Blanking method, apparatus, and medium

Abstract

The application discloses a discharging method, device and medium, comprising: acquiring real-time state data of a target device discharging process; wherein the real-time state data comprises a current discharging weight, a target discharging amount and a remaining error; inputting the real-time state data into a pre-trained reinforcement learning model, so that the reinforcement learning model maps the real-time state data into an action vector containing a rotating speed control value and a stage flag value; parsing a stage switching instruction and a rotating speed adjusting instruction from the action vector; according to the stage switching instruction, controlling the target device to discharge in stages according to at least two different discharging stages; and according to the rotating speed adjusting instruction, adjusting the driving rotating speed of the target device to control the discharging flow of each discharging stage. The application adaptively generates stage switching and rotating speed instructions through reinforcement learning, realizes precise segmented discharging and flow adjustment, and improves the discharging precision.

Description

Technical Field

[0001] This application belongs to the field of intelligent control technology, and in particular relates to a feeding method, system, equipment, medium and product. Background Technology

[0002] In large-scale farming, feed costs account for 60% to 70% of the total farming cost. Insufficient feed dispensing precision directly leads to a high feed waste rate, which has become a core factor restricting the improvement of farming efficiency.

[0003] In related technologies, timed feeding or PID control based on weighing feedback is commonly used to achieve feed dispensing. However, feed types are diverse (pellets, powders, mixtures), with significant differences in density, moisture content, and flowability. Existing control methods are difficult to adapt to these differences, and fixed control parameters cannot be dynamically adjusted according to real-time operating conditions, resulting in significant accuracy drift over long-term operation. Therefore, how to achieve high-precision, adaptive, and low-drift intelligent feeding control in complex feeding scenarios with multiple materials and disturbances has become an urgent technical problem to be solved in this field. Summary of the Invention

[0004] In view of the above, this application provides a feeding method, system, device, medium and product to address some of the deficiencies involved in the background art.

[0005] In a first aspect, embodiments of this application provide a material feeding method, including: Acquire real-time status data of the target equipment's material feeding process; wherein, the real-time status data includes the current feeding weight, target feeding quantity, and remaining error; The real-time state data is input into a pre-trained reinforcement learning model so that the reinforcement learning model maps the real-time state data into an action vector containing speed control values ​​and stage flag values. The stage switching command and speed adjustment command are parsed from the action vector; According to the stage switching instruction, the target device is controlled to perform staged feeding in at least two different feeding stages; and... According to the speed adjustment command, the drive speed of the target device is adjusted to control the material flow rate of each feeding stage.

[0006] In a second aspect, embodiments of this application provide a feeding system, comprising: The data acquisition module is configured to acquire real-time status data of the target device's material feeding process; wherein, the real-time status data includes the current material feeding weight, the target material feeding quantity, and the remaining error; The data mapping module is configured to input the real-time state data into a pre-trained reinforcement learning model, so that the reinforcement learning model maps the real-time state data into an action vector containing speed control values ​​and stage flag values. The instruction generation module is configured to: parse stage switching instructions and speed adjustment instructions from the action vector; The drive feeding module is configured to: control the target device to perform phased feeding in at least two different feeding stages according to the stage switching instruction; and adjust the drive speed of the target device according to the speed adjustment instruction to control the feeding flow rate of each feeding stage.

[0007] In a third aspect, this application provides a computer device including one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and executed by the one or more processors, and the one or more programs include instructions for performing the method described in the first aspect.

[0008] In a fourth aspect, this application provides a non-volatile computer-readable storage medium containing a computer program that, when executed by one or more processors, causes the one or more processors to perform the method described in the first aspect.

[0009] In a fifth aspect, this application provides a computer program product comprising one or more computer programs that, when executed by one or more processors, implement the method described in the first aspect.

[0010] The feeding method, system, equipment, medium, and product provided in this disclosure combine deep reinforcement learning with a multi-stage vector control strategy, transforming the feeding process from being driven by a single parameter to being driven by data-driven adaptive decision-making. This effectively solves the problems of low feeding accuracy, serious feed waste, high dependence on manual intervention, and poor adaptability to different materials in traditional methods, achieving a significant improvement in feeding accuracy and intelligence. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic diagram of a model training process provided in an embodiment of this application is shown; Figure 2 A schematic flowchart of a feeding method provided in an embodiment of this application is shown; Figure 3 This illustration shows a closed-loop feedback correction process according to an embodiment of the present application; Figure 4 This illustration shows a schematic diagram of a blockage detection and vibration-based arch breaking process provided in an embodiment of this application; Figure 5 This paper shows a schematic diagram of the structure of a feeding system provided in an embodiment of this application; Figure 6 A schematic diagram of a more specific device hardware structure provided in an embodiment of this application is shown. Detailed Implementation

[0013] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0014] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0015] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0016] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0017] The term "in response to" as used herein refers to a state in which a corresponding event occurs or a condition is satisfied. It will be understood that the timing of subsequent actions performed in response to such event or condition is not necessarily strongly correlated with the time when the event occurs or the condition is met. For example, in some cases, subsequent actions may be performed immediately upon the occurrence of the event or the fulfillment of the condition; while in others, they may be performed some time after the occurrence of the event or the fulfillment of the condition.

