology, and optimize the production line collaborative scheduling model based on the execution results.

[0078] The present application has the following advantages:

[0079] 1. The present application can accurately capture uncertain factors in the production environment and quickly generate corresponding scheduling instructions by collecting and analyzing the state data of each production line and constructing a multi-dimensional data tensor based on the state data, thereby improving the ability to respond to complex and variable production conditions, reducing the time for manual information collection and decision-making, reducing the influence of personal experience and subjective judgment on decision-making, and ensuring the consistency and accuracy of decision-making.

[0080] 2. The present application can generate more reasonable and efficient production plans by constructing and training the production line collaborative scheduling model, thereby improving production efficiency, optimizing resource allocation and reducing waste. The production line collaborative scheduling model uses multi-dimensional data tensors and machine learning algorithms to comprehensively consider factors such as equipment status, material supply, personnel allocation and market demand, realizes accurate prediction and decision-making, and thus enables the factory to better adapt to complex and variable production environments and market demands.

[0081] 3. The present application transmits the scheduling instructions by using wireless communication network technology, realizes remote monitoring and management of the production line, greatly improves the response speed and operation flexibility, thereby meeting the needs of the manufacturing industry, enabling management personnel to obtain real-time state information of the production line, and enabling them to take prompt measures when problems are found, thereby maintaining efficient production coordination and control and further improving the competitiveness and operating efficiency of the enterprise. BRIEF DESCRIPTION OF DRAWINGS

[0082] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0083] Figure 1 is a flowchart of an intelligent factory dynamic collaborative scheduling method based on big data according to an embodiment of the present application;

[0084] Figure 2is a principle block diagram of a big data-based smart factory dynamic collaborative scheduling system according to an embodiment of the present application.

[0085] In the drawings:

[0086] 1, data collection and preprocessing module; 2, model construction and training module; 3, scheduling instruction generation and transmission module; 4, execution monitoring and model optimization module. DETAILED DESCRIPTION

[0087] To further illustrate the embodiments, the present application provides drawings which are part of the disclosure of the present application, mainly used to illustrate the embodiments, and can explain the operating principle of the embodiments in conjunction with the related description of the specification. With reference to these contents, those skilled in the art should understand other possible implementations and advantages of the present application.

[0088] According to an embodiment of the present application, a big data-based smart factory dynamic collaborative scheduling method and system are provided.

[0089] The present application will be further described in conjunction with the drawings and specific embodiments, as shown, according to an embodiment of the present application, a big data-based smart factory dynamic collaborative scheduling method, the method comprising the following steps: Figure 1

[0090] S1, according to the running state of each production line in the smart factory, the state data corresponding to the production line is obtained, and a multi-dimensional data tensor is established by using the state data.

[0091] Specifically, according to the running state of each production line in the smart factory, the state data corresponding to the production line is obtained, and a multi-dimensional data tensor is established by using the state data, comprising the following steps:

[0092] S11, according to the running state of each production line in the smart factory, the dimension of the state data is determined.

[0093] It should be noted that the dimension of the state data includes: time dimension (collecting production data at different time points, including each time node in each production cycle), equipment dimension (device state data of each production line, such as temperature, pressure, speed, working time, etc.), product dimension (product information involved in the production process, such as product batch, product specification, process parameters, etc.), production state dimension (running state of each production line, such as normal, fault, maintenance, etc., and its corresponding running efficiency), and environment dimension (external environmental conditions, such as factory temperature, humidity, etc. Data that may affect production efficiency).

[0094] S12, the state data is collected from each production line by using the pre-set data interface, and the state data is preprocessed.

[0095] ​It should be noted that the data interface includes API endpoints, request parameters, and response structures.

[0096] In addition, the collected state data is pre-processed, including data cleaning (removing duplicate, error or invalid data), data conversion (such as data type conversion, unit conversion, etc.) and data normalization (converting data to a uniform magnitude or range to facilitate subsequent analysis).

[0097] S13, based on the pre-processed state data, integrating the state data of each process of each production line to obtain the preliminary running state information of the corresponding production line.

[0098] It should be noted that for a certain processing process on a certain production line, the equipment state data (such as temperature, pressure, speed, etc.), product dimension data (such as product batch, product specification, etc.) and production state data (such as normal, fault, maintenance, etc.) of the process need to be integrated; based on the integrated process state data, generate the preliminary running state information of the corresponding production line, the preliminary running state information includes timestamp, equipment number, equipment state (such as running, shutdown, fault, etc.), product batch, product specification, process parameter, production line efficiency, factory temperature, humidity, etc.

[0099] S14, analyzing the preliminary running state information, identifying state indicators that can reflect the running state of the production line, and combining the identified state indicators according to the determined dimensions to establish a multi-dimensional data matrix of the production line.

