Intelligent control method and system for pressure regulation of energy accumulator

Abstract

The invention discloses an energy accumulator pressure regulation intelligent control method and system. The method comprises the steps that historical operation data are collected; time sequence features of the data are extracted, and feature data are obtained through calculation; constructing a physical information neural network, and constructing a physical constraint loss function training model; performing real-time data acquisition and preprocessing, and performing real-time monitoring through a sensor group to obtain data; performing prediction based on the trained physical information neural network to obtain a predicted future pressure sequence; and defining a dynamic weight multi-objective optimization objective function, and solving by adopting a numerical optimization algorithm to obtain an optimal control sequence. According to the energy accumulator pressure regulation intelligent control method and system, it is guaranteed that the predicted pressure sequence result conforms to the thermodynamic law, meanwhile, the target function is defined by combining the pressure tracking error, the energy consumption variable and the valve abrasion variable, the predicted pressure sequence output by the physical information neural network is optimized, the timeliness and accuracy are greatly improved, and the method and system are suitable for popularization and application. And the production quality is improved.

Description

Technical Field

[0001] The present invention belongs to the field of accumulator pressure regulation, and in particular relates to an accumulator pressure regulation intelligent control method and system. Background Art

[0002] In hydraulic systems, accumulators store and release energy to mitigate pressure fluctuations, ensuring stable operation of the entire system. Accumulators absorb excess energy in the hydraulic system and release it when needed to maintain pressure balance. This energy management is crucial to improving the efficiency and reliability of hydraulic systems, especially in applications with frequent load changes or requiring rapid response.

[0003] However, traditional accumulator pressure regulation methods have some significant limitations. First, these methods often rely on manual monitoring and adjustment, which not only increases operational complexity but can also lead to response delays, as manual operations cannot match the response speed of automated systems. Second, traditional methods have a slow response speed, which is particularly evident in the face of rapidly changing load conditions. This can prevent the system from adjusting pressure in a timely manner, thus affecting performance. Furthermore, due to their lack of flexibility, traditional methods are also limited in their efficiency in energy recovery and reuse, which not only wastes energy but also can increase system operating costs.

[0004] Therefore, the above problems need to be solved urgently. Summary of the Invention

[0005] Purpose of the invention: In order to overcome the above shortcomings, the purpose of the present invention is to provide an intelligent control method and system for accumulator pressure regulation. By constructing a physical constraint loss function to train a physical information neural network, it is ensured that the predicted pressure sequence results conform to the laws of thermodynamics and are consistent with actual production; at the same time, by combining pressure tracking error, energy consumption variables and valve wear variables to define the objective function, the predicted pressure sequence output by the physical information neural network is optimized, which greatly improves the timeliness and accuracy and improves the production quality.

[0006] Technical Solution: To achieve the above objectives, the present invention provides an intelligent control method for accumulator pressure regulation, comprising:

[0007] S1): Collect historical operation data and pre-process the collected data;

[0008] S2): Extract the time series features of the data to obtain the original time series data, calculate the theoretical pressure deviation, and fuse it with the original time series data to obtain feature data. By extracting the time series features of the data and combining them with the theoretical pressure deviation, the dynamic characteristics of the system can be better captured.

[0009] Furthermore, it also includes S3): constructing a physical information neural network and constructing a physical constraint loss function training model; using feature data as input to complete the training of the physical information neural network; the model does not rely solely on data during training, but can also utilize physical knowledge to achieve better performance.

[0010] Furthermore, S301): the physical constraint loss function embeds the theoretical pressure deviation in the loss function to ensure that the prediction result conforms to the laws of thermodynamics; the physical constraint loss function is used to ensure that the prediction result conforms to the laws of thermodynamics, thereby improving the prediction accuracy of the model.

[0011] Furthermore, it also includes S4): real-time data collection and preprocessing, which is to obtain data through real-time monitoring of the sensor group and preprocess the collected data; through real-time data collection and preprocessing, the current status of the system can be obtained in time.

[0012] Furthermore, it also includes S5): calling S2 to obtain the theoretical pressure deviation, and making a prediction based on the trained physical information neural network to obtain a predicted future pressure sequence; predicting the future pressure sequence based on the trained model to provide support for real-time control.

