A centralized energy storage power station heat dissipation system

Abstract

The application relates to the technical field of energy storage heat dissipation, and discloses a centralized energy storage power station heat dissipation system. A data acquisition module of the system acquires historical operation data under different altitudes and generates a historical cooling parameter set; a feature analysis module extracts features from the historical operation data to obtain a historical feature parameter set, and a trend change feature matrix is established according to the historical feature parameter set; a pressure regulation module constructs a pressure control space based on the feature matrix, determines a feature early warning index set by calculating the parameter aggregation degree of historical abnormal events; a heat dissipation execution module extracts real-time feature parameters to establish a state vector, calculates the position deviation degree of the state vector and the early warning index set to generate a real-time pressure regulation factor. The pressure regulation module combines the real-time regulation factor, the historical feature parameters and the trend matrix to generate a pressure control model, outputs the cooperative control parameters of a variable frequency screw pump and a gear pump, realizes accurate regulation of the working pressure of cooling liquid, and guarantees efficient and stable operation of the energy storage power station.

Description

Technical Field

[0001] This invention relates to the field of energy storage heat dissipation technology, specifically a centralized energy storage power station heat dissipation system. Background Technology

[0002] With the rapid development of the new energy industry, centralized energy storage power stations, as core facilities for energy storage and dispatch, are playing an increasingly important role in grid peak shaving and renewable energy consumption. These power stations typically consist of a large number of energy storage battery packs, converters, and supporting electrical equipment. During operation, they generate a large amount of heat, which, if not dissipated in a timely and effective manner, will directly affect the operating status of the equipment.

[0003] The heat dissipation effect of centralized energy storage power stations is closely related to equipment operating efficiency and service life. During the charging and discharging process of battery packs, excessively high temperatures can lead to abnormal chemical reaction rates and accelerate material aging. Power electronic equipment such as converters, operating in high-temperature environments for extended periods, are prone to problems such as decreased insulation performance and component parameter drift. Therefore, the heat dissipation system becomes a critical component in ensuring the stable operation of energy storage power stations.

[0004] Existing energy storage power stations primarily employ air cooling or liquid cooling for heat dissipation. Air-cooled systems are simple in structure, but their heat dissipation efficiency is significantly affected by ambient temperature and airflow distribution. In high-altitude areas, the reduced air pressure leads to decreased air density, resulting in a significant reduction in heat dissipation capacity. Liquid cooling systems remove heat through coolant circulation, and variable frequency pumps are commonly used to adjust the flow rate to control the heat dissipation effect. However, existing control logic is mostly based on real-time temperature feedback and lacks analysis of historical operating patterns.

[0005] In practical applications, energy storage power stations operate in complex and variable environments. Differences in air pressure and temperature at different altitudes alter the physical properties of the coolant, affecting heat dissipation efficiency. Traditional cooling systems often rely on experience-based control parameters, making it difficult to adapt to the effects of altitude variations. Furthermore, existing systems rely heavily on single threshold judgments to identify abnormal operating conditions, failing to incorporate trend characteristics from historical data, leading to false alarms or missed alarms and resulting in delayed heat dissipation regulation. In addition, the coordinated control of variable frequency screw pumps and gear pumps lacks a dynamic optimization mechanism, making it difficult to achieve precise flow and pressure matching during load fluctuations, resulting in energy waste or insufficient heat dissipation. These problems make existing cooling systems unable to meet the high requirements of centralized energy storage power stations in terms of adaptability, stability, and economy. Summary of the Invention

[0006] The purpose of this invention is to provide a centralized energy storage power station heat dissipation system to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a centralized energy storage power station heat dissipation system, the system comprising:

[0008] The data acquisition module acquires historical operating data of the energy storage power station under different altitude conditions and generates a set of historical cooling parameters based on the historical operating data.

[0009] The feature analysis module performs feature extraction processing on the historical cooling parameter set to obtain a historical feature parameter set, and establishes a trend change feature matrix based on the trend of the historical feature parameter set over time.

[0010] The pressure regulation module constructs a pressure control space based on the trend change feature matrix, calculates the parameter clustering degree of historical abnormal events of the energy storage power station in the pressure control space, and determines a set of feature early warning indicators based on the parameter clustering degree.

[0011] The heat dissipation execution module extracts real-time feature parameters based on the real-time operation data of the energy storage power station, establishes a real-time feature state vector, and calculates the position deviation with the feature warning index set within the pressure control space to generate a real-time pressure adjustment factor.

[0012] The pressure regulation module generates a pressure control model based on the real-time pressure regulation factor, the historical feature parameter set, and the trend change feature matrix, and outputs the coordinated control parameters of the variable frequency screw pump and the gear pump; the heat dissipation execution module adjusts the working pressure of the coolant based on the coordinated control parameters.

[0013] Preferably, the data acquisition module acquires historical operating data of the energy storage power station under different altitude conditions, and generates a historical cooling parameter set based on the historical operating data. Specifically, this includes: acquiring historical operating data of the energy storage power station in a preset altitude range through a barometer and an altitude sensor; performing data validity screening based on the variation range and fluctuation characteristics of coolant temperature, pressure, and flow parameters in the historical operating data; and classifying and integrating the screened historical operating data according to parameter type to generate a historical cooling parameter set containing ambient air pressure parameters, coolant boiling point parameters, and pipeline pressure drop parameters.

