Multi-machine operation energy-saving intelligent control system of air blower

Abstract

The invention relates to the field of multi-machine operation control systems, and discloses a multi-machine operation energy-saving intelligent control system for an air blower, and the system comprises a data collection and fusion module which is used for collecting and fusing operation parameters and environment parameters of the air blower; the prediction modeling module constructs a system load prediction model based on historical data; the multi-agent decision module generates a control strategy according to the current state; the collaborative optimization module carries out non-cooperative strategy coordination among the multiple air blowers; and the parameter updating module updates the prediction model and the control strategy through the control error and training feedback. All the modules work cooperatively, and efficient energy-saving control and dynamic optimization of the system are achieved. The multi-agent collaborative optimization technology is adopted, the overall energy efficiency of the system is guaranteed, meanwhile, the load fluctuation of a single air blower is reduced to the maximum extent, the service life of equipment is prolonged, the problem of disordered scheduling when the load fluctuation is large in a traditional control method is solved, and therefore energy consumption is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the field of multi-machine operation control systems, and particularly to an energy-saving intelligent control system for multi-machine operation of blowers. Background Art

[0002] In modern industry and daily life, energy conservation and system optimization have always been major technical requirements. As one of the devices with large energy consumption, blowers are widely used in various industrial applications. Traditional single-device control methods often fail to effectively coordinate the operation of multiple devices, resulting in low system operation efficiency, high energy consumption, and inability to adapt to external environmental changes.

[0003] In the prior art, optimization strategies based on single-machine control or static model predictive systems are usually adopted to manage device operation. The single-machine control method mainly achieves energy-saving effects by adjusting the operation parameters of a single device, but it lacks a global perspective and cannot take into account the collaborative optimization between multiple devices. The static model predictive system operates through a pre-set model and has poor adaptability. Although they can achieve good energy-saving effects under certain stable conditions, they often perform poorly in an environment with large load fluctuations and cannot automatically correct model errors, resulting in the control strategy being unable to flexibly respond to the changing system state.

[0004] However, there are still some obvious deficiencies in the prior art in practical applications. First, the strategy based on single-machine control cannot coordinate the operation of multiple devices, resulting in uneven load distribution within the system, thereby causing energy waste and equipment wear. Second, the static model predictive method lacks adaptability to long-term changes in the system. When the environment or device state changes, the control accuracy of the system drops significantly. Finally, the fault tolerance mechanism in the existing control methods is weak. Once a device fails, manual intervention is often required, affecting the automation degree and stability of the system. For this reason, those skilled in the art have proposed an energy-saving intelligent control system for multi-machine operation of blowers to solve the above problems. Summary of the Invention

[0005] In view of the deficiencies of the prior art, the present invention provides an energy-saving intelligent control system for multi-machine operation of blowers, which solves the problems of poor coordination of multiple devices, poor system adaptability, and insufficient fault recovery ability in the prior art.

[0006] To achieve the above object, the present invention is realized through the following technical solutions: An energy-saving intelligent control system for multi-machine operation of blowers, comprising:

[0007] A data acquisition and fusion module, configured to collect the operation parameters and environmental parameters of the blower and perform fusion processing;

[0008] A prediction modeling module, connected to the data acquisition and fusion module, for constructing a system load prediction model based on historical input and output data;

[0009] A multi-agent decision-making module, connected to the data acquisition and fusion module and the prediction modeling module, for generating a control strategy based on the current state vector;

[0010] A collaborative optimization module, connected to the multi-agent decision-making module, for coordinating non-cooperative strategies among multiple blowers;

[0011] A parameter update module, connected to the prediction modeling module and the multi-agent decision-making module, for updating the prediction model and control strategy based on the control error and training feedback.

[0012] Preferably, the data acquisition and fusion module includes:

[0013] An operating parameter acquisition unit, for acquiring the pressure value, flow value, rotational speed value, energy consumption value, and vibration spectrum data of each blower;

[0014] An environmental parameter acquisition unit, for acquiring the external environmental temperature value of the system and the real-time electricity price;

[0015] A data fusion processing unit, for performing principal component analysis on the vibration spectrum data, generating a dimensionality-reduced feature vector, and constructing a system state vector, where the state vector includes the pressure value, flow value, rotational speed value, energy consumption value, vibration feature vector, environmental temperature value, and electricity price.

[0016] Preferably, the prediction modeling module includes:

[0017] A step response modeling unit, for constructing a response matrix of the control input to the system output;

[0018] A free response estimation unit, for calculating the system output trend in the absence of a control input;

[0019] A multi-step prediction generation unit, for combining the control input increment and the free response, and outputting the system predicted load at several future moments.

