Special lubricating oil production temperature control system based on machine learning

Abstract

The invention discloses a special lubricating oil production temperature control system based on machine learning. According to the invention, through a dynamic multi-objective decision-making mechanism of the intelligent prediction and optimization module, thermodynamic characteristic changes caused by variables such as formula switching and environment disturbance in the lubricating oil production process can be self-adapted. According to the hybrid modeling method integrating data driving and a physical mechanism, not only is the temperature control precision of + / -0.1 DEG C in a key reaction stage guaranteed, but also the thermal compensation intensity can be intelligently adjusted in the heating and cooling process, and energy consumption and equipment thermal stress are effectively balanced. A built-in process knowledge migration framework of the system remarkably shortens the debugging period of new formula production, the production parameter adaptation efficiency of aerospace-grade special lubricating oil is improved by several times, meanwhile, microsecond-grade collaborative response of all the modules is achieved through a time-sensitive network, it is ensured that temperature fluctuation is restrained within a narrow interval of process requirements, and the production efficiency of the aerospace-grade special lubricating oil is improved. And a systematic guarantee is provided for the quality stability and the process reliability of special lubricating oil production.

Description

Technical Field

[0001] The present invention belongs to the technical field of lubricating oil production, and specifically relates to a temperature control system for the production of special lubricating oil based on machine learning. Background Art

[0002] The temperature control system for lubricating oil production is a crucial part in the lubricating oil production process. Through precise temperature monitoring and control, it ensures that the lubricating oil is mixed, reacted, and refined at an appropriate temperature, thereby guaranteeing product quality. This system usually consists of temperature sensors, control units, actuators, heat exchange equipment, etc. It can collect temperature data in the production process in real time and automatically adjust heating or cooling devices through the PID control algorithm to maintain a constant process temperature. The stability and accuracy of the temperature control system for lubricating oil production directly affect the performance and stability of the lubricating oil, and are of great significance for improving production efficiency, reducing energy consumption, and extending the service life of equipment.

[0003] However, the existing technologies have obvious limitations in the field of temperature control for special lubricating oil production. The traditional control system relies on the fixed-parameter PID algorithm and is difficult to dynamically adapt to formula changes and environmental disturbances, resulting in frequent temperature fluctuations exceeding the process threshold. The debugging mode dominated by manual experience is time-consuming and laborious, and repeated trial and error is required for the production of new formulas, seriously restricting the R & D efficiency of high-end lubricating oil. Summary of the Invention

[0004] The purpose of the present invention is to provide a temperature control system for the production of special lubricating oil based on machine learning in order to solve the above-mentioned problems.

[0005] The technical solution adopted by the present invention is as follows: A temperature control system for the production of special lubricating oil based on machine learning, the system includes: a data acquisition module, a central processing module, an intelligent prediction and optimization module, an execution module, a communication module, and a human-computer interaction module;

[0006] The intelligent prediction and optimization module is internally provided with a data preprocessing module, a time series prediction model module, a dynamic multi-objective optimization module, and a rule library module;

[0007] The output end of the sensor array of the data acquisition module is connected to the RS485 bus interface of the communication module through an industrial Ethernet, and the preprocessed data output port of its edge computing node is directly connected to the real-time data input end of the central processing module;

[0008] The core computing unit of the central processing module is connected to the NPU accelerator of the intelligent prediction and optimization module through a PCIe high-speed channel, and the control parameter output end of the multi-objective optimizer is connected to the input end of the PLC controller of the execution module through a 4 - 20mA analog signal;

[0009] The feedback signal output end of the execution module is returned to the closed-loop verification interface of the data acquisition module through the photoelectric isolator, and is connected to the industrial switch of the communication module through the Modbus-TCP protocol;

[0010] The HMI operation panel of the human-machine interaction module is bidirectionally connected to the instruction parsing unit of the central processing module through the OPC UA protocol, the real-time data input terminal of the AR display terminal is directly connected to the gigabit optical fiber interface of the communication module, and the mobile terminal interaction platform is wirelessly connected to the IoT gateway of the communication module through the 5G base station;

[0011] The time-sensitive network master station of the communication module synchronizes and coordinates the clock signals of each module, and its data distribution unit provides the distributed data stream required for cross-domain feature fusion for the intelligent prediction and optimization module.