[0014] Preferably, the feature analysis module performs feature extraction processing on the historical cooling parameter set to obtain a historical feature parameter set, and establishes a trend change feature matrix based on the time-varying trend of the historical feature parameter set. Specifically, this includes: performing wavelet packet decomposition processing on the historical cooling parameter set to extract frequency domain energy feature parameters; establishing the historical feature parameter set according to the altitude gradient; calculating the change trend values ​​of the historical feature parameter set within a specified time window; and arranging the change trend values ​​in a time series to form a trend change feature matrix, where the rows of the trend change feature matrix represent different historical feature parameters and the columns represent the corresponding time nodes.

[0015] Preferably, the pressure regulation module constructs a pressure control space based on the trend change feature matrix, calculates the parameter clustering degree of historical abnormal events of the energy storage power station in the pressure control space, and determines a feature early warning index set based on the parameter clustering degree. Specifically, this includes: determining the dimension of the pressure control space based on the number of historical feature parameters in the trend change feature matrix; calculating the clustering density value of historical coolant boiling events in the pressure control space based on a density clustering algorithm; setting a density threshold based on the clustering density value; and selecting historical feature parameters corresponding to feature positions exceeding the density threshold as the feature early warning index set.

[0016] Preferably, the heat dissipation execution module extracts real-time feature parameters based on the real-time operation data of the energy storage power station, establishes a real-time feature state vector, and calculates the positional deviation between the real-time feature state vector and the feature warning indicator set within the pressure control space to generate a real-time pressure adjustment factor. Specifically, this includes: collecting coolant pressure data in real time through a pressure sensor; performing wavelet packet transform processing on the coolant pressure data to form a real-time feature state vector; calculating the spatial Manhattan distance between the real-time feature state vector and the feature warning indicator set; and generating a real-time pressure adjustment factor based on the spatial Manhattan distance.

[0017] Preferably, the pressure regulation module generates a pressure control model based on the real-time pressure regulation factor, the historical feature parameter set, and the trend change feature matrix, and outputs coordinated control parameters for the variable frequency screw pump and the gear pump. Specifically, this includes: inputting the real-time pressure regulation factor and the historical feature parameter set into a gradient boosting decision tree model; outputting the speed regulation coefficient of the variable frequency screw pump and the flow compensation coefficient of the gear pump based on the historical trend values ​​of the trend change feature matrix; and generating coordinated control parameters based on the speed regulation coefficient and the flow compensation coefficient.

[0018] Preferably, the heat dissipation execution module adjusts the working pressure of the coolant according to the coordinated control parameters, specifically including: controlling the output pressure of the variable frequency screw pump according to the speed adjustment coefficient; adjusting the auxiliary flow of the gear pump based on the flow compensation coefficient; and delivering pressurized coolant to the liquid cooling plate through a double-layer vacuum insulated pipeline.

[0019] Preferably, the heat dissipation execution module further includes an abnormal response unit, specifically including: activating the PTC heater to preheat the coolant when the real-time pressure regulation factor exceeds a preset threshold; and dynamically adjusting the power level of the PTC heater according to the temperature difference parameter of the plate heat exchanger.

[0020] Preferably, the pressure regulation module further includes a dynamic correction unit, specifically comprising: receiving the actual coolant pressure data from the heat dissipation execution module in real time; calculating the root mean square error between the actual pressure data and the target pressure value; and correcting the weight parameters of the gradient boosting decision tree model based on the root mean square error.

[0021] Preferably, the feature analysis module further includes a pattern recognition unit, specifically comprising: using the isolated forest algorithm to detect unexpected fluctuation patterns in the historical cooling parameter set; and updating the anomaly marker bits of the trend change feature matrix according to the unexpected fluctuation patterns.

[0022] Compared with the prior art, the beneficial effects of the present invention are:

[0023] This centralized energy storage power station's cooling system, through the coordinated operation of multiple modules, forms a complete intelligent cooling control system, effectively addressing cooling demands under different operating conditions. The data acquisition module focuses on historical operating data under different altitude conditions, generating a set of historical cooling parameters that provides comprehensive foundational information for subsequent analysis. This data covers changes in key parameters such as temperature, pressure, and flow rate in different altitude environments, enabling the system to fully consider the impact of altitude differences on the cooling process, overcoming the limitations of traditional cooling control that ignores environmental factors.

[0024] The feature analysis module extracts features from historical cooling parameter sets and establishes a trend change feature matrix, enabling precise capture of the evolution of heat dissipation parameters over time. By mining the hidden trend features in historical data, the system can identify parameter change patterns under normal operating conditions, providing a reliable reference for subsequent anomaly detection and pressure adjustment. Compared to traditional single-parameter monitoring, this trend analysis-based approach better reflects the overall operating status of the cooling system and helps to detect potential heat dissipation anomalies in advance.

[0025] The pressure control space constructed by the pressure regulation module, combined with parameter clustering analysis of historical abnormal events, determines a set of characteristic early warning indicators, enhancing the ability to identify abnormal operating conditions. By calculating the clustering of historical abnormal parameters in the control space, the system can accurately define the distribution range of abnormal parameters and form targeted early warning indicators. This enables the cooling system to quickly identify parameter changes deviating from the normal range during operation, promptly triggering the adjustment mechanism and preventing cooling failure due to the accumulation of anomalies.

