Hydraulic pump station dynamic cooling method based on intelligent algorithm

Abstract

The application relates to the technical field of heat dissipation of hydraulic pump stations, and discloses a dynamic cooling method for a hydraulic pump station based on an intelligent algorithm. The method first acquires a multi-source sensor data set of a hydraulic pump station, which comprises temperature distribution signals, pressure fluctuation signals and a flow change curve; then the corresponding data is processed through a multi-scale feature fusion algorithm and a modal decomposition algorithm respectively to obtain thermal anomaly features and a pressure disturbance atlas, and the flow change curve is processed to generate a flow time sequence correlation sequence; then the data is input into an adaptive decision model to generate a dynamic cooling control vector; and finally, a collaborative optimization algorithm is used to construct an optimal cooling regulation path to output a dynamic cooling instruction. The application realizes comprehensive and accurate monitoring and intelligent cooling control of the hydraulic pump station, effectively improves the operation stability and reliability of the pump station, reduces energy consumption, adapts to various working conditions, and has significant application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of heat dissipation technology for hydraulic pump stations, in particular to a dynamic cooling method for hydraulic pump stations based on intelligent algorithms. BACKGROUND

[0002] In industrial production, hydraulic pump stations, as key power equipment, are widely used in mechanical manufacturing, metallurgy, mining and many other fields. Its stable operation is crucial to the reliability and efficiency of the entire production system. However, during the operation of the hydraulic pump station, a large amount of heat will be generated due to mechanical friction, hydraulic oil flow resistance, etc. If it cannot be cooled in time and effectively, it will cause a series of serious problems.

[0003] With the continuous improvement of industrial automation, the power density and operating load of hydraulic pump stations continue to increase, and traditional cooling methods are increasingly difficult to meet the needs. For example, the mechanical ventilation cooling method, the regulation of ventilation volume is usually fixed or rough, and cannot be flexibly adjusted according to the actual temperature change of the pump station. When the pump station load is low and the heat is less, excessive ventilation will cause energy waste; while in high load operation of the pump station, the ventilation volume may be insufficient, resulting in poor cooling effect. Similarly, if the water cooling system cannot accurately control the flow and temperature of the cooling liquid, uneven cooling may occur, and the local overheating problem cannot be effectively solved.

[0004] From the perspective of monitoring and control, the traditional method can only obtain limited pump station operating parameters, such as single-point temperature measurement, and cannot fully grasp the complex thermal distribution and pressure fluctuation inside the pump station. This one-sided data acquisition method makes the subsequent cooling decision lack accuracy and scientificity, and it is difficult to achieve efficient cooling control of the pump station. In the complex scene of multi-working condition operation, the heat generation law and heat dissipation demand of the pump station are constantly changing, and the limitations of traditional cooling technology are more prominent, which seriously restricts the performance improvement of the hydraulic pump station and the efficient and stable operation of industrial production.

[0005] With the development trend of industrial intelligence, the demand for intelligent and precise cooling control of hydraulic pump stations is increasingly urgent. However, there is a lack of a mature and effective dynamic cooling solution based on intelligent algorithms in the market, which cannot meet the strict requirements of modern industry for efficient operation and reliable heat dissipation of hydraulic pump stations. Therefore, it has become a key problem in this field to develop an innovative dynamic cooling method for hydraulic pump stations. SUMMARY

[0006] The purpose of the present application is to provide a dynamic cooling method for hydraulic pump stations based on intelligent algorithms to solve the problems raised in the background.

[0007] To achieve the above object, the present application provides the following technical scheme: a hydraulic pump station dynamic cooling method based on an intelligent algorithm, the method comprising:

[0008] Obtaining a multi-source sensor data set of the hydraulic pump station; the multi-source sensor data includes a temperature distribution signal, a pressure fluctuation signal and a flow variation curve; the temperature distribution signal includes a pump body surface temperature and a real-time cooling liquid temperature, and the pressure fluctuation signal includes a pump cavity pressure peak value and a fluctuation frequency;

[0009] Based on the temperature distribution signal, thermal anomaly features are extracted by a multi-scale feature fusion algorithm, including local overheating areas, temperature gradient change rates and heat dissipation efficiency parameters;

[0010] According to the pressure fluctuation signal, a pressure disturbance atlas is generated by a modal decomposition algorithm, including a pressure fluctuation main frequency component and a harmonic energy distribution;

[0011] The flow variation curve is subjected to phase alignment processing to generate a flow time sequence correlation sequence;

[0012] The thermal anomaly features, the pressure disturbance atlas and the flow time sequence correlation sequence are input into an adaptive decision model to generate a dynamic cooling control vector;

[0013] Based on the dynamic cooling control vector, an optimal cooling regulation path is constructed by a collaborative optimization algorithm to output a hydraulic pump station dynamic cooling instruction.