[0012] In a preferred embodiment, the data acquisition module is internally provided with a distributed sensor array and an edge computing node; high-precision infrared temperature sensing units are deployed at key positions of the reactor, and a star topology is adopted to ensure full coverage of the measurement points. Each sensing unit has a built-in self-calibration circuit to eliminate environmental interference; the edge computing node is equipped with a dedicated signal processing chip, which performs millisecond-level timestamp alignment on the raw data, removes mechanical vibration noise through a composite filtering algorithm, and extracts temperature change rate and local thermal gradient characteristic parameters at the same time; the sensing unit and the edge node are directly connected with an anti-interference coaxial cable.

[0013] In a preferred embodiment, the central processing module is internally provided with a general-purpose CPU and a neural network processor; the CPU unit runs a real-time operating system, is responsible for logic control, task scheduling and security monitoring, and synchronously processes device status data from 20 parallel threads; the neural network processor has a built-in dedicated matrix operation unit, executes prediction model reasoning in a pipeline manner, and exchanges intermediate calculation results with the CPU through a memory sharing mechanism.

[0014] In a preferred embodiment, the data preprocessing module deploys a dedicated feature extraction engine with a built-in sliding window analysis algorithm to calculate the temperature change rate in real time, and combines wavelet transform technology to remove high-frequency noise in the sensor signal; based on the production characteristics of lubricating oil, the feature library presets 8 process parameter extraction templates such as thermal inertia coefficient and gradient mutation threshold, and dynamically adjusts the feature combination weights through an adaptive weighting strategy.

[0015] In a preferred embodiment, a dual-scale memory network architecture is provided inside the timing prediction model module; the network body is composed of a double-layer gated recurrent unit. The first layer captures the minute-level temperature fluctuation pattern, and the second layer analyzes the hourly-level production cycle characteristics. Residual connections are used between layers to prevent gradient vanishing. A sliding time window mechanism is integrated at the model input end, and each time a production data slice containing 30 consecutive minutes is processed. The output end generates a predicted trajectory for the next 5 minutes through mapping by a fully connected layer.

[0016] In a preferred embodiment, the dynamic multi-objective optimization module adopts a hybrid intelligent algorithm architecture. The core realizes multi-objective collaborative decision-making through a three-order optimization framework. The first order is the reinforcement learning real-time response layer, which quickly generates an initial control strategy based on the proximal policy optimization algorithm and captures the instantaneous disturbance characteristics of the production process through an online learning mechanism. The second order is the multi-objective genetic algorithm optimization layer, which uses the improved NSGA-III algorithm to screen the optimal solution set in the Pareto front and introduces an adaptive crossover and mutation operator to balance the global search and local development capabilities. The third order is the physical constraint verification layer, which relies on the safety boundary condition library constructed by the knowledge graph to perform double verification of the thermodynamic feasibility and equipment tolerance of the optimization results, and realizes real-time trade-off of temperature control accuracy, energy consumption efficiency, and equipment life through a dynamic weight allocation mechanism, where the weight coefficient is automatically adjusted according to the production stage, giving priority to ensuring temperature stability during the isothermal reaction period and focusing on energy-saving optimization during the heating and cooling stages. The dynamic weight adaptive formula is as follows:

[0017]

[0018] In the formula, \(W_t\) represents the dynamic weight of temperature accuracy within the time window \(t\), \(\eta\) is the initial weight base, \(\gamma\) is the time decay coefficient, \(\Delta T_{max}\) is the maximum allowable temperature deviation threshold, \(\Delta T_t\) is the current actual temperature deviation, and \(\alpha\) is the sensitivity adjustment factor. Through the combined action of the exponential decay function and the real-time temperature deviation, this formula enables the system to automatically strengthen the temperature control weight during sudden disturbances and increase the proportion of the energy-saving weight under stable operating conditions, achieving adaptive balance of multiple objectives.

