Concrete mixing plant scheduling strategy generation method and system combined with deep learning

Abstract

The application provides a concrete mixing station scheduling strategy generation method and system combined with deep learning, and relates to the technical field of deep learning. First, concrete production task data set and mixing station equipment operation state data stream are acquired. Then, a concrete scheduling deep decision network is used to mine the implicit dependency relationship between the two, and an implicit dependency relationship strength matrix is generated. Based on the implicit dependency relationship strength matrix and a preset scheduling constraint rule library, an initial scheduling strategy set is generated. A conflict perception module is used to detect conflicts and dynamically adjust the initial scheduling strategy set, so as to obtain an optimized scheduling strategy set. Finally, equipment control instructions and material feeding timing instructions are generated according to the optimized scheduling strategy set and are executed. The application can realize automatic and intelligent scheduling of the concrete mixing station production process, improve production efficiency and quality, and reduce costs and risks.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, and more specifically, to a method and system for generating scheduling strategies for concrete mixing plants that incorporates deep learning. Background Technology

[0002] In the production and operation of concrete batching plants, a scientific and reasonable scheduling strategy is crucial for improving production efficiency, ensuring concrete quality, and reducing production costs. Traditional scheduling methods for concrete batching plants mainly rely on manual experience and simple rule settings. While manual scheduling can handle some routine production tasks based on experience to a certain extent, it is prone to decision-making errors and fails to achieve globally optimal scheduling arrangements when faced with complex and ever-changing production scenarios, such as the simultaneous issuance of multiple concrete production tasks with different specifications and delivery times, and the real-time changes in the operating status of batching plant equipment.

[0003] Meanwhile, while existing rule-based scheduling systems can allocate production tasks and equipment according to preset rules, these rules are often fixed and lack the flexibility to adapt to various dynamic factors in actual production. For example, they cannot dynamically adjust the task execution sequence and equipment allocation scheme based on the real-time operating parameters of the mixing host and material conveying device, making it difficult to effectively handle scheduling conflicts caused by sudden equipment failures or urgent order changes. This often leads to problems such as low production efficiency and uneven equipment utilization. Summary of the Invention

[0004] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for generating a scheduling strategy for a concrete mixing plant using deep learning, the method comprising: Acquire a concrete production task dataset and a batching plant equipment operation status data stream. The concrete production task dataset contains multiple concrete production task orders, and the batching plant equipment operation status data stream contains a real-time collected sequence of operating parameters of the mixing host and a sequence of working status signals of the material conveying device. The concrete scheduling deep decision network processes the concrete production task dataset and the batching plant equipment operation status data stream to mine implicit dependencies, generating an implicit dependency strength matrix between concrete production task orders and batching plant equipment. The dimension of the implicit dependency strength matrix corresponds to the number of concrete production task orders and the number of batching plant equipment. An initial scheduling strategy is generated based on the implicit dependency strength matrix and the preset scheduling constraint rule base. This generates a set of initial scheduling strategies for concrete production that includes the mapping relationship between task execution order and equipment allocation. The preset scheduling constraint rule base includes a threshold for continuous working time of equipment and a threshold for order delivery time window. The conflict perception module of the concrete scheduling deep decision network processes the initial scheduling strategy set of concrete production to detect and dynamically adjust strategy conflicts, thereby obtaining the optimized scheduling strategy set of concrete production after conflict resolution. The conflict perception module is trained based on historical scheduling conflict case data. Based on the set of optimized scheduling strategies for concrete production, a sequence of control instructions for concrete batching plant equipment and a sequence of material delivery timing instructions are generated. The sequence of control instructions for concrete batching plant equipment and the sequence of material delivery timing instructions are then sent to the corresponding batching plant control unit for execution.

[0005] Furthermore, embodiments of the present invention also provide a concrete mixing plant scheduling strategy generation system incorporating deep learning, comprising: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described method for generating a deep learning-integrated concrete mixing plant scheduling strategy by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, and the processor executing the machine-executable instructions, causing the computer device to execute the above-described method for generating a concrete mixing plant scheduling strategy incorporating deep learning.

[0007] Based on the above, by acquiring concrete production task datasets and batching plant equipment operation status data streams, a deep decision network for concrete scheduling is used to mine the implicit dependencies between concrete production task orders and batching plant equipment, generating an implicit dependency strength matrix. Based on the implicit dependency strength matrix and a preset scheduling constraint rule base, an initial scheduling strategy is generated. This strategy considers both the inherent connection between production tasks and equipment and follows the actual constraints of equipment operation and order delivery. Through the conflict perception module of the deep decision network for concrete scheduling, conflict detection and dynamic adjustment of the initial scheduling strategy can effectively solve the scheduling conflict problem caused by the dynamic changes in production tasks and equipment status, resulting in an optimized scheduling strategy set after conflict resolution. This improves the robustness and adaptability of the scheduling strategy. Finally, based on the optimized scheduling strategy set, equipment control command sequences and material delivery timing command sequences are generated and executed, realizing automated and intelligent scheduling of the concrete batching plant production process. This significantly improves production efficiency, equipment utilization, and concrete quality, while reducing production costs and operational risks. Attached Figure Description

[0008] Figure 1This is a schematic diagram of the execution flow of the concrete mixing plant scheduling strategy generation method combined with deep learning provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of the concrete mixing plant scheduling strategy generation system combining deep learning provided in an embodiment of the present invention. Detailed Implementation

[0010] Figure 1 This is a flowchart illustrating a method for generating scheduling strategies for concrete mixing plants that incorporates deep learning, provided in one embodiment of the present invention. A detailed description follows.

[0011] Step S110: Obtain a concrete production task dataset and a batching plant equipment operation status data stream. The concrete production task dataset contains multiple concrete production task orders, and the batching plant equipment operation status data stream contains a real-time collected sequence of operating parameters of the mixing host and a sequence of working status signals of the material conveying device.

[0012] In this embodiment, the concrete production task dataset comes from the batching plant order management system. This dataset contains multiple concrete production task orders, each with a unique identifier. For example, the identifier for a concrete production task order is ORD001, and this order includes information such as concrete grade attributes, volume attributes, pouring location attributes, and transportation distance attributes. The batching plant equipment operation status data stream is collected in real time by sensors installed at various locations in the batching plant. Among them, the mixing host operating parameter sequence records parameters such as the instantaneous value of the motor's three-phase current at a fixed sampling frequency (e.g., 10 times per second), and the material conveying device operating status signal sequence includes belt speed feedback signals and motor torque current signals.

[0013] During the data collection process, for data involving privacy sensitivity, such as order customer information, data anonymization technology is used to replace sensitive information such as the customer's specific name and contact information with anonymous identifiers to ensure data privacy and security.

[0014] Step S120: The concrete production task dataset and the batching plant equipment operation status data stream are processed by the concrete scheduling deep decision network to mine implicit dependencies and generate an implicit dependency strength matrix between concrete production task orders and batching plant equipment. The dimension of the implicit dependency strength matrix corresponds to the number of concrete production task orders and the number of batching plant equipment.

[0015] In this embodiment, the deep decision network for concrete scheduling aims to uncover the implicit dependencies between concrete production task orders and batching plant equipment. It receives a concrete production task dataset and a batching plant equipment operating status data stream as input. After feature processing and computation, it outputs an implicit dependency strength matrix. The number of rows in this matrix equals the number of concrete production task orders, and the number of columns equals the number of batching plant equipment. Each element in the matrix represents the dependency strength between the corresponding order and the equipment.

[0016] Step S121: The input feature transformation layer of the concrete scheduling deep decision network processes each concrete production task order in the concrete production task dataset for feature encoding, and converts the concrete grade attribute, volume attribute, pouring location attribute and transportation distance attribute contained in each concrete production task order into a task feature vector of a unified dimension.

[0017] In this embodiment, the input feature transformation layer encodes features for each concrete production task order. Taking order ORD001 as an example, its concrete grade attribute is C30, its volume attribute is 50 cubic meters, its pouring location attribute is "basement floor slab", and its transportation distance attribute is 10 kilometers. The input feature transformation layer first processes the above attributes. For the concrete grade attribute, C30 is converted into a numerical vector according to a preset mapping relationship; the volume attribute is directly used as a numerical feature; the pouring location attribute is converted into a vector through one-hot encoding; and the transportation distance attribute is also used as a numerical feature. Then, the above-processed attribute features are combined and nonlinearly transformed through a fully connected layer to finally generate a task feature vector with a fixed dimension (e.g., 128 dimensions).

[0018] Step S122: The equipment state coding layer of the concrete scheduling depth decision network processes the data stream of the batching plant equipment operation status to extract time-series features, extracts the motor current fluctuation pattern features of the mixing host and the belt speed stability features of the material conveying device, and combines them to generate an equipment state feature vector.

[0019] In this embodiment, the equipment status encoding layer is responsible for extracting temporal features from the data stream of the mixing plant's equipment operating status. Taking the operating parameter sequence of a certain mixing host (equipment identifier M001) as an example, this operating parameter sequence includes the instantaneous values ​​of the motor's three-phase current. The equipment status encoding layer first processes the above current values ​​to extract the motor current fluctuation pattern features; simultaneously, for the operating status signal sequence of the material conveying device (equipment identifier C001), it extracts the belt speed stability features. Then, these two types of features are combined and further processed to generate an equipment status feature vector.

[0020] Step S1221: Separate the operating parameter sequence of the mixing host from the operating status data stream of the mixing plant equipment. The operating parameter sequence of the mixing host includes the instantaneous values ​​of the three-phase current of the motor recorded at a fixed sampling frequency.

[0021] In this embodiment, the data stream of the mixing plant equipment operation status is a mixed data stream containing operation data of various devices. It is necessary to extract the operating parameter sequence of the mixing host, which is used to record the operating parameters of the mixing host. Taking the mixing host M001 as an example, its operating parameter sequence records the instantaneous current values ​​of phases A, B, and C of the motor at a sampling frequency of 10 times per second. These data are arranged in chronological order to form a three-dimensional sequence data, where each time point corresponds to three instantaneous current values.

[0022] Step S1222: Perform coordinate transformation on the instantaneous values ​​of the three-phase current of the motor to convert the three-phase current values ​​into amplitude and phase angle representations.

[0023] In this embodiment, the instantaneous values ​​of the three-phase current of the separated stirring host M001 motor need to be transformed using coordinates. A combination of Clarke transform and Park transform is used to convert the three-phase current values ​​from a three-phase coordinate system to a two-phase rotating coordinate system, thereby obtaining the current amplitude and phase angle. For example, after coordinate transformation, the current amplitude at a certain moment is I_m, and the phase angle is θ. This representation is more conducive to subsequent feature extraction.

[0024] Step S1223: Calculate the rate of change sequence of motor current amplitude between consecutive sampling points, perform sliding window statistics on the rate of change sequence, and calculate the variance of the rate of change within each window as an index of current fluctuation intensity.

[0025] In this embodiment, after obtaining the current amplitude sequence, the rate of change of amplitude between consecutive sampling points is calculated. For example, for the current amplitude sequence I_m1, I_m2, ..., I_mn, the rate of change sequence is (I_m2-I_m1) / Δt, (I_m3-I_m2) / Δt, ..., (I_mn-I_m(n-1)) / Δt, where Δt is the sampling time interval. Then, a sliding window is set with a window size of W (e.g., 100 sampling points), and the rate of change sequence is statistically analyzed using the sliding window. The variance of the rate of change within each window is calculated, and this variance value is the current fluctuation intensity index, used to measure the fluctuation of the current amplitude.

