Intelligent heat dissipation system of oil-free digital energy air compression station

Abstract

The invention provides an intelligent heat dissipation system of an oil-free digital energy air compression station, which relates to the technical field of heat dissipation of air compression stations and comprises a sensor module, a correlation analysis module and a heat dissipation execution module which are in communication connection. Statistical correlation, causal correlation and time correlation between each operation parameter and temperature are analyzed, a physical mechanism is introduced, a hybrid correlation model is constructed, a predicted temperature change trend is combined, specific operation parameters influencing the temperature change at the current moment are analyzed, specific equipment is positioned according to the operation parameters, and a heat dissipation strategy is executed. Through a closed loop of model prediction, parameter attribution, equipment positioning and accurate heat dissipation, data-driven correlation analysis and causal reasoning of a physical mechanism are combined, rapid positioning from temperature abnormity to a specific heating source is realized, and low efficiency of traditional equipment-by-equipment troubleshooting is avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of air compression station heat dissipation, and particularly relates to an intelligent heat dissipation system for an oil-free digital energy air compression station. BACKGROUND

[0002] In the compression process of an oil digital energy air compression station, oil participates in air compression and absorbs heat generated by compression, and then the oil is cooled through an oil cooler and the like. The cooled oil is recycled to the compressor to continue heat dissipation. The heat source is relatively concentrated. In addition, the oil also plays a role in lubricating the running parts in the main machine. However, a small amount of oil remains in the compressed air after oil-gas separation, which pollutes the compressed air. At the same time, the treatment of waste oil is also an environmental problem.

[0003] In order to solve this problem, an oil-free digital energy air compression station emerges as the times require. In the compression process, oil is not needed to participate, and heat dissipation mainly relies on air, water and other media to take away heat through air flow or water circulation to overcome high temperature generated in the compression process. Although oil reduction reduces environmental pollution, an additional heat dissipation structure needs to be designed, and the heat source is relatively dispersed.

[0004] Currently, there are many drawbacks in the heat dissipation of the oil-free air compression station. Common passive heat dissipation methods, such as simply relying on fans with fixed rotating speed or cooling water with constant flow for heat dissipation, lack perception and response to the actual running state of the equipment, resulting in low heat dissipation efficiency. Although some air compression stations adopt a simple temperature control system, the heat dissipation is triggered based on a single temperature threshold, which cannot accurately identify the root cause of temperature abnormalities, and is easy to cause "excessive heat dissipation" or "insufficient heat dissipation", which not only wastes energy, but also may shorten the service life of the equipment due to continuous high temperature in the key position. In addition, the traditional method is difficult to cope with the influence of multiple parameters coupling under complex working conditions, for example, when the load of the equipment fluctuates and the environmental temperature changes, it is difficult to accurately determine the specific influence of each operating parameter on the temperature, resulting in difficult fault location and high maintenance cost.

[0005] Therefore, how to accurately locate the heating position through the perception and response to the actual running state of the equipment and execute accurate heat dissipation strategy is a technical problem to be solved at present. SUMMARY

[0006] To solve the above problems, the present application provides an intelligent heat dissipation system for an oil-free digital energy air compression station, which can analyze the statistical correlation, causal correlation and time correlation between each operating parameter and temperature, introduce a physical mechanism, construct a mixed correlation model, and analyze the specific operating parameter affecting the temperature change at the current time according to the predicted temperature change trend, locate the specific equipment according to the operating parameter, and execute the corresponding heat dissipation strategy.

[0007] To achieve the above purpose, the technical scheme adopted by the present application is as follows: The application provides an oil-free digital energy air compression station intelligent heat dissipation system, comprising a sensor module, an association analysis module and a heat dissipation execution module which are sequentially communicatively connected. The sensor module is used for collecting various operation parameters of the air compression station equipment through multiple types of sensors and pre-processing the collected data. The association analysis module is used for analyzing statistical correlation, causal correlation and time correlation between various operation parameters and temperature, and constructing a hybrid association model by comprehensively analyzing the correlation results. The heat dissipation execution module is used for combining operation parameters, reversely analyzing the predicted temperature change trend, determining specific operation parameters affecting the current temperature change, positioning to specific heating equipment or parts according to the operation parameters, and performing heat dissipation treatment.

[0008] In the sensor module, the pre-processing includes data cleaning, data conversion, data alignment and synchronization and data labeling.

[0009] In the association analysis module, before performing correlation analysis, feature extraction is first performed on the pre-processed data to extract more physically meaningful features from the original sensor data; the analysis object is determined, and the target variable is the temperature parameter of the key equipment of the air compression station, and the independent variable is various operation parameters.

[0010] The association analysis module comprises a statistical correlation analysis unit, a causal correlation analysis unit and a time correlation analysis unit. The statistical correlation analysis unit is used for analyzing statistical correlation between various operation parameters and temperature. The causal correlation analysis unit is used for analyzing causal correlation between various operation parameters and temperature, establishing a causal diagram and obtaining a chain conduction path. The time correlation analysis unit is used for analyzing time correlation between various operation parameters and temperature.

[0011] In the statistical correlation analysis unit, the following steps are included: When analyzing the correlation between numerical value type parameters and temperature, the correlation coefficient is calculated, and the calculated value is compared with a preset threshold value, so as to judge the correlation strength between data; When analyzing the correlation between type parameters and temperature, the method of variance analysis is adopted, the mean value difference of temperature under different operation modes is compared, and the influence significance of the mode type on temperature is judged; According to the analysis result of statistical correlation, the operation parameters strongly correlated with temperature are retained, and a statistical correlation analysis model based on linear regression is constructed to analyze the statistical association between various operation parameters and temperature. According to the linear relationship between the current operating parameter and the temperature, the current operating parameter is input into a corresponding linear model, a model outputs a predicted temperature, if the predicted temperature exceeds a preset threshold, a heat dissipation strategy adjustment is triggered, and meanwhile, new data is collected regularly and model parameters are updated online using a stochastic gradient descent algorithm.