[0026] The heat dissipation execution module calculates and generates a real-time pressure adjustment factor based on the positional deviation between real-time feature parameters and the early warning indicator set, achieving precise heat dissipation control. Comparative analysis of the real-time feature state vector and the early warning indicator set quantifies the deviation between the current heat dissipation state and the normal state, providing a precise basis for pressure adjustment. This dynamic adjustment method can quickly respond to changes in equipment operating status, ensuring that the coolant pressure always matches the heat dissipation requirements.

[0027] The pressure regulation module generates a pressure control model that outputs coordinated control parameters for the variable frequency screw pump and gear pump, optimizing their operational coordination. By combining historical characteristics, trend matrices, and real-time adjustment factors, the control model can dynamically adjust the operating parameters of the two pumps according to different operating conditions, achieving coordinated optimization of flow and pressure. This coordinated control mode fully leverages the performance advantages of different pump types, ensuring effective heat dissipation while reducing energy consumption and improving the operational economy of the heat dissipation system. Overall, through the organic integration of its modules, the system achieves intelligent, precise, and efficient heat dissipation, adapting to the complex and variable operating environment of centralized energy storage power stations. Attached Figure Description

[0028] Figure 1 This is a timing diagram of the centralized energy storage power station heat dissipation system described in this invention;

[0029] Figure 2 A flowchart illustrating the operation of the feature analysis module;

[0030] Figure 3 A flowchart for determining the characteristic early warning index set of the pressure regulation module;

[0031] Figure 4 A flowchart for generating real-time pressure regulation factors for the heat dissipation execution module. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] Please see Figure 1 This invention provides a centralized energy storage power station heat dissipation system, the system comprising:

[0034] Dynamic optimization control of heat dissipation pressure is achieved through multi-module collaboration. The data acquisition module obtains historical operating data of the energy storage power station under different altitude conditions, including environmental parameters and cooling system parameters. This historical data is filtered and classified to generate a structured set of historical cooling parameters. The feature analysis module extracts features from this set, identifies key operating characteristics, and analyzes their changes over time, forming a quantified trend change feature matrix. The pressure regulation module constructs a multi-dimensional pressure control space based on this matrix, analyzes the distribution density of historical abnormal events within this space, and determines a set of key monitoring indicators for early warning. During real-time operation, the heat dissipation execution module continuously collects cooling system operating data, extracts real-time feature parameters, constructs a state vector, maps it to the pressure control space, calculates its spatial deviation from the feature warning indicator set, and generates a real-time pressure regulation factor reflecting the current system risk. The pressure regulation module integrates this real-time pressure regulation factor, the historical feature parameter set, and the trend change feature matrix, using a machine learning model to calculate the optimal coordinated control parameters for the variable frequency screw pump and gear pump. Finally, the heat dissipation execution module precisely adjusts the coolant working pressure based on these coordinated control parameters to achieve dynamic optimization of heat dissipation efficiency.

[0035] Example 1: See Figure 2The data acquisition module deploys a multi-type sensor network on the physical structure of the energy storage power station's cooling system. Environmental pressure monitoring uses a piezoresistive sensor, which measures the change in resistance of a semiconductor material under pressure to obtain pressure values. The range is set from 50 kPa to 110 kPa, with an accuracy of 0.1% of full scale, and data is collected every five seconds. Altitude measurement uses an integrated module that simultaneously receives elevation signals from a satellite positioning system and pressure-altitude data calculated based on a standard atmospheric model. When the difference between the two methods exceeds five meters, a calibration procedure is automatically initiated. Coolant temperature monitoring uses a platinum resistance thermometer, which utilizes the characteristic of platinum wire resistance changing with temperature. The thermometer is installed covering the inlet and outlet pipes of the liquid cooling plate, with a measurement range of -40°C to 150°C and an error not exceeding ±0.1°C. Main pipeline pressure monitoring uses a piezoelectric sensor, which generates an electrical signal when subjected to pressure. The range is 0 to 2.5 MPa, with a nonlinearity error of less than 0.05%. Electromagnetic flow meters are used for branch pipeline flow measurement. Their working principle is based on the induced electromotive force generated when a conductive liquid flows through a magnetic field. Pipe diameters range from 25 mm to 80 mm, with an accuracy class of 0.5. Pump current monitoring is achieved through Hall effect current sensors, utilizing the principle of generating Hall voltage based on changes in the magnetic field around a current-carrying conductor. The range is 0 to 100 amperes, with a response time less than one millisecond. The system is divided into data acquisition zones according to altitude, with each 500-meter interval forming an independent unit, such as 0 to 500 meters, 500 to 1000 meters, etc. Within each zone, sensors continuously collect historical operating data at a frequency of ten times per second. Recorded items include timestamps accurate to milliseconds, equipment number, air pressure readings, inlet and outlet temperatures, main pipeline pressure, flow rates in each branch, and current values ​​of the main and auxiliary pumps. Data validity screening employs a two-stage verification process: the first stage filters based on physical limits, such as temperature readings exceeding -40°C to 150°C, which are considered invalid; the second stage compares the data against a historical statistical database for the corresponding altitude range. Data points are deemed abnormal if the coolant temperature change rate exceeds 5°C per second, the pressure value is 30% higher than the historical high for that altitude, or the flow rate drops sharply from over 300 liters per minute to below 50 liters per minute within one second and then recovers. Valid data after screening is categorized by parameter type: ambient air pressure parameters are stored directly as raw measurements; coolant boiling point parameters are obtained by querying a pre-established table of boiling point and air pressure correspondences, generated based on fluid thermodynamics equations; and pipeline pressure drop parameters are calculated using fluid mechanics formulas based on real-time flow data, precisely measured pipeline length, calibrated pipeline inner diameter, and engineering design surface roughness, before outputting the calculated friction loss. The resulting historical cooling parameter set is stored in a hierarchical data format. Each record contains fields such as collection time, altitude region code, ambient air pressure, coolant boiling point, and main pipeline pressure drop, establishing a database architecture with time and altitude as dual search dimensions.This storage method can clearly distinguish the system characteristics under different altitude conditions, such as the low boiling point characteristics recorded in high-altitude areas and the changes in pump power under low-pressure environments.