[0018] For clarity, before describing the specific technical solutions of the embodiments of this application, the technical terms involved in the embodiments of this application will be explained first.

[0019] Real-time status data refers to the set of data collected in real time by the target device during the material feeding process, used to describe the current material feeding condition. In the embodiments of this application, the real-time status data includes the current material feeding weight, the target material feeding quantity, and the remaining error, which are used to provide a basis for decision-making for the reinforcement learning model.

[0020] Residual error: refers to the difference between the target material quantity and the current material weight, used to characterize the degree of deviation between the current material feeding progress and the target. In the embodiments of this application, the residual error is the core basis for determining the switching of the material feeding stage.

[0021] Reinforcement learning model: refers to an intelligent decision-making model trained based on reinforcement learning algorithms, used to generate corresponding control actions based on the input real-time state data. In this embodiment, the reinforcement learning model is a proximal policy optimization (PPO) model, which includes a policy network and a value network.

[0022] Policy network: refers to the neural network in a reinforcement learning model used to map state data to action data. In this embodiment, the policy network is used to output an action vector containing speed control values ​​and stage flag values ​​based on real-time state data.

[0023] Value network: refers to the neural network in a reinforcement learning model used to evaluate the value of the current state. In the embodiments of this application, the value network is used to output state value estimates to assist in the training and optimization of the policy network.

[0024] Action vector: refers to the set of multi-dimensional control commands output by the reinforcement learning model. In the embodiments of this application, the action vector includes a speed control value and a stage flag value, which are used to adjust the drive speed and switch the feeding stage, respectively.

[0025] Rotational speed control value: refers to the control parameter output by the reinforcement learning model used to set the driving rotational speed of the target device. In this embodiment, the rotational speed control value is parsed to generate a rotational speed adjustment command, which is used to control the material flow rate at each feeding stage.

[0026] Stage flag value: refers to the identifier parameter output by the reinforcement learning model used to indicate the switching of the material feeding stage. In the embodiments of this application, the stage flag value is parsed to generate a stage switching instruction, which is used to control the target equipment to switch between different material feeding stages.

[0027] Weighing feedback data: refers to the current weight data of the target equipment collected in real time by the weighing sensor during the feeding process. In the embodiments of this application, the weighing feedback data is used to update the real-time status data and realize closed-loop correction of the feeding process.

[0028] Simulation interaction data: refers to simulated interaction data generated based on a simulation model of the material feeding process, used for offline pre-training of reinforcement learning models. In the embodiments of this application, the simulation interaction data includes simulation state data and simulation actions.

[0029] Actual interaction data refers to the real interaction data collected by the target device during the actual material feeding process. In this embodiment, the actual interaction data is used for online iterative optimization of the reinforcement learning model.

[0030] As described in the background section, the relevant technologies have many problems in the context of livestock feeding. Specifically: First, the timed feeding method in the relevant technology relies entirely on manually set fixed time and fixed rotation speed for feeding. Although it is simple to operate, it cannot adapt to the significant differences in density, moisture content, and particle shape of different feed types. The actual output of the same set parameters can vary by 5% to 8% in different batches of feed, resulting in a high feed waste rate and failing to meet the needs of large-scale farming for precise feeding.

[0031] Secondly, the single PID closed-loop control method in related technologies uses the feedback signal of the weighing sensor to adjust the feeding process in real time. However, the control parameters (proportional, integral, and derivative coefficients) need to be manually tuned and remain fixed. This makes it difficult to adapt to dynamic working conditions such as changes in silo level and fluctuations in feed flowability. When the environment or material characteristics change, repeated manual parameter adjustments are required, resulting in unstable control performance.

[0032] Furthermore, the traditional open-loop control method in related technologies uses a frequency converter to drive the motor to feed at a constant speed, which lacks real-time sensing and adaptive adjustment capabilities. When a slight blockage occurs at the discharge port, the feed brittles, or the material level in the silo drops suddenly, the system cannot automatically identify the abnormality and adjust the control strategy. It can only rely on manual inspection and intervention, which results in a delayed response and low efficiency.