[0019] The equipment degradation compensation formula is:

[0020]

[0021] In the formula, \(C_d\) is the equipment degradation compensation amount, \(NN\) is the cumulative number of equipment working cycles, \(\mu\) and \(\sigma\) respectively represent the normal distribution parameters of equipment performance attenuation, \(\beta\) is the material fatigue coefficient, \(\tau\) is the thermal stress cycle strength, and \(k\) is the non-linear correction exponent.

[0022] In a preferred embodiment, the underlying storage structure of the rule library module decomposes more than 200 pieces of expert experience into atomic rule units. The inference engine in the middle layer supports multi-condition parallel matching, and the dynamic loader at the top layer activates corresponding rule clusters according to the production stage. The exception handling strategy adopts a hierarchical management mode. The first-level rules directly block dangerous operations, the second-level rules trigger compensation control, and the third-level rules generate manual review prompts. A two-way verification channel is established between the rule library and the prediction model. When the model output deviates from process common sense, a correction coefficient is automatically injected, and at the same time, newly occurred abnormal cases are fed back to the rule library for iterative update.

[0023] In a preferred embodiment, a dynamic compensation type power drive system is internally provided in the execution module, which includes a multi-stage actuator collaborative working mechanism. The electric heating unit adopts pulse width modulation technology, and the output accuracy adjustment at the 0.1% level is realized through the IGBT power module. The cooling system is equipped with a fast-response solenoid valve to control the liquid nitrogen injection flow. The drive controller is built-in with a feedback compensation algorithm, which compares the set value with the actual output of the actuator in real time. When a deviation of more than 0.5% is detected, the second-order PID compensation is automatically started. The module is equipped with a double safety interlock, and immediately switches to the preset safety mode under communication interruption or over-limit working conditions to ensure that the temperature fluctuation of the reaction kettle is always within the process allowable range.

[0024] In a preferred embodiment, the backbone network of the communication module adopts a dual-ring topology structure, deploys a nanosecond-level clock synchronization mechanism, and aligns the clock signals of each module through the Precision Time Protocol. The data transmission is divided into three independent virtual channels: the real-time control instruction channel ensures an end-to-end transmission delay of 10 ms, the monitoring data channel provides a bandwidth of 200 Mbps to carry the sensor data stream, and the maintenance channel realizes remote diagnosis and firmware upgrade. The network switching device is built-in with a traffic shaping engine, which preferentially ensures the transmission certainty of temperature control instructions and can still maintain a zero packet loss rate for key data packets under 90% network load.

[0025] In a preferred embodiment, the operation console of the human-machine interaction module is equipped with a laser holographic projection device, which projects the three-dimensional temperature field of the reaction kettle onto the operation area in the form of color gradients in real time and supports gesture recognition to realize dynamic parameter adjustment. The data display layer of the human-machine interaction module adopts a multi-screen fusion technology. The main monitoring screen presents the global temperature trend curve, and the auxiliary screen displays the real-time values of 16 key points. The mobile terminal of the human-machine interaction module accesses the system through digital certificate authentication, realizes hierarchical management of control permissions under the 5G network, and the operation records are stored using blockchain technology.

[0026] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are:

[0027] 1. In the present invention, through the dynamic multi-objective decision-making mechanism of the intelligent prediction and optimization module, it is possible to adapt to the changes in thermodynamic characteristics brought about by variables such as formula switching and environmental disturbances during the lubricating oil production process. Its hybrid modeling method that combines data-driven and physical mechanisms not only ensures the temperature control accuracy of ±0.1°C in the key reaction stage but also can intelligently adjust the heat compensation intensity during the heating and cooling processes, effectively balancing energy consumption and equipment thermal stress. The built-in process knowledge transfer framework in the system significantly shortens the commissioning cycle for the production of new formulas, increases the adaptation efficiency of production parameters for aerospace-grade special lubricating oils by several times, and at the same time extends the service life of the key reaction kettle through the equipment degradation prediction and compensation mechanism, reducing maintenance costs.