[0036] The feature analysis module performs signal feature extraction on the received structured dataset. The processing utilizes wavelet packet decomposition, a mathematical method for finely dividing signals according to frequency components. For time-series data of three core parameters—temperature, pressure, and flow rate—a fourth-order Daubechies wavelet basis function is selected for three-level decomposition. Each level of decomposition generates two sets of coefficients: a low-frequency component and a high-frequency component. The three-level processing ultimately divides the signal into eight frequency sub-bands. The energy value of each sub-band is calculated, which is the average of the squared values ​​of all data points within that frequency band. Further calculation of the band energy entropy is performed. This indicator reflects the uniformity of energy distribution across different frequency bands. The calculation process involves first determining the proportion of energy in each frequency band to the total energy, and then summing these values ​​according to the information entropy formula. Simultaneously, the frequency band number with the highest energy and its percentage are recorded. For example, in the altitude range of 1000 to 1500 meters, the energy proportion of the coolant pressure signal in the 0.5 Hz to 1 Hz frequency band reaches 45% during a specific time period, with an entropy value of 1.2. These values ​​are extracted as feature parameters. The system establishes independent feature datasets based on altitude: each altitude range corresponds to a dataset containing 32 feature indicators, including temperature entropy values, pressure main frequency band numbers, and maximum energy percentage of flow rate at all time points within that region. Trend analysis employs a sliding time window mechanism, with each window fixed at 600 seconds (ten minutes) and sliding for 60 seconds (one minute) at a time. Within each analysis window, linear regression calculations are performed on each feature parameter sequence. For example, for the pressure specific frequency band energy values ​​in the 1500-2000 meter altitude range, a straight line is fitted using the least squares method, and the resulting slope is used as the quantified value of the trend. The slope calculation results are retained to three decimal places, with positive values ​​indicating an upward trend and negative values ​​indicating a downward trend. For example, the slope of the coolant boiling point parameter within a certain window is -0.002, indicating that the parameter is decreasing at a rate of 0.002 degrees Celsius per minute. The slope values ​​of all feature parameters across all time windows are arranged in matrix rows according to feature number order and in matrix columns according to window start time order, constructing a trend change feature matrix. Matrix elements correspond to the rate of change of a specific feature within a specific time window. The matrix is ​​stored in a sparse format, with approximately 35% of the data being effective. A compressed row storage strategy is employed to reduce memory usage. Each column of data is associated with a precise timestamp and altitude code, and the row index and feature parameter lookup table are stored independently in a configuration file. This matrix structure can intuitively display the evolution of different features over time; for example, certain vibration features exhibit periodic fluctuation patterns in high-altitude regions.

[0037] During the construction and maintenance of the trend change feature matrix, the system implements full-process quality control. Time synchronization uses a network time protocol to unify the clocks of all sensors, with deviations controlled within ±10 milliseconds. For the differences in sampling frequencies among different sensors (e.g., temperature once per second, pressure ten times per second), sampling is uniformly downsampled to one data point per second. The data missing handling mechanism includes multiple strategies: for a single missing data point, linear interpolation of adjacent data points is used; for consecutive missing data points less than 120 seconds (20% of the window length), the average value of the feature in adjacent altitude regions is used to supplement the missing data; for missing data points exceeding 120 seconds, the data in that window is discarded. Matrix update and maintenance employ a double-buffering mechanism: a memory buffer stores the most recent 72 hours of data, and every 60 minutes (one hour) of accumulated data is used to generate an incremental package written to disk; disk files are divided into directories by week to store complete matrix snapshots. The data access interface supports searching by time range and filtering by feature type. To improve computational efficiency, the slope calculation of feature parameters uses parallel processing technology: the 600-second window of data is divided into six 100-second sub-blocks, allocated to different processor cores to simultaneously calculate the local slope, and the final result is weighted averaged according to the principle of higher weight in the central region of the window. The system performs an automatic check weekly: by statistically analyzing the average and standard deviation of the slopes of various characteristic parameters, it monitors for systematic deviations. If a slope value for a particular characteristic exceeds 0.01 for three consecutive days, a data review process is triggered, and historical matrix data is recalculated if necessary. This maintenance mechanism requires continuous tracking of the system's operational status, timely removal of invalid data, and supplementation with new records to ensure the matrix accurately reflects the actual operating patterns of the cooling system, especially the characteristic differences under different altitude conditions, such as the frequency fluctuations unique to high-altitude regions.