[0033] Furthermore, although the combined control schemes in related technologies simply superimpose timing control and PID control, they lack a unified collaborative decision-making mechanism between the various control modes. They merely switch between different control methods under different conditions, failing to achieve true collaborative optimization. At the same time, traditional schemes do not introduce intelligent decision-making and online iteration mechanisms, making it impossible to continuously optimize control strategies based on historical material feeding data. After long-term operation, factors such as equipment wear and environmental changes can lead to ineffective suppression of accuracy drift.

[0034] Therefore, how to overcome the inherent defects of a single control method in complex material feeding scenarios with multiple materials, multiple disturbances, and multiple operating conditions, and achieve adaptive switching and precise coordinated control of flow rate during the feeding stage, thereby eliminating long-term accuracy drift, has become a technical problem that urgently needs to be solved in this field.

[0035] In large-scale farming, feed costs account for 60% to 70% of the total cost. Related technologies typically employ timed feeding or PID control based on weighing feedback to achieve feed delivery. However, these methods struggle to adapt to the significant differences in density, moisture content, and flowability among different feed types. Furthermore, fixed control parameters cannot be dynamically adjusted according to real-time operating conditions, leading to significant accuracy drift over long-term operation.

[0036] To achieve precise control of the material feeding process, this application introduces a reinforcement learning model. This model adaptively generates material feeding control commands based on real-time status data, replacing the traditional fixed parameter control method. This allows the material feeding strategy to be dynamically adjusted according to changes in working conditions, improving material feeding accuracy and environmental adaptability.

[0037] Figure 1 This illustration shows a schematic diagram of a model training process provided in an embodiment of this application. Figure 1 As shown, the reinforcement learning model is trained in the following way: In some optional embodiments, simulation interaction data is generated based on a simulation model of the target equipment's material unloading process. This simulation interaction data may include simulation state data, such as the current simulated unloading weight, the target simulated unloading quantity, and the remaining simulation error. The advantage of the simulation model is that it can generate a large number of training samples in a risk-free environment, accelerating model convergence.

[0038] In some optional embodiments, an initial policy network is obtained through offline pre-training using simulated interactive data. Specifically, simulated state data is input into the policy network to be trained, causing the policy network to output simulated actions; simulated state data is also input into a pre-defined value network, causing the value network to output state value estimates. The value network is used to evaluate the quality of the current state, providing a reference benchmark for optimizing the policy network.

[0039] The reward value is calculated based on the simulated actions and state data. The reward function can be a multi-objective comprehensive reward function, simultaneously constraining material feeding accuracy, overshoot, response time, and process stability, so that the policy network gradually learns the optimal control strategy that balances efficiency and accuracy during training. For example, a positive reward is given when the material feeding error is small, there is no overshoot, and the speed is fast; a negative reward is given when the material feeding error is large or overshoot occurs.

[0040] The advantage estimate is calculated based on the state value estimate and the reward value. The advantage estimate measures how good or bad the current action is relative to the average level, and it can be calculated as the difference between the reward value and the state value estimate. The parameters of the policy network are updated using the advantage estimate. Specifically, the PPO-Clip objective function can be used to limit the policy update magnitude to the range [1-ε, 1+ε], where ε=0.2, to avoid policy abrupt changes that could lead to training instability. This process is repeated until the convergence condition is met (e.g., the average reward change is less than 3% over 100 consecutive training epochs), resulting in the trained initial policy network.

[0041] In some optional embodiments, an initial policy network is deployed to the target device, and actual interaction data is continuously collected during the actual material feeding process of the target device. The actual interaction data may include real-time status data of the target device, actions output by the policy network, the new status after executing the actions, and reward values.

[0042] In some optional embodiments, the initial policy network is iteratively optimized based on actual interaction data to obtain a trained reinforcement learning model. Through online iteration, the model can adapt to nonlinear and time-varying disturbances in the real environment (such as feed type switching, changes in silo level, equipment wear, etc.), achieving continuous optimization of the control strategy.

[0043] Figure 2 This illustration shows a flowchart of a feeding method provided in an embodiment of this application; as shown Figure 2 As shown, the material cutting method in this application specifically includes: S102: Obtain real-time status data of the target equipment during the material unloading process.

[0044] To achieve precise control of the material feeding process, it is first necessary to acquire real-time status data that characterizes the current feeding conditions, providing a data foundation for subsequent model decisions. This real-time status data includes the current feeding weight, target feeding quantity, and remaining error.