[0014] Preferably, the thermal anomaly features are extracted by a multi-scale feature fusion algorithm, including:

[0015] The temperature distribution signal is subjected to wavelet denoising processing to generate denoised temperature field data;

[0016] Based on a pre-defined pump body thermodynamic benchmark model library, a local overheating area is identified by a multi-resolution matching algorithm, and a temperature anomaly sub-area is divided;

[0017] A temperature change slope of the temperature anomaly sub-area is extracted by a gradient calculation algorithm, and a heat dissipation efficiency deviation value is generated in combination with standard heat dissipation parameters in the benchmark model library;

[0018] The heat dissipation efficiency deviation value, the temperature gradient change rate and the local overheating area are encoded as structured thermal anomaly features.

[0019] Preferably, the pressure disturbance atlas is generated by a modal decomposition algorithm, including:

[0020] The pressure fluctuation signal is subjected to amplitude normalization processing to eliminate noise components with an amplitude below a threshold value;

[0021] Separate the main frequency components of pressure fluctuations based on the empirical mode decomposition algorithm, and calculate the harmonic energy proportion of each component;

[0022] Generate a disturbance level according to the harmonic energy proportion and a preset pressure stability threshold;

[0023] Correlate the disturbance level with the harmonic energy distribution to form a two-dimensional pressure disturbance map.

[0024] Preferably, the adaptive decision-making model comprises a feature dimension reduction module and a dynamic coupling module, and the feature dimension reduction module comprises:

[0025] The heat dissipation efficiency parameter in the thermal anomaly feature is standardized to obtain a first dimension reduction vector, the disturbance level in the pressure disturbance map is discretized and encoded to generate a second dimension reduction vector, and the flow time series correlation sequence is analyzed by window sliding to extract the flow fluctuation trend to obtain a third dimension reduction vector; and the first dimension reduction vector, the second dimension reduction vector and the third dimension reduction vector are combined into a low-dimensional decision sequence through a feature fusion layer.

[0026] Preferably, the optimal cooling regulation path is constructed by a cooperative optimization algorithm, comprising:

[0027] The node coordinates are initialized according to the positions of the cooling execution units, and an edge weight matrix is generated based on the response priority weight;

[0028] The dynamic cooling control vector is taken as a node attribute, and the edge weight matrix is composed of the energy consumption cost and the priority weight of the regulation sequence;

[0029] The optimal regulation value of each node is iteratively calculated by a multi-objective optimization equation, and the edge weight matrix is updated;

[0030] An optimal cooling regulation instruction sequence covering all nodes is generated according to the updated edge weight matrix.

[0031] Preferably, the construction method of the pump body thermodynamic benchmark model library comprises:

[0032] Thermodynamic simulation data of the hydraulic pump station under multiple working conditions are collected, and benchmark temperature distribution and heat dissipation efficiency characteristics are extracted;

[0033] The heat dissipation efficiency characteristics are interpolated under multiple working conditions to generate a multi-state benchmark model;

[0034] The benchmark models are classified according to the pump body types, and a thermodynamic parameter database is associated;

[0035] The classified benchmark models are stored as a benchmark model library, and the models are updated regularly based on measured data.

[0036] Preferably, the parameter optimization method of the empirical mode decomposition algorithm comprises:

[0037] Calculate initial decomposition layer number and stop criterion threshold according to historical pressure data distribution;

[0038] Select the parameter with the highest matching degree of the decomposition result and the measured pressure data by traversing the parameter combination through cross-validation algorithm;

[0039] Adjust the decomposition layer number and the stop criterion threshold dynamically according to the matching degree, and optimize the disturbance level division precision.

[0040] Preferably, the dynamic coupling module comprises:

[0041] Align the low-dimensional decision sequence with the time window to generate a dynamic correlation matrix;

[0042] Extract the correlation features between the execution units through the multi-head attention mechanism to generate a spatial coupling matrix;

[0043] Perform tensor product operation on the dynamic correlation matrix and the spatial coupling matrix to generate multi-dimensional fusion features;

[0044] Superimpose the multi-dimensional fusion features and the original low-dimensional decision sequence through the skip connection to output the dynamic cooling control vector.

[0045] Preferably, the construction method of the multi-objective optimization equation comprises:

[0046] Define the adjustment cost between nodes as a linear combination of energy consumption cost and response priority weight;

[0047] Initialize the adjustment value of each node to the initial state value, and the starting point adjustment value to zero;

[0048] Calculate the optimal adjustment value of each node based on the predecessor node through the iteration formula, and record the optimal predecessor node;

[0049] Generate a complete cooling adjustment instruction sequence according to the optimal predecessor node.