[0012] In the causal correlation analysis unit, the causal correlation between each operating parameter and the temperature is determined in combination with a physical mechanism, including the following steps: Operating parameters strongly correlated with the temperature screened by the statistical correlation analysis are recorded as a feature set, and the target variable is set as the temperature. In combination with a preliminary analysis of the physical mechanism of the air compression station, the potential causal relationship direction is determined, an initial causal hypothesis graph is established, and direct causal relationships and indirect causal relationships are marked. A causal correlation analysis model based on a causal random forest is constructed and trained to determine the causal effect strength of each operating parameter. The effect significance test is used to determine whether there is a direct causal relationship between the operating parameter and the temperature, and the physical consistency is used to verify whether the causal effect direction is reasonable. The result of the direct causal relationship analysis is further analyzed to determine whether there is an intermediate variable, and the indirect causal relationship is mined through recursive analysis. Based on the direct causal relationship and the indirect causal relationship, a causal relationship graph is constructed, and the causal direction between nodes is marked. The trained model and the causal path are verified. The model and the causal graph are optimized through key path priority sorting, intervention strategy design, and model iteration.

[0013] In the causal correlation analysis unit, a causal correlation analysis model based on a causal random forest is constructed and trained, including the following steps: Given an intervention variable, operating parameters are divided into an intervention group and a control group, and individual causal effect values are calculated. The strongly correlated parameters are regarded as intervention variables respectively, and the causal effect of one operating parameter is analyzed each time, with the remaining operating parameters as covariates. For each intervention variable, a binary treatment variable is constructed. The data set is randomly divided into a training set and a test set; the training set is used to fit the causal random forest model, and the optimization target is to minimize the mean square error of the estimated causal effect and the true effect; the model estimates the average causal effect in each leaf node through recursive division of the feature space. For each sample, the conditional causal effect estimate value of the intervention variable on the temperature is output. The causal effect strength of each operating parameter is evaluated through permutation importance or SHAP value.

[0014] In the time correlation analysis unit, the following steps are included: Time series data is input, including a running parameter sequence and a temperature sequence; Stationarity test is performed on the input data; A time correlation analysis model based on a vector autoregressive model is constructed and trained; The time correlation of each running parameter and temperature is determined according to the sign and size of the regression coefficient; The time series causal chain is verified, and dynamic response analysis is performed through the impulse response function.

[0015] In the correlation analysis module, after the correlation analysis, the analysis results of statistical correlation, causal correlation and time correlation are fused and analyzed, and a hybrid correlation model is constructed through weighted fusion.

[0016] In the heat dissipation execution module, the execution process includes the following steps: The predicted temperature output by the hybrid correlation model and the predicted temperature change trend in the future period of time are obtained, and the abnormal temperature rise area is identified; The SHAP value or feature importance ranking is used to quantify the contribution of each running parameter to the current temperature change; The running parameters with high contribution are screened as key influence parameters, and pseudo-correlation parameters are excluded in combination with the air compression station process logic; Based on the established causal diagram, the path is traced back, the time leading relationship between the running parameters is determined through impulse response analysis, and the running parameter with the highest contribution and clear causal chain is determined as the dominant parameter of the current temperature change; The mapping relationship between the running parameters and the equipment is established, and the specific equipment is located based on the spatial positioning algorithm and heat conduction path analysis; Different heat dissipation strategies are adopted for the located different equipment, and continuous monitoring is performed.

[0017] The beneficial effects of the present application are as follows: The present application provides an oil-free digital energy air compression station intelligent heat dissipation system, which collects running parameters by using multiple types of sensors and performs preprocessing, analyzes the statistical correlation, causal correlation and time correlation between each running parameter and temperature, introduces physical mechanism, constructs a hybrid correlation model, analyzes the specific running parameters affecting the temperature change at the current time in combination with the predicted temperature change trend, and locates the specific equipment according to the running parameters to execute the corresponding heat dissipation strategy. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a schematic diagram of an intelligent heat dissipation system of an oil-free digital energy air compression station. DETAILED DESCRIPTION

[0019] Referring to Figure 1 The present application relates to an intelligent heat dissipation system of an oil-free digital energy air compression station, comprising a sensor module, an association analysis module, a prediction model construction module and a heat dissipation execution module connected in sequence; The sensor module is used for collecting various operating parameters of the air compression station equipment through multiple types of sensors, and pre-processing the collected data. High-precision temperature sensors are installed at key heat generating parts of the air compressor motor, compression chamber, cooler inlet and outlet, and positions prone to heat accumulation such as gas storage tanks and pipelines, to sense the temperature changes of each part of the equipment in real time. Pressure sensors are deployed at the air inlet and air outlet of the air compressor and at the key nodes of the pipeline to monitor the gas pressure. The pressure measurement range can be set according to the actual working pressure range of the air compression station. Flow sensors are installed at the inlet and outlet pipes of the cooling medium (such as water in the water cooling system and air in the air cooling system) to monitor the flow of the cooling medium in real time. For rotating equipment such as air compressor motors and fan motors, speed sensors are installed to monitor the motor speed in real time, providing a basis for analyzing the equipment operating conditions and energy consumption. The speed measurement range is determined according to the rated speed of the motor.

[0020] The data collected by all sensors is transmitted to the data center through wired (such as industrial Ethernet) or wireless (such as LoRa, NB-IoT, etc. low-power wide-area network technology) methods for preprocessing. The preprocessing includes data cleaning, data conversion, data alignment and synchronization, and data labeling.

[0021] The data cleaning includes detecting and processing abnormal values, filling missing values, and filtering noise.

[0022] Detecting and processing abnormal values: Calculate the mean and standard deviation, and use the 3σ principle (about 99.7% of the data falls within the mean ± 3 times the standard deviation under normal distribution) to identify abnormal values that deviate significantly from the normal range. For parameters such as temperature and pressure that change continuously over time, use sliding window or exponential smoothing method to detect sudden changes at adjacent time points (such as sudden temperature jump beyond reasonable range). If the abnormal value is caused by temporary sensor failure or transmission interference (such as transient pulse noise), it can be corrected by the mean or median of adjacent time points (for example, replace the abnormal value with the average temperature of the previous 1 minute and the next 1 minute). For abnormal data that cannot be corrected (such as continuous error value caused by sensor damage), directly eliminate the data point and ignore it in subsequent interpolation.