[0028] 2. In the present invention, from the high-precision sensing array in the data acquisition layer to the dynamic compensation drive in the execution layer, each module achieves microsecond-level collaborative response through the time-sensitive network, ensuring that the temperature fluctuation is suppressed within the narrow range required by the process. The augmented reality human-machine interface converts the complex heat transfer process into a visual three-dimensional field map, enabling operators to intuitively grasp the overall production status and quickly locate potential risks. The exception handling module combines the process common sense in the rule library and the early warning ability of the prediction model, and can trigger multi-level interlock protection 30 seconds before the temperature gets out of control, greatly improving the production safety redundancy. This design concept that integrates precise control, intelligent decision-making, and safety protection provides a systematic guarantee for the quality stability and process reliability of special lubricating oil production. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is the overall system block diagram of the present invention;

[0030] Figure 2 is the system block diagram of the intelligent prediction and optimization module in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0031] In order to make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0032] Embodiment:

[0033] Referring to Figure 1-2 ,

[0034] A temperature control system for the production of special-purpose lubricating oil based on machine learning, the system includes: a data acquisition module, a central processing module, an intelligent prediction and optimization module, an execution module, a communication module, and a human-machine interaction module;

[0035] The intelligent prediction and optimization module is internally provided with a data preprocessing module, a time series prediction model module, a dynamic multi-objective optimization module, and a rule library module;

[0036] The output end of the sensor array of the data acquisition module is connected to the RS485 bus interface of the communication module through industrial Ethernet, and the preprocessed data output port of its edge computing node is directly connected to the real-time data input end of the central processing module;

[0037] The core operation unit of the central processing module is connected to the NPU accelerator of the intelligent prediction and optimization module through the PCIe high-speed channel, and the control parameter output end of the multi-objective optimizer is connected to the input end of the PLC controller of the execution module with a 4-20mA analog signal;

[0038] The feedback signal output end of the execution module returns to the closed-loop verification interface of the data acquisition module through an optoelectronic isolator, and at the same time accesses the industrial switch of the communication module through the Modbus-TCP protocol;

[0039] The HMI operation panel of the human-machine interaction module is bidirectionally connected to the instruction parsing unit of the central processing module through the OPC UA protocol. The real-time data input end of its AR display terminal is directly connected to the gigabit fiber interface of the communication module, and the mobile interaction platform wirelessly accesses the IoT gateway of the communication module through the 5G base station;

[0040] The time-sensitive network (TSN) master station of the communication module synchronizes and coordinates the clock signals of each module, and its data distribution unit provides the distributed data stream required for cross-domain feature fusion for the intelligent prediction and optimization module.

[0041] A distributed sensor array and an edge computing node are set inside the data acquisition module. High-precision infrared temperature sensing units are deployed at key positions of the reactor, and a star topology is adopted to ensure full coverage of the measurement points. Each sensing unit is built with a self-calibration circuit to eliminate environmental interference. The edge computing node is equipped with a dedicated signal processing chip to align the original data with millisecond-level timestamps, strip mechanical vibration noise through a composite filtering algorithm, and at the same time extract characteristic parameters such as temperature change rate and local thermal gradient. The sensing unit and the edge node are directly connected by anti-interference coaxial cables to ensure that the measurement accuracy of 0.001°C can still be maintained in a strong electromagnetic environment. The preprocessed standard data packets are uploaded to the central processing module through a dual-redundancy channel.

[0042] Inside the central processing module, there is a general-purpose CPU and a neural network processor. The CPU unit runs a real-time operating system, is responsible for logical control, task scheduling, and security monitoring, and synchronously processes device status data from 20 parallel threads. The neural network processor is built with a dedicated matrix operation unit, which executes prediction model inference in a pipelined manner and exchanges intermediate calculation results with the CPU through a memory sharing mechanism. A ring data bus is deployed inside the module, supporting the transmission of 2.4 GB of real-time production data per second, completing the entire process from data reception to instruction generation within the temperature control cycle, and strictly controlling the processing delay within 5 milliseconds.

[0043] The data preprocessing module deploys a dedicated feature extraction engine, which incorporates a sliding window analysis algorithm to calculate the temperature change rate in real time and combines wavelet transform technology to strip high-frequency noise from the sensor signal. For the production characteristics of lubricating oil, the feature library pre-sets 8 process parameter extraction templates such as the thermal inertia coefficient and the gradient mutation threshold, and dynamically adjusts the weight of the feature combination through an adaptive weighting strategy. The preprocessed standardized data stream adopts timestamp alignment technology to ensure that multi-dimensional parameters are synchronized within a millisecond-level error range, providing an input matrix with strict temporal alignment for the subsequent prediction model.