[0050] Preferably, the node of the optimal cooling adjustment path represents a cooling execution unit, and the edge represents the adjustment order and the response priority weight.

[0051] Compared with the prior art, the present application has the following advantages:

[0052] The hydraulic pump station dynamic cooling method based on the intelligent algorithm has many significant beneficial effects. In the data acquisition and analysis layer, by obtaining a hydraulic pump station multi-source sensor data set covering temperature distribution signals, pressure fluctuation signals and flow change curves, comprehensive and accurate monitoring of the pump station running state is realized. Among them, the pump body surface temperature and the real-time temperature of the cooling liquid in the temperature distribution signal can directly reflect the heating and heat dissipation of the pump station; the pump cavity pressure peak value and fluctuation frequency of the pressure fluctuation signal are helpful to analyze the influence of the pressure change in the pump station on heating; and the flow change curve reflects the flow characteristics of the hydraulic oil and the heat dissipation correlation. The multi-scale feature fusion algorithm is used to extract thermal anomaly features such as local overheating area, temperature gradient change rate and heat dissipation efficiency parameters, which can sensitively capture the thermal state anomaly of the pump station and provide key basis for accurate cooling. The modal decomposition algorithm generates a pressure disturbance atlas to show the pressure fluctuation main frequency component and harmonic energy distribution, which can help to understand the pressure fluctuation law and assist in optimizing the cooling strategy.

[0053] In the decision and control link, the adaptive decision model integrates the thermal anomaly features, the pressure disturbance atlas and the flow time sequence correlation sequence, and generates an accurate dynamic cooling control vector through a feature dimension reduction module and a dynamic coupling module. The feature dimension reduction module standardizes, discretizes and encodes different types of data features and extracts trends, effectively reducing the data dimension and retaining key information, and improving the decision efficiency. The dynamic coupling module fully excavates the correlation features between the execution units through time window alignment and multi-head attention mechanism, so that the decision is more in line with the actual demand. Based on this, the collaborative optimization algorithm constructs an optimal cooling regulation path, comprehensively considers the cooling execution unit position, response priority weight and energy consumption cost and other factors, and outputs a dynamic cooling instruction that can accurately control each cooling execution unit, realizing efficient and energy-saving cooling.

[0054] From the overall operation benefit, the present application greatly improves the stability and reliability of the hydraulic pump station operation. The accurate dynamic cooling effectively avoids equipment failure caused by overheating, reduces downtime maintenance time and reduces maintenance cost. In terms of energy saving, the cooling system is intelligently adjusted according to the actual running state, avoiding energy waste of traditional methods and improving energy utilization rate. Moreover, the method has strong adaptability and can flexibly adjust the cooling strategy under various working conditions, and is widely applicable to different types of hydraulic pump stations, providing a solid guarantee for efficient and stable operation of industrial production, and effectively promoting the intelligent development of the hydraulic pump station heat dissipation technology. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 The working principle diagram of the hydraulic pump station dynamic cooling method based on the intelligent algorithm described in the present application;

[0056] Figure 2 The flow chart for generating a pressure disturbance atlas by a modal decomposition algorithm;

[0057] Figure 3 Flowchart for adaptive decision model feature dimensionality reduction module;

[0058] Figure 4 Flowchart for adaptive decision model dynamic coupling module. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0060] Please refer to Figures 1-4 The present application provides a hydraulic pump station dynamic cooling method based on intelligent algorithm, aiming to realize efficient dynamic cooling control of the hydraulic pump station through multi-source sensor data acquisition and intelligent algorithm processing. The specific implementation scheme of the present application is described in detail below.

[0061] A multi-source sensor data set of the hydraulic pump station is obtained. The multi-source sensor data includes temperature distribution signals, pressure fluctuation signals and flow variation curves. Among them, the temperature distribution signals include the surface temperature of the pump body and the real-time temperature of the cooling liquid, which can intuitively reflect the thermal state of the hydraulic pump station; the pressure fluctuation signals include the peak value and fluctuation frequency of the pump cavity pressure, which can be used to analyze the pressure change; the flow variation curve reflects the dynamic change of the fluid flow in the hydraulic pump station. Through real-time collection of various sensors, these data provide a basis for subsequent analysis and processing.

[0062] Based on the temperature distribution signals, thermal anomaly features are extracted through a multi-scale feature fusion algorithm. The thermal anomaly features include local overheating areas, temperature gradient change rates and heat dissipation efficiency parameters. By analyzing the temperature distribution signals, the possible local overheating areas, temperature gradient and heat dissipation efficiency parameters can be accurately identified, so that a clearer understanding of the thermal abnormality of the hydraulic pump station can be obtained.