[0026] Step S1224: Perform spectrum analysis on the motor current amplitude, extract the power value of the main frequency component and its trend of change over time, as the characteristic of load periodic fluctuation.

[0027] In this embodiment, a spectral analysis is performed on the motor current amplitude sequence. A Fast Fourier Transform (FFT) is used to convert the time-domain current amplitude sequence to the frequency domain, obtaining the power spectrum of different frequency components. Then, the dominant frequency component with the highest energy in the power spectrum is identified, and its power value is recorded. Simultaneously, a sliding window method is used to observe the change in the power value of the dominant frequency component over time, and this trend is used as a characteristic of the load's periodic fluctuations. For example, if the power value of the dominant frequency component exhibits a periodic increase and decrease over a period of time, this trend reflects the periodic changes in the load on the mixing host.

[0028] Step S1225: Separate the working status signal sequence of the material conveying device from the data stream of the operating status of the mixing plant equipment. The working status signal sequence of the material conveying device includes the belt speed feedback signal and the motor torque current signal.

[0029] In this embodiment, similar to separating the operating parameter sequence of the mixing host, the working status signal sequence of the material conveying device is separated from the operating status data stream of the mixing plant equipment. Taking the material conveying device C001 as an example, the working status signal sequence of the material conveying device includes the belt speed feedback signal (in revolutions per minute) and the motor torque current signal (in amperes). These signals are also recorded at a fixed sampling frequency (e.g., 5 times per second) to form a two-dimensional sequence data.

[0030] Step S1226: Calculate the mean and standard deviation of the belt speed feedback signal within a set time period, and use the ratio of the standard deviation to the mean as the belt speed stability coefficient.

[0031] In this embodiment, for the belt speed feedback signal of the material conveying device C001, a time period (e.g., 5 minutes) is set, and the mean μ and standard deviation σ of the speed signal within this time period are calculated. Then, the ratio of the standard deviation σ to the mean μ (σ / μ) is used as the belt speed stability coefficient. The smaller the belt speed stability coefficient, the more stable the belt speed; conversely, the larger the value, the greater the speed fluctuation.

[0032] Step S1227: Analyze the phase lag relationship between the motor torque current signal and the belt speed feedback signal, and calculate the lag time constant as a characteristic of the transmission system response delay.

[0033] In this embodiment, the phase lag relationship between the motor torque current signal and the belt speed feedback signal of the material conveying device C001 is analyzed. Through cross-correlation analysis, the time delay between the two signals is found; that is, how long it takes for the belt speed feedback signal to change after the motor torque current signal changes. This time delay value is used as the lag time constant, which reflects the response delay characteristics of the transmission system.

[0034] Step S1228: Perform feature scaling on the current fluctuation intensity index, the load periodic fluctuation characteristics, the belt speed stability coefficient, and the transmission system response delay characteristics so that all feature values ​​are within the same order of magnitude range.

[0035] In this embodiment, to ensure the accuracy and effectiveness of subsequent feature processing, feature scaling is required for the extracted current fluctuation intensity index, load periodic fluctuation characteristics, belt speed stability coefficient, and transmission system response delay characteristics. A standardization method is used: each feature is subtracted from its mean and then divided by its standard deviation, resulting in a mean of 0 and a standard deviation of 1 for each feature. This adjusts all feature values ​​to the same order of magnitude. For example, the original value range of the current fluctuation intensity index might be between 0 and 100; after standardization, its value range will be adjusted to between -3 and 3.

[0036] Step S1229: Assemble the scaled current fluctuation intensity index, the load periodic fluctuation characteristics, the belt speed stability coefficient, and the transmission system response delay characteristics in a preset order to generate a one-dimensional array.

[0037] In this embodiment, the preset order is current fluctuation intensity index, load periodic fluctuation characteristics, belt speed stability coefficient, and transmission system response delay characteristics. These four scaled features are concatenated in this order to form a one-dimensional array. For example, if the scaled value of the current fluctuation intensity index is a, the scaled value of the load periodic fluctuation characteristics is b, the scaled value of the belt speed stability coefficient is c, and the scaled value of the transmission system response delay characteristics is d, then the concatenated one-dimensional array is [a, b, c, d].

[0038] Step S12210: Input the one-dimensional array into a fully connected neural network layer for feature compression and nonlinear transformation, output a fixed-length device state feature vector, and perform standardization processing on the device state feature vector.

[0039] In this embodiment, the concatenated one-dimensional array is input into a fully connected neural network layer. This fully connected neural network layer includes an input layer, hidden layers, and an output layer. The hidden layers use the ReLU activation function for non-linear transformation. Through the processing of this network layer, the one-dimensional array is compressed into a fixed-length (e.g., 64-dimensional) device state feature vector. Then, the output device state feature vector is standardized so that the mean of each dimension is 0 and the standard deviation is 1, to facilitate processing by subsequent network layers.

[0040] Step S123: Input the task feature vector and the equipment state feature vector into the cross-attention calculation layer of the concrete scheduling deep decision network, and calculate the bidirectional attention weight value between the task feature vector corresponding to each concrete production task order and the equipment state feature vector corresponding to each batching plant equipment.

[0041] In this embodiment, the cross-attention calculation layer receives task feature vectors and equipment state feature vectors as input. For each concrete production task order's task feature vector, it performs interactive calculations with the equipment state feature vector of each batching plant device. By calculating the similarity between the two, a bidirectional attention weight value is obtained. For example, for the task feature vector of order ORD001 and the equipment state feature vector of device M001, an attention weight value is calculated between them, which represents the degree of association between order ORD001 and device M001.

[0042] Step S124: Construct an initial association weight graph between tasks and equipment based on the bidirectional attention weight value. The nodes of the initial association weight graph include concrete production task order nodes and mixing plant equipment nodes. The edge weights between nodes are assigned by the bidirectional attention weight value.

[0043] In this embodiment, concrete production task orders and batching plant equipment are used as nodes to construct an initial association weight graph. The weight of the edge between each order node and equipment node is the bidirectional attention weight value calculated in step S123. For example, the edge weight between order node ORD001 and equipment node M001 is 0.8, and the edge weight between order node ORD001 and equipment node M002 is 0.3, etc., thus forming a graph structure in which the association strength is represented by weight.

[0044] Step S125: The initial association weight graph is processed by the multi-layer graph convolution module of the concrete scheduling depth decision network to propagate higher-order relationships, and to capture the indirect dependencies between concrete production task order nodes through shared equipment and the collaborative relationships between batching plant equipment nodes through common tasks.

[0045] In this embodiment, the multi-layer graph convolution module processes the initial association weight graph to achieve the propagation of higher-order relationships. Through multi-layer graph convolution operations, this module enables information transfer and fusion between order nodes and device nodes, thereby capturing the indirect dependencies between orders generated by shared devices and the collaborative relationships between devices generated by shared tasks. For example, if both order ORD001 and order ORD002 require the use of device M001, the indirect dependency between ORD001 and ORD002 can be captured through the processing of the multi-layer graph convolution module.

[0046] Step S1251: The multi-layer graph convolution module loads the initial association weight graph generated by the cross-attention calculation layer. In the initial association weight graph, each concrete production task order node is associated with a task feature vector containing transportation distance attributes, and each batching plant equipment node is associated with an equipment state feature vector containing motor current fluctuation pattern features.

[0047] In this embodiment, the multi-layer graph convolution module first loads the initial association weight graph. In this graph, each concrete production task order node (such as ORD001) is associated with a task feature vector, which contains attribute information such as the transportation distance of the order; each batching plant equipment node (such as M001) is associated with an equipment status feature vector, which contains feature information such as the motor current fluctuation pattern of the equipment.

[0048] Step S1252: In the first propagation layer of the multi-layer graph convolution module, according to the connection edges of the initial association weight graph, the equipment state feature vector of each batching plant equipment node is sent to the concrete production task order node directly connected to it, and each concrete production task order node receives the feature message from its candidate batching plant equipment node.

[0049] In this embodiment, the first propagation layer handles message passing from device nodes to order nodes. Based on the connection edges in the initial association weight graph, for example, if device node M001 is connected to order nodes ORD001 and ORD003, then the device state feature vector of device node M001 will be passed to order nodes ORD001 and ORD003 respectively. Each order node will receive feature messages from all the device nodes directly connected to it.

[0050] Step S1253: For each concrete production task order node, the first propagation layer aggregates all incoming equipment status feature vector messages, calculates a weighted sum based on the connection edge weights, and generates the first layer equipment influence features for each concrete production task order node.

[0051] In this embodiment, for each order node, such as the ORD001 order node, it receives feature messages from device nodes M001 and M002. The first propagation layer performs a weighted summation of these feature messages based on the weights of the connecting edges. For example, the edge weight between device node M001 and order node ORD001 is 0.8, and its device state feature vector is V1; the edge weight between device node M002 and order node ORD001 is 0.3, and its device state feature vector is V2. Therefore, the first-layer device influence feature of order node ORD001 is 0.8*V1 + 0.3*V2.

[0052] Step S1254: The first propagation layer fuses the first-layer equipment influence features of each concrete production task order node with its own task feature vector containing the transportation distance attribute, and generates the first-layer updated feature vector of each concrete production task order node through a fully connected layer.

[0053] In this embodiment, the first-layer device influence feature of the order node is fused with its own task feature vector. For example, the first-layer device influence feature of the ORD001 order node is F1, and its own task feature vector is T1. F1 and T1 are concatenated to obtain a new vector. This new vector is then input into a fully connected layer, and after nonlinear transformation, the updated first-layer feature vector U1 is generated. This updated first-layer feature vector integrates the device influence and the order's own attribute information.

[0054] Step S1255: In the second propagation layer of the multi-layer graph convolution module, the updated feature vector of each concrete production task order node is sent to other concrete production task order nodes that are indirectly connected by sharing the same candidate mixing plant equipment for secondary message transmission. The transmission path is through the shared mixing plant equipment node.

[0055] In this embodiment, the second propagation layer enables indirect message passing between order nodes. For example, both order nodes ORD001 and ORD002 are connected to device node M001. The updated feature vector U1 of order node ORD001 will first be passed to device node M001, and then from device node M001 to order node ORD002. Similarly, the updated feature vector U2 of order node ORD002 will also be passed to order node ORD001 via device node M001. In this way, indirect message passing is achieved between order nodes through shared device nodes.

[0056] Step S1256: For each concrete production task order node, the second propagation layer aggregates the updated feature vector messages from other indirectly connected concrete production task order nodes, generating the second layer inter-order dependency features for each concrete production task order node.

[0057] In this embodiment, the ORD001 order node receives U2 transmitted from the ORD002 order node via the M001 device node, as well as feature messages that may be transmitted from other indirectly connected order nodes. The second propagation layer aggregates the aforementioned incoming feature messages, for example, by using a simple average or weighted average method, to generate a second-layer inter-order dependency feature D1, which reflects the indirect dependencies between orders.

[0058] Step S1257: The second propagation layer simultaneously transmits the updated first-layer feature vector of each concrete production task order node to the batching plant equipment node connected to it in reverse message transmission. Each batching plant equipment node receives feature messages from all concrete production task order nodes it serves.