[0020] Preferably, the multi-agent decision-making module includes:

[0021] A state processing unit, for inputting the state vectors of each blower into their respective policy networks;

[0022] A policy generation unit, for outputting the control actions of the blower, where the control actions include start / stop signals and rotational speed change amounts;

[0023] A reward construction unit, for constructing a local reward value for a single blower based on the energy consumption value, rotational speed deviation value, and start / stop times.

[0024] Preferably, the collaborative optimization module includes:

[0025] A revenue function construction unit for defining a non - cooperative revenue function of multiple blowers, where the revenue function is constructed based on their respective reward functions and system prediction errors;

[0026] An optimization coordination unit for jointly optimizing strategies using the alternating direction method of multipliers;

[0027] An instruction coordination output unit for outputting the final consistent control strategy among multiple blowers.

[0028] Preferably, the parameter update module includes:

[0029] A model update unit for real - time correction of the response matrix using prediction errors;

[0030] A strategy training unit for training the control strategy network based on simulation data;

[0031] A strategy synchronization unit for synchronously updating the policy parameters obtained from offline training to the control network in the edge device.

[0032] Preferably, the dimensionality - reduced feature vector of the data fusion processing unit is the product of the vibration spectrum data and the principal component matrix, and the calculation formula is:

[0033] v pca = W·v;

[0034] where: v is the original vibration spectrum vector; W is the principal component dimensionality - reduction weight matrix; v pca is the output dimensionality - reduced vibration feature vector.

[0035] Preferably, the output prediction value of the multi - step prediction generation unit satisfies the following expression:

[0036] y pred = G·Δu + y free ;

[0037] where: G represents the step - response matrix obtained through the response matrix construction unit; Δu represents the input increment vector in the future prediction period; y free is the free - response vector; y pred is the future prediction output.

[0038] Preferably, the reward value of the reward construction unit is calculated according to the following expression:

[0039]

[0040] where: r iis the reward value for the blower i; e i is the energy consumption value of the blower i; e base is the reference energy consumption; ω i is the current rotational speed of the blower i; is the average rotational speed; ω max is the maximum allowable rotational speed; δ is the start / stop change identifier of the blower i between the current step and the previous time step. If the start / stop state changes, it is 1; otherwise, it is 0; α, β, γ are preset weight factors, all of which are positive numbers.

[0041] Preferably, the response matrix of the model update unit is updated through the following expression:

[0042]

[0043] where: G new represents the updated response matrix; G old represents the original response matrix; η is the learning rate parameter, dimensionless, and can be dynamically adjusted according to actual applications; y real is the output vector actually observed by the system; y pred is the predicted vector output by the prediction modeling module; Δu is the increment vector of the control input.

[0044] The present invention provides an intelligent energy-saving control system for multi-unit operation of blowers. It has the following beneficial effects:

[0045] 1. The present invention adopts the multi-agent collaborative optimization technology. While ensuring the overall energy efficiency of the system, it minimizes the load fluctuations of a single blower to the greatest extent, improves the service life of the equipment. Compared with the existing single optimization control scheme, the present invention effectively avoids the negative impact of the "selfish" behavior of a single machine on the overall system, solves the problem of disorderly scheduling of traditional control methods under large load fluctuations, and thus significantly reduces energy consumption.

[0046] 2. The present invention introduces an incremental correction online update mechanism. By real-time monitoring and adjusting the prediction model, it ensures the stability and adaptability of the system during long-term operation. This is different from the static model update method in the prior art, avoiding the accumulation of model deviations and successfully solving the problem of reduced control effect caused by model distortion in traditional methods.

[0047] 3. The combination of offline training and online update of the strategy of the present invention enables the control strategy to be continuously optimized according to the actual operation situation of the system and to be synchronized with each intelligent agent of the system. Different from the scheme in the prior art that can only rely on a single data source for update, the present invention can make full use of historical data and real-time feedback, avoid the problem of lag in strategy update, and improve the decision-making efficiency.