[0063] According to the pressure fluctuation signals, a pressure disturbance atlas is generated through a modal decomposition algorithm. The pressure disturbance atlas includes the main frequency components and harmonic energy distribution of the pressure fluctuation. By processing the pressure fluctuation signals with the modal decomposition algorithm, the main frequency components of the pressure fluctuation and the energy distribution of each harmonic can be obtained, which provides a basis for evaluating the pressure stability and potential disturbance factors.

[0064] The flow change curve is phase-aligned to generate a flow time sequence correlation sequence. The phase alignment processing can eliminate the inconsistency in the flow change curve caused by time difference and other factors, so that the flow data has better correlation in the time sequence, facilitating subsequent analysis of the rules and trends of flow changes.

[0065] The thermal anomaly features, pressure disturbance atlas and flow time sequence correlation sequence are input into an adaptive decision model to generate a dynamic cooling control vector. The adaptive decision model can comprehensively analyze these different types of data features, generate a control vector for dynamic cooling of the hydraulic pump station according to preset rules and algorithms, and provide a decision basis for subsequent cooling regulation.

[0066] Based on the dynamic cooling control vector, an optimal cooling regulation path is constructed by a collaborative optimization algorithm to output a dynamic cooling instruction for the hydraulic pump station. The collaborative optimization algorithm is based on the dynamic cooling control vector, considers factors such as the location of the cooling execution unit, energy consumption cost, response priority, etc., constructs an optimal cooling regulation path, and finally outputs specific dynamic cooling instructions to achieve effective cooling control of the hydraulic pump station.

[0067] The technical solutions of the present application will be further described in detail below in combination with specific embodiments.

[0068] Embodiment 1:

[0069] In practical applications, the operating environment of the hydraulic pump station is complex and variable, and obtaining accurate and reliable multi-source sensor data is the basis for effective cooling control. For temperature distribution signals, high-precision temperature sensors can be used, which are installed at key positions on the surface of the pump body and in the circulating pipeline of the cooling liquid. For example, select positions on the pump body surface where the temperature is prone to rise, such as the pump shaft, the sealing part, etc., and install thermocouple sensors to collect the pump body surface temperature in real time. For the real-time temperature of the cooling liquid, install thermistor sensors on the inlet and outlet pipelines of the cooling liquid to accurately measure the temperature of the cooling liquid at different positions.

[0070] The collection of pressure fluctuation signals depends on pressure sensors. The pressure sensors are installed inside the pump cavity to ensure accurate measurement of the pump cavity pressure peak and fluctuation frequency. The flow change curve can be obtained by an electromagnetic flowmeter installed on the main pipeline of the hydraulic system to monitor the flow change in real time.

[0071] After collecting multi-source sensor data, it needs to be pre-processed to improve data quality. For temperature distribution signals, due to the possibility of noise interference in the sensor measurement process, wavelet denoising processing can be used. The principle of wavelet denoising is to use wavelet transform to decompose the signal into sub-bands of different frequencies, and then according to the energy distribution characteristics of noise and signal in different sub-bands, the threshold value of the noise coefficient is processed to remove the noise component and generate denoised temperature field data. Let the original temperature distribution signal be , and the wavelet coefficient obtained after wavelet transform is , where is the scale parameter, is the translation parameter. By setting a suitable threshold , the wavelet coefficient is processed: when , the processed wavelet coefficient is ; when , let . Through inverse wavelet transform, the denoised temperature field data can be obtained.

[0072] Based on the pre-defined pump body thermodynamic benchmark model library, a multi-resolution matching algorithm is used to identify local overheating areas. The pump body thermodynamic benchmark model library contains pump body thermodynamic models under various working conditions, and each model has a corresponding benchmark temperature distribution. The multi-resolution matching algorithm compares the denoised temperature field data with the temperature distribution in the benchmark model, from coarse to fine, step by-step matching, finds the area with larger difference from the current temperature field data, and identifies it as a local overheating area, and divides it into temperature anomaly sub-regions. For example, set the temperature deviation threshold to , when the temperature of a certain area in the denoised temperature field is greater than than the temperature of the corresponding area in the corresponding benchmark model, mark the area as a local overheating area.

[0073] The gradient calculation algorithm is used to extract the temperature change slope of the temperature anomaly sub-region. In the temperature anomaly sub-region, select multiple sampling points, and according to the temperature values and position information of adjacent sampling points, calculate the temperature change slope. Let the temperature of sampling point be , and the position coordinate be , then the temperature change slope can be calculated by formula . Combined with the standard heat dissipation parameters in the benchmark model library, the heat dissipation efficiency deviation value is calculated. Let the standard heat dissipation parameter be , and the actual calculated heat dissipation parameter be , then the heat dissipation efficiency deviation value .