[0059] In this embodiment, the second propagation layer also performs reverse message passing from order nodes to device nodes. For example, order node ORD001 passes its updated first-layer feature vector U1 to device node M001, and order node ORD003 also passes its updated first-layer feature vector U3 to device node M001. In this way, device node M001 will receive feature messages passed from all the order nodes connected to it.

[0060] Step S1258: The second propagation layer aggregates all incoming first-layer updated feature vector messages from different concrete production task order nodes for each batching plant equipment node, and generates task collaborative load features for each batching plant equipment node.

[0061] In this embodiment, for device node M001, it receives U1 from order node ORD001 and U3 from order node ORD003. The second propagation layer aggregates the above feature messages, for example, by calculating their weighted sum (the weights can be determined according to the association strength between the order and the device), to generate task collaborative load feature L1, which reflects the collaborative load among multiple orders served by the device.

[0062] Step S1259: The second propagation layer concatenates the second-layer inter-order dependency features of each concrete production task order node with its updated first-layer feature vector, and generates the final output concrete production task order node embedding vector through another fully connected layer. This concrete production task order node embedding vector encodes the indirect dependencies generated through shared devices.

[0063] In this embodiment, the second-layer inter-order dependency feature D1 of the order node is concatenated with its first-layer updated feature vector U1 to obtain a vector that integrates indirect dependencies and previously updated features. This vector is then input into another fully connected layer, and after processing, the final order node embedding vector E1 is generated. This order node embedding vector contains the indirect dependency information generated by the order through the shared device.

[0064] Step S12510: The second propagation layer concatenates the task collaboration load features of each batching plant equipment node with its initial equipment state feature vector, and generates the final output batching plant equipment node embedding vector through another fully connected layer. The batching plant equipment node embedding vector encodes the collaborative relationship generated through the common task.

[0065] In this embodiment, the task collaboration load feature L1 of the device node is concatenated with its initial device state feature vector V to obtain a vector that integrates the collaboration load and the initial state features. This vector is then input into another fully connected layer, and after processing, the final device node embedding vector F1 is generated. This device node embedding vector contains the collaboration relationship information generated by the devices through shared tasks.

[0066] Step S126: Calculate the cosine similarity value between any concrete production task order node and any mixing plant equipment node based on the node embedding vector output by the multi-layer graph convolution module, as the refined association strength.

[0067] In this embodiment, the multi-layer graph convolution module outputs order node embedding vectors and device node embedding vectors. For any order node (e.g., ORD001) embedding vector E1 and any device node (e.g., M001) embedding vector F1, the cosine similarity between them is calculated. The cosine similarity is calculated by dividing the inner product of the two vectors by the product of their magnitudes. The result is the refined association strength. The greater the refined association strength, the stronger the association between the order and the device.

[0068] Step S127: Integrate the bidirectional attention weight value and the refined correlation strength value, calculate the final correlation strength value by weighted summation, and organize the final correlation strength value into a two-dimensional matrix structure according to the order of concrete production task orders and the order of batching plant equipment.

[0069] In this embodiment, the bidirectional attention weight value obtained in step S123 and the refined association strength value obtained in step S126 are fused. A weight parameter α (e.g., 0.5) is set, and the final association strength value is α * bidirectional attention weight value + (1-α) * refined association strength value. Then, according to the order of concrete production task orders (e.g., ORD001, ORD002, ORD003...) as the rows of the matrix and the order of batching plant equipment (e.g., M001, M002, M003...) as the columns of the matrix, the final association strength value is filled into the corresponding matrix positions to form a two-dimensional matrix structure.

[0070] Step S128: Perform row-direction normalization and column-direction normalization on the two-dimensional matrix structure so that the distribution of the associated strength value corresponding to each concrete production task order and the distribution of the associated strength value corresponding to each batching plant equipment both satisfy the probability distribution characteristics. Output the two-dimensional matrix structure that has completed the two-way normalization process as the implicit dependency strength matrix between concrete production task orders and batching plant equipment.

[0071] In this embodiment, the two-dimensional matrix is ​​first normalized in the row direction. That is, for each row (each order), all elements are divided by the sum of the elements in that row, ensuring the sum of the elements in each row is 1, satisfying the probability distribution characteristics. Then, column-direction normalization is performed. For each column (each device), all elements are divided by the sum of the elements in that column, again ensuring the sum of the elements in each column is 1. After these two-way normalization processes, the resulting two-dimensional matrix is ​​the implicit dependency strength matrix. Each element in the matrix represents the dependency strength between the order and the device, considering both row and column probability distributions.

[0072] Step S130: Generate an initial scheduling strategy based on the implicit dependency strength matrix and the preset scheduling constraint rule base, and generate a set of initial scheduling strategies for concrete production that includes the mapping relationship between task execution order and equipment allocation. The preset scheduling constraint rule base includes the equipment continuous working time threshold and the order delivery time window threshold.

[0073] In this embodiment, the preset scheduling constraint rule base is a database storing scheduling-related constraints, including equipment continuous working time thresholds (e.g., the continuous working time threshold for a certain mixing host is 4 hours) and order delivery time window thresholds (e.g., the delivery time window for order ORD001 is from 8:00 AM to 10:00 AM). Combining the implicit dependency strength matrix and the preset scheduling constraint rule base, an initial scheduling strategy is generated. This initial scheduling strategy includes the order of task execution and the mapping relationship of which device each task is assigned to.

[0074] Step S131: Extract the equipment association strength distribution vector corresponding to each concrete production task order from the implicit dependency strength matrix.

[0075] In this embodiment, each row of the implicit dependency strength matrix corresponds to a concrete production task order. Therefore, extracting the elements of each row yields the equipment association strength distribution vector corresponding to that order. For example, the row vector corresponding to order ORD001 is [0.6, 0.3, 0.1], indicating that the association strengths of this order with equipment M001, M002, and M003 are 0.6, 0.3, and 0.1, respectively.

[0076] Step S132: Based on the equipment association strength distribution vector, select a preset number of mixing plant equipment with the highest association strength for each concrete production task order to form a candidate equipment set.

[0077] In this embodiment, the preset quantity can be set according to the actual situation, such as 3 units. For the device association strength distribution vector [0.6, 0.3, 0.1] of order ORD001, the top 3 devices with the highest association strength (assuming there are 3 devices here), namely M001, M002, and M003, are selected to form the candidate device set for this order.

[0078] Step S133: Based on the order delivery time window threshold in the preset scheduling constraint rule base, calculate a time urgency coefficient for each concrete production task order. The time urgency coefficient is equal to the reciprocal of the current time to the deadline of the order delivery time window and the weighted combination of the corresponding order volume. Sort all concrete production task orders in descending order according to the time urgency coefficient to generate a priority queue of concrete production task orders.

[0079] In this embodiment, assuming the current time is 7:00 AM and the delivery window deadline for order ORD001 is 10:00 AM, the time difference between the current time and the deadline is 3 hours, and its reciprocal is 1 / 3. The order volume is 50 cubic meters. The weight of the reciprocal of time is set to 0.6, and the weight of the order volume is set to 0.4. Therefore, the time urgency coefficient is 0.6*(1 / 3) + 0.4*50 (Note that in actual calculations, the order volume needs to be normalized to ensure consistency with the dimensions of the reciprocal of time; for example, dividing the order volume by the largest order volume yields the normalized volume). After calculating the time urgency coefficients for all orders, the orders are sorted in descending order to generate a priority queue. Orders with higher priority are scheduled first.

[0080] Step S134: Process each concrete production task order in sequence according to the priority queue, and pre-allocate the mixing plant equipment that is currently idle and has the earliest estimated task completion time from the candidate equipment set of the currently processed concrete production task order.

[0081] In this embodiment, the priority queue order is ORD001, ORD003, ORD002... Order ORD001 is processed first, and the current status of each device in its candidate device set {M001, M002, M003} is checked. Assume that device M001 is currently idle, with an estimated completion time of 1 hour; device M002 is processing other tasks and is expected to become idle in 1.5 hours; device M003 is idle, with an estimated completion time of 1.2 hours. Therefore, device M001 is selected for pre-allocation because it is currently idle and has the earliest estimated task completion time.

[0082] Step S135: During the pre-allocation process, query the equipment continuous working time threshold in the preset scheduling constraint rule library. If the cumulative planned working time of the selected mixing plant equipment exceeds the equipment continuous working time threshold, mark the mixing plant equipment as temporarily unavailable and remove it from the candidate equipment set of the current concrete production task order.

[0083] In this embodiment, assuming the continuous working time threshold for device M001 is 4 hours, its current cumulative planned working time has reached 3.5 hours, while the estimated task execution time for order ORD001 is 1 hour, then the cumulative planned working time will reach 4.5 hours, exceeding the threshold. At this point, device M001 is marked as temporarily unavailable and removed from the candidate device set of order ORD001, and a suitable device is selected from the remaining candidate devices.

[0084] Step S136: Construct a sequence of allocation pairs based on the pre-allocation results. Each allocation pair contains a concrete production task order identifier and an assigned batching plant equipment identifier.

[0085] In this embodiment, after pre-allocation, order ORD001 is assigned to device M002, order ORD003 is assigned to device M001, order ORD002 is assigned to device M003, etc., thereby constructing an allocation pair sequence of [(ORD001, M002), (ORD003, M001), (ORD002, M003), ...].

[0086] Step S137: Based on the task receiving order and estimated task execution time of the batching plant equipment in the allocation sequence, simulate and generate timeline occupancy information of the batching plant equipment. The timeline occupancy information is presented in the form of equipment timeline segments.

[0087] In this embodiment, for device M001, the assigned task order is ORD003, ORD005, ... The estimated start time of order ORD003 is the current time (7:00 AM), and the estimated execution time is 1.5 hours, so its time period is 7:00-8:30; the estimated start time of order ORD005 is 8:30, and the estimated execution time is 2 hours, so its time period is 8:30-10:30. Arranging the above time periods in chronological order forms a timeline segment for device M001, such as [7:00-8:30, 8:30-10:30, ...].

[0088] Step S138: Calculate the estimated start time and estimated end time of each concrete production task order based on the equipment timeline segment, and generate a preliminary scheduling time plan.

[0089] In this embodiment, the estimated start and end times of corresponding orders can be determined based on the time period occupied by each task in the device timeline segment. For example, if the time period occupied by order ORD003 in the M001 device timeline segment is 7:00-8:30, then the estimated start time of order ORD003 is 7:00 and the estimated end time is 8:30. The estimated start and end times of all orders are collected, arranged in chronological order, and a preliminary scheduling time plan is generated.

[0090] Step S1381: Traverse the sequence of allocation pairs, and for each allocation pair in the sequence of allocation pairs, obtain the identifier of the mixing plant equipment to which it is allocated.

[0091] In this embodiment, the sequence of allocation pairs [(ORD001, M002), (ORD003, M001), (ORD002, M003), ...] is traversed. For the first allocation pair (ORD001, M002), the device identifier M002 is obtained; for the second allocation pair (ORD003, M001), the device identifier M001 is obtained, and so on.

[0092] Step S1382: Query the equipment timeline segment corresponding to the equipment identifier of the mixing plant. The equipment timeline segment records the set of time periods that the mixing plant equipment has been planned to be occupied. Find the end time point of the last occupied time period in the equipment timeline segment and take the end time point as the earliest start time point of the current concrete production task order on the mixing plant equipment.

[0093] In this embodiment, taking the allocation pair (ORD001, M002) as an example, we query the timeline segment of device M002. Assuming that the current segment is [6:00-7:00, 7:00-7:45], and the end time of the last occupied time period is 7:45, then the earliest start time of order ORD001 on device M002 is 7:45.