[0023] Missing value filling: For scenarios where data is missing for a short period of time and the change trend is linear (such as a brief loss of cooling water flow rate), we use linear fitting to fill in the missing values ​​using the values ​​at the previous and next time points. For parameters with large fluctuations, such as temperature, we construct smooth spline curves to fill in the missing values, avoiding the abrupt changes of linear interpolation.

[0024] Noise filtering: For high-frequency noise (such as sensor signal fluctuations caused by compressor vibration), a Butterworth low-pass filter or moving average filter is used to smooth the signal curve while preserving low-frequency trend characteristics. For non-stationary signals (such as sudden changes in data during startup and shutdown), the signal is decomposed through wavelet transform, removing high-frequency noise components and reconstructing the signal.

[0025] Normalize or standardize the cleaned data to make them have the same dimension and scale. Normalization maps the data to the interval [0, 1], or standardizes the data to a standard normal distribution with a mean of 0 and a standard deviation of 1 to eliminate the influence of dimension and facilitate subsequent model processing.

[0026] Because different sensor types may have different sampling frequencies (for example, a temperature sensor collects data once per second, while a flow sensor collects data once every 5 seconds), timestamp alignment is necessary to bring the data into the same time frame. Specifically, high-frequency data is downsampled (for example, by taking the average every minute) to match the sampling frequency of low-frequency data; low-frequency data is upsampled (for example, by interpolating data at intermediate time points) to match the timestamps of high-frequency data.

[0027] The data is annotated with status labels based on the start / stop signals and operating mode switching signals (e.g., the compressor switches from full load to no load) from the air compressor station equipment. For example, data is labeled with states such as "startup phase," "normal operation," and "shutdown cooling" to avoid model training errors caused by mixing data from different operating phases.

[0028] The correlation analysis module is used to analyze the statistical correlation, causal correlation and time correlation between various operating parameters and temperature, and to construct a hybrid correlation model based on the correlation analysis results; In the association analysis module, before performing correlation analysis, feature extraction is first required on the preprocessed data to extract more physically meaningful features from the raw sensor data. For example, the temperature change rate (∆T / ∆t) can reflect the heating rate of the device; the pressure fluctuation coefficient (standard deviation / mean) can characterize the impact of airflow stability on heat generation; and the cooling efficiency index (cooling medium inlet and outlet temperature difference × flow rate) can quantify the real-time performance of the cooling system to enhance the data's ability to characterize temperature changes.

[0029] When determining the analysis object, the target variable (dependent variable) is determined as the temperature parameter (such as the temperature of the compressor motor, the temperature of the bearing, the temperature of the cooling medium inlet and outlet, etc.) of the key equipment of the air compression station; the independent variable is various operating parameters, such as the equipment operating state parameter (compressor speed, inlet pressure, outlet flow, motor power, loading rate, etc.), environmental parameter (environmental temperature, humidity, cooling water flow rate, air volume, etc.), time parameter (operation time, start-stop cycle, etc.), etc.

[0030] The correlation analysis module includes a statistical correlation analysis unit, a causal correlation analysis unit, and a time correlation analysis unit. The statistical correlation analysis unit is configured to analyze the statistical correlation between various operating parameters and temperature, and specifically includes the following steps: When analyzing the correlation between numerical parameters and temperature, the correlation coefficient is calculated, and the calculated value is compared with a preset threshold value, so as to determine the correlation strength between the data.

[0031] The Pearson correlation coefficient is suitable for linear relationship, and is used to calculate the correlation coefficient r between the operating parameter and the temperature, and the value range is [-1, 1], and the closer the absolute value is to 1, the stronger the linear correlation is. The correlation coefficient threshold range is set, for example, when |r|>0.7, it is a strong correlation (such as motor power and motor temperature), 0.3≤|r|≤0.7 is a moderate correlation (such as cooling water flow rate and bearing temperature), and |r|<0.3 is a weak correlation (such as non-key pipeline pressure). The parameters strongly correlated with temperature are reserved, such as compressor inlet pressure, outlet flow, motor current, cooling water flow rate, and environmental humidity.

[0032] The Spearman rank correlation coefficient is suitable for non-linear relationship or non-normal distribution data (such as the segmented linear relationship between the device loading rate and the temperature), and is used to calculate the correlation based on the parameter rank, set the corresponding threshold, determine the correlation strength, and reserve the parameters strongly correlated with the temperature.

[0033] When analyzing the correlation between the type parameters and the temperature, the method of analysis of variance (ANOVA) is used, that is, by comparing the mean value difference of the temperature under different operating modes (such as full load / half load), the significance of the influence of the mode type on the temperature is determined (for example, if the p value of F test is less than 0.05, it is considered that the working condition type is significantly correlated with the temperature).

[0034] According to the analysis result of the statistical correlation, the operating parameters strongly correlated with the temperature are reserved, and a statistical correlation analysis model based on linear / non-linear regression is constructed to analyze the statistical correlation between the operating parameters and the temperature. For linear correlation, a multiple linear regression model (MLR) is constructed, and the model formula is: ; where T t represents the temperature at time t (target variable), represents the i-th operating parameter at time t, β i represents the regression coefficient, estimated by least squares, the regression coefficient represents the degree of influence of the independent variable on the dependent variable, n is the number of operating parameters, and ϵ t represents a random error term, which follows N(0, σ 2 ). The model is evaluated by the coefficient of determination R 2 , and the closer R 2 is to 1, the better the model fits.

[0035] For nonlinear correlation, a polynomial regression model is constructed, and the model formula is: ; where β i and γ i represent the regression coefficients, respectively, and the quadratic term is introduced to capture the nonlinear relationship (e.g., after the cooling air volume exceeds the critical value, the temperature decreases slowly, showing a quadratic function relationship), and the problem is transformed into a multivariate linear regression problem by variable substitution (let Z=X 2 ).