[0044] The time series prediction model module is internally set with a dual-scale memory network architecture. The main body of the network consists of a double-layer gated recurrent unit. The first layer captures the minute-level temperature fluctuation pattern, and the second layer analyzes the hour-level production cycle characteristics. Residual connections are used between layers to prevent gradient vanishing. The sliding time window mechanism is integrated at the input end of the model, and each time it processes a production data slice containing 30 consecutive minutes. The output end generates the prediction trajectory for the next 5 minutes through a fully connected layer mapping. During the training phase, the knowledge distillation technology is adopted to convert expert experience into soft label constraints, enabling the model to comply with the process physical laws while ensuring the prediction accuracy, and compressing the prediction response time within 800 milliseconds.

[0045] The dynamic multi-objective optimization module adopts a hybrid intelligent algorithm architecture, and its core is a third-order optimization framework to achieve multi-objective collaborative decision-making. The first order is the reinforcement learning real-time response layer, which quickly generates an initial control strategy based on the proximal policy optimization algorithm and captures the instantaneous perturbation characteristics of the production process through an online learning mechanism; the second order is the multi-objective genetic algorithm optimization layer, which uses the improved NSGA-III algorithm to screen the optimal solution set in the Pareto front and introduces an adaptive crossover and mutation operator to balance the global search and local development capabilities; the third order is the physical constraint verification layer, which relies on the safety boundary condition library constructed by the knowledge graph to perform double verification of the thermodynamic feasibility and equipment tolerance of the optimization results. This module innovatively integrates data-driven and mechanism models, and realizes real-time trade-offs among temperature control accuracy, energy consumption efficiency, and equipment life through a dynamic weight allocation mechanism, where the weight coefficients are automatically adjusted according to the production stage. For example, during the isothermal reaction period, temperature stability is given priority, and during the heating and cooling stages, energy-saving optimization is emphasized.

[0046] Among them, the dynamic weight adaptation formula is as follows:

[0047]

[0048] In the formula, \(W_t\) represents the dynamic weight of temperature accuracy within the time window \(t\), \(\eta\) is the initial weight base, \(\gamma\) is the time decay coefficient, \(\Delta T_{max}\) is the allowable maximum temperature deviation threshold, \(\Delta T_t\) is the current actual temperature deviation, and \(\alpha\) is the sensitivity adjustment factor. Through the combined action of the exponential decay function and the real-time temperature deviation, this formula enables the system to automatically strengthen the temperature control weight during sudden disturbances and increase the proportion of energy-saving weight under stable operating conditions, achieving an adaptive balance of multiple objectives.

[0049] The equipment degradation compensation formula is as follows:

[0050]

[0051] In the formula, \(C_d\) is the equipment degradation compensation amount, \(NN\) is the cumulative number of working cycles of the equipment, \(\mu\) and \(\sigma\) respectively represent the normal distribution parameters of equipment performance attenuation, \(\beta\) is the material fatigue coefficient, \(\tau\) is the thermal stress cycle strength, and \(k\) is the nonlinear correction exponent. This model quantifies the equipment performance attenuation process through an integral form and predicts the remaining life in combination with thermal stress historical data, enabling the optimizer to compensate the heating power or cooling rate in advance during decision-making, thereby reducing the impact of thermal shock on the equipment life.

[0052] The underlying storage structure of the rule base module decomposes more than 200 expert experiences into atomic rule units. The middle-layer inference engine supports multi-condition parallel matching, and the top-layer dynamic loader activates the corresponding rule clusters according to the production stage. The exception handling strategy adopts a hierarchical management mode. The first-level rules directly block dangerous operations, the second-level rules trigger compensation control, and the third-level rules generate manual review prompts. A two-way verification channel is established between the rule base and the prediction model. When the model output deviates from the process common sense, a correction coefficient is automatically injected, and at the same time, newly occurred abnormal cases are fed back to the rule base for iterative update.