[0094] Step S1383: Based on the volume attribute of the current concrete production task order and the rated production rate of the mixing plant equipment, calculate the pure mixing operation time required to complete the concrete production task order.

[0095] In this embodiment, the volume attribute of order ORD001 is 50 cubic meters, and the rated production rate of equipment M002 is 20 cubic meters / hour. Therefore, the pure mixing operation time is 50 / 20=2.5 ​​hours.

[0096] Step S1384: Based on historical data statistics, obtain the average material preparation time and average unloading time corresponding to the concrete grade attribute of the concrete production task order. Add the pure mixing operation time, the average material preparation time and the average unloading time to obtain the total estimated task execution time of the current concrete production task order.

[0097] In this embodiment, the concrete grade attribute of order ORD001 is C30. According to historical data, the average material preparation time for C30 concrete is 0.5 hours, and the average unloading time is 0.3 hours. Therefore, the total estimated task execution time is 2.5 + 0.5 + 0.3 = 3.3 hours.

[0098] Step S1385: Add the earliest possible start time to the total estimated task execution time to obtain the estimated end time of the current concrete production task order. Check whether the estimated end time exceeds the delivery time window threshold of the concrete production task order. If it does, adjust the earliest possible start time forward by a time offset, where the time offset is equal to the length of the time exceeding the threshold.

[0099] In this embodiment, the earliest possible start time for order ORD001 is 7:45, the total estimated task execution time is 3.3 hours, and the estimated end time is 7:45 + 3.3 hours = 11:03. The delivery time window threshold for this order is from 8:00 AM to 10:00 AM. 11:03 AM exceeds the deadline of 10:00 AM by 1 hour and 3 minutes. Therefore, the earliest possible start time is adjusted forward by 1 hour and 3 minutes, and the adjusted start time is 7:45 - 1 hour and 3 minutes = 6:42.

[0100] Step S1386: Take the adjusted start time as the final estimated start time of the current concrete production task order, and recalculate the new estimated end time by adding the final estimated start time to the total estimated task execution time.

[0101] In this embodiment, the adjusted start time is 6:42, the total estimated task execution time is 3.3 hours, and the new estimated end time is 6:42 + 3.3 hours = 10:00. This time does not exceed the delivery time window threshold.

[0102] Step S1387: Record the time interval consisting of the final estimated start time and the new estimated end time as the scheduling time block of the concrete production task order, update the scheduling time block to the equipment timeline segment corresponding to the mixing plant equipment, and extend the length of the equipment timeline segment.

[0103] In this embodiment, the scheduling time block for order ORD001 is [6:42-10:00]. This time block is added to the timeline segment of device M002, and the updated timeline segment is [6:00-7:00, 7:00-7:45, 6:42-10:00] (Note that the time order needs to be adjusted here. In actual applications, the timeline segments will be sorted according to the order of time).

[0104] Step S1388: Repeat the above steps for the next allocation pair in the allocation pair sequence until all allocation pairs have calculated scheduling time blocks.

[0105] In this embodiment, each allocation pair in the allocation pair sequence is processed sequentially according to the above steps, a scheduling time block is calculated for each order, and the timeline segment of the corresponding device is updated.

[0106] Step S1389: Collect all scheduling time blocks of concrete production task orders, sort them according to the order of estimated start time, associate the sorted scheduling time block list with the corresponding allocation pair sequence, and generate a structured preliminary scheduling time plan.

[0107] In this embodiment, the scheduling time blocks of all orders are collected, such as [6:42-10:00] for ORD001, [7:00-8:30] for ORD003, [8:00-11:00] for ORD002, etc., and sorted according to the estimated start time points as ORD001 (6:42), ORD003 (7:00), ORD002 (8:00)... Then, the sorted scheduling time block list is associated with the allocation pair sequence to form a preliminary scheduling time plan. This preliminary scheduling time plan clarifies the start time, end time and the allocated equipment for each order.

[0108] Step S139: Combine all allocation pairs with their corresponding scheduling time plans to generate multiple independent scheduling strategy units.

[0109] In this embodiment, each allocation pair is combined with the corresponding scheduling time block to form a scheduling strategy unit. For example, the allocation pair (ORD001, M002) and the scheduling time block [6:42-10:00] are combined to form a scheduling strategy unit, which includes information such as order identifier, device identifier, start time and end time.

[0110] Step S1310: Group the scheduling strategy units according to the pouring location attribute of the concrete production task order, and assign the scheduling strategy units corresponding to the concrete production task orders of the same pouring location to the same scheduling strategy group.

[0111] In this embodiment, the pouring location attribute of order ORD001 is "basement floor slab", the pouring location attribute of order ORD003 is also "basement floor slab", and the pouring location attribute of order ORD002 is "main structural column". Therefore, the scheduling strategy units of ORD001 and ORD003 are assigned to the "basement floor slab" scheduling strategy group, and the scheduling strategy unit of ORD002 is assigned to the "main structural column" scheduling strategy group.

[0112] Step S1311: Perform time-sequence arrangement and resource conflict check on the scheduling strategy units in each scheduling strategy group, and generate scheduling strategy fragments based on the continuity of device timeline occupancy.

[0113] In this embodiment, taking the "basement floor slab" scheduling strategy group as an example, it includes scheduling strategy units ORD001 and ORD003. They are arranged according to their estimated start times as ORD001 (6:42-10:00) and ORD003 (7:00-8:30). Checking for resource conflicts reveals that ORD001 and ORD003 are allocated to different devices (M002 and M001), indicating no device resource conflict. Based on the continuity of device timeline occupancy, a "basement floor slab" scheduling strategy fragment is generated, containing the scheduling information for these two orders.

[0114] Step S1312: Integrate and connect the scheduling strategy fragments generated by different scheduling strategy groups to generate a global initial scheduling strategy set for concrete production. Each strategy in the initial scheduling strategy set for concrete production contains a mapping relationship between task execution order and equipment allocation.

[0115] In this embodiment, scheduling strategy fragments generated by scheduling strategy groups such as "basement floor slab" and "main structural columns" are integrated. The fragments are then linked according to the chronological order of tasks to ensure a reasonable overall task execution sequence and clear equipment allocation mapping. The final generated initial scheduling strategy set for concrete production contains multiple scheduling strategies, each clearly defining the task execution order and equipment allocation.

[0116] Step S140: The initial scheduling strategy set for concrete production is processed by the conflict perception module of the concrete scheduling deep decision network to detect and dynamically adjust strategy conflicts, so as to obtain the optimized scheduling strategy set for concrete production after conflict resolution. The conflict perception module is trained based on historical scheduling conflict case data.

[0117] In this embodiment, the conflict perception module is an important component of the concrete scheduling deep decision network. It analyzes the initial scheduling strategy set to detect potential conflicts, such as excessive equipment switching conflicts and material supply intermittent conflicts. This conflict perception module is trained based on historical scheduling conflict case data, enabling it to identify different types of conflict patterns and generate corresponding adjustment suggestions. It dynamically adjusts the initial scheduling strategy, ultimately obtaining an optimized scheduling strategy set after conflict resolution.

[0118] Step S141: Input the initial scheduling strategy set for concrete production into the strategy parsing layer of the conflict perception module. The strategy parsing layer decomposes each scheduling strategy into task sequence segments and equipment timeline segments.

[0119] In this embodiment, the strategy parsing layer receives the initial set of scheduling strategies for concrete production and decomposes each scheduling strategy. For example, a certain scheduling strategy includes the execution order of tasks A, B, and C and their corresponding equipment allocation. The strategy parsing layer decomposes the strategy into task sequence segments [A, B, C] and equipment timeline segments corresponding to each equipment. For example, the timeline segment of equipment M001 is [the time period of A, the time period of C], and the timeline segment of equipment M002 is [the time period of B].

[0120] Step S142: The conflict pattern matching submodule of the conflict perception module calculates the similarity of the conflict feature patterns in the task sequence segment, equipment timeline segment, and historical scheduling conflict case data to identify potential conflict types, including equipment over-switching conflicts and material supply intermittent conflicts.

[0121] In this embodiment, the conflict pattern matching submodule compares the decomposed task sequence segments and device timeline segments with conflict feature patterns in historical scheduling conflict case data. For example, historical cases may contain conflict feature patterns of devices frequently switching tasks within a short period. By calculating the similarity between the frequency and time interval of device switching in the current task sequence segment and this pattern, if the similarity exceeds a set threshold, it is identified as an excessive device switching conflict. Similarly, by analyzing timeline segments related to material supply and comparing them with feature patterns of intermittent material supply conflicts, intermittent material supply conflicts are identified.

[0122] Step S143: For the identified equipment over-switching conflict, the conflict perception module generates a device task merging suggestion, merging multiple concrete production task orders that are adjacent in time and use the same batching plant equipment into a continuous production batch.

[0123] In this embodiment, when an excessive switching conflict is identified, the conflict awareness module analyzes the orders and equipment involved in the conflict and generates a task merging suggestion. For example, if equipment M001 needs to execute three adjacent orders ORD005, ORD006, and ORD007 within a short period of time, and there is an excessive switching conflict, then a suggestion is generated to merge these three orders into a single continuous production batch.

[0124] Step S1431: Locate the batching plant equipment identifier involved in the equipment over-switching conflict and the identifiers of multiple concrete production task orders executed sequentially on that batching plant equipment.

[0125] In this embodiment, the conflict pattern matching submodule determines that the device involved in the device over-switching conflict is M001, and the order identifiers executed sequentially on this device are ORD005, ORD006, and ORD007.

[0126] Step S1432: Calculate the task switching time interval between the multiple concrete production task orders. The task switching time interval includes equipment cleaning time and equipment reset time. If the sum of the task switching time intervals exceeds the preset switching time tolerance threshold, it is determined that there is an excessive equipment switching conflict, and task merging analysis is triggered.

[0127] In this embodiment, the switching time interval between ORD005 and ORD006 is 0.5 hours for equipment cleaning time and 0.2 hours for equipment reset time, totaling 0.7 hours; the switching time interval between ORD006 and ORD007 is also 0.7 hours, for a total switching time interval of 1.4 hours. The preset switching time tolerance threshold is 1 hour. 1.4 hours exceeds the threshold, indicating an excessive equipment switching conflict, triggering task merging analysis.

[0128] Step S1433: Analyze the concrete grade attributes of multiple concrete production task orders identified as equipment over-switching conflicts, query the preset concrete grade compatibility rule table, and check whether the concrete grade attributes of the multiple concrete production task orders identified as equipment over-switching conflicts are defined as allowing continuous production without thorough equipment cleaning.

[0129] In this embodiment, the concrete grade attribute of ORD005, ORD006, and ORD007 is C30. According to the concrete grade compatibility rule table, C30 is defined as allowing continuous production without thorough cleaning of the equipment, only requiring simple rinsing.

[0130] Step S1434: Analyze the pouring location attributes of multiple concrete production task orders identified as equipment over-switching conflicts, and check whether the engineering coordinates or transportation paths associated with the pouring location attributes of multiple concrete production task orders identified as equipment over-switching conflicts meet the preset proximity determination rules.

[0131] In this embodiment, the pouring location attributes of ORD005, ORD006, and ORD007 are all "5th floor slab of Building 3". Their associated engineering coordinates are very close, and their transportation paths are basically the same, which meets the preset proximity judgment rules.