[0074] The heat dissipation efficiency deviation value, temperature gradient change rate and local overheating area are encoded as a structured thermal anomaly feature. For example, the position information of the local overheating area, the numerical range of the temperature gradient change rate and the positive and negative and size of the heat dissipation efficiency deviation value and other information can be encoded in a binary coded manner to form a unified structured feature vector, facilitating subsequent processing.

[0075] Embodiment 2:

[0076] The pressure fluctuation signal plays an important role in the operation analysis of the hydraulic pump station. The amplitude normalization processing of the collected pressure fluctuation signal can eliminate the influence of the difference of different signal amplitudes on the subsequent analysis. The amplitude normalization method is to map the amplitude of the pressure fluctuation signal to a specific interval, such as . Let the original pressure fluctuation signal be , and its amplitude range is , then the normalized signal can be calculated by the formula . In the normalization process, the noise components with amplitudes below the threshold value are removed. The threshold value can be determined by analyzing a large amount of historical data according to the actual situation. For example, it is found through statistical analysis that the signal with amplitude below (the value is only an example, and the actual value needs to be determined according to the specific situation) is usually noise, which can be removed.

[0077] Separate the pressure fluctuation main frequency component based on the empirical mode decomposition algorithm (EMD). EMD algorithm is an adaptive signal decomposition method, which can decompose complex signals into several intrinsic mode functions (IMF). Let the pressure fluctuation signal be , and after EMD algorithm decomposition, get IMF components , . Each IMF component has different frequency characteristics, and the IMF component with lower frequency usually contains the main trend and main frequency component of the signal. The energy of each IMF component is calculated, and the harmonic energy ratio of each component is calculated.

[0078] Generate the disturbance level according to the harmonic energy ratio and the preset pressure stability threshold. The preset pressure stability threshold can be determined according to the design requirements of the hydraulic pump station and the actual operation experience. For example, when the harmonic energy ratio is greater than the threshold value , it is considered that the disturbance level corresponding to the frequency component is high; when is between the threshold value and ( ), the disturbance level is medium; when Less than a threshold value When the disturbance level is low.

[0079] The disturbance level is associated with the harmonic energy distribution to form a two-dimensional pressure disturbance atlas. In the two-dimensional coordinate system, the abscissa represents the frequency, and the ordinate represents the harmonic energy proportion. According to the disturbance level of different frequency components, different colors or marks are used to distinguish, so as to intuitively show the main frequency component and harmonic energy distribution of the pressure fluctuation, and facilitate the analysis of the characteristics and stability of the pressure fluctuation.

[0080] Example 3:

[0081] The phase alignment processing of the flow change curve is crucial for accurately analyzing the flow change law. During the operation of the hydraulic pump station, the flow change curves collected at different times may have phase differences due to various factors. The purpose of phase alignment processing is to eliminate these differences and make the flow data have better comparability and correlation in time sequence.

[0082] First, determine the reference flow curve. The flow curve during a stable operation period can be selected as the reference curve, denoted as . For other flow change curves that need to be aligned , calculate the phase difference with the reference curve. A commonly used method is to use the cross-correlation function to calculate the phase difference. The cross-correlation function , by finding the delay time corresponding to the maximum value of the cross-correlation function, the phase difference of the two curves can be obtained.

[0083] According to the calculated phase difference, the flow change curve is translated. If the phase difference is positive, it means is lagging behind , and is translated forward by time units; if the phase difference is negative, then is translated backward by time units. The flow change curve after translation is basically aligned in phase with the reference curve.

[0084] The phase-aligned flow change curves are arranged in time sequence to generate a flow time sequence correlation sequence. In the process of generating the flow time sequence correlation sequence, the flow data can also be further processed, such as filtering, interpolation, etc., to improve the quality and stability of the data. For example, using the moving average filtering method, the flow data is smoothed to remove high-frequency noise in the data. Let the original flow data be , the flow data after moving average filtering can be calculated by the formula , where The size of the sliding window. Through such processing, the changing trend of traffic can be better reflected, providing more reliable data support for subsequent analysis and decision-making.