[0013] Furthermore, the method further includes S6): defining a dynamic weight multi-objective optimization objective function and solving it using a numerical optimization algorithm to obtain an optimal control sequence; defining a dynamic weight multi-objective optimization objective function, and being able to dynamically adjust the optimization objective according to the real-time state, thereby achieving a more flexible control strategy;

[0014] S601): Analyze sensor data in real time and obtain a target pressure sequence based on a static rule base;

[0015] S602): Calculate the mean square error between the predicted pressure sequence and the target pressure sequence to obtain a pressure tracking error;

[0016] S603): Setting energy consumption variables and valve wear variables, and defining an objective function based on the pressure tracking error, energy consumption variables, and valve wear variables;

[0017] S604): A numerical optimization algorithm is used to solve and determine the optimal control sequence. During the optimization process, not only pressure tracking error is considered, but also energy consumption variables and valve wear variables are incorporated. This achieves multi-objective optimization, ensuring that the control strategy not only meets pressure control requirements but also strikes a balance between energy consumption and equipment life. This allows the optimal trade-off between multiple objectives to be found, thereby improving overall performance, reducing operating costs, and extending service life.

[0018] Furthermore, the physical constraint loss function Loss formula described in S301 is as follows:

[0019] Loss=MSE(y pred ,y true )+λ·MSE(p pred ·V actual ,nRT actual )

[0020] Among them, y pred is the future pressure sequence predicted by the model; true is the actual observed future pressure series; p pred is the pressure value at the current moment predicted by the model; V actual is the actual volume of the gas chamber, which is calculated by dividing the liquid injection by the release volume; nR is the gas constant, where n is the amount of gas substance and R is the universal gas constant; T actual is the gas temperature measured by the sensor; λ is the weight coefficient of the physical constraint term. By incorporating the ideal gas law as a constraint term into the loss function, we ensure that the model predictions conform not only to the data but also to known physical laws, thereby improving the model's physical consistency. Furthermore, because the model relies not only on data but also on the laws of physics, it improves the model's predictive accuracy and robustness.

[0021] Furthermore, the formula for the pressure tracking error J1 in S602 is as follows:

[0022]

[0023] Among them, p pred,k is the pressure value of the kth step predicted by the model; p target,k is the target pressure value at step k. The pressure tracking error J1 provides a quantitative indicator to evaluate the closeness between the pressure predicted by the model and the target pressure. By calculating the pressure tracking error J1 in real time, the prediction error can be quickly identified.

[0024] Furthermore, the formulas for the energy consumption variable J2 and the valve wear variable J3 in S603 are as follows:

[0025]

[0026] Among them, u k is the valve opening at step k, obtained through data acquisition;

[0027] The objective function J formula is as follows:

[0028] J=w1J1+w2J2+w3J3

[0029] Here, w1, w2, and w3 are the weights for the pressure tracking error J1, the energy consumption variable J2, and the valve wear variable J3, respectively. By incorporating the energy consumption variable J2 and the valve wear variable J3, the hydraulic system's energy efficiency and the equipment's service life can be factored into the objective function J, improving the overall effectiveness of the optimization. Furthermore, the weights w1, w2, and w3 allow decision-makers to customize the balance between performance (such as pressure tracking accuracy) and costs (such as energy consumption and equipment wear) based on their specific circumstances, ensuring flexibility in the control approach.

[0030] Furthermore, the formula for the theoretical pressure deviation Δp in S2 is as follows:

[0031]

[0032] Among them, p actual is the pressure measured by the sensor; T actual is the actual temperature measured by the sensor; V actual The volume of the gas chamber is inferred from the change in liquid volume. Calculating the theoretical pressure deviation can help predict the future pressure change trend of the system and provide a basis for decision-making in the model.

[0033] Furthermore, the physical information neural network in S3 includes an input layer, two LSTM layers, and an output layer; the input layer, two LSTM layers, and output layer are sequentially connected. LSTMs contain one or more memory cells that can store long-term information and access it when needed, effectively processing time series features and improving accuracy.

[0034] Furthermore, S5 also includes setting a threshold value. If the theoretical pressure deviation exceeds the threshold value, an alarm is triggered. By setting the threshold value, pressure deviation can be monitored in real time to ensure that the pressure is always within a safe and effective range, reducing unexpected downtime caused by equipment failure and ensuring production continuity and stability.