[0048] 4. During the output process of the control instruction, the present invention adopts a fault tolerance mechanism, which can promptly abort the error instruction when the system gives an abnormal feedback, ensuring the stable operation of the system. Compared with the traditional method, the present invention enhances the robustness of the system. Especially in the case of high load or equipment failure, it can avoid the global problems caused by single-point failures, thereby improving the overall reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the system architecture of the present invention;

[0050] Figure 2 It is a schematic diagram of the architecture of the data acquisition and fusion module of the present invention;

[0051] Figure 3 It is a schematic diagram of the architecture of the prediction modeling module of the present invention;

[0052] Figure 4 It is a schematic diagram of the architecture of the multi-agent decision-making module of the present invention;

[0053] Figure 5 It is a schematic diagram of the architecture of the collaborative optimization module of the present invention;

[0054] Figure 6 It is a schematic diagram of the architecture of the parameter update module of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0056] Please refer to the attached Figure 1 - attached Figure 6 , the embodiment of the present invention provides an intelligent control system for energy-saving multi-machine operation of a blower, including:

[0057] A data acquisition and fusion module, configured to collect the operation parameters and environmental parameters of the blower and perform fusion processing;

[0058] Specifically, in the intelligent energy-saving control system for multi-unit operation of blowers involved in this embodiment, the data acquisition and fusion module, as the information input end of the system, undertakes the functions of fully collecting and preprocessing the real-time status and environmental information of the blowers. This module is not only closely coupled with the downstream prediction modeling module but also directly affects the model accuracy and control strategy effectiveness of the overall system. Therefore, when implementing the system, it is necessary to ensure that this module has the ability to collect data with high frequency and high precision and can perform real-time dimensionality reduction and fusion on high-dimensional data streams to form a complete state description.

[0059] Generally, the data acquisition and fusion module includes an operating parameter acquisition unit, an environmental parameter acquisition unit, and a data fusion processing unit. These units cooperate with each other to jointly complete the real-time monitoring of the blower group and the construction of the state vector. In a possible implementation manner of the present invention, this module first collects the key operating parameters of each blower in real time through a sensor network, then integrates the data from different sources, and performs dimensionality reduction analysis to provide a unified input for subsequent model prediction.

[0060] As an option, the system can adopt a variety of hardware such as industrial-grade pressure sensors, flow meters, Hall speed sensors, and watt-hour meters to achieve a comprehensive monitoring of the operating status of the blowers. For example, the acquisition of flow can use a turbine flow meter or an electromagnetic flow meter to ensure measurement accuracy; the acquisition of vibration spectrum data is completed through a high-sensitivity acceleration sensor array, and the vibration frequency range can reach 20 Hz to 20 kHz to meet the needs of mechanical health monitoring.

[0061] In this embodiment, the operating parameter acquisition unit is mainly used to obtain the following data:

[0062] The pressure value p of the blower, with the unit of kilopascal (kPa);

[0063] The flow value q, with the unit of cubic meters per hour (m 3 / h);

[0064] The rotational speed value n, with the unit of revolutions per minute (rpm);

[0065] The real-time energy consumption e, with the unit of kilowatt (kW);

[0066] The vibration spectrum vector v, which is a multi-dimensional array representing the acceleration values of the blower at different frequency points.

[0067] In some embodiments, to ensure real-time performance, the acquisition period of pressure and flow data is set to 1 second, and the vibration spectrum data can adopt a batch sampling mode with a window of 10 seconds to achieve both real-time and accurate state perception.

[0068] The environmental parameter acquisition unit is used to acquire the key influencing factors in the working environment of the blower, specifically including:

[0069] The external temperature T, with the unit of degree Celsius (°C);

[0070] The real-time electricity price E, with the unit of yuan per kilowatt-hour (yuan / kWh).

[0071] Among them, the electricity price data is obtained through the API interface with the grid side, and the update frequency is usually set to be refreshed every 15 minutes.

[0072] The role of the data fusion and processing unit is to combine the above operating parameters with the environmental parameters and perform dimensionality reduction on the high-dimensional vibration spectrum vector v. Specifically, the principal component analysis (PCA) method is used to reduce the dimensionality of the vibration spectrum data. The calculation steps of the principal component analysis include: first, perform mean normalization on the original vibration spectrum matrix, then extract the most important eigenvectors through singular value decomposition (SVD), and reconstruct the state vector after dimensionality reduction.

[0073] In a possible implementation, assume that the dimension of the original vibration spectrum vector v is 1024 dimensions. The system uses the PCA weight matrix W (with the dimension of 20×1024) obtained through training to reduce its dimension to a 20-dimensional feature vector. Its mathematical expression is as follows:

[0074] v pca = W·v;

[0075] Where: v is the original vibration spectrum vector; W is the principal component dimensionality reduction weight matrix, with the dimension of m×n. In this example, m = 20 and n = 1024; v pca is the output dimensionality-reduced vibration feature vector.