[0100] It should be noted that in the preliminary running state information, key state indicators that can reflect the running state of the production line are identified, which are usually closely related to production efficiency, equipment performance, product quality, etc. For example: equipment efficiency, which measures the proportion of actual running time to total time of equipment within a certain time, reflecting the utilization rate of equipment; product pass rate, which measures the proportion of the number of qualified products to the total number of products on the production line, reflecting the product quality control level; process parameter stability: which measures the fluctuation of process parameters (such as temperature, pressure, speed, etc.) in the production process, reflecting the stability of the production process; production line efficiency, which measures the proportion of actual output to theoretical output of the production line within a certain time, reflecting the overall efficiency of the production line.

[0101] The identified state indicators are combined according to the time dimension, equipment dimension, product dimension, production state dimension and environment dimension to form a multi-dimensional data matrix of the production line, the specific implementation steps are as follows:

[0102] Time dimension: divide the state indicators according to different time points or time periods, such as daily, weekly, monthly, etc., to observe the performance of the production line in different time periods; equipment dimension: divide the state indicators according to different equipment or equipment types, such as machine tools, production lines, detection equipment, etc., to understand the performance of each equipment; product dimension: divide the state indicators according to different product batches, product specifications or product types to analyze the impact of different products on the production line; production state dimension: divide the state indicators according to the running state of the production line, such as normal, fault, maintenance, etc., to evaluate the stability and reliability of the production line; environmental dimension: divide the state indicators according to external environmental conditions, such as factory temperature, humidity, etc., to explore the influence of environmental factors on production efficiency.

[0103] The identified state indicators are combined according to the determined dimensions to establish a multi-dimensional data matrix of the production line, for example, a data matrix containing time dimension, equipment dimension and product dimension is shown in Table 1:

[0104] Table 1 Multi-dimensional data matrix

[0105]

[0106] S15, repeat steps S13 to S14 to obtain the multi-dimensional data matrix corresponding to each production line.

[0107] S16, summarize the multi-dimensional data matrix corresponding to each production line to establish a multi-dimensional data tensor.

[0108] S2, based on the multi-dimensional data tensor, construct a production line collaborative scheduling model, and use the gradient boosting machine algorithm to train the production line collaborative scheduling model, and obtain the final production line collaborative scheduling model according to the training result.

[0109] Based on the multi-dimensional data tensor, a production line collaborative scheduling model is constructed, and the gradient boosting machine algorithm is used to train the production line collaborative scheduling model, and the final production line collaborative scheduling model is obtained according to the training result, including the following steps:

[0110] S21, use principal component analysis algorithm to process the multi-dimensional data tensor to extract feature data of each process of the production line, and integrate the feature data to obtain a feature data set.

[0111] Specifically, using principal component analysis algorithm to process the multi-dimensional data tensor to extract feature data of each process of the production line, and integrating the feature data to obtain a feature data set includes the following steps:

[0112] S211, according to the properties and structural characteristics of each process of the production line, classify the multi-dimensional data tensor, and construct a corresponding data structure model.

[0113] It should be noted that each process on the production line is analyzed to understand its physical properties (such as equipment type, size, power, etc.) and process characteristics (such as machining accuracy, processing time, temperature requirements, etc.); the logical relationship and structural characteristics between processes are sorted out, such as serial relationship, parallel relationship, feedback relationship, etc.; according to the process attributes and structural characteristics, the dimensions of the multi-dimensional data tensor are determined.

[0114] For example, according to the time dimension (such as date, shift), the spatial dimension (such as production line position, process sequence), the equipment dimension (such as equipment type, equipment number), the product dimension (such as product type, product batch) and the like, the multi-dimensional data tensor is classified, the processes with similar attributes or functions are classified into the same category, and different tensor categories are formed; for each classified tensor, an array, a matrix, a tensor or the like is used to construct a corresponding data structure model, and a corresponding data type, length, range or the like is defined as needed.

[0115] S212, analyze the demand for collaborative scheduling of the production line, select a number of state data associated with the scheduling demand in the data structure model, and establish a candidate feature data set.

[0116] It should be noted that the demand for collaborative scheduling of the production line is analyzed, the scheduling target is determined, the key process is identified, and the scheduling constraint is determined; the state data related to the scheduling target, the key process and the scheduling constraint is selected from the data structure model, including the equipment state (such as running, shutdown, failure, etc.), the product state (such as product batch, product specification, production progress, etc.), the personnel state (such as personnel configuration, skill level, etc.) and the material state (such as material supply, inventory, etc.); the selected state data is sorted into a candidate feature data set, and each feature corresponds to one or more state data.