[0038] Example 2: See Figure 3The pressure regulation module constructs a multi-dimensional pressure control space based on a trend change feature matrix. The number of dimensions in this space is directly equal to the total number of historical feature parameters contained in the trend change feature matrix. For example, when the matrix contains 12 feature parameters, the space has a 12-dimensional structure, with each dimension corresponding to the trend value of a feature parameter. The system retrieves coolant boiling events recorded by the energy storage power station from the historical event database. These events contain precise timestamps and altitude information. For each abnormal event, the system extracts the column vector from the trend change feature matrix corresponding to its occurrence time, which represents the trend values ​​of all feature parameters at that time, and maps this vector as a data point to the multi-dimensional pressure control space. Each axis of the spatial coordinate system represents the trend of a specific feature parameter. For example, the first axis corresponds to the rate of change of coolant pressure in the main frequency band between 500-1000 meters above sea level, and the second axis represents the rate of change of flow entropy between 1000-1500 meters above sea level. The coordinate values ​​of the data point are the trend values ​​of that feature at the time of the abnormal event. The system uses spatial indexing technology (such as R-trees) to manage these data points, supporting fast range queries. To optimize computational efficiency, the spatial construction process performs dimensionality reduction preprocessing: principal component analysis is performed on the trend change feature matrix, and principal components with a cumulative contribution rate exceeding 85% are selected as new coordinate axes. The original high-dimensional space is then projected to a low-dimensional space, for example, reducing it from 12 dimensions to 5 dimensions. The projected space retains the main feature change information while significantly reducing the computational complexity of subsequent calculations. After the spatial construction is completed, a spatial description file is generated, recording the dimension definitions, coordinate ranges, data point distribution, and dimensionality reduction transformation matrix parameters.

[0039] Within the pressure control space, the system performs cluster density analysis on historical anomalous events, employing the density-based clustering algorithm DBSCAN to calculate the distribution characteristics of data points. This algorithm requires setting two key parameters: neighborhood radius. and minimum points . The value is dynamically calculated based on the spatial scale, taking 1.5 times the median of the Euclidean distance between all pairs of data points. Set to the integer part of the square root of the total number of exceptions. During algorithm execution, a random unprocessed data point is selected, and its... All points within the neighborhood, i.e., points less than or equal to the given point. The point. If the number of points in the neighborhood reaches If a point is found to be a core point, a new cluster is created, and all points within that neighborhood are recursively added to the same cluster; otherwise, it is marked as a noise point. After processing all points, the clustering results and the density index of each cluster are output. For each cluster, its cluster density value is calculated. The volume is defined as the number of data points contained in the cluster divided by the volume of its smallest enclosing sphere in space. An approximate algorithm is used for volume calculation: after determining the extreme coordinates of points within the cluster, the product of the differences in coordinates across all dimensions is taken as the volume of the enclosing hypercube. Cluster density value. It reflects the spatial concentration of anomalies. The system is based on the spatial concentration of all clusters. Value distribution sets density threshold Take all The upper quartile of the value is the 75th percentile. Filter out... Value greater than The high-density clusters are used to extract the feature parameters corresponding to all data points contained in these clusters. The final generated feature warning index set is a list of feature parameter indices.

[0040] After the feature-based early warning indicator set is generated, the system establishes a dynamic update mechanism. When a new historical anomaly event is added, its corresponding feature trend vector is added as a new data point to the pressure control space. DBSCAN clustering and density analysis are re-executed every 10 newly added data points to update the feature-based early warning indicator set. The spatial structure itself is also periodically optimized: a full principal component analysis is performed on the trend change feature matrix every quarter, the dimensionality reduction matrix is ​​recalculated based on the latest data, and the spatial coordinate axis definition is adjusted. To verify the effectiveness of the spatial model, the system sets up a cross-validation process: 20% of historical anomalies are randomly retained as a test set, and the remaining 80% of the data is used to construct the space and generate an early warning indicator set. The system checks whether the feature vectors corresponding to the test set events fall into high-density regions. If the proportion of test set events falling into high-density regions is lower than a set standard (e.g., 85%), spatial parameter adjustments are automatically triggered (e.g., increasing the percentage of events falling into high-density regions). (Values ​​or added spatial dimensions). The feature-based early warning indicator set output is a machine-readable configuration file, containing a list of early warning feature numbers, the weight coefficient of each feature in space, and the last update timestamp. The operation and maintenance interface provides visualization tools, supporting 3D projection display of the data point distribution and high-density area locations in the pressure control space.