[0036] The preprocessed and feature-extracted data is divided into training set and test set, the training set data is used to train the constructed linear / nonlinear model, and the test set data is used for verification, after the model is trained, the evaluation index is calculated to evaluate the model, such as root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R 2 ), if R 2 >0.8 and the residual distribution is random, the model is qualified, according to the Occam razor principle, the model with simple structure is preferred.

[0037] In practical application, according to the linear / nonlinear relationship between the current operating parameters and the temperature, the current operating parameters are input into the corresponding linear / nonlinear model, the model outputs the predicted temperature, if the predicted temperature exceeds the preset threshold, the heat dissipation strategy is adjusted. At the same time, new data is collected regularly, and the model parameters are updated online using stochastic gradient descent algorithm (SGD).

[0038] For multiple collinearity, calculate the variance inflation factor, if VIF i >10, the variable is removed or principal component analysis is used. Draw a scatter plot of residual-predicted value to check whether there is heteroscedasticity; if the residual shows regularity, higher order polynomial or variable transformation is needed. The collinearity problem can be alleviated by regularization method.

[0039] For example, when predicting the temperature of the air compressor, a linear model is established: T = 65 + 0.03P + 0.15I - 0.2Q + 0.5T env ; where P is the pressure (bar), I is the current (A), Q is the cooling air volume (M 3 / min), and T env is the ambient temperature (℃). The regression coefficients in front of each operating parameter are standardized values, such as the regression coefficient of pressure and temperature is 0.03, which means that for every unit increase in pressure, the temperature is expected to increase by 0.03 units, indicating that pressure has a positive effect on temperature.

[0040] On the test set, R 2 = 0.87, RMSE = 1.8℃, and the temperature threshold risk can be predicted 30 minutes in advance.

[0041] When the operating parameter is cooling power, a nonlinear model of cooling power P cool and flow rate v is established: ; where T in is the inlet water temperature (℃). By taking the derivative, the extreme point v * = 5.8 m / s is found, at which the cooling efficiency is highest, guiding the control of the frequency conversion water pump.

[0042] The causal correlation analysis unit is configured to analyze the causal correlation between each operating parameter and the temperature to obtain a chain conduction path. Correlation only indicates the degree of association between variables, and causal relationship analysis needs further verification of whether the operating parameter is a driving factor of temperature change. For strongly correlated operating parameters, the causal relationship is determined by combining the physical mechanism of the air compression station, including the following steps: The operating parameters strongly correlated with the temperature selected by statistical correlation analysis are denoted as a feature set X = {X1, X2, …, X n}, and the target variable is set as the temperature Y. Preliminary analysis is combined with the physical mechanism of the air compression station to determine the potential causal relationship direction, establish an initial causal hypothesis graph, and label direct causal relationships and indirect causal relationships. For example, increasing the compressor power → generating heat through mechanical friction → increasing the cylinder temperature; increasing the cooling water temperature → decreasing the heat dissipation efficiency → increasing the temperature of the cooled gas; insufficient intake air flow → increasing the compression ratio → increasing the exhaust gas temperature. Among them, the direct causal relationship can be denoted as X i → Y, and the indirect causal relationship can be denoted as X i → X j → Y.

[0043] A causal correlation analysis model based on causal random forest is constructed and trained to determine the causal effect strength of each operating parameter; Given the intervention variable D i (whether to intervene in the operating parameter X i ), the operating parameters are divided into intervention and control groups, D i =1 for the intervention group and D i =0 for the control group, and the individual causal effect (ATE) is: τ(x)=E[Y(1)-Y(0)|X=x], where Y(1) is the potential outcome after intervention (temperature), Y(0) is the potential outcome without intervention, and X=x represents the value of the given feature vector X.

[0044] Strongly correlated parameters X i are considered as intervention variables, and the causal effect of each operating parameter is analyzed, with the remaining operating parameters as covariates.

[0045] For each intervention variable X i , a binary treatment variable is constructed, for example, if X i exceeds the threshold, then D i =1, otherwise D i =0.

[0046] Randomly divide the data set into training and test sets to ensure that the intervention and control groups are balanced in terms of covariate distribution. Use the training set to fit the causal random forest model, with the optimization goal being to minimize the mean square error of the estimated causal effect and the true effect; the model recursively divides the feature space and estimates the average causal effect in each leaf node.

[0047] For each sample x, output the conditional causal effect estimate of the intervention variable X i on the temperature Y .

[0048] Evaluate the causal effect strength of each operating parameter by permutation importance or SHAP value.

[0049] Determine whether there is a direct causal relationship between the operating parameter and the temperature through effect significance test, and verify whether the causal effect direction is reasonable through physical consistency; Effect significance test: through bootstrap or random permutation test, determine whether the estimated causal effect is significantly different from zero, if > threshold and p-value < 0.05, then X i has a direct causal relationship with Y.

[0050] Physical consistency verification: Combined with the air compression station mechanism, determine whether the direction of the causal effect is reasonable. For example, an increase in cooling water temperature should lead to an increase in gas temperature, and the causal effect is positive.

[0051] Further analyze the results of direct causal relationship analysis to determine whether there are intermediate variables, and explore indirect causal relationships through recursive analysis and conditional causal effect comparison; Recursive analysis: For confirmed direct causal relationship X i →Y, further analyze whether there is an intermediate variable X j Make X i →X j → Y. For example, if the compressor power X1 directly affects the cylinder temperature Y, and the increase in power leads to an increase in the equipment load factor X2, and the load factor X2 affects the temperature Y through the heat dissipation efficiency, then there is a chain path X1→X2→Y.

[0052] Conditional Causal Effect Comparison: Fixed Intermediate Variable X j , compare intervention variables X i If the effect of temperature Y before and after is weakened, it indicates that there is an indirect path. The mathematical expression is: fixed Original, then X i →X j →Y may be true.

[0053] Based on direct and indirect causal relationships, a causal relationship diagram is constructed, and the causal direction (effect intensity) between each node (operating parameter) is marked; For example, there is an indirect causal device load rate X2 between the compressor power X1 and the cylinder temperature Y. Then the causal effect intensity between the compressor power X1 and the cylinder temperature Y is: the sum of the effect intensity of the compressor power X1 and the device load rate X2, the effect intensity of the device load rate X2 and the cylinder temperature Y, and the effect intensity of the compressor power X1 and the cylinder temperature Y.