[0076] It should be particularly noted that the weight matrix of W is obtained through training using a large amount of historical data in the offline stage. The training process includes steps such as solving the covariance matrix and eigenvalue decomposition, ensuring that the dimensionality-reduced feature vector can maximize the retention of the main information in the vibration spectrum.

[0077] Specifically, the finally formed state vector x includes the following components:

[0078] x = [p, q, n, e, T, E, v pca ;

[0079] Where: p is the current pressure value; q is the current flow value; n is the current rotational speed value; e is the current energy consumption value; T is the external temperature value; E is the current real-time electricity price; v pca is the dimensionality-reduced vibration spectrum feature vector.

[0080] This structure enables the downstream module to receive a state vector of a fixed length without repeatedly decoding the high-dimensional original signal, thus greatly reducing the computational pressure.

[0081] In a preferred implementation, a multi-threaded processing mechanism is provided inside the data fusion processing unit, ensuring that in the scenario of high-concurrency data, each data source can be collected in a timely manner and merged into the state vector. As an extensible solution, the edge computing unit can also be integrated on the blower side to perform real-time dimensionality reduction of the vibration spectrum and push the state vector to the central control system through the industrial bus.

[0082] In some embodiments, the system communicates with the remote server using the TCP / IP network protocol to ensure that environmental parameters such as real-time electricity price data can be synchronized regularly. The blower status data is reported in real time through the local local area network to avoid data delay caused by external network fluctuations.

[0083] Generally, the data acquisition and fusion module of the present invention can be applied not only to the blower cluster within a single factory but also extended to the remote monitoring scenario of blowers across regions, realizing broader data integration and processing.

[0084] A prediction modeling module, connected to the data acquisition and fusion module, is used to construct a system load prediction model based on historical input-output data;

[0085] Specifically, in the intelligent energy-saving control system for multi-unit operation of blowers involved in the present invention, the prediction modeling module, as the downstream of the data acquisition and fusion module, undertakes the dynamic prediction task of the system load and is the basis for the implementation of the entire control strategy. This module directly receives the state vector input from the data acquisition and fusion module and outputs the predicted load data for guiding control decisions.

[0086] Generally, the prediction modeling module includes a response matrix construction unit, a free response estimation unit, and a multi-step prediction calculation unit, which together constitute the dynamic prediction mechanism of the system. Specifically, this module mainly extracts the dynamic characteristics of the system from historical input-output data, constructs a prediction model, and generates multi-step prediction results through this model for use by the multi-agent decision-making module.

[0087] In this embodiment, the response matrix construction unit is used to establish a dynamic prediction relationship based on the system historical data. As an option, this unit first receives the historical control input sequence {Δu(k)} of the blower and the corresponding system response output {y(k)}. By processing the input-output sequence, the step response matrix G of the system is solved using the least squares estimation method or the recursive least squares algorithm.

[0088] In a possible implementation, assume that the input increment vector of the system at time k is Δu(k), and the corresponding output vector is y(k). The step response matrix G of the system is defined as:

[0089]

[0090] where: g mn represents the step response coefficient between the m-th output and the n-th input; m is the dimension of the output variable; n is the dimension of the input variable.

[0091] These step response parameters are obtained through system identification training, reflecting the dynamic influence relationship of the input on the output.

[0092] The free response estimation unit is used to predict the natural evolution trajectory of the future state of the system when there is no new input increment currently. Generally, this unit uses the current state vector x(k) and historical output data to estimate the free response vector y free . In some embodiments, the calculation of the free response can be implemented through a recursive model. For example, a first-order or high-order difference model is used to model the output dynamics.

[0093] In this embodiment, the estimation of the free response follows the following expression:

[0094] y free (k + 1) = A·y(k) + B·x(k);

[0095] where: A is the state transition matrix; B is the state input matrix; y(k) is the output vector at time k; x(k) is the state vector at time k.

[0096] This structure not only considers the inertia of the system but also combines the potential influence of the current environmental change on the system output.

[0097] The multi-step prediction calculation unit is responsible for combining the response matrix and the free response to generate the complete future prediction output. Specifically, this unit uses the control input increment vector Δu within the prediction period to calculate the prediction output through the following formula:

[0098] y pred = G·Δu + y free ;

[0099] where: G represents the step response matrix obtained through the response matrix construction unit; Δu represents the input increment vector for the future prediction period, with a dimension of n×1; y free is the free response vector, with a dimension of m×1; y pred is the future prediction output, with a dimension of m×1.

[0100] As an option, in implementation, the prediction time step can be set according to the actual working conditions, such as set to 5 steps, 10 steps or more, to meet the multi-step prediction requirements. This module allows dynamic adjustment of the prediction step length to adapt to the real-time requirements of system operation.