[0117] S213, using principal component analysis algorithm to process the candidate feature data set, and extracting the feature data of each process of the production line.

[0118] It should be noted that the covariance matrix between the candidate feature data set is calculated to reflect the correlation between the features; the eigenvalue decomposition is performed on the covariance matrix to obtain the eigenvalue and the corresponding eigenvector; according to the size of the eigenvalue, the eigenvectors corresponding to a number of larger eigenvalues are selected as principal components; the principal component space is constructed using the selected principal components; the feature data in the principal component space is explained to determine the meaning and physical meaning represented by each principal component, and the feature data of each process of the production line is extracted from the principal component space. These feature data can be coordinate values in the principal component space, or feature indicators after further processing.

[0119] S214, according to the data flow of each process of the preset production line and the scheduling rule, the extracted feature data is integrated to obtain a feature data set.

[0120] Specifically, the data flow formula of each process of the production line is:

[0121]

[0122] In the formula, n(t) represents the data flow of each process of the production line at time t;

[0123] t represents the extraction time of the feature data;

[0124] N represents the total number of feature data;

[0125] i represents the i-th feature data; i

[0126] M represents the candidate feature data set;

[0127] λ represents the oscillation amplitude existing in the data flow;

[0128] p represents the attribute of the feature data;

[0129] c represents the non-production data in each process of the production line.

[0130] It should be noted that the scheduling rule includes the priority of the process, the availability of the equipment, the configuration of the personnel, etc.; for example, assuming that a production line includes three processes: A (cutting), B (welding), and C (spraying), then:

[0131] (1) Data flow and scheduling rule

[0132] Data flow: raw materials (input) → A process (cutting) → intermediate products (output) → B process (welding) → semi-finished products (output) → C process (spraying) → finished products (output).

[0133] Scheduling rule: A process has the highest priority, B process has the second priority, and C process has the lowest priority.

[0134] (2) Feature data integration

[0135] Extract the feature data of A process: cutting speed, cutting accuracy, cutting time, etc.

[0136] Extract the feature data of B process: welding current, welding temperature, welding time, etc.

[0137] ​The feature data of the extraction C process includes spraying speed, spraying thickness, spraying color, etc.

[0138] The feature data is integrated into a data set, and the source and purpose of each data are labeled.

[0139] (3) Feature data set storage

[0140] The integrated feature data set is stored in the database, and clear table structure and field description are set.

[0141] S22, based on the feature data set, design and build a production line collaborative scheduling model, and the production line collaborative scheduling model takes the feature data set as input and the scheduling instructions of each production line as output.

[0142] Specifically, based on the feature data set, the production line collaborative scheduling model is designed and built, and the production line collaborative scheduling model takes the feature data set as input and the scheduling instructions of each production line as output, including the following steps:

[0143] S221, based on the feature data set, design the number of nodes of the input layer to match the dimension of the feature data set;

[0144] S222, design the structure of the hidden layer to process the input feature data set and further extract the input feature, and the structure of the hidden layer includes the number of layers, the number of neurons in each layer and the activation function;

[0145] S223, according to the requirements of the production line collaborative scheduling, design the number of nodes of the output layer to correspond to and generate the required number of scheduling instructions of each production line;

[0146] S224, use the deep learning framework to connect the designed input layer, hidden layer and output layer to build the production line collaborative scheduling model.

[0147] S23, use gradient boosting machine algorithm to train the production line collaborative scheduling model to get the final production line collaborative scheduling model.

[0148] Specifically, the gradient boosting machine algorithm is used to train the production line collaborative scheduling model to get the final production line collaborative scheduling model, including the following steps:

[0149] S231, divide the feature data set into training set and test set, and initialize the parameters of the production line collaborative scheduling model.

[0150] S232, traverse each training data in the training set, calculate the gradient of the loss function using the parameters of the current production line collaborative scheduling model, and combine the binary ant colony algorithm to optimize the gradient calculation.

[0151] Specifically, the process involves iterating through each training data point in the training set, calculating the gradient of the loss function using the parameters of the current production line collaborative scheduling model, and optimizing the gradient calculation using the binary ant colony algorithm. This includes the following steps:

[0152] S2321. Iterate through each training data in the training set, and calculate the gradient value of the loss function with respect to the current parameters after the previous iteration based on the parameters of the current production line collaborative scheduling model for each training data.

[0153] It should be noted that the formula for calculating the gradient value is:

[0154]

[0155] In the formula, Indicates the first m The gradient value of the loss function with respect to the parameters is calculated in the next iteration; i Indicates the first i One training data set; I S ( · ) represents Fisher information for the parameters; Indicates the first m-1 Parameters calculated after the next iteration (i.e., after the previous iteration) σ The value; ▽ S ( · () represents the gradient of the loss function in the parameter space with respect to the parameters; Indicates the first m-1 Parameters calculated after the next iteration (i.e., after the previous iteration) The value; y i Indicates the first i The true value of each training data point.