[0132] Step S1435: Analyze the volume attributes of multiple concrete production task orders identified as equipment over-switching conflicts, and calculate whether the total volume after merging exceeds the maximum allowable volume limit of the batching plant equipment in single-batch continuous production mode.

[0133] In this embodiment, the volume of ORD005 is 30 cubic meters, the volume of ORD006 is 25 cubic meters, and the volume of ORD007 is 20 cubic meters, with a combined total volume of 75 cubic meters. The maximum allowable volume of equipment M001 in single-batch continuous production mode is limited to 80 cubic meters, and 75 cubic meters does not exceed the limit.

[0134] Step S1436: If the concrete grade attributes of multiple concrete production task orders identified as being involved in equipment over-switching conflicts are compatible, the pouring location attributes are similar, and the combined volume does not exceed the limit, then an equipment task merging suggestion is generated. The equipment task merging suggestion includes a list of identifiers of the concrete production task orders to be merged and information on the suggested merged production batches. The equipment task merging suggestion specifies that the estimated start time of the first concrete production task order among the multiple concrete production task orders identified as being involved in equipment over-switching conflicts is used as the start time of the batch. The equipment task merging suggestion calculates the total production time corresponding to the total volume after merging, and adds the start time of the batch to the total production time to obtain the end time of the batch. The equipment task merging suggestion cancels the planned equipment cleaning and reset steps between the multiple concrete production task orders identified as being involved in equipment over-switching conflicts, and converts the original task switching time interval into continuous production time.

[0135] In this embodiment, since ORD005, ORD006, and ORD007 meet the merging conditions, a device task merging suggestion is generated. The list of order identifiers to be merged is [ORD005, ORD006, ORD007]. The start time of the merged production batch is the estimated start time of ORD005, 8:00. The total volume is 75 cubic meters, the rated production rate of equipment M001 is 20 cubic meters / hour, the total production time is 75 / 20 = 3.75 hours, and the end time of the batch is 8:00 + 3.75 hours = 11:45. The device cleaning and reset steps between ORD005 and ORD006, and between ORD006 and ORD007 are cancelled, and the original switchover time interval of 1.4 hours is converted into continuous production time.

[0136] Step S1437: The equipment task merging suggestion is sent to the strategy adjustment submodule. The strategy adjustment submodule modifies the task sequence fragments and equipment timeline fragments in the initial scheduling strategy set for concrete production according to the equipment task merging suggestion. The modified task sequence fragments will replace multiple independent concrete production task orders identified as equipment over-switching conflicts with a single task node representing a merged batch. The modified equipment timeline fragments will merge the original multiple discrete occupancy intervals into a continuous occupancy interval with a longer duration.

[0137] In this embodiment, after receiving the device task merging suggestion, the policy adjustment submodule modifies the initial scheduling policy set. In the task sequence segment, [ORD005, ORD006, ORD007] are replaced with a merged batch task node ORD-M001. In the timeline segment of device M001, the original three discrete occupancy intervals (e.g., 8:00-9:30, 10:00-11:15, 11:45-12:30) are merged into a continuous occupancy interval 8:00-11:45.

[0138] Step S1438: The strategy adjustment submodule recalculates the material delivery sequence of each sub-order within the merged batch, so that the material supply is adapted to the continuous production rhythm. After the merging process, the original equipment over-switching conflict is resolved.

[0139] In this embodiment, the strategy adjustment submodule recalculates the material delivery sequence of ORD005, ORD006, and ORD007 within the merged batch based on the total volume and production duration of the merged batch. For example, it allocates the material delivery volume according to the proportion of the total volume and adjusts the delivery time according to the rhythm of continuous production to ensure that the material supply matches the production process. Through the above processing, equipment M001 no longer needs to frequently switch tasks, and the conflict of excessive equipment switching is resolved.

[0140] Step S144: For the identified material supply intermittent conflicts, the conflict perception module analyzes the material types and supply device status involved in the conflict, generates material delivery timing optimization suggestions, and adjusts the start time of the material preparation stage for different concrete production task orders.

[0141] In this embodiment, when a material supply intermittent conflict is identified, such as a concentrated demand for a certain aggregate within a short period, causing the supply device to be unable to supply it in a timely manner, the conflict perception module analyzes the type of aggregate and the current status of the corresponding supply device (such as aggregate bins, conveyor belts, etc.). Then, based on the supply capacity of the supply device and the material demand, it generates material delivery timing optimization suggestions, such as advancing or delaying the start time of the material preparation stage for some orders to avoid peak material demand and supply intermittent.

[0142] Step S145: Based on the equipment task merging suggestion and the material delivery timing optimization suggestion, the strategy adjustment submodule of the conflict perception module performs a partial rearrangement of the task execution order in the initial scheduling strategy set for concrete production.

[0143] In this embodiment, the strategy adjustment submodule combines equipment task merging suggestions and material delivery timing optimization suggestions to make local adjustments to the task execution order in the initial scheduling strategy set. For example, the merged batch task node ORD-M001 is adjusted to a more suitable time position, and the start time of related orders is adjusted according to the material delivery timing optimization suggestions, making the overall task execution order more reasonable and reducing conflicts.

[0144] Step S146: The strategy adjustment submodule introduces resource relaxation variables during the local reordering process. The resource relaxation variables allow for fine-tuning of the task start time under the premise of satisfying the preset scheduling constraint rule base, thereby alleviating resource contention.

[0145] In this embodiment, the resource slack variable is an allowed time adjustment range, such as ±15 minutes. During the local rescheduling process, for some tasks with resource contention, the strategy adjustment submodule can fine-tune the start time of the task within the allowed range of the resource slack variable, such as delaying the start time of a task by 10 minutes to avoid conflicts with other tasks in terms of equipment or material resources, while ensuring that the adjusted time still meets the requirements such as the delivery time window in the preset scheduling constraint rule base.

[0146] Step S147: After the local rearrangement is completed, the feasibility verification submodule of the conflict perception module simulates the execution of the adjusted scheduling strategy to check whether new conflicts are generated or the preset scheduling constraint rule base is violated.

[0147] In this embodiment, the feasibility verification submodule simulates execution according to the adjusted scheduling strategy, tracking the execution process of each task, equipment occupancy, and material supply. It checks for new equipment conflicts, material supply conflicts, etc., and verifies whether all tasks can be completed within their delivery time windows and whether constraints such as continuous equipment operating time are met.

[0148] Step S148: If the feasibility verification submodule detects a new conflict, the steps of conflict pattern matching, generating adjustment suggestions and local rearrangement are repeated until no new conflicts are generated and all constraints are met.

[0149] In this embodiment, if the feasibility verification submodule discovers a new conflict during the simulation execution, such as the adjusted task execution order causing an excessive switching conflict on another device, the conflict pattern matching submodule will be triggered again to identify the conflict. The conflict perception module generates a new adjustment suggestion, and the strategy adjustment submodule performs a new round of local rearrangement. This cycle continues until no new conflicts are generated and all constraints are met.

[0150] Step S149: The efficiency evaluation submodule of the conflict perception module calculates the overall efficiency index of the scheduling strategy after conflict resolution. The overall efficiency index includes equipment utilization and average order waiting time.

[0151] In this embodiment, the efficiency evaluation submodule calculates the overall efficiency index of the scheduling strategy after conflict resolution. Equipment utilization is calculated as the ratio of actual equipment operating time to total available time. For example, if the total operating time of all equipment is 100 hours and the total available time is 120 hours, the equipment utilization rate is approximately 100 / 120 ≈ 83.3%. The average order waiting time is calculated by dividing the sum of the waiting times of all orders by the number of orders. The waiting time is the interval between the order arrival time and the start of execution time.

[0152] Step S1410: If the overall efficiency index does not reach the preset threshold, the efficiency evaluation submodule triggers a strategy fragment replacement operation, retrieves a scheduling strategy fragment similar to the current working condition from the historical high-efficiency scheduling strategy library, replaces the strategy fragment in the current scheduling strategy that does not reach the preset efficiency index, and outputs the final version of the concrete production optimization scheduling strategy set after conflict resolution after iterative adjustment and optimization.

[0153] In this embodiment, the preset equipment utilization threshold is 85%, and the average order waiting time threshold is 0.5 hours. If the calculated equipment utilization rate is 83.3%, which does not reach the threshold, the efficiency evaluation submodule triggers a strategy fragment replacement operation. It retrieves efficient scheduling strategy fragments from the historical efficient scheduling strategy library that are similar to the current batching plant's equipment status, order type, and other operating conditions, and replaces the strategy fragments with lower equipment utilization rates in the current scheduling strategy with the retrieved efficient fragments. Then, conflict detection and feasibility verification are performed again. After multiple iterations of adjustment and optimization, until the overall efficiency index reaches the preset threshold, the final set of optimized scheduling strategies for concrete production is output.

[0154] Step S150: Generate a sequence of control instructions for concrete batching plant equipment and a sequence of material delivery timing instructions based on the set of optimized scheduling strategies for concrete production, and send the sequence of control instructions for concrete batching plant equipment and the sequence of material delivery timing instructions to the corresponding batching plant control unit for execution.

[0155] In this embodiment, the set of optimized scheduling strategies for concrete production determines the execution order of tasks, equipment allocation, and time schedule. Based on this information, specific equipment control instructions and material delivery timing instructions are generated. Equipment control instructions include starting and stopping the mixing host, adjusting its speed, and controlling the operation of the material conveying device; material delivery timing instructions specify the delivery time and quantity of various materials. The above instruction sequence is sent to the mixing plant control unit, such as the mixing tower programmable logic controller and the material control programmable logic controller, which drives the corresponding actuators to complete the concrete production operation.

[0156] Step S151: Analyze the first scheduling strategy in the set of optimized scheduling strategies for concrete production, and extract the equipment allocation mapping relationship and task execution time points contained in the scheduling strategy.

[0157] In this embodiment, the set of optimized scheduling strategies for concrete production may include multiple scheduling strategies. First, the first scheduling strategy is parsed. The equipment allocation mapping relationship is extracted from this strategy, such as assigning the ORD-M001 batch task to equipment M001, and assigning the ORD002 order to equipment M003, etc. At the same time, the task execution time points are extracted, such as the start time of ORD-M001 at 8:00 and the end time at 11:45, and the start time of ORD002 at 9:00 and the end time at 12:00, etc.

[0158] Step S152: Determine the identifier of the mixing host and the identifier of the material conveying device to be started according to the equipment allocation mapping relationship.

[0159] In this embodiment, according to the equipment allocation mapping relationship, ORD-M001 is assigned to device M001, and ORD002 is assigned to device M003. Therefore, the identifiers of the mixing host that need to be started are M001 and M003. At the same time, the identifiers of the material conveying devices that are matched with these mixing hosts, such as aggregate conveying device C001 and powder conveying device C002, also need to be determined and started.

[0160] Step S153: Calculate the start time of the material weighing stage, the start time of the mixing stage, and the start time of the unloading stage for each concrete production task order based on the task execution time points.

[0161] In this embodiment, the ORD-M001 task execution time is 8:00-11:45, with a total production time of 3.75 hours. Assuming that the material weighing stage accounts for 20% of the total time, the mixing stage accounts for 60%, and the unloading stage accounts for 20%, then the material weighing stage starts at 8:00, lasts for 0.75 hours, and ends at 8:45; the mixing stage starts at 8:45, lasts for 2.25 hours, and ends at 11:15; and the unloading stage starts at 11:15, lasts for 0.75 hours, and ends at 11:45.