[0053] The execution module is internally equipped with a dynamic compensation type power drive system, which includes a multi-stage actuator collaborative working mechanism. The electric heating unit adopts pulse width modulation technology, and the output accuracy adjustment at the 0.1% level is achieved through the IGBT power module. The cooling system is equipped with a fast-response solenoid valve to control the liquid nitrogen injection flow. The drive controller is built-in with a feedback compensation algorithm, which compares the set value with the actual output of the actuator in real time. When a deviation of more than 0.5% is detected, the second-order PID compensation is automatically started. The module is equipped with a dual safety interlock, which immediately switches to the preset safety mode under communication interruption or over-limit conditions, ensuring that the temperature fluctuation of the reaction kettle is always within the process allowable range.

[0054] The backbone network of the communication module adopts a dual-ring topology structure, deploys a nanosecond-level clock synchronization mechanism, and aligns the clock signals of each module through the Precision Time Protocol. Data transmission is divided into three independent virtual channels: the real-time control instruction channel ensures an end-to-end transmission delay of 10 ms, the monitoring data channel provides a bandwidth of 200 Mbps to carry the sensor data stream, and the maintenance channel enables remote diagnosis and firmware upgrade. The network switching device is built with a traffic shaping engine to prioritize ensuring the transmission determinism of temperature control instructions and maintain a zero packet loss rate for critical data packets even under 90% network load.

[0055] The operating console of the human-machine interaction module is equipped with a laser holographic projection device, which projects the three-dimensional temperature field of the reactor onto the operation area in the form of a color gradient in real time, and supports gesture recognition to achieve dynamic parameter adjustment. The data display layer of the human-machine interaction module adopts a multi-screen fusion technology. The main monitoring screen presents the global temperature trend curve, and the auxiliary screens display the real-time values of 16 key points. The mobile terminal of the human-machine interaction module accesses the system through digital certificate authentication, realizes hierarchical management of control permissions under the 5G network, and the operation records are stored using blockchain technology to ensure that the production process data cannot be tampered with and is traceable throughout the process.

[0056] It can be known from the above that:

[0057] In the present invention, through the dynamic multi-objective decision-making mechanism of the intelligent prediction and optimization module, it can adapt to the changes in thermodynamic characteristics brought about by variables such as formula switching and environmental disturbances during the lubricating oil production process. Its hybrid modeling method that combines data-driven and physical mechanisms not only ensures a temperature control accuracy of ±0.1°C in the critical reaction stage but also can intelligently adjust the thermal compensation intensity during the heating and cooling processes, effectively balancing energy consumption and equipment thermal stress. The built-in process knowledge migration framework in the system significantly shortens the commissioning cycle for new formula production, improves the adaptation efficiency of production parameters for aerospace-grade special lubricating oil by several times, and at the same time extends the service life of the critical reactor through the equipment degradation prediction and compensation mechanism, reducing maintenance costs.

[0058] In the present invention, from the high-precision sensing array in the data acquisition layer to the dynamic compensation drive in the execution layer, each module achieves microsecond-level collaborative response through the time-sensitive network, ensuring that the temperature fluctuation is suppressed within a narrow range required by the process. The augmented reality human-machine interface transforms the complex heat transfer process into a visual three-dimensional field map, enabling operators to intuitively grasp the overall production status and quickly locate potential risks. The abnormal handling module combines the process common sense in the rule library and the early warning ability of the prediction model to trigger multi-level interlock protection 30 seconds before the temperature gets out of control, greatly enhancing the production safety redundancy. This design concept that integrates precise control, intelligent decision-making, and safety protection provides a systematic guarantee for the quality stability and process reliability of special lubricating oil production.