[0054] Validate the trained model and causal paths; Model validity verification: The prediction error and effect estimation stability of causal random forest were evaluated by k-fold cross validation.

[0055] Counterfactual testing: Design virtual intervention scenarios (such as forcibly lowering the cooling water temperature) to verify whether the temperature changes predicted by the model conform to the physical laws of the air compressor station.

[0056] Physical verification of causal paths: Combine thermodynamic laws (such as conservation of energy) and fluid mechanics principles to verify the physical reasonableness of each step in the chain path. For example, the increase in device load rate leading to insufficient heat dissipation area conforms to the physical logic of "load-heat dissipation efficiency-temperature". Invite domain experts to review the causal diagram and correct unreasonable relationships caused by data bias or model assumptions.

[0057] Optimization of model and causal diagram through critical path prioritization, intervention strategy design, and model iteration.

[0058] Critical path prioritization: Prioritize causal paths according to causal effect strength |τ| and path length, determine the chain path with the greatest impact on temperature, and focus on running optimization.

[0059] Intervention strategy design: Based on causal effect estimation, develop parameter control schemes and simulate intervention effects through historical data.

[0060] Model iteration: Regularly update data, incorporate new operating parameters, and retrain causal random forests to adapt to device aging or operating condition changes.

[0061] The time correlation analysis unit is configured to analyze the time correlation between each operating parameter and the temperature. To analyze the time correlation between operating parameters and temperature, we need to consider the time series dependence, dynamic correlation characteristics of time series data, and the physical mechanism of the air compression station, focusing on the lag effect and dynamic causal chain between variables. By quantifying the lag effect strength and leading relationship, we reveal the correlation mechanism between operating parameters and temperature in the time dimension, providing a basis for system optimization and prediction. This embodiment uses a vector autoregressive model (VAR) for analysis, including the following steps: The time series data includes operating parameter sequences and temperature sequences. Stationarity test is performed on the input data, such as using the unit root test (ADF test) to determine whether the sequence is stationary. If not, difference processing is used to convert it to a stationary sequence. Construct a time correlation analysis model based on the vector autoregressive model and train it. The mathematical expression of the model is: ; Where Y t represents the current temperature, Y t-i represents the lagged value of the temperature, X t-i represents the lagged value of the operating parameter, α i and β i represent the regression coefficients, respectively, measuring the influence of lagged items on the current temperature, p represents the lag order, and the optimal order is determined by the AIC / BIC criterion, and ϵt represents random error term, which needs to satisfy white noise assumption.

[0062] Select the order that makes the AIC / BIC value minimum among different p values, and determine it as the lag order p. Use the least squares method (OLS) to estimate the regression coefficients α i , β i , and obtain the quantitative relationship between each lag term and temperature. Test the model through residual test (check if the residual is white noise, through Ljung-Box test) and stability test (if the characteristic root is within the unit circle, the model is stable, through AR root test).

[0063] According to the sign and size of the regression coefficient β i , judge the time correlation of each operating parameter and temperature; Specifically, the positive and negative of the regression coefficient β i reflects the direction of correlation (positive / negative), and the absolute value size reflects the influence strength, and the significant non-zero regression coefficient β i indicates that there is a time correlation between the operating parameter corresponding to the lag order and the temperature.

[0064] Verify through the time causal chain, and analyze the dynamic response through the impulse response function.

[0065] Time causal chain verification: for example, in the air compression station, "intake flow X1" and "compressor power X2" may change before "exhaust temperature Y", through the significance of the lag coefficient in the time correlation analysis model, it can be judged whether the lag term of the intake flow X1 or the compressor power X2 has a significant impact on the exhaust temperature Y, and the reasonableness of the correlation is verified combined with physical logic (such as power increase → compression work increase → temperature rise).

[0066] Dynamic response analysis: on the basis of the model, through the impulse response function (IRF), analyze the dynamic influence path of a unit shock of a certain operating parameter on temperature, if the temperature rises significantly in the future k period after the operating parameter increases, it indicates that there is a positive time correlation, and the lag time is k.

[0067] In the correlation analysis module, after the correlation analysis, the analysis results of statistical correlation, causal correlation and time correlation are fused and analyzed, and a hybrid correlation model is constructed through weighted fusion method.

[0068] For the convenience of understanding and calculation, the temperature output by statistical correlation, causal correlation and time correlation analysis is uniformly represented by the letter T, and the model formula is: ; Among them, represents the total prediction result, represents the prediction value of the statistical correlation analysis model, represents the temperature response of the causal correlation analysis model (T t +ATE), represents the predicted value of the time-dependent analysis model, ω1, ω2, and ω3 represent weight coefficients, satisfying ω1+ω2+ω3=1, determined by cross-validation. The weight values ​​are updated using an adaptive update strategy.

[0069] The mixed association model is trained based on the data after preprocessing and feature extraction, and the weight coefficients are optimized. When validating the model, for the regression model, the root mean square error (RMSE) can be used to reflect the average deviation between the predicted value and the true value; the mean absolute error (MAE) can avoid the excessive influence of outliers on the evaluation; the coefficient of determination (R 2 ), with values ​​ranging from 0 to 1, where the closer to 1, the better the model fit; for time series models, the mean absolute percentage error (MAPE) can also be used to measure the relative error of the predicted value (e.g., MAPE = 5% means that the average prediction error is 5% of the true value).

[0070] The real-time prediction process is: Input the current operating parameter X into the statistical correlation analysis model, causal correlation analysis model and time correlation analysis model respectively. t ; Statistical correlation analysis model output ; The causal correlation analysis model calculates the causal effect and superimposes it to obtain ; The time-dependent analysis model is based on historical sequence prediction ; The final prediction is obtained by weighted fusion of statistical association, causal association and temporal association results ; like >threshold, the cooling strategy adjustment is triggered.