[0101] In some embodiments, to improve the prediction accuracy, an adaptive update mechanism is integrated inside the prediction modeling module. When it is detected that the dynamic characteristics of the system change, the retraining process of the response matrix can be automatically triggered, so as to maintain the timeliness and accuracy of the prediction model. As a preferred method, a sliding window strategy can be adopted to update the historical sequence when new data arrives, ensuring the timeliness of the model training data.

[0102] Generally, data is exchanged between the prediction modeling module and the multi-agent decision-making module through a standardized interface, such as REST API or an internal memory sharing mechanism, to minimize the overall system latency.

[0103] In one implementation solution, to ensure stable operation in high-concurrency scenarios, the prediction modeling module adopts a parallel computing architecture and uses matrix parallel multiplication to accelerate the prediction calculation process to support the real-time control requirements of large-scale blower systems.

[0104] A multi-agent decision-making module, connected to the data acquisition and fusion module and the prediction modeling module, is used to generate a control strategy based on the current state vector;

[0105] Specifically, in the energy-saving intelligent control system for multi-unit operation of blowers of the present invention, the multi-agent decision-making module is located downstream of the prediction modeling module, receives the prediction output and the state vector, and generates control actions for each blower. Through a distributed agent mechanism, this module enables each blower to have independent decision-making capabilities and coordinate with the global strategy at the same time, so as to achieve flexible and refined operation control.

[0106] Generally, the multi-agent decision-making module includes a state encoding unit, a policy output unit, and a reward function generation unit. These units cooperate closely to complete the dynamic control decision of blower startup and shutdown and speed. The output of this module is the local control action for each blower, which is then further fused into a global optimal strategy by the cooperative optimization module.

[0107] In this embodiment, the state encoding unit is used to receive the state vector x provided by the prediction modeling module i This state vector includes the pressure, flow rate, speed, energy consumption, ambient temperature, electricity price of the blower, and the vibration spectrum characteristics after dimensionality reduction by principal component analysis. In one possible implementation, this state encoding unit inputs x i into the policy network to extract state features and use them for subsequent control action calculations.

[0108] As an option, the policy network can adopt a deep neural network structure, including an input layer, several hidden layers, and an output layer. The input layer receives the state vector, the hidden layers are used for non-linear feature extraction, and the output layer generates continuous or discrete values of control actions. In this embodiment, the output of the policy output unit includes:

[0109] Start-stop control signal s i , whose value range is {0, 1}, indicating whether the blower i is enabled;

[0110] Speed adjustment increment Δn i , with the unit of revolutions per minute (rpm).

[0111] In a preferred implementation, the control action of the blower can be regarded as a two-dimensional action vector:

[0112] a i =[s i , Δn i ;

[0113] where: s i represents the start-stop state of the blower i; Δn i represents the target speed increment of the blower i.

[0114] The reward function generation unit is used to construct an individual reward function for each blower to guide the policy network to learn an energy-saving and efficient control strategy. Generally, this reward function combines multiple indicators such as energy consumption reduction, speed consistency, and start-stop behavior smoothness to ensure that the system control is both energy-saving and smooth.

[0115] In this embodiment, the calculation formula of the reward function r i is as follows:

[0116]

[0117] where: r i is the reward value of the blower i; e i is the energy consumption value of the blower i (unit: kW); e base is the reference energy consumption (unit: kW); ω i is the current speed of the blower i (unit: rpm); is the average speed (unit: rpm); ω max is the maximum allowable speed (unit: rpm); δ is the start-stop change identifier of the blower i between the current step and the previous time step, which is 1 if the start-stop state changes, otherwise 0; α, β, γ are preset weight factors, all of which are positive numbers.

[0118] As an implementation, the first term α·(e base -ei ) represents the energy-saving driving force. The second term is used to punish the behavior of the rotational speed deviating from the global average, and the third term is used to suppress the equipment loss caused by frequent start-stop.

[0119] In some embodiments, the policy output unit further includes an action limit mechanism to ensure that the rotational speed adjustment increment Δn output i does not exceed the preset safety threshold range. For example, set Δn i ∈[-50, +50] rpm to avoid exceeding the mechanical safety range of the blower.

[0120] Specifically, in the multi-agent structure, the agents of each blower are independent of each other. The system adopts a centralized training and distributed execution mechanism, enabling each agent to share environmental information during the training phase, while making autonomous decisions based only on local states during the execution phase. In a possible implementation, the policy network is trained through the Deep Deterministic Policy Gradient (DDPG) algorithm and can handle continuous action spaces.