[0045] In this embodiment, the target device can be a feeder, a feeder, or other aquaculture equipment with feeding functions, and can be equipped with components such as a weighing sensor, a controller, and a drive motor. The weighing sensor is used to collect the current weight of the fed material in real time, the controller is used to perform model inference and logical judgment, and the drive motor is used to adjust the feeding speed according to control commands.

[0046] In some optional embodiments, the real-time state data may also include the target device's feeding rate, error change rate, material type code, and frequency, thereby forming a seven-dimensional state vector containing the current feeding weight, target feeding quantity, remaining error, feeding rate, error change rate, material type code, and frequency. This seven-dimensional state vector covers the complete information chain of "target—current status—deviation—trend—operating condition—execution," providing sufficient decision-making basis for the reinforcement learning model.

[0047] In some optional embodiments, the residual error is the difference between the current feeding weight and the target feeding amount, used to characterize the degree of deviation between the current feeding progress and the target. It is understood that a larger residual error indicates that the current feeding amount is further from the target, and a higher flow rate is needed to shorten the feeding time; a smaller residual error indicates that the current feeding amount is close to the target, and a lower flow rate is needed to avoid overshoot.

[0048] In some optional embodiments, the current feed weight can be acquired in real time by a weighing sensor. The material type code can be a uniquely encoded value obtained after thermally encoding pelleted feed, powdered feed, or mixed feed, enabling the model to perceive the differences in flowability of different materials. The current frequency of the actuator can be the operating frequency of the drive motor at the current moment, reflecting the current operating status of the actuator.

[0049] S104: Input the real-time state data into a pre-trained reinforcement learning model so that the reinforcement learning model maps the real-time state data into an action vector containing speed control values ​​and stage flag values.

[0050] After acquiring real-time state data, this data can be used as input to provide a pre-trained reinforcement learning model for decision-making and reasoning.

[0051] In this embodiment, the reinforcement learning model can be a Proximal Policy Optimization (PPO) model, which includes a policy network and a value network. The policy network is used to output corresponding actions based on the input state data, and the value network is used to evaluate the value of the current state to assist training. The advantages of using the PPO model are its good training stability, fast convergence speed, and ability to handle continuous action spaces, making it suitable for the continuous adjustment requirements of the feeding speed.

[0052] In some optional embodiments, real-time state data (such as the seven-dimensional state vector mentioned above) is input into a policy network. The policy network performs layer-by-layer feature extraction and nonlinear transformation on the real-time state data and outputs an action vector. The dimension of the action vector corresponds to the sum of the number of speed control values ​​and stage flag values. The policy network may include an input layer, a first hidden layer, and a second hidden layer connected in sequence.

[0053] Furthermore, the input layer receives a seven-dimensional state vector, the first hidden layer performs a first linear transformation on the input data and then processes it through an activation function, the second hidden layer performs a second linear transformation on the output of the first hidden layer and then processes it through an activation function, the output layer performs a third linear transformation on the output of the second hidden layer, and finally outputs an action vector.

[0054] In some optional embodiments, the action vector is a six-dimensional action vector. The first dimension of the six-dimensional action vector outputs a speed control value, ranging from 0 to 3000 rpm; the fourth dimension outputs a stage flag value. The stage switching command is obtained by discretizing the fourth dimension output value (e.g., taking the category with the highest probability). The value of the stage switching command can be a coarse input stage, a fine input stage, or a compensation stage.

[0055] S106: Extract the stage switching command and speed adjustment command from the action vector.

[0056] After the policy network outputs action vectors, it is necessary to parse out the specific control instructions that can be executed.

[0057] In some optional embodiments, the speed control value and the stage flag value are parsed from the six-dimensional motion vector. The speed control value is directly used as a speed adjustment command to set the drive speed of the target equipment; the stage flag value is discretized and used as a stage switching command to indicate the current unloading stage that the target equipment should enter.

[0058] The discretization of the stage flag value can be achieved by taking the maximum value, that is, taking the class with the largest probability among all classes output by the policy network as the current stage switching instruction.

[0059] S108: According to the stage switching instruction, control the target device to perform staged feeding in at least two different feeding stages.

[0060] After parsing the stage switching instruction, the target equipment can be controlled to switch between different feeding stages according to the instruction, realizing multi-stage vector control of "coarse feeding → fine feeding → compensation → stabilization".

[0061] In some optional embodiments, when the remaining error exceeds a first threshold, the target device is controlled to enter the coarse feeding stage. The coarse feeding stage involves rapid material feeding at a high flow rate, with the aim of quickly reducing the remaining error.

[0062] In some optional embodiments, when the remaining error is less than or equal to a first threshold and greater than a second threshold, the target device is controlled to enter the fine feeding stage. The fine feeding stage uses a medium flow rate to continue approaching the target while ensuring accuracy.