[0035] The present invention also provides an accumulator pressure regulation intelligent control system for implementing the accumulator pressure regulation intelligent control method, including: a data collection module, a pressure prediction module and an optimization control module; the data collection module is connected to the pressure prediction module to transmit real-time data and historical data to the pressure prediction module; the data collection module is also connected to the optimization control module to transmit real-time data to the optimization control module; the pressure prediction module is connected to the optimization control module to transmit pressure prediction results to the optimization control module. The data collection module focuses on data collection and preprocessing to ensure data quality and availability; the pressure prediction module focuses on data-based modeling and prediction, using historical and real-time data to make accurate pressure predictions; the optimization control module focuses on making optimization decisions based on prediction results and real-time data to generate optimal control strategies. The three modules have clear division of labor, so that each module can be independently developed and optimized, simplifying the complexity of the system.

[0036] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0037] 1. The present invention provides an intelligent control method and system for accumulator pressure regulation. By constructing a physical constraint loss function to train a physical information neural network, the predicted pressure sequence results are ensured to conform to thermodynamic laws and be consistent with actual production.

[0038] 2. The present invention provides an intelligent control method and system for accumulator pressure regulation. By defining the objective function by combining pressure tracking error, energy consumption variables and valve wear variables, the objective function is optimized, and the predicted pressure sequence output by the physical information neural network is optimized, thereby greatly improving the timeliness and accuracy and enhancing the production quality.

[0039] 3. Invent an intelligent control method and system for accumulator pressure regulation. The optimization algorithm based on data and models can provide operators with a scientific basis for decision-making, and can quickly provide the optimal control strategy even in complex situations, thereby improving the operating efficiency of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a step diagram of an intelligent control method for accumulator pressure regulation according to the present invention;

[0041] Figure 2 This is an architectural diagram of an accumulator pressure regulation intelligent control system described in the present invention. DETAILED DESCRIPTION

[0042] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0043] Example

[0044] In this embodiment, Figure 1 The present invention discloses an intelligent control method for accumulator pressure regulation, comprising:

[0045] S1): Collect historical operation data and pre-process the collected data;

[0046] S2): Extract the time series features of the data to obtain the original time series data, calculate the theoretical pressure deviation, and fuse it with the original time series data to obtain feature data;

[0047] S3): Construct a physical information neural network and a physical constraint loss function training model; use the feature data as input to complete the training of the physical information neural network;

[0048] S301): The physical constraint loss function embeds a theoretical pressure deviation in the loss function to ensure that the prediction result conforms to the laws of thermodynamics;

[0049] S4): Real-time data acquisition and preprocessing: acquiring data through real-time monitoring by the sensor group and preprocessing the collected data;

[0050] S5): Call S2 to obtain the theoretical pressure deviation, and make a prediction based on the trained physical information neural network to obtain the predicted future pressure sequence;

[0051] S6): Define the dynamic weight multi-objective optimization objective function and use the numerical optimization algorithm to solve it to obtain the optimal control sequence;

[0052] S601): Analyze sensor data in real time and obtain a target pressure sequence based on a static rule base;

[0053] S602): Calculate the mean square error between the predicted pressure sequence and the target pressure sequence to obtain a pressure tracking error;

[0054] S603): Setting energy consumption variables and valve wear variables, and defining an objective function based on the pressure tracking error, energy consumption variables, and valve wear variables;

[0055] S604): A numerical optimization algorithm is used to solve the problem and obtain an optimal control sequence.

[0056] Specifically, the timing features in S2 include pressure sequence, temperature sequence, flow sequence and valve opening; the pressure sequence is the sampling value of the past N seconds, such as once per second, with a total history window of 60 seconds; the temperature sequence is the liquid temperature; the flow sequence is the liquid injection rate; and the valve opening is the control signal.

[0057] Specifically, the data preprocessing in S1 and S4 includes data cleaning, sliding window segmentation and normalization processing;

[0058] The data cleaning includes but is not limited to removing abnormal values caused by sensor failures such as pressure suddenly dropping to negative values and interpolating missing values;

[0059] The sliding window segmentation includes setting the input window, output window and defining the sliding step. The input window takes the time step t as the starting point and takes the data of the past 60 seconds as the input feature. The output window takes the pressure values of the next 10 seconds from t+1 to t+10 seconds as the label. The sliding step is 1 second to generate continuous samples.

[0060] The normalization process performs Min-Max normalization on each feature column.

[0061] Specifically, in S2, the theoretical pressure deviation is spliced to the original time series data to achieve data fusion and obtain feature data.