[0071] For example, suppose the following parameters are collected at a certain moment: Compressor speed X1 = 3000 rpm, cooling water flow rate X2 = 5 m / s, ambient temperature X3 = 30°C.

[0072] The prediction results of each model are: Statistical correlation analysis model: T 1 =75.3℃; Causal correlation analysis model: If the flow rate increases to 6m / s, ATE = -2.1℃, the predicted temperature T 2 =73.2℃; Time-dependent analysis model: Considering historical trends, predicting temperature T 3 =76.8℃; Fusion results (weights ω1=0.4, ω2=0.3, ω3=0.3) T 0 =0.4×75.3+0.3×73.2+0.3×76.8=75.06℃; System decision: The current temperature is close to the threshold (76°C), and the cooling water flow rate is automatically increased to 6 m / s. The temperature is expected to drop to 73.2°C to avoid the risk of equipment overheating.

[0073] The heat dissipation execution module is used to perform reverse analysis on the predicted temperature change trend in combination with the operating parameters, determine the specific operating parameters that affect the current temperature change, and locate the specific heating equipment or parts according to the operating parameters for heat dissipation processing.

[0074] The execution process includes the following steps: Obtain the predicted temperature output by the hybrid correlation model, as well as the predicted temperature change trend over a period of time in the future, and identify areas of abnormal warming; Use SHAP values ​​or feature importance ranking to quantify the contribution of each operating parameter to the current temperature change; When calculating the SHAP value, it is first necessary to set a temperature baseline value. In this embodiment, the baseline temperature is set to 60°C.

[0075] Select the operating parameters with the highest contribution as key influencing parameters, and combine them with the process logic of the air compression station to eliminate pseudo-correlated parameters; For example, if the sensor measurement delay causes the "exhaust temperature" to be strongly correlated with the "downstream valve opening", but the valve opening does not directly generate heat physically, this parameter needs to be eliminated and the parameters that conform to the causal logic need to be retained.

[0076] Based on the established causal graph, path tracing is performed. Through impulse response analysis, the time leading relationship between operating parameters is determined. For example, if power changes lead temperature changes by 5-10 minutes, the operating parameter with the highest contribution and clear causal chain is selected as the leading parameter of the current temperature change. Establish a mapping relationship between operating parameters and equipment / parts, and locate specific equipment / parts based on spatial positioning algorithms and heat conduction path analysis; Establish a parameter table of "equipment / position-monitoring parameter-sensor location-physical heat generation mechanism", for example, "compressor cylinder-cylinder temperature, compression power-cylinder surface-gas compression power, mechanical friction heat generation". Based on sensor location backtracking, if the key parameter is "compressor power", according to the table, it is known that the parameter corresponds to the compressor cylinder and the motor, and further comparing the temperature prediction values of the two positions, if the cylinder temperature rises more, it is located as the insufficient heat dissipation of the compressor cylinder; if the motor winding temperature is abnormal, it is located as the motor failure (such as bearing wear, winding short circuit).

[0077] Combined with the layout of the air compression station, the heat transfer path is analyzed. For example, the cooling water flow rate decreases→the heat exchange efficiency of the cooler decreases→the exhaust temperature of the compressor rises→the heat is conducted to the downstream storage tank through the pipeline, causing the wall temperature of the storage tank to rise. At this time, the dominant parameter is "cooling water flow rate", but the direct heat generation position is the cooler and the compressor cylinder.

[0078] For different equipment / positions located, targeted heat dissipation strategies are taken, and continuous monitoring is carried out.

[0079] For example, for the cooler (equipment), clean the heat exchange tube bundle, increase the cooling water flow rate, and replace the aging cooling medium. After implementing the heat dissipation measures, the temperature change trend is continuously monitored, the predicted value is compared with the actual value, the error rate is calculated, and if the error rate is >10%, it indicates that the positioning is inaccurate or the measures are ineffective, and the mapping relationship between the operating parameters and the equipment / positions needs to be rechecked. Add new fault case data, update the parameter-temperature correlation rules, and improve the accuracy of subsequent reverse analysis.

[0080] Air cooling system adjustment: if the problem located is related to the air cooling system, such as fan speed abnormality causing insufficient heat dissipation. When the temperature rises, the frequency converter of the fan motor is controlled to increase the fan speed and increase the air flow to enhance the heat dissipation effect.

[0081] Water cooling system adjustment: for the water cooling system, if the cooling medium flow rate is detected to be reduced, the frequency device of the water pump can be controlled to increase the water pump speed and increase the water flow. At the same time, check the valve state of the pipeline, if the valve opening is insufficient, automatically adjust the valve opening to the appropriate position. If it is found that the cooling effect of the cooler is reduced, it may be that the cooler is fouled, at which time the online cleaning system (such as chemical cleaning or physical cleaning device) can be started to clean the cooler to restore its heat exchange capacity. Equipment operation adjustment: if the temperature change is caused by the high load pressure of the air compressor, the operation mode of the air compressor can be adjusted according to the actual gas demand. For example, some air compressors are switched to unloaded operation state to reduce gas production and reduce load pressure, thereby reducing heat generation. At the same time, check whether there is leakage or abnormal gas use in the gas using equipment, repair the leakage point in time or adjust the operation parameters of the gas using equipment, optimize the gas balance, and reduce the overall load of the air compressor. Intelligent linkage control: each heat dissipation device and actuator can work cooperatively. For example, when temperature rise is detected, the air cooling system and the water cooling system can simultaneously adjust the heat dissipation according to their respective adjustment strategies, and cooperate with the air compressor operation mode adjustment to form an organic whole. Through intelligent linkage, rapid and efficient heat dissipation is achieved to ensure that the temperature of the air compression station equipment is always within a safe and stable range.

[0082] Using Tableau, PowerBI or industrial Internet of Things (IIoT) data visualization platform, real-time display of running parameter contribution, temperature field distribution and equipment positioning results. Deploy lightweight models locally to realize real-time warning and rapid response of temperature anomalies (such as completing parameter analysis-equipment positioning-alarm pushing within 5 minutes), through model prediction-parameter attribution-equipment positioning-precise cooling closed loop, combining data-driven correlation analysis and physical mechanism causal reasoning, realizing rapid positioning of specific heat sources from temperature anomalies, avoiding the inefficiency of traditional device-by-device troubleshooting.