[0156] S2322. Based on the calculated gradient values, define a discrete search space for the binary ant colony algorithm, and randomly generate parameter solutions within the discrete search space using binary encoding rules as the initial population.

[0157] S2323. Use the loss function to evaluate the fitness of each parameter solution in the initial population, and select the parameter solution according to the fitness function.

[0158] It should be noted that the formula for the loss function is:

[0159]

[0160] In the formula, L ( θ () represents the loss function; θ These represent the parameters of the current production line collaborative scheduling model; N This represents the total number of training data;i denotes a natural number; l denotes mean square error; denotes the predicted value of the production line collaborative scheduling model under the parameters θ to the input x i . x i denotes the i-th training data; i i denotes the i-th training data. y i The formula of the fitness function is:

[0161] The formula of the fitness function is:

[0162]

[0163] In the formula, F ( θ ) denotes the fitness of the parameters θ ; L ( θ ) denotes the loss function; β denotes the loss function.

[0164] In addition, a certain number of parameter solutions are selected as candidate solutions for subsequent crossover and mutation operations according to the fitness function, which can be realized by sorting and selecting the top N parameter solutions, or sampling according to the probability distribution of the fitness function.

[0165] S2324, combine the selected parameter solutions with the roulette selection algorithm for crossover processing to obtain new parameter solutions, and perform inverse mutation according to the fitness function to further explore the discrete search space.

[0166] It should be noted that for the selected parameter solutions, the roulette selection algorithm is used for pairing, and the crossover operation is performed between the paired parameter solutions. The crossover operation can be single-point crossover, double-point crossover or uniform crossover, etc. The inverse mutation operation is performed on the new parameter solutions generated after the crossover operation, a section of binary bits in the parameter solution is randomly selected and reversed (i.e. 0 becomes 1 and 1 becomes 0), and the purpose of the inverse mutation is to further explore the discrete search space and increase the diversity of the population, so as to possibly find better parameter solutions.

[0167] S2325, assign an initial pheromone value to each parameter solution path, and place several ants in the discrete search space to select the next moving direction according to the current pheromone distribution and gradient value.

[0168] S2326, update the pheromone distribution of the discrete search space according to the moving path of the ant and the fitness of the parameter solution.

[0169] ​Specifically, the pheromone distribution update formula is:

[0170]

[0171] In the formula, τ ij denotes the path i to the path j ;

[0172] t denotes the time;

[0173] n denotes a natural number;

[0174] ρ denotes the pheromone evaporation coefficient;

[0175] denotes the pheromone increment of the path i to the path j in this cycle.

[0176] S2327, repeat steps S2323 to S2326 until the preset iteration number is reached, and obtain the optimal parameter solution.

[0177] S233, combine the gradient and the feature data in the feature data set to establish a new training set, train the current weak decision tree using the new training set, and determine the optimal scaling factor.

[0178] It should be noted that the current weak decision tree is trained using the new training set. During the training process, the weak decision tree divides the sample space according to the feature data and gradient information, and finds the best division point to minimize the loss function. After training the weak decision tree, the optimal scaling factor is determined through cross-validation or similar methods to adjust the output of the weak decision tree.

[0179] S234, update the parameters of the production line collaborative scheduling model according to the output of the weak decision tree and the optimal scaling factor, and use the updated parameters as the starting parameters for the next iteration.

[0180] S235, repeat the above steps S232 to S234 until the preset iteration number is reached.

[0181] S236, integrate all the weak decision trees trained in the iterations according to the corresponding scaling factors to obtain the production line collaborative scheduling model after training.

[0182] S237, evaluate the trained production line collaborative scheduling model using the test set to obtain the final production line collaborative scheduling model.

[0183] S3, input the collected real-time state data into the production line collaborative scheduling model, output the scheduling instructions of each production line, and transmit the scheduling instructions to the intelligent factory control terminal by using wireless communication network technology.

[0184] Specifically, inputting the collected real-time state data into the production line collaborative scheduling model, outputting the scheduling instructions of each production line, and transmitting the scheduling instructions to the intelligent factory control terminal by using wireless communication network technology include the following steps:

[0185] S31, collecting real-time state data from each production line through a pre-set data interface, and pre-processing the real-time state data;

[0186] S32, extracting feature data from the pre-processed real-time state data, establishing a real-time feature data set, and inputting the real-time feature data set into the production line collaborative scheduling model;

[0187] S33, the production line collaborative scheduling model outputs the scheduling instructions of each production line according to the input feature data set;

[0188] S34, using wireless communication network technology, transmitting the generated scheduling instructions to the control terminal of the intelligent factory.