[0085] Embodiment 4:

[0086] The feature dimension reduction module in the adaptive decision model is the key to processing high-dimensional data and improving model efficiency. For the heat dissipation efficiency parameter in the heat anomaly feature, standardization is an important means to make different features have the same scale, which helps subsequent analysis and fusion. Assuming that the heat dissipation efficiency parameter is , the mean is denoted by , and the standard deviation is , then the first dimension reduction vector can be calculated according to the formula . Through this standardization processing, the heat dissipation efficiency parameters of different orders of magnitude can be converted to the same dimension, so that the parameter has a unified standard in subsequent model calculation, avoiding the influence of order of magnitude difference on model effect.

[0087] For the disturbance level in the pressure disturbance map, a second dimension reduction vector is generated by using discrete coding. For example, the disturbance level is divided into low, medium and high levels, represented by numbers , , respectively, and then these numbers are binary coded. If the disturbance level is low, it is coded as "00"; if it is medium, it is coded as "01"; if it is high, it is coded as "10". In this way, this binary vector becomes the second dimension reduction vector . This coding method can convert qualitative disturbance level information into a digital form that is easy for computers to process, facilitating subsequent operation and analysis in the model.

[0088] In processing the traffic time series correlation sequence, the traffic fluctuation trend is extracted by window sliding analysis to obtain the third dimension reduction vector. The specific operation is to set a fixed size window on the traffic time series correlation sequence, and the window will slide one by one according to the time sequence. Assuming that the window size is set to , in the first window, the traffic data is . First, calculate the mean of the traffic in the window, the formula is , the mean can reflect the average level of the traffic in the window; then calculate the standard deviation of the traffic in the window, the formula is , the standard deviation can reflect the dispersion degree of the traffic data in the window, that is, the fluctuation. Combine the mean and standard deviation into a vector, which is the third dimension reduction vector By window sliding analysis, representative fluctuation characteristics can be extracted from the traffic time series correlation sequence to provide a basis for subsequent decision-making.

[0089] Finally, the first reduced dimension vector , the second reduced dimension vector and the third reduced dimension vector are merged into a low-dimensional decision sequence by using a feature fusion layer. The feature fusion layer can use a simple splicing method, that is, the , , are sequentially connected to form a new vector; or a weighted fusion method can be used, according to the importance of different features to give corresponding weights. Assuming , , are the weights of , , , and satisfy , then the low-dimensional decision sequence can be calculated by the formula . This weighted fusion method can highlight the role of important features according to the actual situation, so that the low-dimensional decision sequence can better reflect the key information of the data, thereby improving the decision-making accuracy and running efficiency of the model.

[0090] Embodiment 5:

[0091] The dynamic coupling module of the adaptive decision model plays an important role in mining the correlation characteristics between data and optimizing the generation of dynamic cooling control vectors. The first step is to align the low-dimensional decision sequence in time windows to generate a dynamic correlation matrix. The specific method is to set a fixed-size time window on the low-dimensional decision sequence, and the time window will slide along the time sequence. Assuming that the low-dimensional decision sequence is , and the size of the time window is set to . In the first time window, the sequence data is ; in the first time window, the sequence data is . The dynamic correlation matrix is constructed by calculating the correlation between the low-dimensional decision sequences in different time windows. Here, the correlation coefficient is used to measure the correlation, and the calculation formula of the correlation coefficient is , where and represent the mean values of the sequences in the first and the first time window, respectively. Arrange all the correlation coefficients between the time windows into a matrix, and this matrix is the dynamic correlation matrix . The dynamic correlation matrix It can reflect the degree of correlation between low-dimensional decision sequences at different time stages, providing data connection information in the time dimension for subsequent analysis.

[0092] Next, a multi-head attention mechanism is used to extract the correlation features between execution units, thereby generating a spatial coupling matrix. Assuming a low-dimensional decision sequence... After linear transformation, we obtain , , Three matrices. For the first matrix in a multi-head attention mechanism... First, calculate the attention score, using the formula: , here yes The dimensions of the matrix are determined; then the weights are obtained by processing the matrix using the softmax function. ;No. The output of the size is Finally, put all the heads (assuming there are) By concatenating the outputs of (the two units) together, we obtain the spatial coupling matrix. ,Right now Multi-head attention mechanisms can simultaneously focus on information from different locations. Through parallel computation of multiple heads, they can more comprehensively capture the complex interrelationships between execution units, including the spatial coupling matrix. It reflects the connection between execution units in a spatial dimension, providing rich spatial information for subsequent feature fusion.

[0093] Next, the dynamic correlation matrix With spatial coupling matrix Tensor multiplication is performed to generate multidimensional fused features. Assume a dynamic correlation matrix. The dimension is Spatial coupling matrix The dimension is Their tensor product results The dimension is The specific calculation method is as follows: Tensor multiplication operations can deeply fuse information from the time and spatial dimensions, enabling multidimensional fusion features. It also includes data correlation information in time and space, reflecting the system's status more comprehensively.