[0083] The following is an example of a food processing workshop's oil-free digital energy air compression station, where the compressor cylinder temperature abnormally rises during operation.

[0084] Through temperature sensor, the compressor cylinder surface temperature is 92℃ (safety threshold is 95℃), pressure sensor collects the inlet pressure 1.2MPa, outlet pressure 8.5MPa, flow sensor collects the cooling water inlet flow 50L / min, outlet flow 48L / min (abnormal difference >5%), and speed sensor collects the compressor rotor speed 2900rpm.

[0085] Extract feature parameters, including temperature change rate, cooling efficiency index, and pressure fluctuation coefficient.

[0086] Through statistical correlation analysis, the correlation coefficient of cylinder temperature and cooling water flow is r=-0.82 (strong negative correlation), the correlation coefficient of cylinder temperature and compressor speed is r=0.75 (strong positive correlation), and the correlation coefficient of cylinder temperature and exhaust pressure is r=0.3 (moderate correlation). Retain strong correlation parameters, and through linear regression model, the predicted temperature is 91.5℃, with a deviation of 0.5℃ from the measured value.

[0087] Taking the cooling water flow as the intervention variable, divide the intervention group and the control group, and calculate the average causal effect ATE=-3.2℃ (i.e. the temperature decreases by 3.2℃ for every 10L / min increase in flow). Through causal diagram verification, cooling water flow→cooling efficiency→cylinder temperature (direct causal chain), without intermediate variables.

[0088] In the time correlation analysis, the VAR model lag order p = 2 is set, the regression coefficient shows that the lag 1 period of cooling water flow has an impact coefficient β =-0.45 (p < 0.01) on temperature, the lag 2 period of rotating speed has an impact coefficient β = 0.32 on temperature, and the lag 2 period of exhaust pressure has an impact coefficient β = 0.32 (p < 0.05) on temperature, that is, the changes of pressure and rotating speed lag behind the cooling water flow (lag 1 period) in the impact on temperature, indicating that the impact on temperature has a delay, and it is not an immediate dominant factor. Through pulse response analysis, after the cooling water flow is reduced by 10%, the cylinder temperature rises significantly by 1.5℃ after 5 minutes.

[0089] The weight coefficient values of the mixed correlation model are set, and the final current temperature prediction value is weighted and calculated as 91.8℃.

[0090] Based on the final temperature predicted by the mixed model, combined with the temperature change rate, the temperature rise rate caused by insufficient flow in the causal effect, and the prediction of the time series model for the lag effect, it is predicted that the temperature will rise to 95.3℃ after 10 minutes, exceeding the safety threshold of 95℃.

[0091] The current measured temperature 92℃ deviates from the T mixed value by 0.2℃, which is within the normal range, but the prediction trend shows that it will soon exceed the limit, triggering the reverse analysis process.

[0092] Through SHAP value analysis of the contribution of each parameter to the difference between the current temperature 92℃ and the baseline value (assuming 60℃), the SHAP value is calculated and sorted to get cooling water flow (SHAP = 2.1) > rotating speed (SHAP = 1.8) > exhaust pressure (SHAP = 0.9), and the dominant parameter is determined as cooling water flow, which needs to be analyzed first.

[0093] According to the causal correlation analysis model, it is obtained that the cooling water flow increases by 10 L / min, and the temperature decreases by 3.2℃, and the causal diagram verifies that "flow ↓→ cooling efficiency ↓→ temperature ↑" is a direct causal chain without intermediate variables. The time correlation analysis model shows that the lag 1 period (5 minutes ago) of cooling water flow β =-0.45, that is, the current temperature is sensitive to the change of flow 5 minutes ago, which confirms that the reduction of flow is the leading factor of temperature rise.

[0094] According to the mapping relationship between the established operating parameters and the equipment / position, combined with the physical layout analysis of the air compression station, the heat conduction path is determined, that is, flow ↓→ cylinder water jacket temperature ↑→ cylinder surface temperature ↑, and the measured cooling efficiency index 343 L・℃ / min < normal average 400 L・℃ / min, which confirms the insufficient heat dissipation.

[0095] Heat dissipation strategy execution: Start the standby cooling water pump, increase the flow to 60 L / min (frequency converter frequency from 40 Hz to 50 Hz), check the pipeline valve, find that the outlet valve opening is only 50%, automatically adjust to 80%, turn on the axial flow fan above the cylinder (speed from 1500 rpm to 2000 rpm), enhance air convection, at the same time reduce the compressor load rate from 85% to 75%, the speed is reduced to 2800 rpm, reduce heat production.

[0096] Effect verification: Through real-time monitoring, it is found that the cylinder temperature is reduced to 88℃ after 30 minutes, the cooling water flow is stable at 58L / min, the cooling efficiency index is restored to 410L・℃ / min, the deviation between the predicted value and the measured value of the mixed model is 0.8℃, the error rate is <10%, and the strategy is effective.

[0097] System optimization: Add this fault data (92℃ temperature rise case) to the training set, update the causal random forest model, strengthen the flow-temperature correlation weight, according to the response data this time, the cooling water flow abnormal threshold is lowered from 5% to 3%, the early warning is triggered.

[0098] The above embodiments only describe the preferred embodiments of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements of the technical solutions of the present application made by ordinary engineering technicians in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. An oil-free digital energy air compressor station intelligent heat dissipation system, characterized in that: It includes a sensor module, a correlation analysis module and a heat dissipation execution module which are sequentially communicatively connected; The sensor module is used to collect various operating parameters of the air compressor station equipment through multiple types of sensors and pre-process the collected data; The correlation analysis module is used to analyze the statistical correlation, causal correlation and time correlation between various operating parameters and temperature, and to construct a hybrid correlation model based on the correlation analysis results; The heat dissipation execution module is used to perform reverse analysis on the predicted temperature change trend in combination with the operating parameters, determine the specific operating parameters that affect the current temperature change, and locate the specific heating equipment or parts according to the operating parameters for heat dissipation processing.