[0041] A spatial partitioning strategy is introduced in the density threshold setting stage, dividing the pressure control space into several subspaces. The number of subspaces is equal to the square root of the total number of feature parameters. A local density threshold is calculated independently for each subspace. The global density threshold is the weighted average of the thresholds of all subspaces, with the weight determined by the proportion of historical abnormal events contained within each subspace. This strategy avoids threshold bias caused by uneven feature distribution. Subspace partitioning uses a kd-tree algorithm, recursively dividing the space along the median of the feature dimensions until the number of subspaces reaches the target. For each subspace, the average nearest neighbor distance of its internal data points is calculated as a local density reference, and then combined with the subspace's positional weight in the overall space to determine the final threshold. The system calculates the weight coefficients of the feature parameters in the early warning indicator set using the following formula:

[0042]

[0043] in: Indicates the first The weight coefficients of each feature parameter, The total number of high-density clusters, It is the first The cluster density value of each cluster. It is an indicator function (when the feature) Appearing in clusters The weights are set to 1 if the result is satisfactory, and 0 otherwise. The denominator is the maximum sum of the weights of all feature parameters, used for normalization. The weight calculation results are persistently stored and recalculated with each space update. The system records the confidence index of each weight coefficient, and triggers a manual review process when the confidence level falls below a threshold.

[0044] Example 3: See Figure 4 The cooling execution module continuously monitors the coolant pressure during system operation, acquiring pressure data in real time through a piezoelectric pressure sensor installed in the middle of the main pipeline. The sensor's range covers 0 to 2.5 MPa, and the sampling frequency is set to 100 times per second to capture rapid fluctuations. The raw pressure signal undergoes digital filtering, using a low-pass filter with a cutoff frequency of 5 Hz to eliminate high-frequency noise interference. The filtered signal is then input to a wavelet packet transform processor, which performs a three-level decomposition using the same fourth-order Daubechies wavelet basis function as in the historical data analysis phase. The decomposition process divides the pressure signal into eight frequency sub-bands: 0 to 12.5 Hz and 12.5 to 25 Hz. The energy value, average of the sum of squares of the signal amplitude, and energy entropy for each sub-band are calculated, reflecting the uniformity of energy distribution. The system selects eight feature parameters that are completely consistent with the historical feature parameter set definition: the first five are key frequency band energy values ​​determined based on historical analysis of sensitive frequency bands; the last three are the full-band energy entropy, the main frequency band number, and the main frequency band energy percentage. These real-time calculated feature parameters are arranged in a fixed order, forming an eight-dimensional real-time feature state vector. This vector is updated every 0.1 seconds and stored in a circular buffer for subsequent analysis. For example, when the system detects that the coolant is running at an altitude of 2000 meters, it will continuously output vector data containing frequency characteristics specific to that altitude.

[0045] After the real-time feature state vector is generated, the system maps it to a pre-constructed pressure control space. This space has the same dimension as the vector, with each coordinate axis corresponding to a feature parameter. The system retrieves a feature-based early warning indicator set, containing the coordinates of high-risk feature locations extracted from historical anomalies, and calculates the spatial distance between the current vector and each early warning point in the indicator set. The distance calculation uses the Manhattan distance formula:

[0046]

[0047] in: Represents the Manhattan distance value. The real-time feature state vector is at the 1st... Dimension value, Is a certain warning point at the 1st The distance is calculated in parallel across eight dimensions, summing the absolute values ​​of the differences in each dimension. The system iterates through all warning points in the warning indicator set (typically between 50 and 200) and records the minimum distance value. and average distance value Real-time pressure regulation factor Calculated using a piecewise function: when hour ;when hour ;when hour Threshold and Based on historical data statistics, Take 1.2 times the radius of the area where the early warning points are clustered. Take 2.5 times. Factor The value ranges from 0.0 to 1.0, with a larger value indicating that the current state is closer to a historical abnormal pattern. The calculation result is updated every 0.5 seconds and transmitted to the pressure regulation module via the data bus.

[0048] The pressure regulation module receives real-time pressure regulation factors. Then, collaborative control decisions are made by combining historical feature parameter sets and trend change feature matrices. The module calls a pre-trained gradient boosting decision tree model (GBDT), which contains 200 regression trees with a maximum depth of 8 layers. The model input is a 21-dimensional feature vector: the first 8 dimensions are the current real-time feature state vector, and the 9th dimension is... The first 12 dimensions are the associated features extracted from the trend change feature matrix, including the mean, variance, and deviation from historical data for the current altitude range over the past 10 minutes. The model output includes two key parameters: the speed regulation coefficient of the variable frequency screw pump. (Floating-point number) and the flow compensation coefficient of the gear pump (Floating-point number). The coefficient calculation process is based on the principle of ensemble learning. Each decision tree makes a judgment based on the input features, and the final output is obtained by weighted averaging of the predictions of all trees. For example, when the input features show that the pressure main frequency band energy in the 1500-meter altitude area is continuously rising and... At that time, the model may output (Indicates an increase of 15% in RPM) and (This indicates that the gear pump provides 80% of the rated compensated flow). Coordinated control parameters. and The data is transmitted to the execution unit of the heat dissipation execution module via industrial Ethernet, with a transmission cycle of 1 second. The model automatically performs incremental training every 24 hours, updating the tree structure weights using 3,000 newly generated sets of running data each day to maintain adaptability to changes in system characteristics.