[0062] In this embodiment, the physical constraint loss function Loss formula described in S301 is as follows:

[0063] Loss=MSE(y pred ,y true )+λ·MSE(p pred ·V actual ,nRT actual )

[0064] Among them, y pred is the future pressure sequence predicted by the model; true is the actual observed future pressure series; p pred is the pressure value at the current moment predicted by the model; V actual is the actual volume of the gas chamber, which is calculated by dividing the liquid injection by the release volume; nR is the gas constant, where n is the amount of gas substance and R is the universal gas constant; T actual is the gas temperature measured by the sensor; λ is the weight coefficient of the physical constraint term.

[0065] Specifically, in the loss function, the theoretical pressure deviation Δp is implicitly embedded; by forcing the model to predict p pred Satisfy p pred V actual ≈nRT actual , ensuring that the prediction results conform to the laws of thermodynamics;

[0066] For example, if the model predicts p pred =100bar, and according to the real-time temperature T actual and volume V actualIf the calculated theoretical pressure deviates far from this value, the physical constraint term in the loss function will penalize this deviation and force the model to correct the predicted value.

[0067] In this embodiment, the formula for the pressure tracking error J1 in S602 is as follows:

[0068]

[0069] Among them, p pred,k is the pressure value of the kth step predicted by the model; p target,k is the target pressure value of step k.

[0070] Specifically, p target,k The system analyzes the pressure, temperature and flow data of the sensor in real time and compares them with the preset target values in the static rule base to obtain the corresponding pressure sequence; the static rule base presets several operating modes (such as "gear shifting", "high load" and "leakage") and corresponding sensor data thresholds.

[0071] In this embodiment, the formulas for the energy consumption variable J2 and the valve wear variable J3 in S603 are as follows:

[0072]

[0073] Among them, u k is the valve opening at step k, obtained through data acquisition;

[0074] The objective function J formula is as follows:

[0075] J=w1J1+w2J2+w3J3

[0076] Among them, w1, w2 and w3 are the weights of pressure tracking error J1, energy consumption variable J2 and valve wear variable J3 respectively.

[0077] Specifically, the sum of pump power + valve power can be used as the valve opening u k An optional alternative is as follows; the pump power is related to the flow rate and is proportional to the square of the speed; the valve power is related to the amplitude of the valve opening adjustment, such as frequent and large valve movements will increase energy consumption.

[0078] In particular, w1, w2, and w3 can introduce smooth transition as a preferred method to achieve dynamic weight adjustment and increase the amount of calculation in exchange for higher recommendation accuracy.

[0079] In this embodiment, Figure 1 , the formula for the theoretical pressure deviation Δp in S2 is as follows:

[0080]

[0081] Among them, p actual is the pressure measured by the sensor; T actual is the actual temperature measured by the sensor; V actual It is the volume of the gas chamber inferred from the change in liquid volume.

[0082] Specifically, the model input includes not only the original time series data (pressure, temperature, flow, valve opening), but also the theoretical pressure deviation Δp as a key feature. The input dimensions are:

[0083] [Pressure, temperature, flow, valve opening, theoretical pressure deviation].

[0084] In this embodiment, Figure 1 The physical information neural network in S3 includes an input layer, two LSTM layers and an output layer; the input layer, two LSTM layers and the output layer are connected in sequence.

[0085] Specifically, the number of neurons in each LSTM layer is set to 64 as a preferred method to ensure the calculation depth while avoiding excessive calculation complexity; the fully connected layer outputs the pressure sequence for the next 10 seconds.

[0086] In this embodiment, Figure 1 , the S5 also includes setting a threshold value, and triggering an alarm if the theoretical pressure deviation exceeds the threshold value.

[0087] Specifically, the theoretical pressure deviation Δp is calculated in real time, and the threshold is set to 2% as a preferred embodiment. If the absolute value of the theoretical pressure deviation Δp exceeds the threshold, the following actions are triggered:

[0088] (1) Alarm: Notify operation and maintenance personnel to check for system leaks or sensor failures;

[0089] (2) Control fallback: suspend model predictive control and switch to PID control until the deviation recovers.

[0090] The present invention also discloses an intelligent control system for accumulator pressure regulation, which is used to implement the intelligent control method for accumulator pressure regulation, and is characterized in that it includes: a data collection module, a pressure prediction module and an optimization control module; the data collection module is connected to the pressure prediction module to transmit real-time data and historical data to the pressure prediction module; the data collection module is also connected to the optimization control module to transmit real-time data to the optimization control module; the pressure prediction module is connected to the optimization control module to transmit pressure prediction results to the optimization control module.