[0189] It needs to be noted that in the intelligent factory, the production state data is collected in real time through the pre-set data interface on each production line, and the state data includes equipment running state, product processing progress, production quality data, environmental parameters (such as temperature, humidity) and the like, and the state data is pre-processed; the pre-processed real-time state data is input into the trained production line collaborative scheduling model as input, and the production line collaborative scheduling model will calculate the scheduling instructions of each production line according to the input data, the scheduling instructions including the start, stop, acceleration, deceleration and other operations of the production line, as well as the coordination and cooperation between different production lines; using wireless communication network technology (such as Wi-Fi, Bluetooth, Zigbee, etc.), the calculated scheduling instructions are transmitted to the control terminal of the intelligent factory in real time, ensuring the timeliness and accuracy of the scheduling instructions, so as to realize real-time monitoring and scheduling of the production line.

[0190] S4, the intelligent factory control terminal receives and executes the scheduling instructions, monitors the execution results of the scheduling instructions by using Internet of Things and monitoring technology, and optimizes the production line collaborative scheduling model based on the execution results.

[0191] It needs to be noted that the intelligent factory control terminal receives and executes the scheduling instructions, monitors the execution results of the scheduling instructions by using Internet of Things and monitoring technology, and optimizes the production line collaborative scheduling model based on the execution results, including:

[0192] The smart factory control end receives scheduling instructions from the production line collaborative scheduling model through wireless communication network technology; the control end analyzes the received scheduling instructions and converts them into specific control signals, which are sent to the corresponding production line, including adjusting production speed, changing production order, starting or stopping equipment, etc. The automated equipment on the production line executes the corresponding operations according to the control signals to realize the collaboration and optimization of the production process.

[0193] Using Internet of Things technology, real-time collection of production line state data, including equipment operating conditions, product quality, production efficiency, etc. These data are transmitted to the control end, through data analysis and visualization, to help managers understand the operation of the production line and the execution effect of the scheduling instructions in real time; based on the monitoring data, the execution effect of the scheduling instructions is evaluated (comparing the production efficiency, product quality, resource utilization rate, etc. before and after execution, and analyzing the impact of scheduling instructions on the production process); through evaluation, determine whether the scheduling instructions have achieved the expected effect, and whether there are areas that need to be improved; based on the evaluation results, optimize the production line collaborative scheduling model (including adjusting model parameters, adding new feature data, etc.).

[0194] The optimized model is redeployed to the production environment, and real-time state data is continuously collected for monitoring and evaluation; through continuous iteration and update, gradually improve the production line collaborative scheduling model, realize the continuous optimization and collaboration of the production line.

[0195] In specific embodiments, the smart factory control end receives scheduling instructions transmitted from the production line collaborative scheduling model, such as device ID, operation type (start, stop, adjust parameters, etc.), operation parameters (such as temperature setting, speed adjustment value) and timestamp.

[0196] Step one, analysis of instruction content

[0197] The control end analyzes the instructions, extracts the required operation information, and verifies the timestamp to ensure the timeliness of the instructions, as follows:

[0198] Scheduling instruction: the device of production line Line_1 needs to run at 1800 RPM during the completion period of 14:00:00-16:00:00;

[0199] Execution time: 14:00:00-16:00:00;

[0200] Device status: the device runs at 1800 RPM after the specified time.

[0201] Step two, execution of scheduling instructions

[0202] The control end transmits the parsed instructions to the production line equipment through OPCUA, Modbus, or other industrial communication protocols. For example, in the case of a PLC (Programmable Logic Controller), the control end issues device control instructions such as adjusting motor speed, changing production parameters, etc. The devices perform corresponding operations according to the received instructions. For example, device 1 adjusts the speed to 1800 RPM, and device 2 heats to the set temperature of 150°C.

[0203] During the scheduling execution process, IoT devices (such as sensors and monitoring devices) monitor the execution in real time to ensure that the devices correctly respond to the control instructions. The collected data includes whether the devices execute the instructions, and the state changes after execution (such as temperature, speed, power consumption, etc.).

[0204] Step three, monitoring and feedback of execution results

[0205] Through SCADA (Supervisory Control and Data Acquisition System) and IoT sensors, the control end continuously monitors the operation of the production line equipment. The collected data includes the state of the equipment (such as whether it starts or stops), speed, temperature, power consumption, etc., as follows:

[0206] Speed adjustment result of device 1: 1800 RPM;

[0207] Temperature adjustment result of device 2: 150°C;

[0208] Device energy consumption: 5kWh;

[0209] Workshop environment temperature: 25°C.