[0049] During the real-time feature state vector construction phase, the system implements a signal quality monitoring mechanism. When an abnormal abrupt change in frequency band energy is detected during wavelet packet decomposition, with the rate of change exceeding 50% in adjacent calculation cycles, the sensor calibration procedure is automatically triggered: the solenoid valve is controlled to generate a pressure step signal lasting 100 milliseconds, and the sensor's dynamic characteristics are verified by comparing the theoretical response curve with the actual measured value. If the deviation exceeds 5%, a sensor fault alarm is generated and the system switches to the backup sensor channel. Vector transmission uses CRC-32 checksums to ensure data integrity; if the checksum fails, the three most recent cached data are resent.

[0050] The Manhattan distance calculation process incorporates feature weight optimization. This is based on pre-stored weight coefficients from the feature-based early warning indicator set. (Values ​​range from 0 to 1), weighted adjustments are made to the distance formula:

[0051]

[0052] in: The Manhattan distance between the real-time feature state vector and a single warning point. The real-time feature state vector at the th The value of dimension, Feature-based early warning indicators set at a certain early warning point on the 1st Dimensional coordinates, weighting coefficients Generated by the pressure regulation module in Example 2, it reflects the differences in the importance of each characteristic parameter in historical anomalies. For example, the weight of the energy characteristic in the main pressure frequency band is usually above 0.9, while the weight of the secondary frequency band characteristic may be only 0.3. This weighting mechanism makes the distance calculation more focused on key risk characteristics and avoids interference from secondary characteristics.

[0053] The training dataset for the gradient boosting decision tree model includes historical operation records and manual intervention records. Positive samples are selected from feature data from 5 minutes prior to the occurrence of historical abnormal events and labeled as requiring pressurization. Negative samples are selected from data from stable system operation periods and labeled as maintaining operation. The training objective is to minimize the mean squared error between the output values ​​of control parameters and the actual adjustment actions of maintenance personnel. An A / B version mechanism is used during model deployment: the new version of the model runs in parallel for 72 hours in shadow mode. When its output parameters consistently differ from the original version by more than 10%, manual review is triggered. Only after the review is passed is it switched to the primary version. This design prevents uncontrollable risks from being introduced by model updates.

[0054] The collaborative control parameter output interface includes safety limiting logic. (Speed ​​regulation coefficient) Constrained within the range of 0.8 to 1.5, corresponding to 80% to 150% of the rated speed, the flow compensation coefficient... The limit is set between 0 and 1.2. When the model output value exceeds the limit, the boundary value is taken and an over-limit alarm is generated. The parameter transmission message uses the industry-standard Modbus TCP protocol and includes a timestamp, parameter value, and data quality identifier reflecting the reliability of the input features. After receiving the message, the execution unit begins adjustment action with a delay of no more than 20 milliseconds.

[0055] Example 4: After receiving the coordinated control parameters, the heat dissipation execution module initiates the pressure regulation process and obtains the speed regulation coefficient through the industrial Ethernet interface. and flow compensation coefficient The transmission cycle is 1 second. The variable frequency screw pump control unit includes an ABB ACS880 series variable frequency drive with a rated power of 22kW. The drive has a built-in PID controller and receives... After setting the value, perform speed conversion: Multiply by the pump's base speed (1500 rpm) to get the target speed. For example, when At that time, the target speed was 1800 rpm. The driver samples the actual motor speed 50 times per second, and uses closed-loop control to stabilize the speed error within ±5 rpm. The gear pump control unit uses a servo motor drive with a rated flow rate of 300 L / min. Flow compensation coefficient. Directly converted to target traffic (Unit: L / min). The servo controller adjusts the motor speed in real time with 0.2-level accuracy feedback from a high-precision flow meter, ensuring the deviation between the actual flow rate and the target flow rate is less than ±2%. During dual-pump coordinated operation, the system monitors the standard deviation of pressure change within a 10ms time window of the main pipeline pressure fluctuation rate. When the fluctuation rate exceeds 0.05MPa / s, the pressure damping algorithm is automatically activated. and A sinusoidal compensation signal with an amplitude of 0.1 and a frequency of 2Hz is superimposed on the base to suppress fluid pulsation.

[0056] The coolant delivery system employs a double-layer vacuum-insulated piping system. The inner pipe is made of 316L stainless steel with an outer diameter of 28mm and a wall thickness of 1.5mm; the outer pipe is made of 304 stainless steel with an outer diameter of 42mm. A vacuum is evacuated between the two layers. The system is filled with microporous insulation material. Pipe connections are argon-arc welded to ensure sealing, with a designed heat loss rate of less than 5W per meter. Pressurized coolant, ranging from 0.8 to 1.6 MPa, is delivered to the liquid-cooled plate via insulated piping. The liquid-cooled plate features a serpentine flow channel design with a cross-section of 8mm × 3mm, and its surface is sandblasted to enhance heat transfer. Field measurements at an altitude of 3000 meters show that the temperature rise of the coolant from the pump outlet to the liquid-cooled plate inlet does not exceed 0.8℃, verifying the effectiveness of the insulation design. The piping system is equipped with stress monitoring sensors (fiber optic grating type), which trigger a maintenance alarm when the displacement of the mounting bracket exceeds ±1mm or an abnormal temperature gradient is detected.