[0121] Generally, the multi-agent decision module and the collaborative optimization module exchange control action data through shared memory or a fast data bus to ensure that the control loop is completed within a millisecond-level time delay.

[0122] As a preferred solution, a policy replay buffer can also be integrated inside the module to store historical state-action-reward sequences for subsequent offline retraining to enhance the robustness of the policy.

[0123] In some embodiments, the module implements an action smoothing mechanism, that is, a low-pass filter is introduced between two adjacent decisions to make the rotational speed adjustment signal Δn i smoother in the time series and reduce mechanical shock.

[0124] The collaborative optimization module, connected to the multi-agent decision module, is used to perform non-cooperative policy coordination among multiple blowers;

[0125] Specifically, in the intelligent energy-saving control system for multi-machine operation of blowers of the present invention, the collaborative optimization module, as the downstream of the multi-agent decision module, is responsible for integrating the local control actions of each blower and performing joint optimization processing at the system level to solve the optimal overall control strategy. A high-coupling relationship is formed between this module and the multi-agent decision module, and the coordination of local and global policies is achieved through real-time data exchange.

[0126] Generally, the co-optimization module includes a revenue function construction unit, an equilibrium solving unit, and a control instruction output unit. These units complete the collaborative scheduling of the multi-machine control of the blower through orderly logical cooperation. Specifically, the core of this module is to combine the rewards of local agents with the global system goals to form a strategy optimization problem under the multi-agent game framework.

[0127] In this embodiment, the revenue function construction unit is used to construct the non-cooperative revenue function of each blower according to the control actions output by the multi-agent decision module. As an option, the revenue function not only considers the individual rewards but also incorporates the system load balance constraint into the revenue model to encourage the local actions to be consistent with the global goals.

[0128] In a possible implementation, assume that the individual reward value of blower i is r i , and the system-wide load deviation is defined as ΔL. Then the revenue function U i can be expressed as:

[0129] U i = r i - λ·|ΔL|;

[0130] Where: U i is the final revenue value of blower i; r i is the individual reward value of blower i (unit: dimensionless, from the output of the reward function of the multi-agent module); ΔL is the deviation between the current total load of the system and the target load (unit: kW); λ is the load constraint weight coefficient, with the unit of dimensionless parameter.

[0131] This structure couples the individual rewards with the global load, effectively suppressing the degradation of system performance caused by the "selfish" behavior of a single blower.

[0132] The equilibrium solving unit is used to find the optimal solution of the control actions of each blower under the above revenue function framework. Generally, this unit adopts distributed optimization algorithms such as the alternating direction method of multipliers (ADMM) to achieve the collaborative convergence of solutions among multiple agents through an iterative process.

[0133] In this embodiment, the basic update steps of ADMM are as follows:

[0134] For each agent i, while fixing the actions of other blowers, update its own action a i to minimize the local loss function;

[0135] Perform Lagrange multiplier update to correct the system consistency constraint;

[0136] Repeat the above steps until the system meets the convergence conditions.

[0137] As an implementation, the update formula of ADMM can be expressed as:

[0138]

[0139] Where: is the control action vector of blower i in the (k + 1)-th iteration; z (k) is the system average action vector; u (k) is the scaled Lagrange multiplier of blower i; ρ is the penalty parameter used to balance the local loss and the global constraint.

[0140] Specifically, the control action vector a i includes the start-stop signal s i and the rotational speed increment Δn i , that is:

[0141] a i = [s i , Δn i ;

[0142] In a preferred implementation, the equilibrium solving unit dynamically adjusts the value of ρ in each iteration to accelerate the algorithm convergence, especially when the load fluctuates greatly, the response speed can be significantly improved.

[0143] The control instruction output unit is used to distribute the optimized global optimal control strategy to each blower execution unit after the equilibrium solving unit finishes the solution. Generally, this unit uses an industrial bus or a high-efficiency network interface to implement data distribution, ensuring a response delay of milliseconds.

[0144] In some embodiments, the control instruction output unit is integrated with a fault tolerance mechanism. When the system detects that a certain blower feedback is abnormal, it can output an abort signal according to the current strategy to avoid affecting the execution of the overall control strategy due to a single point of failure.

[0145] As a preferred method, this module can also introduce other constraint terms into the profit function, such as noise suppression or temperature control, to enhance the scalability and robustness of the system. For example, the profit function can be extended to:

[0146] U i = r i - λ1|ΔL| - λ2|T i - T ref |;

[0147] Where: T i is the motor temperature of blower i (unit: °C); T ref is the target reference temperature (unit: °C); λ1, λ2 are weight parameters used to balance multiple constraints.