[0210] Collect feedback data after the execution of the scheduling instructions, including whether the execution is successful, execution time, and device response. For example, if the device fails to adjust the speed due to some reason, the feedback data will record the abnormal information. The collected feedback data is uploaded to the central scheduling system, which analyzes the efficiency and accuracy of the scheduling execution based on these feedback data. The feedback data includes execution success rate, device response time, and energy consumption data.

[0211] Step four, optimization of the scheduling model

[0212] The control end transmits the feedback data of the execution to the central system, which analyzes which factors affect the scheduling effect based on the real-time state of the production equipment, production parameters, and environmental factors, as follows:

[0213] Instruction execution time: 1.5 seconds;

[0214] Speed adjustment error: ±10 RPM;

[0215] Device response time: 0.5 seconds.

[0216] Based on the feedback-based execution data, the prediction accuracy of the scheduling model is evaluated. For example, if the speed adjustment predicted by the scheduling model does not achieve the expected effect, the system will mark this data point and include it in the training data for subsequent model optimization. If the feedback data indicates that the production conditions have changed (such as large fluctuations in ambient temperature), the scheduling model can be fine-tuned or retrained to adapt to the new production environment. Through the feedback loop mechanism, a continuous optimization cycle is formed between the scheduling instructions and the production state. The model will be continuously updated based on real-time data and feedback to ensure that future scheduling decisions are more accurate and efficient.

[0217] For example:

[0218] (1) Device execution status:

[0219] Device 1 successfully adjusts the speed, with a feedback speed of 1800 RPM and no error;

[0220] Device 2 successfully adjusts the temperature, with a feedback temperature of 150°C and an error range of ±1°C;

[0221] The entire scheduling process takes 2 seconds.

[0222] (2) Energy consumption and efficiency analysis:

[0223] The total energy consumption during device adjustment is 5kWh, which is higher than the historical data (6kWh);

[0224] The execution time of the scheduling instruction is shortened by 10%, indicating that the model optimization has a significant effect.

[0225] (3) Model optimization results:

[0226] The model is retrained with new feedback data, taking into account changes in device response time, adjusting some parameters, and the optimized model improves the prediction accuracy by 5% in subsequent scheduling.

[0227] As shown in Figure 2 , according to another embodiment of the present application, a big data-based smart factory dynamic collaborative scheduling system is also provided, which comprises a data collection and preprocessing module 1, a model construction and training module 2, a scheduling instruction generation and transmission module 3, and an execution monitoring and model optimization module 4;

[0228] The data collection and preprocessing module 1, the model construction and training module 2, the scheduling instruction generation and transmission module 3, and the execution monitoring and model optimization module 4 are sequentially connected;

[0229] The data collection and preprocessing module 1 is used to obtain the state data corresponding to the production line according to the running status of each production line in the smart factory, and to establish a multi-dimensional data tensor using the state data;

[0230] The model construction and training module 2 is configured to construct a production line collaborative scheduling model based on a multi-dimensional data tensor and train the production line collaborative scheduling model by using a gradient boosting machine algorithm, and obtain a final production line collaborative scheduling model according to a training result;

[0231] The scheduling instruction generation and transmission module 3 is configured to input the collected real-time state data into the production line collaborative scheduling model, output scheduling instructions of each production line, and transmit the scheduling instructions to a smart factory control terminal by using a wireless communication network technology.

[0232] The execution monitoring and model optimization module 4 is configured to receive and execute the scheduling instructions by the smart factory control terminal, monitor an execution result of the scheduling instructions by using an Internet of Things and monitoring technology, and optimize the production line collaborative scheduling model based on the execution result.

[0233] In summary, by means of the above technical solutions of the present application, by collecting and analyzing the state data of each production line and constructing a multi-dimensional data tensor based on the state data, the uncertain factors in the production environment can be accurately captured, and the corresponding scheduling instructions can be quickly generated, which not only improves the ability to cope with complex and variable production conditions, but also reduces the time for manual information collection and decision-making, reduces the influence of personal experience and subjective judgment on decision-making, and ensures the consistency and accuracy of decision-making. By constructing and training the production line collaborative scheduling model, more reasonable and efficient production plans can be generated, thereby improving production efficiency, optimizing resource allocation and reducing waste. The production line collaborative scheduling model uses a multi-dimensional data tensor and a machine learning algorithm to comprehensively consider factors such as equipment state, material supply, personnel allocation and market demand, realizes accurate prediction and decision-making, and thus enables the factory to better adapt to complex and variable production environments and market demands. By transmitting the scheduling instructions by using a wireless communication network technology, remote monitoring and management of the production line are realized, which greatly improves the response speed and operation flexibility, thereby meeting the needs of the manufacturing industry, enabling management personnel to obtain real-time state information of the production line, and enabling them to take prompt measures when problems are found, maintain efficient production coordination and control, and further improve the competitiveness and operating efficiency of the enterprise.