[0162] Step S154: Based on the start time of the material weighing stage, generate door opening control commands and weighing instrument start commands for different aggregate bins and powder bins, and generate a material weighing control command sub-sequence.

[0163] In this embodiment, the material weighing stage of ORD-M001 begins at 8:00 AM. The aggregate bins to be used are G001 (gravel) and G002 (sand), and the powder bins are P001 (cement) and P002 (fly ash). At 8:00 AM, commands are generated to open the doors of bins G001, G002, P001, and P002, as well as to activate the corresponding weighing instruments. These commands form a material weighing control command sub-sequence in chronological order.

[0164] Step S155: Based on the start time of the stirring stage, generate motor start command, water supply valve opening command and admixture pump start command for the specified stirring host, and generate stirring host control command sub-sequence.

[0165] In this embodiment, the stirring stage of ORD-M001 begins at 8:45, and the designated stirring host is M001. At 8:45, the motor start command (setting the speed to a specified value), the water supply valve open command (setting the flow rate to a specified value), and the admixture pump start command (setting the pump speed to a specified value) for the M001 main unit are generated. These commands constitute the stirring host control command sub-sequence.

[0166] Step S156: Based on the start time of the unloading stage, generate instructions to control the opening of the unloading gate of the mixing host and signals to control the positioning of the concrete transport vehicle, and generate a sub-sequence of unloading control instructions.

[0167] In this embodiment, the unloading phase of ORD-M001 starts at 11:15. It generates an instruction to control the opening of the unloading gate of the M001 host (the opening degree is a specified value) and sends a positioning signal instruction to the concrete transport vehicle, instructing the transport vehicle to wait for unloading at a specified position. These instructions constitute the unloading control instruction subsequence.

[0168] Step S157: Integrate and time-align the material weighing control instruction sub-sequence, the mixing host control instruction sub-sequence, and the unloading control instruction sub-sequence according to the chronological order of task execution time points to generate a complete concrete mixing plant equipment control instruction sequence.

[0169] In this embodiment, the material weighing control instruction subsequence (starting at 8:00), the mixing host control instruction subsequence (starting at 8:45), and the unloading control instruction subsequence (starting at 11:15) are integrated in chronological order, ensuring that each instruction is aligned in time to form a complete sequence of control instructions for the concrete mixing plant equipment. For example, the 8:00 silo door opening instruction, the 8:45 motor start instruction, and the 11:15 unloading door opening instruction are arranged in chronological order.

[0170] Step S158: Based on the volume attributes of the concrete production task order corresponding to each task scheduling time block in the concrete production optimization scheduling strategy set and the required proportions of various materials, calculate the quantity of various materials required for the concrete production task, locate and accumulate the demand of each material on the time axis according to the start time of its corresponding material weighing stage, and generate a material demand time curve describing the demand quantity of various materials at different time points.

[0171] In this embodiment, the total volume of the ORD-M001 batch is 75 cubic meters, and the concrete mix ratio is cement:sand:aggregate:water:admixture = 1:1.8:3.5:0.5:0.02. The calculated cement requirement is 75*(1 / (1+1.8+3.5+0.5+0.02)) = 75*(1 / 6.82), the sand requirement is 11*1.8, and so on, to obtain the requirements of other materials. Then, based on the material weighing stage starting at 8:00, the requirements of various materials are accumulated on the time axis from 8:00. Assuming the weighing stage lasts 0.75 hours, the material demand gradually increases from 0 to the total demand during the period from 8:00 to 8:45, generating a material demand time curve.

[0172] Step S159: Calculate the material replenishment trigger time and replenishment amount based on the material demand time curve and the current storage status of the material conveying device.

[0173] In this embodiment, the current storage capacity of aggregate bin G001 (gravel) is 20 cubic meters. The required amount of gravel for batch ORD-M001 is 11 * 3.5 = 38.5 cubic meters, which is insufficient. According to the material demand time curve, 38.5 cubic meters of gravel are needed between 8:00 and 8:45. The current storage capacity of 20 cubic meters can last until 8:00 + (20 / 38.5) * 0.75 hours. Therefore, the material replenishment trigger time is set to 8:10, and the replenishment amount is 38.5 - 20 cubic meters to ensure that replenishment is completed before the gravel runs out.

[0174] Step S1510: Based on the material replenishment trigger time and replenishment amount, generate instructions to control the material conveying device to start, stop and speed adjustment, and generate a material delivery timing instruction sequence.

[0175] In this embodiment, for the replenishment of stones, a command to start the stone conveying device is generated at 8:10, and the rotation speed is set to ensure that the replenishment amount of 18.5 cubic meters is completed between 8:10 and 8:23. When the conveying amount reaches 18.5 cubic meters, a stop command is generated. Similarly, similar calculations and command generation are performed for other materials to form a material delivery timing command sequence.

[0176] Step S1511: Encapsulate the control command sequence of the concrete mixing plant equipment and the material delivery timing command sequence into a unified format scheduling command message, and send the scheduling command message to the corresponding mixing plant control unit through the industrial communication network. The mixing plant control unit includes a mixing tower programmable logic controller and a material central control programmable logic controller. The mixing plant control unit receives and parses the scheduling command message and drives the corresponding actuator to complete the concrete production operation.

[0177] In this embodiment, the control command sequence of the concrete mixing plant equipment and the material delivery timing command sequence are encapsulated according to a preset communication protocol format to form a scheduling command message. This message is then sent to the programmable logic controller (PLC) of the mixing plant and the central material control PLC via an industrial communication network such as an industrial Ethernet network. Upon receiving the message, the control unit parses it, extracts the various commands, and then drives the corresponding motors, valves, pumps, and other actuators to operate according to the command requirements, completing the concrete production operation.

[0178] For example, the method may also include: step S210: collecting historical concrete production task datasets, historical batching plant equipment operation status data streams and their corresponding manually formulated or actually executed scheduling strategy result data to form a training sample set.

[0179] In this embodiment, a historical concrete production task dataset over a past period is collected, containing a large amount of order information; a historical batching plant equipment operation status data stream, recording the equipment's operating parameters at different times; and corresponding manually formulated or actually executed scheduling strategy results data, including task allocation, execution order, and time plan. The above data is organized according to a specific format to form a training sample set, which is used to train the deep decision-making network for concrete scheduling.

[0180] Step S220: Perform feature encoding on the historical concrete production task dataset in the training sample set that is consistent with the current production environment to generate a set of historical task feature vectors.

[0181] In this embodiment, the same feature encoding method as in step S121 is used to process each order in the historical concrete production task dataset, converting its attributes such as concrete grade, volume, pouring location, and transportation distance into a historical task feature vector of a unified dimension. All these vectors form a set of historical task feature vectors.

[0182] Step S230: Extract time-series features from the historical mixing plant equipment operation status data stream in the training sample set that are consistent with the current production environment, and generate a set of historical equipment status feature vectors.

[0183] In this embodiment, the same time-series feature extraction method as step S122 is used to process the historical equipment operation status data stream of the mixing plant, extract the motor current fluctuation pattern features of the mixing host and the belt speed stability features of the material conveying device, and combine them to generate historical equipment status feature vectors. All these vectors form a set of historical equipment status feature vectors.

[0184] Step S240: Input the set of historical task feature vectors and the set of historical equipment state feature vectors into the initial version of the concrete scheduling deep decision network to be trained, wherein the initial version has randomly initialized network weight parameters.

[0185] In this embodiment, the initial version of the concrete scheduling deep decision network to be trained has the same structure as the final applied network structure, but the network's weight parameters are randomly initialized. The set of historical task feature vectors and the set of historical equipment state feature vectors are used as inputs to this initial version of the network.

[0186] Step S250: Process the historical task feature vector set and the historical device state feature vector set through the initial version of the cross-attention calculation layer and multi-layer graph convolution module, and output the predicted implicit dependency strength matrix.

[0187] In this embodiment, the cross-attention computation layer and multi-layer graph convolution module of the initial version of the concrete scheduling deep decision network process the input set of historical task feature vectors and set of historical equipment state feature vectors, simulate the implicit dependency relationship between orders and equipment, and output the predicted implicit dependency strength matrix.

[0188] Step S260: Based on the predicted implicit dependency strength matrix and the scheduling policy result data in the training sample set, construct a first loss function. The first loss function measures the difference between the predicted dependency strength and the actual dependency reflected in the scheduling policy result.

[0189] In this embodiment, the actual dependencies between orders and devices are extracted from the scheduling strategy result data of the training sample set. For example, if an order is actually assigned to a certain device, then the actual dependency strength between them is 1, and for other devices it is 0 (or set according to the tightness of the actual assignment). The first loss function adopts the mean squared error loss function to calculate the mean squared error between the predicted implicit dependency strength matrix and the actual dependency matrix.

[0190] Step S270: Input the predicted implicit dependency strength matrix and the preset scheduling constraint rule base into a simulation scheduler. The simulation scheduler follows the same logic as the initial scheduling strategy generation to generate a simulation scheduling strategy.

[0191] In this embodiment, the simulated scheduler uses the predicted implicit dependency strength matrix and the preset scheduling constraint rule base to generate a simulated scheduling strategy according to the same logic as generating the initial scheduling strategy in step S130.

[0192] Step S280: Based on the simulated scheduling strategy and the scheduling strategy result data in the training sample set, construct a second loss function. The second loss function measures the difference between the simulated scheduling strategy and the real scheduling strategy in terms of task order and equipment allocation.

[0193] In this embodiment, the second loss function is the cross-entropy loss function, which compares the differences between the simulated scheduling strategy and the real scheduling strategy in terms of task execution order and device allocation. For example, for task order, the cross-entropy loss between simulated and real sorting is calculated; for device allocation, the cross-entropy loss between simulated and real allocation is calculated, and then the two are weighted and summed as the second loss function.

[0194] Step S290: The first loss function and the second loss function are weighted and combined to obtain the total training loss function. The gradient descent optimization algorithm is used to calculate the gradient of the total training loss function with respect to all weight parameters of the concrete scheduling depth decision network. The weight parameters of the concrete scheduling depth decision network are updated according to the gradient to reduce the value of the total training loss function.

[0195] In this embodiment, the weight of the first loss function is set to 0.4, the weight of the second loss function is set to 0.6, and the total training loss function is 0.4 * first loss function + 0.6 * second loss function. The Adam optimizer is used as the gradient descent optimization algorithm, with an initial learning rate of 0.001 and a batch size of 32. The gradient of the total training loss function with respect to all network weight parameters is calculated, and then the weight parameters are updated according to the direction of gradient descent, gradually decreasing the value of the total training loss function.

[0196] Step S2100: Repeat the steps of forward prediction, loss calculation, backpropagation and weight update until the total training loss function value converges to a stable state or reaches the preset training rounds. Save the weight parameters of the concrete scheduling deep decision network after training convergence to obtain a concrete scheduling deep decision network that can be used for actual scheduling.

[0197] In this embodiment, the process of continuously inputting historical data into the network for forward prediction, calculating the total training loss function, calculating the gradient through backpropagation, and updating the weight parameters is repeated. The preset training epochs are 100. When the training reaches 100 epochs or the value of the total training loss function changes by less than 0.001 within 10 consecutive epochs, the training is considered to have converged. The network weight parameters at this time are saved, and the trained concrete scheduling deep decision network is obtained.