[0059] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0060] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A temperature control system for the production of special lubricating oil based on machine learning, characterized in that: The system includes: a data acquisition module, a central processing module, an intelligent prediction and optimization module, an execution module, a communication module, and a human-machine interaction module; Inside the intelligent prediction and optimization module, there are a data preprocessing module, a time series prediction model module, a dynamic multi-objective optimization module, and a rule base module; The output end of the sensor array of the data acquisition module is connected to the RS485 bus interface of the communication module through an industrial Ethernet, and the preprocessed data output port of its edge computing node is directly connected to the real-time data input end of the central processing module; The core operation unit of the central processing module is connected to the NPU accelerator of the intelligent prediction and optimization module through a PCIe high-speed channel. The control parameter output end of the multi-objective optimizer is connected to the input end of the PLC controller of the execution module with a 4-20mA analog signal; The feedback signal output end of the execution module returns to the closed-loop verification interface of the data acquisition module through an opto-isolator, and at the same time accesses the industrial switch of the communication module through the Modbus-TCP protocol; The HMI operation panel of the human-machine interaction module is bidirectionally connected to the instruction parsing unit of the central processing module through the OPC UA protocol. The real-time data input end of its AR display terminal is directly connected to the gigabit fiber interface of the communication module, and the mobile interaction platform wirelessly accesses the IoT gateway of the communication module through a 5G base station; The time-sensitive network master station of the communication module synchronizes and coordinates the clock signals of each module, and its data distribution unit provides the distributed data stream required for cross-domain feature fusion for the intelligent prediction and optimization module.

2. The temperature control system for the production of special lubricating oil based on machine learning according to claim 1, characterized in that: Inside the data acquisition module, there is a distributed sensor array and an edge computing node; a high-precision infrared temperature sensing unit is deployed at the key positions of the reactor, and a star topology structure is adopted to ensure full coverage of the measurement points. Each sensing unit is built with a self-calibration circuit to eliminate environmental interference; the edge computing node is equipped with a dedicated signal processing chip to align the original data with millisecond-level timestamps, strip mechanical vibration noise through a composite filtering algorithm, and extract characteristic parameters such as temperature change rate and local thermal gradient at the same time; the sensing unit and the edge node are directly connected by an anti-interference coaxial cable.

3. A temperature control system for the production of special lubricating oil based on machine learning as claimed in claim 1, characterized in that: Inside the central processing module, there are a general CPU and a neural network processor; the CPU unit runs a real-time operating system, is responsible for logic control, task scheduling, and security monitoring, and synchronously processes the device status data from 20 parallel threads; the neural network processor is built with a dedicated matrix operation unit, executes the prediction model inference in a pipeline manner, and exchanges intermediate calculation results with the CPU through a memory sharing mechanism.

4. A temperature control system for the production of special lubricating oil based on machine learning according to claim 1, characterized in that: The data preprocessing module deploys a dedicated feature extraction engine, which is built with a sliding window analysis algorithm to calculate the temperature change rate in real time, and combines wavelet transform technology to strip high-frequency noise from the sensor signal; for the lubricating oil production characteristics, the feature library pre-sets 8 process parameter extraction templates such as thermal inertia coefficient and gradient mutation threshold, and dynamically adjusts the feature combination weight through an adaptive weighting strategy.

5. The temperature control system for the production of special lubricating oil based on machine learning according to claim 1, characterized in that: The internal of the time series prediction model module is set with a dual-scale memory network architecture; the network body is composed of a two-layer gated recurrent unit. The first layer captures the minute-level temperature fluctuation law, and the second layer analyzes the hourly-level production cycle characteristics. The layers are connected by residual connections to prevent gradient disappearance; the model input end integrates a sliding time window mechanism, and each time it processes a production data slice containing 30 consecutive minutes, and the output end generates a prediction trajectory for the next 5 minutes through a fully connected layer mapping.