[0063] In some optional embodiments, when the remaining error is less than a third threshold, the target device is controlled to enter the compensation phase. The compensation phase involves feeding material sequentially at small flow rates, with the aim of accurately approximating the target with minute steps and avoiding overshoot.

[0064] In some optional embodiments, when the remaining error is within a preset range (e.g., within ±2g), the target device is controlled to enter a stabilization phase and the feeding stops. The purpose of the stabilization phase is to confirm that the feeding result is stable and then lock in the final weight.

[0065] S110: Adjust the drive speed of the target device according to the speed adjustment command to control the feeding flow rate of each feeding stage.

[0066] After determining the current feeding stage, it is also necessary to adjust the drive speed according to the speed adjustment command to control the actual feeding flow rate at each stage.

[0067] In some optional embodiments, during the coarse feeding stage, the target device is controlled to operate at a first rotational speed and feed material at a first flow rate; during the fine feeding stage, the target device is controlled to operate at a second rotational speed and feed material at a second flow rate; and during the compensation stage, the target device is controlled to operate at a third rotational speed and replenish material sequentially at a third flow rate. The first flow rate is greater than the second flow rate, and the third flow rate is calculated by the reinforcement learning model based on real-time state data.

[0068] By controlling the corresponding stages and rotation speed, a coordinated optimization is achieved: "rapidly approaching the large flow rate during coarse feeding, improving accuracy with medium flow rate during fine feeding, and gradually replenishing with small flow rate to compensate."

[0069] As mentioned earlier, the reinforcement learning model generates speed adjustment and stage switching commands based on the current real-time state data, and the target equipment executes the material feeding operation accordingly. However, in the actual feeding process, due to various factors such as material flowability fluctuations, mechanical vibration, and weighing lag, there is inevitably a deviation between the actual feeding quantity and the expected value. If feeding is executed solely based on the result of a single model inference without dynamic adjustments based on the actual weighing results, this deviation will continue to accumulate, making it difficult to guarantee the final feeding accuracy.

[0070] Based on this, the embodiments of this application introduce a closed-loop feedback correction mechanism throughout the entire material feeding process. Figure 3 This illustration shows a closed-loop feedback correction process according to an embodiment of this application. Figure 3 As shown, the mechanism includes the following steps.

[0071] In some optional embodiments, weighing feedback data of the target device is acquired. The weighing feedback data is the current material weight collected in real time during the material unloading process of the target device. Real-time status data is updated based on the weighing feedback data; for example, the weighing feedback data is used as the new current material weight, and status parameters such as the remaining error are recalculated.

[0072] In some optional embodiments, the updated real-time state data is re-input into the reinforcement learning model, enabling the model to iteratively generate new speed adjustment commands and stage switching commands. Based on these new commands, the target equipment is controlled to unload materials. Through this continuous "perception → decision → execution → feedback" cycle, real-time closed-loop correction of the unloading process is achieved, gradually bringing the actual unloading amount closer to the target amount.

[0073] Furthermore, to suppress flow fluctuations during the feeding process on a shorter timescale, embodiments of this application also provide a dual-closed-loop PID correction mechanism at the execution layer. In some optional embodiments, during the feeding process, the speed loop controls the stability of the drive speed of the target device to stabilize the feeding flow rate, while the weight loop dynamically adjusts the speed loop setting value based on the deviation between the actual feeding weight and the target feeding amount.

[0074] The speed loop and weight loop constitute a dual-closed-loop PID control system. The speed loop, as the inner loop, is responsible for quickly responding to changes in rotational speed and suppressing the impact of motor load fluctuations on speed. The weight loop, as the outer loop, is responsible for continuously correcting the feeding progress based on weighing feedback. The reinforcement learning loop handles stage switching and speed setting at the decision-making level, while the PID loop handles speed stabilization and flow correction at the execution level. The two loops work together to ensure feeding accuracy.

[0075] During the feeding process, when the feed has a high moisture content, strong electrostatic adsorption between particles, or when materials accumulate in the feed hopper for a long time, the feed is prone to forming an arched accumulation structure (i.e., "arching" or "bridging") at the feed hopper's constriction section or discharge port. This can lead to blockage of the feeding channel, a sudden drop in the feeding rate, or even a complete halt in feeding. If this blockage is not detected and addressed in time, it can cause feeding failures, resulting in feeding interruptions and impacting livestock production.

[0076] Related technologies typically rely on manual inspection to detect material blockages, which is slow to respond and inefficient. To address this issue, the embodiments of this application introduce an automatic material blockage detection and vibration-based arch-breaking mechanism during the staged material feeding process.