[0234] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for dynamic collaborative scheduling of a smart factory based on big data, characterized in that, The method comprises the following steps: S1, according to the running state of each production line in the smart factory, the state data corresponding to the production line is obtained, and a multi-dimensional data tensor is established by using the state data; S2, a production line collaborative scheduling model is constructed based on the multi-dimensional data tensor, and the production line collaborative scheduling model is trained by using a gradient boosting machine algorithm, and a final production line collaborative scheduling model is obtained according to the training result; S3, input the collected real-time state data into the production line collaborative scheduling model, output the scheduling instruction of each production line, and transmit the scheduling instruction to the smart factory control end by using wireless communication network technology; S4, the smart factory control end receives and executes the scheduling instruction, monitors the execution result of the scheduling instruction by using the Internet of Things and monitoring technology, and optimizes the production line collaborative scheduling model based on the execution result; The production line collaborative scheduling model is constructed based on the multi-dimensional data tensor, and the production line collaborative scheduling model is trained by using a gradient boosting machine algorithm, and a final production line collaborative scheduling model is obtained according to the training result, comprising the following steps: S21, the characteristic data of each process of the production line is extracted by using a principal component analysis algorithm to process the multi-dimensional data tensor, and the characteristic data set is obtained by integrating the characteristic data; S22, based on the characteristic data set, a production line collaborative scheduling model is designed and constructed, wherein the characteristic data set is used as input, and the scheduling instruction of each production line is used as output; S23, the production line collaborative scheduling model is trained by using a gradient boosting machine algorithm to obtain a final production line collaborative scheduling model; The production line collaborative scheduling model is trained by using a gradient boosting machine algorithm to obtain a final production line collaborative scheduling model, comprising the following steps: S231, the characteristic data set is divided into a training set and a test set, and the parameters of the production line collaborative scheduling model are initialized; S232, each training data in the training set is traversed, the gradient of the loss function is calculated by using the parameters of the current production line collaborative scheduling model, and the gradient calculation is optimized by combining a binary ant colony algorithm; S233, a new training set is established by combining the gradient and the characteristic data in the characteristic data set, the current weak decision tree is trained by using the new training set, and the optimal scaling factor is determined; S234, the parameters of the production line collaborative scheduling model are updated according to the output result of the weak decision tree and the optimal scaling factor, and the updated parameters are used as the starting parameters for the next iteration; S235, repeat steps S232 to S234 until the preset number of iterations is reached; S236, the weak decision trees trained in all iterations are integrated according to the corresponding scaling factors to obtain the production line collaborative scheduling model after training; S237, the production line collaborative scheduling model after training is evaluated by using the test set to obtain a final production line collaborative scheduling model. 2.The big data based dynamic collaborative scheduling method of a smart factory according to claim 1, wherein, According to the running state of each production line in the smart factory, the state data corresponding to the production line is obtained, and a multi-dimensional data tensor is established by using the state data, comprising the following steps: S11, according to the running state of each production line in the smart factory, the dimension of the state data is determined; S12, collect state data from each production line by using a pre-set data interface, and pre-process the state data; S13, based on the pre-processed state data, integrate the state data of each process of each production line to obtain preliminary running state information of the corresponding production line; S14, analyze the preliminary running state information, identify state indicators that can reflect the running state of the production line, and combine the identified state indicators according to the determined dimensions to establish a multi-dimensional data matrix of the production line; S15, repeat steps S13 to S14 to obtain the corresponding multi-dimensional data matrix of each production line; S16, summarize the multi-dimensional data matrix corresponding to each production line to establish a multi-dimensional data tensor. 3.The big data based dynamic collaborative scheduling method of a smart factory according to claim 1, wherein, The step of using the principal component analysis algorithm to process the multi-dimensional data tensor to extract feature data of each process of the production line, and integrating the feature data to obtain a feature data set comprises the following steps: S211, classify the multi-dimensional data tensor according to the attributes and structural characteristics of each process of the production line, and construct a corresponding data structure model; S212, analyze the demand for collaborative scheduling of the production line, select a plurality of state data associated with the scheduling demand in the data structure model, and establish a candidate feature data set; S213, use the principal component analysis algorithm to process the candidate feature data set to extract feature data of each process of the production line; S214, according to the data flow and scheduling rules of each process of the production line, integrate the extracted feature data to obtain a feature data set.