[0094] Finally, skip connections are used to fuse multidimensional features. Compared with the original low-dimensional decision sequence The data is superimposed to output a dynamic cooling control vector. Skip connections preserve key information from the original low-dimensional decision sequences, preventing the loss of important data during feature fusion. Dynamic cooling control vector. The calculation formula is In this way, the newly mined multi-dimensional association features are integrated, and the characteristics of the original data are preserved, making the output dynamic cooling control vector more reasonable and accurate, and providing more accurate decision support for the cooling control of the hydraulic pump station.

[0095] Embodiment 6:

[0096] In the process of constructing the optimal cooling adjustment path, the collaborative optimization algorithm comprehensively considers multiple key factors. First, the node coordinates are initialized according to the positions of the cooling execution units, and the edge weight matrix is generated based on the response priority weights. In the hydraulic pump station, the positions of the cooling execution units are different, and their influence on the cooling effect also differs. For example, the cooling execution unit close to the heat source may play a more critical role in the cooling process. Assuming that there are cooling execution units, each cooling execution unit is regarded as a node, and its position coordinates are represented by , where The response priority weight is determined according to the importance and response speed of the cooling execution unit, and the response priority weight of the cooling execution unit to the cooling execution unit is denoted as The element in the edge weight matrix is composed of the energy consumption cost of the adjustment sequence and the response priority weight , that is, The edge weight matrix generated in this way comprehensively considers the energy consumption and response priority, providing basic data for subsequent optimization calculations.

[0097] Next, the dynamic cooling control vector is taken as the node attribute. The dynamic cooling control vector contains decision information for the cooling of the hydraulic pump station, and taking it as the node attribute allows the subsequent optimization process to better integrate the overall cooling demand. Assuming that the dynamic cooling control vector is , its elements are correspondingly assigned to the corresponding nodes as the node attribute values, so that each node carries information related to the cooling decision.

[0098] Then, the optimal adjustment value of each node is calculated by iterating the multi-objective optimization equation, and the edge weight matrix is updated. The construction process of the multi-objective optimization equation is as follows: the adjustment cost between nodes is defined as the linear combination of energy consumption cost and response priority weight. Let the adjustment cost from node to node be , where is the weight coefficient, and its value range is to between the energy consumption cost and the influence of the response priority weight on the adjustment cost. For example, when a larger value indicates that the energy consumption cost is more important in the adjustment cost; and when a smaller value indicates that the response priority weight is more important. The adjustment values of the nodes are initialized as the initial state values, and the start node adjustment value is set to zero. Then, the optimal adjustment value of each node based on the predecessor nodes is calculated by an iterative formula, and the optimal predecessor node is recorded. Assuming that the iterative formula is wherein represents the optimal adjustment value of node , and represents the set of predecessor nodes of node . In each iteration process, the node state is updated according to the calculated optimal adjustment value, and the elements in the edge weight matrix are recalculated according to the new node state, for example, the value of is adjusted according to the changes in the energy consumption cost and the response priority.

[0099] Finally, the optimal cooling adjustment instruction sequence covering all nodes is generated according to the updated edge weight matrix. Starting from the start node, the complete cooling adjustment instruction sequence is derived in reverse according to the recorded optimal predecessor node. This sequence determines the adjustment order and adjustment value of each cooling execution unit, so that the entire cooling adjustment process can meet the cooling demand while reducing energy consumption and improving response speed as much as possible, thereby achieving efficient cooling control of the hydraulic pump station.

[0100] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0101] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for dynamic cooling of a hydraulic pump station based on intelligent algorithms, characterized by, The method comprises the following steps: acquiring a multi-source sensor data set of a hydraulic pump station; the multi-source sensor data comprises a temperature distribution signal, a pressure fluctuation signal, and a flow variation curve; the temperature distribution signal comprises a pump body surface temperature and a real-time cooling liquid temperature, and the pressure fluctuation signal comprises a pump cavity pressure peak value and a fluctuation frequency; based on the temperature distribution signal, extracting thermal anomaly features by a multi-scale feature fusion algorithm, wherein the thermal anomaly features comprise a local overheating area, a temperature gradient change rate, and a heat dissipation efficiency parameter; generating a pressure disturbance atlas by a modal decomposition algorithm according to the pressure fluctuation signal, wherein the pressure disturbance atlas comprises a pressure fluctuation main frequency component and a harmonic energy distribution; performing phase alignment processing on the flow variation curve to generate a flow time sequence correlation sequence; inputting the thermal anomaly features, the pressure disturbance atlas, and the flow time sequence correlation sequence into an adaptive decision model to generate a dynamic cooling control vector; based on the dynamic cooling control vector, constructing an optimal cooling regulation path by a collaborative optimization algorithm to output a hydraulic pump station dynamic cooling instruction; the collaborative optimization algorithm comprises the following steps: initializing node coordinates according to the positions of cooling execution units and generating an edge weight matrix based on response priority weights; using the dynamic cooling control vector as a node attribute, and using the energy consumption cost and priority weight of the regulation sequence to form an edge weight matrix; iteratively calculating the optimal regulation value of each node by a multi-objective optimization equation, and updating the edge weight matrix; generating an optimal cooling regulation instruction sequence covering all nodes according to the updated edge weight matrix.