[0077] Figure 4 This application provides a schematic diagram of a blockage detection and vibration-based arch breaking process. Figure 4 As shown, the mechanism includes the following steps: In some optional embodiments, the feeding rate of the target device is acquired in real time. When the feeding rate is detected to be consistently lower than a preset rate threshold for a first preset time period, or when the current feeding weight remains unchanged for a second preset time period, it is determined that the target device is experiencing a blockage. These two determination conditions complement each other: the former determines blockage based on a decrease in the feeding rate, while the latter determines blockage based on a stagnation in weight. This approach can cover different types and degrees of blockage scenarios, improving the reliability and sensitivity of blockage detection.

[0078] In some optional embodiments, the vibration frequency value is calculated based on the state data of the target device at the moment of material blockage. The state data at the moment of material blockage may include information such as the current material type and the degree of blockage. Different states correspond to different vibration frequencies to achieve a more effective arch-breaking effect.

[0079] In some alternative embodiments, vibration is applied to the target equipment based on the vibration frequency value to disrupt the accumulation structure causing the blockage. The feeding rate can be continuously monitored during vibration, and vibration automatically stops when the feeding rate returns to the normal range. This responsive vibration control enables on-demand start-up and automatic shutdown, avoiding equipment wear and energy waste caused by continuous vibration.

[0080] Compared with related technologies, this application has the following advantages: First, by introducing a reinforcement learning model to adaptively generate stage switching instructions and speed adjustment instructions based on real-time status data, the feeding strategy can be dynamically adjusted according to different material types and different residual errors. This effectively avoids the problem of decreased accuracy in traditional fixed parameter control when material characteristics change, and significantly improves the feeding accuracy under different working conditions.

[0081] Second, by setting up a multi-stage vector control strategy with a coarse feeding stage, a fine feeding stage, a compensation stage, and a stabilization stage, the feeding flow rate gradually transitions from a large flow rate to a small flow rate as it approaches the target feeding amount. This not only ensures feeding efficiency but also effectively suppresses overshoot, achieving coordinated optimization control of "fast approach, slow arrival, and precise stop".

[0082] Third, by acquiring weighing feedback data and continuously updating real-time status data based on the feedback data, the updated status data is re-input into the reinforcement learning model to iteratively generate new control commands, forming a complete closed loop of "perception → decision → execution → feedback", which realizes real-time correction of the feeding process and effectively eliminates the accuracy drift caused by factors such as equipment wear and environmental changes during long-term operation.

[0083] Fourth, through a dual closed-loop PID control system consisting of a speed loop and a weight loop, the drive speed is stably controlled in real time at the execution level. On the basis of the reinforcement learning closed loop at the decision level, the execution level closed-loop correction is added. The two work together to further improve the stability of flow control and the feeding accuracy.

[0084] Fifth, by monitoring the feeding rate in real time, when the feeding rate is continuously lower than the preset threshold or the weight does not change for a long time, the system automatically determines that the material is blocked. Based on the state data at the time of the blockage, the vibration frequency value is calculated and vibration is applied. This realizes the automatic perception and automatic handling of the blockage scenario, avoids the lag of manual inspection, and ensures the continuity and reliability of the feeding process.

[0085] In summary, the feeding method, system, equipment, medium, and product provided in this application effectively solve the technical problems in the prior art, such as low feeding accuracy, poor material adaptability, lack of closed-loop iteration mechanism, long-term operation accuracy drift, and reliance on manual intervention for material blockage. It achieves high-precision, adaptive, and low-drift intelligent feeding control in complex feeding scenarios with multiple materials, multiple disturbances, and multiple working conditions.

[0086] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides a feeding system.

[0087] Figure 5 A schematic diagram of the structure of a feeding system provided in an embodiment of this application is shown.

[0088] refer to Figure 5 As shown, the feeding system includes: The data acquisition module 202 is configured to acquire real-time status data of the target device's material feeding process; wherein the real-time status data includes the current feeding weight, the target feeding quantity, and the remaining error.

[0089] The data mapping module 204 is configured to input the real-time state data into a pre-trained reinforcement learning model, so that the reinforcement learning model maps the real-time state data into an action vector containing speed control values ​​and stage flag values.

[0090] The instruction generation module 206 is configured to parse stage switching instructions and speed adjustment instructions from the action vector.

[0091] The stage switching module 208 is configured to control the target device to perform staged feeding in at least two different feeding stages according to the stage switching instruction.