4. The dynamic collaborative scheduling method of a smart factory based on big data according to claim 3, characterized in that, The data flow formula of each process of the production line is: In the formula, n t ) represents t data flow of each process of the production line at the time point​ t representing the time of extraction of the characteristic data; N denotes the total number of feature data; i represents the first i characteristic data; M representing a candidate feature dataset; The step of designing and constructing a production line collaborative scheduling model based on the feature data set, wherein the production line collaborative scheduling model takes the feature data set as input and the scheduling instructions of each production line as output, comprises the following steps: represents the amplitude of the oscillations present in the data stream; p representing a property of the feature data; c represents non-production data in each process of the production line.

5. The method of claim 1, wherein, S221, based on the feature data set, design the number of nodes of the input layer to match the dimensions of the feature data set; S222, design the structure of the hidden layer to process the input feature data set and further extract the input features, wherein the structure of the hidden layer includes the number of layers, the number of neurons in each layer, and the activation function; S223, according to the demand for collaborative scheduling of the production line, design the number of nodes of the output layer to correspond to and generate the number of scheduling instructions required by each production line; S224, use a deep learning framework to connect the designed input layer, hidden layer and output layer to construct a production line collaborative scheduling model. The step of traversing each training data in the training set, calculating the gradient of the loss function using the parameters of the current production line collaborative scheduling model, and optimizing the gradient calculation by combining the binary ant colony algorithm comprises the following steps:

6. The dynamic collaborative scheduling method of a smart factory based on big data according to claim 1, characterized in that, S2321, traverse each training data in the training set, and calculate the gradient value of the loss function to the current parameters after the previous iteration according to each training data through the parameters of the current production line collaborative scheduling model; S2322, based on the calculated gradient value, define a discrete search space for the binary ant colony algorithm, and randomly generate a parameter solution as an initial population in the discrete search space using a binary coding rule; S2323, evaluate the fitness of each parameter solution in the initial population using the loss function, and select the parameter solution according to the fitness value function; ​ S2324, combine the selected parameter solution with the roulette selection algorithm for cross processing to obtain a new parameter solution, and perform reverse mutation according to the fitness function to further explore the discrete search space; S2325, assign an initial pheromone value to each parameter solution path, and place several ants in the discrete search space to select the next moving direction according to the current pheromone distribution and gradient value; S2326, update the pheromone distribution of the discrete search space according to the moving path of the ant and the fitness of the parameter solution; S2327, repeat steps S2323 to S2326 until a preset iteration number is reached to obtain the optimal parameter solution.

7. The dynamic collaborative scheduling method of a smart factory based on big data according to claim 6, characterized in that, The update formula of the pheromone distribution is: wherein τ ij representing a path i to a path j pheromone concentration on the path; t denotes the time instant; n represents a natural number; ρ represents the pheromone volatility coefficient; represents the amount of information on the path i to the path j in the current cycle. 8.The big data based smart factory dynamic collaborative scheduling method of claim 1, wherein, The steps of inputting the collected real-time state data into the production line collaborative scheduling model, outputting the scheduling instructions of each production line, and transmitting the scheduling instructions to the intelligent factory control end by using wireless communication network technology include the following steps: S31, collect real-time state data from each production line through a pre-set data interface, and pre-process the real-time state data; S32, extract feature data from the pre-processed real-time state data, establish a real-time feature data set, and input the real-time feature data set into the production line collaborative scheduling model; S33, the production line collaborative scheduling model outputs the scheduling instructions of each production line according to the input feature data set; S34, use wireless communication network technology to transmit the generated scheduling instructions to the control end of the intelligent factory. 9.A big data based smart factory dynamic collaborative scheduling system for implementing the big data based smart factory dynamic collaborative scheduling method of any one of claims 1-8. The system includes a data collection and preprocessing module, a model construction and training module, a scheduling instruction generation and transmission module, and an execution monitoring and model optimization module; The data collection and preprocessing module, the model construction and training module, the scheduling instruction generation and transmission module, and the execution monitoring and model optimization module are sequentially connected; The data collection and preprocessing module is used to obtain the state data corresponding to the production line according to the running state of each production line in the intelligent factory, and to establish a multi-dimensional data tensor using the state data; The model construction and training module is used to construct a production line collaborative scheduling model based on the multi-dimensional data tensor, and to train the production line collaborative scheduling model using a gradient boosting machine algorithm to obtain the final production line collaborative scheduling model according to the training result; The scheduling instruction generation and transmission module is used to input the collected real-time state data into the production line collaborative scheduling model, output the scheduling instructions of each production line, and transmit the scheduling instructions to the control end of the intelligent factory by using wireless communication network technology; The execution monitoring and model optimization module is used to receive and execute the scheduling instructions at the control end of the intelligent factory, monitor the execution results of the scheduling instructions using Internet of Things and monitoring technology, and optimize the production line collaborative scheduling model based on the execution results.

Images