6. The temperature control system for the production of special lubricating oil based on machine learning according to claim 1, wherein: The dynamic multi-objective optimization module adopts a hybrid intelligent algorithm architecture, and the core realizes multi-objective collaborative decision-making through a three-order optimization framework; the first order is the reinforcement learning real-time response layer, which quickly generates an initial control strategy based on the proximal policy optimization algorithm, and captures the instantaneous disturbance characteristics of the production process through an online learning mechanism; The second order is the multi-objective genetic algorithm optimization layer, which uses the improved NSGA-III algorithm to screen the optimal solution set in the Pareto front, and introduces an adaptive crossover and mutation operator to balance the global search and local development capabilities; the third order is the physical constraint verification layer, relying on the safety boundary condition library constructed by the knowledge graph, double-checking the optimization results for thermodynamic feasibility and equipment tolerance, and realizing real-time trade-off of temperature control accuracy, energy consumption efficiency and equipment life through a dynamic weight allocation mechanism, where the weight coefficient is automatically adjusted according to the production stage, giving priority to ensuring temperature stability during the isothermal reaction period, and focusing on energy-saving optimization during the heating and cooling stages. The dynamic weight adaptive formula is: In the formula, W_t represents the dynamic weight of temperature accuracy within the time window t, η is the initial weight base, γ is the time decay coefficient, ΔTmax is the threshold of the maximum allowable temperature deviation, ΔTt is the current actual temperature deviation, and α is the sensitivity adjustment factor; through the combined action of the exponential decay function and the real-time temperature deviation, this formula enables the system to automatically strengthen the temperature control weight during sudden disturbances, and increase the proportion of the energy-saving weight under stable operating conditions, realizing the adaptive balance of multiple objectives; The equipment degradation compensation formula is: In the formula, C_d is the equipment degradation compensation amount, NN is the cumulative working cycle number of the equipment, μ and σ respectively represent the normal distribution parameters of the equipment performance attenuation, β is the material fatigue coefficient, τ is the thermal stress cycle strength, and k is the non-linear correction exponent.

7. The temperature control system for the production of special lubricating oil based on machine learning according to claim 1, characterized in that: The underlying storage structure of the rule base module decomposes more than 200 pieces of expert experience into atomic rule units. The middle-layer inference engine supports multi-condition parallel matching, and the top-layer dynamic loader activates the corresponding rule clusters according to the production stage; the exception handling strategy adopts a hierarchical management mode. The first-level rules directly block dangerous operations, the second-level rules trigger compensation control, and the third-level rules generate manual review prompts; a two-way verification channel is established between the rule base and the prediction model. When the model output deviates from the process common sense, a correction coefficient is automatically injected, and at the same time, the newly occurred abnormal cases are fed back to the rule base for iterative update.

8. A temperature control system for the production of special lubricating oil based on machine learning as described in claim 1, characterized in that: The internal of the execution module is provided with a dynamic compensation type power drive system, which includes a multi-stage actuator collaborative working mechanism; the electric heating unit adopts pulse width modulation technology, and realizes the output precision adjustment at the 0.1% level through the IGBT power module. The cooling system is equipped with a fast-response solenoid valve to control the liquid nitrogen injection flow; the drive controller is built-in with a feedback compensation algorithm, which compares the set value with the actual output of the actuator in real time, and automatically starts the second-order PID compensation when a deviation of more than 0.5% is detected; the module is equipped with a dual safety interlock, and immediately switches to the preset safety mode under communication interruption or over-limit working conditions to ensure that the temperature fluctuation of the reactor is always within the process allowable range.

9. The temperature control system for the production of special lubricating oil based on machine learning according to claim 1, characterized in that: The backbone network of the communication module adopts a dual-ring topology structure, deploys a nanosecond-level clock synchronization mechanism, and aligns the clock signals of each module through the Precision Time Protocol; the data transmission is divided into three independent virtual channels: the real-time control instruction channel ensures an end-to-end transmission delay of 10ms, the monitoring data channel provides a bandwidth of 200Mbps to carry the sensor data stream, and the maintenance channel realizes remote diagnosis and firmware upgrade.

10. The temperature control system for the production of special-purpose lubricating oil based on machine learning according to claim 1, characterized in that: The operating platform of the human-computer interaction module is equipped with a laser holographic projection device, which projects the three-dimensional temperature field of the reactor onto the operation area in the form of color gradients in real time, and supports gesture recognition to realize dynamic parameter adjustment; the data display layer of the human-computer interaction module adopts a multi-screen fusion technology, the main monitoring screen presents the global temperature trend curve, and the auxiliary screen displays the real-time values of 16 key points; the mobile terminal of the human-computer interaction module accesses the system through digital certificate authentication, realizes hierarchical management of control permissions under the 5G network, and the operation records are stored using blockchain technology.

Images