[0092] In some optional embodiments, the phase switching module 208 further performs the following steps: When the remaining error is greater than the first threshold, the target device is controlled to enter the coarse-powder stage; When the remaining error is less than or equal to the first threshold and greater than the second threshold, the target device is controlled to enter the fine-throw stage; Wherein, the remaining error is the difference between the current material feeding weight and the target material feeding amount; When the remaining error is less than the third threshold, the target device is controlled to enter the compensation phase; When the remaining error is within the preset range, the target device is controlled to enter a stable phase and the feeding is stopped; And, during the process of controlling the target device to perform staged feeding in at least two different feeding stages, the following continues to be executed: The material feeding rate of the target device is obtained in real time; When the feeding rate is continuously lower than the preset rate threshold for a first preset time, or when the current feeding weight does not change for a second preset time, it is determined that the target equipment is blocked. The vibration frequency value is calculated based on the state data of the target device at the time of material blockage; Vibration is applied to the target device based on the vibration frequency value to disrupt the accumulation structure that causes the blockage.

[0093] The speed adjustment module 210 is configured to adjust the drive speed of the target device according to the speed adjustment command, so as to control the feeding flow rate of each feeding stage.

[0094] In some optional embodiments, the speed regulation module 210 further performs the following steps: When the feeding stage is the coarse feeding stage, the target equipment is controlled to run at a first speed and feed at a first flow rate; When the feeding stage is the fine feeding stage, the target equipment is controlled to run at a second speed and feed at a second flow rate; When the feeding stage is the compensation stage, the target equipment is controlled to run at the third speed and the material is replenished sequentially at the third flow rate. Wherein, the first flow rate is greater than the second flow rate, and the third flow rate is calculated by the reinforcement learning model based on the real-time state data.

[0095] In some optional embodiments, the feeding control system further includes a model training module 212 and an instruction update module 214; The model training module 212 further performs the following steps: Simulation interaction data is generated based on the simulation model of the material unloading process of the target equipment; wherein, the simulation interaction data includes simulation state data; The initial policy network is obtained by offline pre-training using the simulated interaction data; including: Simulation interaction data is generated based on the simulation model of the material unloading process of the target equipment; wherein, the simulation interaction data includes simulation state data; The initial policy network is obtained by offline pre-training using the simulated interaction data. The initial strategy network is deployed to the target device, and actual interaction data is continuously collected during the actual material unloading process of the target device. The initial policy network is iteratively optimized based on the actual interaction data to obtain the trained reinforcement learning model. The initial strategy network is deployed to the target device, and actual interaction data is continuously collected during the actual material unloading process of the target device. The initial policy network is iteratively optimized based on the actual interaction data to obtain the trained reinforcement learning model.

[0096] The instruction update module 214 further performs the following steps: Obtain the weighing feedback data of the target device; wherein, the weighing feedback data is the current unloading weight collected in real time during the unloading process of the target device; The real-time status data is updated based on the weighing feedback data; The updated real-time status data is re-input into the reinforcement learning model so that the reinforcement learning model iteratively generates new speed adjustment commands and stage switching commands. The target device is controlled to unload materials according to the new speed adjustment command and the stage switching command.

[0097] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing this application, the functions of each module can be implemented in one or more software and / or hardware.

[0098] The apparatus described above can be used to implement the corresponding feeding method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0099] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the feeding method described in any of the above embodiments.

[0100] Figure 6 This embodiment illustrates a more specific hardware structure of a computer device, which may include a processor 810, a memory 820, an input / output interface 830, a communication interface 840, and a bus 850. The processor 810, memory 820, input / output interface 830, and communication interface 840 are interconnected internally via the bus 850.

[0101] The processor 810 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the blanking method provided in any of the foregoing embodiments of this specification.

[0102] The memory 820 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 820 can store operating devices and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program code is stored in the memory 820 and is called and executed by the processor 810.

[0103] The input / output interface 830 is used to connect input / output modules to enable information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0104] The communication interface 840 is used to connect the communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0105] Bus 850 includes a pathway for transmitting information between various components of the device, such as processor 810, memory 820, input / output interface 830, and communication interface 840.

[0106] It should be noted that although the above-described device only shows the processor 810, memory 820, input / output interface 830, communication interface 840, and bus 850, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of this specification, and not necessarily all the components shown in the figures.

[0107] The electronic devices described above are used to implement the corresponding feeding methods in any of the foregoing embodiments and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0108] Based on the same inventive concept, corresponding to the methods of any of the above embodiments, this disclosure also provides a non-transitory computer-readable storage medium storing computer instructions for causing the computer to execute the feeding method as described in any of the above embodiments.