[0198] Step S2110: During the operation of the production system, new concrete production task data, batching plant equipment operation status data and corresponding effective scheduling strategies are collected periodically to form new training samples; using the concrete scheduling deep decision network after training convergence as the initial model, supervised training is continued under the original total training loss function using the new training samples to update the network weight parameters and realize incremental learning and adaptive optimization of the model.

[0199] In this embodiment, during the actual production process of the mixing plant, new concrete production task data, equipment operation status data, and corresponding actual scheduling strategies are collected every month to form new training samples. Using the previously trained and converged network as the initial model, the new training samples are used to perform supervised training according to the same total training loss function and optimization algorithm as before. The network weight parameters are fine-tuned so that the network can adapt to new production data and operating conditions, achieving incremental learning and adaptive optimization, and improving the accuracy and adaptability of the scheduling strategy.

[0200] Based on the same inventive concept, please refer to Figure 2 The diagram shows a schematic block diagram of a concrete batching plant scheduling strategy generation system 100 combined with deep learning provided in an embodiment of this application. The concrete batching plant scheduling strategy generation system 100 combined with deep learning may include a communication unit 110, a machine-readable storage medium 120, and a processor 130.

[0201] In this embodiment, the machine-readable storage medium 120 can also be integrated into the processor 130 and can communicate and interact with external systems through the communication unit 110. The machine-readable storage medium 120 stores machine-executable instructions for executing the scheme of this application, and the processor 130 executes the machine-executable instructions stored in the machine-readable storage medium 120 to implement the concrete mixing plant scheduling strategy generation method combined with deep learning provided in the aforementioned method embodiments.

[0202] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for generating scheduling strategies for concrete mixing plants using deep learning, characterized in that, The method includes: Acquire a concrete production task dataset and a batching plant equipment operation status data stream. The concrete production task dataset contains multiple concrete production task orders, and the batching plant equipment operation status data stream contains a real-time collected sequence of operating parameters of the mixing host and a sequence of working status signals of the material conveying device. The concrete scheduling deep decision network processes the concrete production task dataset and the batching plant equipment operation status data stream to mine implicit dependencies, generating an implicit dependency strength matrix between concrete production task orders and batching plant equipment. The dimension of the implicit dependency strength matrix corresponds to the number of concrete production task orders and the number of batching plant equipment. An initial scheduling strategy is generated based on the implicit dependency strength matrix and the preset scheduling constraint rule base. This generates a set of initial scheduling strategies for concrete production that includes the mapping relationship between task execution order and equipment allocation. The preset scheduling constraint rule base includes a threshold for continuous working time of equipment and a threshold for order delivery time window. The conflict perception module of the concrete scheduling deep decision network processes the initial scheduling strategy set of concrete production to detect and dynamically adjust strategy conflicts, thereby obtaining the optimized scheduling strategy set of concrete production after conflict resolution. The conflict perception module is trained based on historical scheduling conflict case data. Based on the set of concrete production optimization scheduling strategies, a sequence of control instructions for concrete batching plant equipment and a sequence of material delivery timing instructions are generated, and the sequence of control instructions for concrete batching plant equipment and the sequence of material delivery timing instructions are sent to the corresponding batching plant control unit for execution. The process involves using a deep decision network for concrete scheduling to process the concrete production task dataset and the batching plant equipment operation status data stream to mine implicit dependencies, generating an implicit dependency strength matrix between concrete production task orders and batching plant equipment. This includes: The input feature transformation layer of the concrete scheduling deep decision network processes each concrete production task order in the concrete production task dataset for feature encoding, and converts the concrete grade attribute, volume attribute, pouring location attribute and transportation distance attribute contained in each concrete production task order into a task feature vector of a unified dimension. The equipment state coding layer of the concrete scheduling deep decision network processes the data stream of the batching plant equipment operation status to extract time-series features, extracts the motor current fluctuation pattern features of the mixing host and the belt speed stability features of the material conveying device, and combines them to generate an equipment state feature vector. The task feature vector and the equipment state feature vector are input into the cross-attention calculation layer of the concrete scheduling deep decision network to calculate the bidirectional attention weight value between the task feature vector corresponding to each concrete production task order and the equipment state feature vector corresponding to each batching plant equipment. An initial association weight graph between tasks and equipment is constructed based on the bidirectional attention weight values. The nodes of the initial association weight graph include concrete production task order nodes and mixing plant equipment nodes, and the edge weights between nodes are assigned by the bidirectional attention weight values. The initial association weight graph is processed by the multi-layer graph convolution module of the concrete scheduling depth decision network to propagate higher-order relationships, capturing the indirect dependencies between concrete production task order nodes through shared equipment and the collaborative relationships between batching plant equipment nodes through common tasks; Based on the node embedding vector output by the multi-layer graph convolution module, the cosine similarity value between any concrete production task order node and any mixing plant equipment node is calculated as a refinement of the association strength. By integrating the bidirectional attention weight value and the refined correlation strength value, the final correlation strength value is calculated by weighted summation. The final correlation strength value is then organized into a two-dimensional matrix structure according to the order of concrete production task orders and the order of batching plant equipment. The two-dimensional matrix structure is subjected to row-direction normalization and column-direction normalization to ensure that the distribution of associated strength values ​​corresponding to each concrete production task order and the distribution of associated strength values ​​corresponding to each batching plant equipment both satisfy the probability distribution characteristics. The two-dimensional matrix structure with completed two-way normalization is output as the implicit dependency strength matrix between concrete production task orders and batching plant equipment.

2. The method for generating a concrete mixing plant scheduling strategy based on deep learning according to claim 1, characterized in that, The equipment state coding layer of the concrete scheduling deep decision network processes the data stream of the batching plant equipment operating status to extract time-series features. It extracts the motor current fluctuation pattern features of the mixing host and the belt speed stability features of the material conveying device, combining them to generate an equipment state feature vector, including: The operating parameter sequence of the mixing host is separated from the data stream of the operating status of the mixing plant equipment. The operating parameter sequence of the mixing host includes the instantaneous values ​​of the three-phase current of the motor recorded at a fixed sampling frequency. The instantaneous values ​​of the three-phase current of the motor are transformed by coordinate transformation to convert the three-phase current values ​​into a representation of amplitude and phase angle; Calculate the rate of change sequence of motor current amplitude between consecutive sampling points, perform sliding window statistics on the rate of change sequence, and calculate the variance of the rate of change within each window as an index of current fluctuation intensity. Spectral analysis is performed on the amplitude of the motor current to extract the power value of the main frequency component and its trend of change over time, which serves as a characteristic of load periodic fluctuation. The working status signal sequence of the material conveying device is separated from the data stream of the operating status of the mixing plant equipment. The working status signal sequence of the material conveying device includes belt speed feedback signal and motor torque current signal. Calculate the mean and standard deviation of the belt speed feedback signal over a set time period, and use the ratio of the standard deviation to the mean as the belt speed stability coefficient; Analyze the phase lag relationship between the motor torque current signal and the belt speed feedback signal, and calculate the lag time constant as a characteristic of the transmission system response delay. The current fluctuation intensity index, the load periodic fluctuation characteristics, the belt speed stability coefficient, and the transmission system response delay characteristics are scaled to bring all characteristic values ​​to the same order of magnitude. The current fluctuation intensity index, the load periodic fluctuation characteristics, the belt speed stability coefficient, and the transmission system response delay characteristics are spliced ​​and scaled in a preset order to generate a one-dimensional array; The one-dimensional array is input into a fully connected neural network layer for feature compression and nonlinear transformation, and a fixed-length device state feature vector is output. The device state feature vector is then standardized.

3. The method for generating a concrete mixing plant scheduling strategy based on deep learning according to claim 1, characterized in that, The process of processing the initial association weight graph through the multi-layer graph convolution module of the concrete scheduling depth decision network to propagate higher-order relationships captures the indirect dependencies between concrete production task order nodes generated through shared equipment and the collaborative relationships between batching plant equipment nodes generated through shared tasks, including: The multi-layer graph convolution module loads the initial association weight graph generated by the cross-attention calculation layer. In the initial association weight graph, each concrete production task order node is associated with a task feature vector containing transportation distance attributes, and each batching plant equipment node is associated with an equipment status feature vector containing motor current fluctuation pattern features. In the first propagation layer of the multi-layer graph convolution module, according to the connection edges of the initial association weight graph, the equipment state feature vector of each batching plant equipment node is sent to the concrete production task order node directly connected to it, and each concrete production task order node receives the feature message from its candidate batching plant equipment node. For each concrete production task order node, the first propagation layer aggregates all incoming equipment status feature vector messages, calculates a weighted sum based on the connection edge weights, and generates the first layer of equipment influence features for each concrete production task order node. The first propagation layer fuses the first-layer equipment influence features of each concrete production task order node with its own task feature vector containing transportation distance attributes, and generates the updated first-layer feature vector of each concrete production task order node through a fully connected layer. In the second propagation layer of the multi-layer graph convolution module, the updated feature vector of each concrete production task order node is sent to other concrete production task order nodes that are indirectly connected by sharing the same candidate mixing plant equipment. The transmission path is through the shared mixing plant equipment node. The second propagation layer aggregates the updated feature vector messages from other indirectly connected concrete production task order nodes for each concrete production task order node, generating the second layer of inter-order dependency features for each concrete production task order node. The second propagation layer simultaneously transmits the updated first-layer feature vector of each concrete production task order node to the batching plant equipment node connected to it in reverse message transmission. Each batching plant equipment node receives feature messages from all concrete production task order nodes it serves. The second propagation layer aggregates all incoming first-layer updated feature vector messages from different concrete production task order nodes for each batching plant equipment node, generating task collaborative load features for each batching plant equipment node. The second propagation layer concatenates the second-layer inter-order dependency features of each concrete production task order node with its updated first-layer feature vector, and generates the final output concrete production task order node embedding vector through another fully connected layer. This concrete production task order node embedding vector encodes the indirect dependencies generated through shared devices. The second propagation layer concatenates the task collaboration load features of each batching plant equipment node with its initial equipment state feature vector, and generates the final output batching plant equipment node embedding vector through another fully connected layer. This batching plant equipment node embedding vector encodes the collaborative relationship generated through the common task.