2. The oil-free digital energy air compressor station intelligent heat dissipation system according to claim 1 is characterized in that: In the sensor module, the preprocessing includes data cleaning, data conversion, data alignment and synchronization, and data marking.

3. The oil-free digital energy air compressor station intelligent heat dissipation system according to claim 1 is characterized in that: In the association analysis module, before performing correlation analysis, it is first necessary to perform feature extraction on the preprocessed data to extract more physically meaningful features from the original sensor data; when determining the analysis object, the target variable is determined to be the temperature parameter of the key equipment of the air compressor station, and the independent variable is determined to be various operating parameters.

4. The oil-free digital energy air compressor station intelligent heat dissipation system according to claim 1 is characterized in that: The association analysis module includes a statistical correlation analysis unit, a causal correlation analysis unit and a time correlation analysis unit; The statistical correlation analysis unit is used to analyze the statistical correlation between various operating parameters and temperature; The causal correlation analysis unit is used to analyze the causal correlation between various operating parameters and temperature, establish a causal graph, and obtain a chain conduction path; The time correlation analysis unit is used to analyze the time correlation between various operating parameters and temperature.

5. The oil-free digital energy air compressor station intelligent heat dissipation system according to claim 4 is characterized in that: In the statistical correlation analysis unit, the following steps are included: When analyzing the correlation between numerical parameters and temperature, the correlation coefficient is calculated and the calculated value is compared with the preset threshold to determine the strength of the correlation between the data; When analyzing the correlation between the classification parameters and temperature, the variance analysis method was used to determine the significance of the impact of the mode type on the temperature by comparing the mean differences of the temperatures under different operating modes. Based on the results of statistical correlation analysis, the operating parameters that are strongly correlated with temperature are retained, and a statistical correlation analysis model based on linear regression is constructed to analyze the statistical correlation between various operating parameters and temperature; Based on the linear relationship between the current operating parameters and temperature, the current operating parameters are input into the corresponding linear model. The model outputs the predicted temperature. If the predicted temperature exceeds the preset threshold, the cooling strategy adjustment is triggered. At the same time, new data is collected regularly, and the model parameters are updated online using the stochastic gradient descent algorithm.

6. The oil-free digital energy air compressor station intelligent heat dissipation system according to claim 4 is characterized in that: In the causal correlation analysis unit, the causal correlation between each operating parameter and temperature is determined in combination with the physical mechanism, including the following steps: The operating parameters that are strongly correlated with temperature and screened out by statistical correlation analysis are recorded as the feature set, and the target variable is set as temperature; Combined with the preliminary analysis of the physical mechanism of the air compressor station, the potential causal relationship direction is clarified, an initial causal hypothesis diagram is established, and direct and indirect causal relationships are marked; Construct and train a causal correlation analysis model based on causal random forest to determine the causal effect strength of each operating parameter; The significance test is used to determine whether there is a direct causal relationship between the operating parameters and the temperature, and the physical consistency test is used to verify whether the direction of the causal effect is reasonable. Further analyze the results of direct causal relationship analysis to determine whether there are intermediate variables and explore indirect causal relationships through recursive analysis; Based on direct and indirect causal relationships, a causal relationship diagram is constructed, and the causal direction between each node is marked; Validate the trained model and causal path; Optimize models and causal diagrams through critical path prioritization, intervention strategy design, and model iteration.

7. The oil-free digital energy air compressor station intelligent heat dissipation system according to claim 6 is characterized in that: In the causal correlation analysis unit, a causal correlation analysis model based on causal random forest is constructed and trained, including the following steps: Given an intervention variable, the operating parameters are divided into intervention and control groups, and the individual causal effect values ​​are calculated; The strongly correlated parameters were treated as intervention variables, and the causal effect of one operating parameter was analyzed at a time, with the remaining operating parameters as covariates; For each intervention variable, a binary treatment variable was constructed; The dataset is randomly divided into training and test sets. The training set is used to fit a causal random forest model, with the optimization objective being to minimize the mean squared error between the estimated causal effect and the true effect. The model recursively partitions the feature space and estimates the average causal effect within each leaf node. For each sample, output the estimated conditional causal effect of the intervention variable on temperature; The strength of the causal effect of each operating parameter was assessed using permutation importance or SHAP value.

8. The oil-free digital energy air compressor station intelligent heat dissipation system according to claim 4 is characterized in that: In the time-correlation analysis unit, the following steps are included: Taking time series data as input, the time series data includes an operating parameter sequence and a temperature sequence; Perform a stationary test on the input data; Construct and train a time correlation analysis model based on a vector autoregression model; The time correlation between each operating parameter and temperature is determined based on the sign and size of the regression coefficient; Verification is performed through the timing causal chain, and dynamic response analysis is performed through the impulse response function.

9. The oil-free digital energy air compressor station intelligent heat dissipation system according to claim 3 is characterized in that: In the association analysis module, after performing correlation analysis, the analysis results of statistical association, causal association and temporal association are fused and analyzed, and a hybrid association model is constructed by a weighted fusion method.

10. The oil-free digital energy air compressor station intelligent heat dissipation system according to claim 1 is characterized in that: In the heat dissipation execution module, the execution process includes the following steps: Obtain the predicted temperature output by the hybrid correlation model, as well as the predicted temperature change trend over a period of time in the future, and identify areas of abnormal warming; Use SHAP values ​​or feature importance ranking to quantify the contribution of each operating parameter to the current temperature change; Select the operating parameters with the highest contribution as key influencing parameters, and combine them with the process logic of the air compression station to eliminate pseudo-correlated parameters; Based on the established causal diagram, path tracing is performed. Through impulse response analysis, the time leading relationship between operating parameters is determined. The operating parameter with the highest contribution and clear causal chain is selected as the dominant parameter of the current temperature change. Establish a mapping relationship between operating parameters and equipment, and locate specific equipment based on spatial positioning algorithms and heat conduction path analysis; Adopt targeted cooling strategies for different located devices and continuously monitor them.

Images