[0109] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0110] The aforementioned non-transitory computer-readable storage media can be any available medium or data storage device that a computer can access, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MOs), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0111] The computer instructions stored in the storage medium of the above embodiments are used to cause the computer to execute the feeding method as described in any of the embodiments in the exemplary method section above, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0112] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0113] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0114] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0115] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A feeding method, characterized in that, include: Acquire real-time status data of the target equipment's material feeding process; wherein, the real-time status data includes the current feeding weight, target feeding quantity, and remaining error; The real-time state data is input into a pre-trained reinforcement learning model so that the reinforcement learning model maps the real-time state data into an action vector containing speed control values ​​and stage flag values. The stage switching command and speed adjustment command are parsed from the action vector; According to the stage switching instruction, the target device is controlled to perform staged feeding in at least two different feeding stages; and... According to the speed adjustment command, the drive speed of the target device is adjusted to control the material flow rate of each feeding stage.

2. The feeding method according to claim 1, characterized in that, The reinforcement learning model is trained through the following steps: Simulation interaction data is generated based on the simulation model of the material unloading process of the target equipment; wherein, the simulation interaction data includes simulation state data; The initial policy network is obtained by offline pre-training using the simulated interaction data. The initial strategy network is deployed to the target device, and actual interaction data is continuously collected during the actual material unloading process of the target device. The initial policy network is iteratively optimized based on the actual interaction data to obtain the trained reinforcement learning model.

3. The feeding method according to claim 2, characterized in that, The initial policy network is obtained by offline pre-training using the simulated interaction data, including: The simulated state data is input into the policy network to be trained, so that the policy network outputs simulated actions; wherein, the simulated state data includes the current simulated material feeding weight, the target simulated material feeding amount, and the remaining simulation error; The simulated state data is input into a preset value network so that the value network outputs a state value estimate. Calculate the reward value based on the simulated actions and the simulated state data; Calculate the advantage estimate based on the state value estimate and the reward value; The parameters of the policy network are updated using the advantage estimate to obtain the trained initial policy network.

4. The feeding method according to claim 1, characterized in that, The control of the target device to perform phased feeding in at least two different feeding stages includes: When the remaining error is greater than the first threshold, the target device is controlled to enter the coarse-powder stage; When the remaining error is less than or equal to the first threshold and greater than the second threshold, the target device is controlled to enter the fine-throw stage; The remaining error is the difference between the current material weight and the target material quantity.

5. The feeding method according to claim 4, characterized in that, The control of the target device to perform phased feeding in at least two different feeding stages also includes: When the remaining error is less than the third threshold, the target device is controlled to enter the compensation phase; When the remaining error is within the preset range, the target device is controlled to enter a stable phase and the feeding is stopped.

6. The feeding method according to claim 1, characterized in that, In the process of controlling the target device to perform staged feeding in at least two different feeding stages, the method further includes: The material feeding rate of the target device is obtained in real time; When the feeding rate is continuously lower than the preset rate threshold for a first preset time, or when the current feeding weight does not change for a second preset time, it is determined that the target equipment is blocked. The vibration frequency value is calculated based on the state data of the target device at the time of material blockage; Vibration is applied to the target device based on the vibration frequency value to disrupt the accumulation structure that causes the blockage.

7. The feeding method according to claim 5, characterized in that, The adjustment of the drive speed of the target device to control the feeding flow rate in each feeding stage includes: When the feeding stage is the coarse feeding stage, the target equipment is controlled to run at a first speed and feed at a first flow rate; When the feeding stage is the fine feeding stage, the target equipment is controlled to run at a second speed and feed at a second flow rate; When the feeding stage is the compensation stage, the target equipment is controlled to run at the third speed and the material is replenished sequentially at the third flow rate. Wherein, the first flow rate is greater than the second flow rate, and the third flow rate is calculated by the reinforcement learning model based on the real-time state data.

8. The feeding method according to claim 1, characterized in that, The method further includes: Obtain the weighing feedback data of the target device; wherein, the weighing feedback data is the current unloading weight collected in real time during the unloading process of the target device; The real-time status data is updated based on the weighing feedback data; The updated real-time status data is re-input into the reinforcement learning model so that the reinforcement learning model iteratively generates new speed adjustment commands and stage switching commands. The target equipment is controlled to unload materials according to the new speed adjustment command and the stage switching command.

9. A computer device, characterized in that, It includes one or more processors, memory; and one or more programs, wherein the one or more programs are stored in the memory and executed by the one or more processors, and the one or more programs include instructions for performing the method of any one of claims 1 to 8.

10. A non-volatile computer-readable storage medium containing a computer program, characterized in that, When the computer program is executed by one or more processors, the one or more processors perform the method according to any one of claims 1 to 8.

Images