[0148] Generally, a dynamic adaptation is achieved between the collaborative optimization module and the parameter update module through a feedback mechanism. Specifically, both the optimization result and the execution status are fed back to the parameter update module in real time for further correcting the model and the strategy.

[0149] In one implementation, this module supports a parallel optimization mechanism, which is particularly suitable for application scenarios where the scale of the blower cluster is large (e.g., more than 50 units), and uses distributed computing to accelerate the global optimization process.

[0150] The parameter update module is connected to the prediction modeling module and the multi-agent decision-making module, and is used to update the prediction model and the control strategy based on the control error and the training feedback.

[0151] Specifically, in the intelligent energy-saving control system for multi-machine operation of blowers of the present invention, the parameter update module is downstream of the collaborative optimization module. Its main function is to monitor in real time the deviation between the system prediction and the actual execution, dynamically correct the model parameters, and periodically update the control strategy. This module forms a two-way coupling relationship with the prediction modeling module and the multi-agent decision-making module to ensure that the system maintains continuous optimization of the prediction accuracy and the control effect during long-term operation.

[0152] Generally, the parameter update module includes an online model correction unit, an offline policy training unit, and a policy synchronization unit. These units work together to construct an update mechanism that is both online adaptive and supports offline enhancement.

[0153] In this embodiment, the online model correction unit is used to dynamically adjust the response matrix G in the prediction modeling module according to the error between the system prediction output and the actual execution output. As an option, when the system detects a significant error, this unit triggers an online update process to suppress the model drift effect.

[0154] In a possible implementation, the online update of the response matrix adopts an incremental correction strategy, and the mathematical expression is as follows:

[0155]

[0156] Where: G new represents the updated response matrix; G old represents the original response matrix; η is the learning rate parameter, dimensionless, and can be dynamically adjusted according to the actual application; y real is the output vector actually observed by the system (unit: kW or m 3 / h, according to the control target); y pred is the prediction vector output by the prediction modeling module; Δu is the incremental vector of the control input (the unit depends on the control dimension and can be rpm, start-stop signal, etc.).

[0157] In general, the learning rate η is set to a positive number less than 1 to avoid instability caused by the model being updated too quickly.

[0158] The policy offline training unit is used to retrain the multi-agent policy network periodically using the simulation dataset. Specifically, during long-term operation, this unit collects system state-action-reward sequences and generates a large-scale training sample library. As an option, an experience replay mechanism can be used to preferentially incorporate historical optimal samples into the training set to accelerate the convergence rate of the policy network.

[0159] In some embodiments, the offline training of the policy network adopts a combination of supervised learning and reinforcement learning. For example, the Deep Deterministic Policy Gradient (DDPG) algorithm can be used to optimize the policy by maximizing the reward function r i The training objective function is:

[0160]

[0161] where: J(θ) represents the expected reward of the policy network parameters θ; x i is the state vector of blower i; π θ (x i ) is the action generated by the policy network; r i is the reward function, which has been defined in the foregoing module.

[0162] The role of the policy synchronization unit is to send the new policy parameters obtained from offline training to the edge computing unit in real time to ensure that the policies are in a synchronized state among all agents in the system. As an implementation method, this unit adopts a dual-channel communication mechanism. On the one hand, it transmits the new policy parameters through the internal network, and on the other hand, it uses a checksum mechanism to confirm the integrity of parameter synchronization.

[0163] Specifically, in a large-scale blower cluster, this unit can adopt a batch push strategy to gradually update the policy parameters by region to reduce the instantaneous network load.

[0164] In general, the parameter update module also has an anomaly detection mechanism. For example, when it is found that the system prediction error suddenly increases due to model update, this module can trigger a fallback mechanism to restore the response matrix to the previous stable version to prevent divergence during online update.

[0165] In some embodiments, this module allows the integration of the rules of an expert system. For example, if the operating environment of the blower suddenly changes (such as a sharp rise in temperature), the function of preferentially adopting a safety policy can be triggered to avoid control anomalies caused by extreme situations.

[0166] As an extended solution, the parameter update module can upload the collected data to the cloud server for further analysis by remote experts and feedback improvement suggestions for future version iterations.