[0091] Specifically, the data collection module needs to integrate and deploy pressure sensors, temperature sensors, flow meters and other sensors to collect pressure, temperature, flow and other data in real time; data exchange between modules needs to be implemented through standardized interfaces, and detailed API documentation should be written for each interface, including the interface name, function, input and output parameters, return value, etc., to ensure data format consistency and transmission reliability; the collected data needs to be stored in a local database for subsequent analysis and use.

[0092] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements can be made without departing from the principles of the present invention. These improvements should also be regarded as the scope of protection of the present invention.

Claims

1. An intelligent control method for accumulator pressure regulation, characterized by: include: S1): Collect historical operation data and pre-process the collected data; S2): Extract the time series features of the data to obtain the original time series data, calculate the theoretical pressure deviation, and fuse it with the original time series data to obtain feature data; S3): Construct a physical information neural network and a physical constraint loss function training model; Take the feature data as input to complete the training of the physical information neural network; S301): The physical constraint loss function embeds a theoretical pressure deviation in the loss function to ensure that the prediction result conforms to the laws of thermodynamics; S4): Real-time data acquisition and preprocessing: acquiring data through real-time monitoring by the sensor group and preprocessing the collected data; S5): Call S2 to obtain the theoretical pressure deviation, and make a prediction based on the trained physical information neural network to obtain the predicted future pressure sequence; S6): Define the dynamic weight multi-objective optimization objective function and use the numerical optimization algorithm to solve it to obtain the optimal control sequence; S601): Analyze sensor data in real time and obtain a target pressure sequence based on a static rule base; S602): Calculate the mean square error between the predicted pressure sequence and the target pressure sequence to obtain a pressure tracking error; S603): Setting energy consumption variables and valve wear variables, and defining an objective function based on the pressure tracking error, energy consumption variables, and valve wear variables; S604): A numerical optimization algorithm is used to solve the problem and obtain an optimal control sequence.

2. The intelligent control method for accumulator pressure regulation according to claim 1, characterized in that: The physical constraint loss function Loss formula described in S301 is as follows: Loss=MSE(y pred ,y true )+λ·MSE(p pred V actual ,nRT actual ) Among them, y pred is the future pressure sequence predicted by the model; true is the actual observed future pressure series; p pred is the pressure value at the current moment predicted by the model; V actual is the actual volume of the gas chamber, which is calculated by dividing the liquid injection by the release volume; nR is the gas constant, where n is the amount of gas substance and R is the universal gas constant; T actual is the gas temperature measured by the sensor; λ is the weight coefficient of the physical constraint term.

3. The intelligent control method for accumulator pressure regulation according to claim 1, characterized in that: The formula of the pressure tracking error J1 in S602 is as follows: Among them, p pred,k is the pressure value of the kth step predicted by the model; p target,k is the target pressure value of step k.

4. The intelligent control method for accumulator pressure regulation according to claim 3, characterized in that: The formulas for the energy consumption variable J2 and the valve wear variable J3 in S603 are as follows: Among them, u k is the valve opening at step k, obtained through data acquisition; The objective function J formula is as follows: J=w1J1+w2J2+w3J3 Among them, w1, w2 and w3 are the weights of pressure tracking error J1, energy consumption variable J2 and valve wear variable J3 respectively.

5. The intelligent control method for accumulator pressure regulation according to claim 1, characterized in that: The formula for the theoretical pressure deviation Δp in S2 is as follows: Among them, p actual is the pressure measured by the sensor; T actual is the actual temperature measured by the sensor; V actual It is the volume of the gas chamber inferred from the change in liquid volume.

6. The intelligent control method for accumulator pressure regulation according to claim 1, characterized in that: The physical information neural network in S3 includes an input layer, two LSTM layers and an output layer; the input layer, two LSTM layers and the output layer are connected in sequence.

7. The intelligent control method for accumulator pressure regulation according to claim 1, characterized in that: The step S5 further includes setting a threshold value, and triggering an alarm if the theoretical pressure deviation exceeds the threshold value.

8. An intelligent control system for accumulator pressure regulation, used to implement an intelligent control method for accumulator pressure regulation according to claims 1 to 7, characterized in that: include: Data collection module, pressure prediction module and optimization control module; The data collection module is connected to the pressure prediction module to transmit real-time data and historical data to the pressure prediction module; the data collection module is also connected to the optimization control module to transmit real-time data to the optimization control module; the pressure prediction module is connected to the optimization control module to transmit pressure prediction results to the optimization control module.

Images