2. The method of claim 1, wherein the method further comprises: the multi-scale feature fusion algorithm comprises the following steps: performing wavelet denoising processing on the temperature distribution signal to generate denoised temperature field data; based on a pre-defined pump body thermodynamic benchmark model library, identifying a local overheating area by a multi-resolution matching algorithm, and dividing a temperature anomaly sub-area; using a gradient calculation algorithm to extract the temperature change slope of the temperature anomaly sub-area, and combining the standard heat dissipation parameters in the benchmark model library to generate a heat dissipation efficiency deviation value; the heat dissipation efficiency deviation value, the temperature gradient change rate, and the local overheating area are encoded as structured thermal anomaly features.

3. The method of claim 1, wherein the method further comprises: the modal decomposition algorithm comprises the following steps: performing amplitude normalization processing on the pressure fluctuation signal to eliminate noise components with an amplitude below a threshold value; based on an empirical mode decomposition algorithm, separating the pressure fluctuation main frequency component, and calculating the harmonic energy proportion of each component; generating a disturbance level according to the harmonic energy proportion and a pre-set pressure stability threshold value; associating the disturbance level with the harmonic energy distribution to form a two-dimensional pressure disturbance atlas.

4. The method of claim 1, wherein the method further comprises: the adaptive decision model comprises a feature dimension reduction module and a dynamic coupling module, and the feature dimension reduction module comprises the following steps: The heat dissipation efficiency parameter in the thermal anomaly feature is standardized to obtain a first dimension reduction vector, the disturbance level in the pressure disturbance pattern is discretely coded to generate a second dimension reduction vector, and the flow time series correlation sequence is analyzed by window sliding to extract a flow fluctuation trend to obtain a third dimension reduction vector; the first dimension reduction vector, the second dimension reduction vector and the third dimension reduction vector are combined into a low-dimensional decision sequence through a feature fusion layer.

5. The method of claim 2, wherein the method further comprises: The construction method of the pump body thermodynamic benchmark model library comprises: Collecting thermodynamic simulation data of the hydraulic pump station under multiple working conditions, extracting baseline temperature distribution and heat dissipation efficiency characteristics; Interpolating the heat dissipation efficiency characteristics under multiple working conditions to generate a multi-state baseline model; Classifying the baseline model according to the pump body type and associating the thermodynamic parameter database; Store the classified baseline model as a baseline model library, and update the model based on the measured data regularly.

6. The method of claim 3, wherein the method further comprises: The parameter optimization method of the empirical mode decomposition algorithm comprises: According to the distribution of historical pressure data, calculate the initial decomposition layer number and the stop criterion threshold; Select the parameter with the highest matching degree of the decomposition result and the measured pressure data by traversing the parameter combination through cross-validation algorithm; Adjust the decomposition layer number and the stop criterion threshold according to the matching degree to optimize the disturbance level division precision.

7. The method of claim 4, wherein the method further comprises: The dynamic coupling module comprises: Align the time window of the low-dimensional decision sequence to generate a dynamic correlation matrix; Extract the correlation features between execution units through a multi-head attention mechanism to generate a spatial coupling matrix; Perform tensor product operation on the dynamic correlation matrix and the spatial coupling matrix to generate multi-dimensional fusion features; Superimpose the multi-dimensional fusion features and the original low-dimensional decision sequence through a skip connection to output a dynamic cooling control vector.

8. The method of claim 1, wherein the method further comprises: The construction method of the multi-objective optimization equation comprises: Define the adjustment cost between nodes as the linear combination of energy consumption cost and response priority weight; Initialize the adjustment value of each node to the initial state value, and the starting point adjustment value to zero; Calculate the optimal adjustment value of each node based on the predecessor node through an iterative formula, and record the optimal predecessor node; Generate a complete cooling adjustment instruction sequence according to the optimal predecessor node.

9. The method of claim 1, wherein the method further comprises: The optimal cooling adjustment path represents the cooling execution unit, and the edge represents the adjustment order and the response priority weight.

Images