[0167] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An energy-saving intelligent control system for multi-unit operation of a blower, characterized in that, It includes: A data acquisition and fusion module, which is used to collect the operating parameters and environmental parameters of the blower and perform fusion processing; A prediction modeling module, connected to the data acquisition and fusion module, which is used to construct a system load prediction model based on historical input and output data; A multi-agent decision-making module, connected to the data acquisition and fusion module and the prediction modeling module, which is used to generate a control strategy based on the current state vector; A collaborative optimization module, connected to the multi-agent decision-making module, which is used to perform non-cooperative strategy coordination among multiple blowers; A parameter update module, connected to the prediction modeling module and the multi-agent decision-making module, which is used to update the prediction model and control strategy based on the control error and training feedback.

2. The multi-machine operation energy-saving intelligent control system of a blower according to claim 1, characterized in that The data acquisition and fusion module includes: An operating parameter acquisition unit, which is used to collect the pressure value, flow value, rotation speed value, energy consumption value and vibration spectrum data of each blower; An environmental parameter acquisition unit, which is used to collect the external environmental temperature value of the system and the real-time electricity price; A data fusion processing unit, which is used to perform principal component analysis on the vibration spectrum data, generate a dimensionality-reduced feature vector, and construct a system state vector, and the state vector includes the pressure value, flow value, rotation speed value, energy consumption value, vibration feature vector, environmental temperature value and electricity price.

3. The multi-machine operation energy-saving intelligent control system of a blower according to claim 1, characterized in that, The prediction modeling module includes: A step response modeling unit, which is used to construct a response matrix of the control input to the system output; A free response estimation unit, which is used to calculate the system output trend in the absence of control input; A multi-step prediction generation unit, which is used to combine the control input increment and the free response and output the system predicted load at several future moments.

4. The multi-machine operation energy-saving intelligent control system of a blower according to claim 1, characterized in that, The multi-agent decision-making module includes: A state processing unit, which is used to input the state vectors of each blower into their respective policy networks; A policy generation unit, which is used to output the control actions of the blower, and the control actions include start-stop signals and rotation speed change amounts; A reward construction unit, which is used to construct a local reward value of a single blower based on the energy consumption value, rotation speed deviation value and start-stop times.

5. The intelligent energy-saving control system for multi-unit operation of a blower according to claim 1, characterized in that The collaborative optimization module includes: A revenue function construction unit, which is used to define a non-cooperative revenue function of multiple blowers, and the revenue function is constructed based on their respective reward functions and system prediction errors; An optimization coordination unit, which is used to perform joint optimization of strategies using the alternating direction method of multipliers; An instruction coordination output unit, which is used to output the final consistent control strategy among multiple blowers.

6. The intelligent energy-saving control system for multi-unit operation of a blower according to claim 1, characterized in that, The parameter update module includes: A model update unit, which is used to perform real-time correction of the response matrix using the prediction error; A policy training unit, which is used to train the control policy network based on simulation data; A policy synchronization unit, which is used to synchronously update the policy parameters obtained from offline training to the control network in the edge device.

7. The intelligent energy-saving control system for multi-unit operation of a blower according to claim 2, characterized in that, The dimensionality-reduced feature vector of the data fusion processing unit is the product of the vibration spectrum data and the principal component matrix, and the calculation formula is: v pca = W·v; Where: v is the original vector of the vibration spectrum; W is the weight matrix for principal component dimensionality reduction; v pca is the output vibration feature vector after dimensionality reduction.

8. The intelligent energy-saving control system for multi-unit operation of a blower according to claim 3, characterized in that, The output predicted value of the multi-step prediction generation unit satisfies the following expression: y pred = G·Δu + y free ; Where: G represents the step response matrix obtained by the response matrix construction unit; Δu represents the input increment vector for the future prediction period; y free is the free response vector; y pred is the future predicted output.

9. The intelligent energy-saving control system for multi-unit operation of a blower according to claim 4, characterized in that, The reward value of the reward construction unit is calculated according to the following expression: where: r i is the reward value of blower i; e i is the energy consumption value of blower i; e base is the reference energy consumption; ω i is the current speed of blower i; is the average speed; ω max is the maximum allowable speed; δ is the start / stop change flag of blower i between the current step and the previous time step, which is 1 if the start / stop state changes, otherwise 0; α, β, γ are preset weight factors, all of which are positive numbers.

10. The multi-machine operation energy-saving intelligent control system of a blower according to claim 6, characterized in that, The response matrix of the model update unit is updated through the following expression: G new = G old + η · (y real - y pred ) · (Δu) T ; Where: G new represents the updated response matrix; G old represents the original response matrix; η is the learning rate parameter, dimensionless, and can be dynamically adjusted according to actual applications; y real is the output vector actually observed by the system; y pred is the predicted vector output by the prediction modeling module; Δu is the incremental vector of the control input.

Images