4. The method for generating a concrete mixing plant scheduling strategy based on deep learning according to claim 1, characterized in that, The process of generating an initial scheduling strategy based on the implicit dependency strength matrix and a preset scheduling constraint rule base generates a set of initial scheduling strategies for concrete production that includes a mapping relationship between task execution order and equipment allocation, including: Extract the equipment association strength distribution vector corresponding to each concrete production task order from the implicit dependency strength matrix; Based on the equipment association strength distribution vector, a preset number of mixing plant equipment with the highest association strength are selected for each concrete production task order to form a candidate equipment set; Based on the order delivery time window threshold in the preset scheduling constraint rule base, a time urgency coefficient is calculated for each concrete production task order. This time urgency coefficient is equal to the reciprocal of the current time to the deadline of the order delivery time window and the weighted combination of the corresponding order volume. All concrete production task orders are sorted in descending order according to the time urgency coefficient to generate a priority queue of concrete production task orders. Each concrete production task order is processed sequentially according to the priority queue, and the mixing plant equipment that is currently idle and has the earliest estimated task completion time is selected from the candidate equipment set of the currently processed concrete production task order for pre-allocation. During the pre-allocation process, the equipment continuous working time threshold in the preset scheduling constraint rule base is queried. If the cumulative planned working time of the selected mixing plant equipment exceeds the equipment continuous working time threshold, the mixing plant equipment is marked as temporarily unavailable and removed from the candidate equipment set of the current concrete production task order. A sequence of allocation pairs is constructed based on the pre-allocation results. Each allocation pair contains a concrete production task order identifier and an assigned batching plant equipment identifier. Based on the task receiving order and estimated task execution time of the batching plant equipment in the allocation sequence, the timeline occupancy information of the batching plant equipment is simulated and generated, and the timeline occupancy information is presented in the form of equipment timeline segments. Based on the equipment timeline segments, the estimated start time and estimated end time of each concrete production task order are calculated to generate a preliminary scheduling time plan. Combine all allocation pairs with their corresponding scheduling time plans to generate multiple independent scheduling strategy units; The scheduling strategy units are grouped according to the pouring location attribute of the concrete production task order, and the scheduling strategy units corresponding to the concrete production task orders of the same pouring location are assigned to the same scheduling strategy group. Perform time-sequential arrangement and resource conflict check on the scheduling policy units within each scheduling policy group, and generate scheduling policy fragments based on the continuity of equipment timeline occupancy. By integrating and connecting scheduling strategy fragments generated from different scheduling strategy groups, a global initial scheduling strategy set for concrete production is generated. Each strategy in the initial scheduling strategy set for concrete production contains a mapping relationship between task execution order and equipment allocation.

5. The method for generating a concrete mixing plant scheduling strategy based on deep learning according to claim 4, characterized in that, The step of calculating the estimated start and end times of each concrete production task order based on the equipment timeline segment, and generating a preliminary scheduling time plan, includes: Traverse the sequence of allocation pairs, and for each allocation pair in the sequence, obtain the identifier of the mixing plant equipment to which it is allocated; Query the equipment timeline segment corresponding to the equipment identifier of the mixing plant. The equipment timeline segment records the set of time periods in which the equipment of the mixing plant has been planned to be occupied. Find the end time point of the last occupied time period in the equipment timeline segment and take the end time point as the earliest start time point of the current concrete production task order on the equipment of the mixing plant. Based on the volume of the current concrete production task order and the rated production rate of the mixing plant equipment, calculate the pure mixing operation time required to complete the concrete production task order. Based on historical data statistics, the average material preparation time and average unloading time corresponding to the concrete grade attribute of the concrete production task order are obtained. The pure mixing operation time, the average material preparation time and the average unloading time are added together to obtain the total estimated task execution time of the current concrete production task order. The earliest possible start time is added to the total estimated task execution time to obtain the estimated end time of the current concrete production task order. The estimated end time is checked to see if it exceeds the delivery time window threshold of the concrete production task order. If it does, the earliest possible start time is adjusted forward by a time offset, which is equal to the length of the time exceeding the threshold. The adjusted start time is taken as the final estimated start time of the current concrete production task order. The new estimated end time is obtained by adding the final estimated start time to the total estimated task execution time. The time interval consisting of the final estimated start time and the new estimated end time is recorded as the scheduling time block of the concrete production task order. The scheduling time block is updated to the equipment timeline segment corresponding to the mixing plant equipment, and the length of the equipment timeline segment is extended. Repeat the above steps for the next allocation pair in the allocation pair sequence until all allocation pairs have calculated scheduling time blocks; Collect all scheduling time blocks for concrete production task orders, sort them according to the order of their estimated start times, associate the sorted list of scheduling time blocks with the corresponding allocation pair sequences, and generate a structured preliminary scheduling time plan.

6. The method for generating a concrete mixing plant scheduling strategy based on deep learning according to claim 1, characterized in that, The process involves using the conflict perception module of the concrete scheduling deep decision network to process the initial concrete production scheduling strategy set, perform strategy conflict detection and dynamic adjustment, and obtain a set of optimized concrete production scheduling strategies after conflict resolution, including: The initial scheduling strategy set for concrete production is input into the strategy parsing layer of the conflict perception module, and the strategy parsing layer decomposes each scheduling strategy into task sequence segments and equipment timeline segments. The conflict pattern matching submodule of the conflict perception module calculates the similarity of the conflict feature patterns in the task sequence fragments, equipment timeline fragments, and historical scheduling conflict case data to identify potential conflict types, including equipment over-switching conflicts and material supply intermittent conflicts. In response to the identified equipment switching conflicts, the conflict perception module generates a device task merging suggestion, which merges multiple concrete production task orders that are adjacent in time and use the same batching plant equipment into a continuous production batch. For the identified material supply intermittent conflicts, the conflict perception module analyzes the types of materials involved in the conflict and the status of the supply devices, generates material delivery timing optimization suggestions, and adjusts the start time of the material preparation stage for different concrete production task orders; Based on the equipment task merging suggestion and the material delivery timing optimization suggestion, the strategy adjustment submodule of the conflict perception module performs a partial rearrangement of the task execution order in the initial scheduling strategy set for concrete production. The strategy adjustment submodule introduces resource relaxation variables during the local reordering process. These resource relaxation variables allow for fine-tuning of the task start time while satisfying the preset scheduling constraint rule base. After the local rearrangement is completed, the feasibility verification submodule of the conflict perception module simulates the execution of the adjusted scheduling strategy to check whether new conflicts are generated or the preset scheduling constraint rule base is violated. If the feasibility verification submodule detects a new conflict, the steps of conflict pattern matching, generating adjustment suggestions and local rearrangement are repeated until no new conflicts are generated and all constraints are met. The efficiency evaluation submodule of the conflict perception module calculates the overall efficiency index of the scheduling strategy after conflict resolution. The overall efficiency index includes equipment utilization and average order waiting time. If the overall efficiency index does not reach the preset threshold, the efficiency evaluation submodule triggers a strategy fragment replacement operation, retrieves a scheduling strategy fragment similar to the current working condition from the historical high-efficiency scheduling strategy library, and replaces the strategy fragment in the current scheduling strategy that does not reach the preset efficiency index. After iterative adjustment and optimization, the final version of the concrete production optimization scheduling strategy set after conflict resolution is output.

7. The method for generating a concrete mixing plant scheduling strategy based on deep learning according to claim 6, characterized in that, In response to identified equipment switching conflicts, the conflict perception module generates a task merging suggestion, combining multiple concrete production task orders that are temporally adjacent and use the same batching plant equipment into a single continuous production batch, including: The location device over-switching conflict involves the batching plant equipment identifier and the identifiers of multiple concrete production task orders executed sequentially on that batching plant equipment; Calculate the task switching time interval between the multiple concrete production task orders, the task switching time interval including equipment cleaning time and equipment reset time; if the sum of the task switching time intervals exceeds the preset switching time tolerance threshold, it is determined that there is an excessive equipment switching conflict, and task merging analysis is triggered. Analyze the concrete grade attributes of multiple concrete production task orders identified as equipment over-switching conflicts, query the preset concrete grade compatibility rule table, and check whether the concrete grade attributes of the multiple concrete production task orders identified as equipment over-switching conflicts are defined as allowing continuous production without thorough equipment cleaning. Analyze the pouring location attributes of multiple concrete production task orders identified as equipment over-switching conflicts, and check whether the engineering coordinates or transportation paths associated with the pouring location attributes of multiple concrete production task orders identified as equipment over-switching conflicts meet the preset proximity determination rules. Analyze the volume attributes of multiple concrete production task orders identified as equipment over-switching conflicts, and calculate whether the total volume after merging exceeds the maximum allowable volume limit of the batching plant equipment in single-batch continuous production mode. If multiple concrete production task orders identified as being involved in an equipment over-switching conflict have compatible concrete grade attributes, similar pouring location attributes, and their combined volume does not exceed the limit, an equipment task merging suggestion is generated. This suggestion includes a list of identifiers for the concrete production task orders to be merged and information on the suggested merged production batch. The suggestion specifies that the estimated start time of the first concrete production task order among the multiple orders identified as being involved in the equipment over-switching conflict is used as the start time of the batch. The suggestion calculates the total production time corresponding to the total volume after merging, and adds the start time of the batch to the total production time to obtain the end time of the batch. The suggestion cancels the planned equipment cleaning and reset steps between the multiple concrete production task orders identified as being involved in the equipment over-switching conflict, converting the original task switching time interval into continuous production time. The proposed equipment task merging is sent to the strategy adjustment submodule. The strategy adjustment submodule modifies the task sequence fragments and equipment timeline fragments in the initial scheduling strategy set for concrete production according to the proposed equipment task merging. The modified task sequence fragments will replace multiple independent concrete production task orders identified as equipment over-switching conflicts with a single task node representing the merged batch. The modified equipment timeline fragments will merge the original multiple discrete occupancy intervals into a continuous occupancy interval with a longer duration. The strategy adjustment submodule recalculates the material delivery sequence of each sub-order within the merged batch, so that the material supply is adapted to the continuous production rhythm. After the merging process, the original equipment switching conflicts are resolved.

8. The method for generating a concrete mixing plant scheduling strategy based on deep learning according to claim 1, characterized in that, The generation of the concrete mixing plant equipment control command sequence and material delivery timing command sequence based on the concrete production optimization scheduling strategy set includes: Analyze the first scheduling strategy in the set of optimized scheduling strategies for concrete production, and extract the equipment allocation mapping relationship and task execution time points contained in the scheduling strategy; The identifiers of the mixing host and the material conveying device to be started are determined based on the equipment allocation mapping relationship. Calculate the start time of the material weighing stage, the start time of the mixing stage, and the start time of the unloading stage for each concrete production task order based on the task execution time points. Based on the start time of the material weighing stage, generate door opening control commands and weighing instrument start commands for different aggregate bins and powder bins, and generate a material weighing control command sub-sequence; Based on the start time of the mixing stage, generate motor start command, water supply valve opening command and admixture pump start command for the specified mixing host, and generate mixing host control command sub-sequence; Based on the start time of the unloading stage, an instruction to control the opening of the unloading gate of the mixing host and a signal instruction to control the positioning of the concrete transport vehicle are generated, and an unloading control instruction subsequence is generated. The material weighing control instruction subsequence, the mixing host control instruction subsequence, and the unloading control instruction subsequence are integrated and time-aligned according to the chronological order of task execution time to generate a complete concrete mixing plant equipment control instruction sequence; Based on the volume attributes of the concrete production task order corresponding to each task scheduling time block in the set of concrete production optimization scheduling strategies and the required proportions of various materials, calculate the quantity of various materials required for the concrete production task, locate and accumulate the demand of each material on the time axis according to the start time of its corresponding material weighing stage, and generate a material demand time curve describing the demand quantity of various materials at different time points. Based on the material demand time curve and the current storage status of the material conveying device, calculate the material replenishment trigger time and replenishment amount; Based on the material replenishment trigger time and replenishment amount, commands are generated to control the material conveying device to start, stop, and adjust its speed, thus generating a material delivery timing command sequence. The control command sequence of the concrete mixing plant equipment and the material delivery timing command sequence are encapsulated into a unified format scheduling command message. The scheduling command message is sent to the corresponding mixing plant control unit through the industrial communication network. The mixing plant control unit includes a mixing tower programmable logic controller and a material central control programmable logic controller. The mixing plant control unit receives and parses the scheduling command message and drives the corresponding actuator to complete the concrete production operation.

9. A system for generating scheduling strategies for concrete mixing plants using deep learning, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the concrete mixing plant scheduling strategy generation method according to any one of claims 1 to 8 via executing the machine-executable instructions.

Images