[0057] The abnormal response unit monitors the real-time pressure regulation factor. ,when The unit activates the PTC heater system when the value exceeds the threshold of 0.85 for 3 consecutive seconds. The heater is installed on the main pipeline of the coolant circulation loop, with three power configurations: 12kW, 18kW, and 24kW. The preheating control logic is based on the temperature difference parameters of the plate heat exchanger. , is defined as the temperature difference between the hot side inlet and the cold side outlet of the heat exchanger.

[0058] Table 1: Relationship between range, power level and preheating temperature target

[0059]

[0060] The system collects data every 10 seconds. The power level is dynamically switched according to the table above. The preheating process employs closed-loop temperature control: a PT100 temperature sensor is installed on the main pipeline. When the coolant temperature reaches the target value ±0.5℃, the heater enters constant temperature mode, and the power is reduced to 30% of the rated value. If the temperature does not reach the target lower limit within 5 minutes after preheating starts, a secondary alarm is triggered, and the gear pump flow compensation coefficient is increased accordingly. Up to 1.1 times. The heater surface temperature is monitored by an infrared thermal imager, and automatically reduces power when it exceeds 120°C.

[0061] An overload protection strategy is implemented in the variable frequency screw pump control. The driver continuously monitors the motor current at a sampling rate of 1kHz. When the current exceeds the rated value (42A), a gradient load reduction is performed: the speed is reduced to 95% of the rated speed in the first second, and to 90% in the second second, until the current returns to normal. If the current exceeds 120% of the rated value for 10 seconds, an emergency stop is triggered. The gear pump servo system is configured to suppress sudden flow changes: the target flow rate change rate is limited to ±50L / min / s to avoid pipeline impact.

[0062] Double-layer vacuum piping enables vacuum monitoring and automatic maintenance. A piezoresistive vacuum sensor (range) is installed in the piping interlayer. to Pa), when the vacuum level is detected to be lower than 100 Pa. The molecular pump unit is started when the pressure reaches Pa, and the high vacuum state is restored within 8 hours. The insulation performance is evaluated using a heat flux meter with a measurement range of 0-100 W / m². Twenty test points are selected along the pipeline every quarter to measure the heat flux value. Pipe sections that exceed the design value by 150% need to have their insulation material replaced.

[0063] The PTC heater system is equipped with redundant control channels. The main control PLC performs routine power regulation, while the backup microcontroller system, based on an ARM Cortex-M7, independently monitors the coolant temperature rise rate. When an abnormal temperature rise rate is detected, the backup system directly cuts off the heater power and switches to safety relay control. Plate heat exchanger temperature difference parameters. The calculation employs a three-sensor verification mechanism: two PT100 sensors are respectively arranged at the hot-side inlet, hot-side outlet, and cold-side outlet, and the average of the calculated values ​​from the two sets of sensors is taken as the final value. The sensor calibration procedure is triggered when the deviation exceeds 0.3℃.

[0064] Example 5: The dynamic correction unit of the pressure regulation module continuously receives actual coolant pressure data from the heat dissipation execution module. The data originates from three redundant pressure sensors installed in the middle section of the main pipeline, with a sampling frequency of ten times per second and a measurement range covering 0 to 2.5 MPa. After eliminating high-frequency interference signals through hardware filtering circuitry, the system processes the raw data using a median filtering algorithm: every ten sampling points form a group, and the median value is output as the valid data point. The target pressure value is calculated in real-time by the central controller based on the current battery charging / discharging power, ambient temperature, and altitude. The calculation model is based on thermodynamic equations and fluid mechanics formulas. The dynamic correction unit performs a deviation analysis between the actual pressure and the target pressure every thirty seconds. The calculation process uses a sliding time window mechanism: the window length is fixed at thirty seconds, containing three hundred valid data points. The root mean square error (RMSE) is calculated by taking the square of the deviation for each data point, then averaging these squares, and taking the square root of this average as the final error value. When the RMSE exceeds a set threshold for three consecutive calculation cycles, the system determines that the current pressure control model has a significant deviation. The correction process employs a gradient descent optimization algorithm: using the root mean square error (RMSE) as the loss function, the partial derivative of this function with respect to the weights of each decision tree in the gradient boosting decision tree model is calculated. The weight parameters are adjusted according to the direction of the derivative with a preset step size. An upper limit is set on the magnitude of the weight adjustment, with a single adjustment not exceeding five percent of the original value. The corrected model parameters take effect immediately, and the version number and correction timestamp are recorded. Model version management uses a rolling storage strategy, retaining historical versions from the last thirty days for fault rollback.

[0065] The pattern recognition unit of the feature analysis module periodically scans historical cooling parameter sets. The dataset is stored in a time-series database, containing complete operation records for the past two years. The scanning cycle is set to run daily during the early morning hours when system load is low. The Isolation Forest algorithm is used to detect unexpected fluctuation patterns. This algorithm constructs a large number of random decision trees to form a forest structure. During the construction of each tree, feature dimensions and segmentation thresholds are randomly selected, and data points are isolated to different branches through recursive segmentation. Abnormal data points, due to feature values ​​deviating from the normal range, are often isolated in shallower tree layers. The algorithm calculates an anomaly score for each data point, with scores ranging from zero to one; higher scores indicate a greater likelihood of an anomaly. The pattern recognition unit sets an anomaly score threshold of 0.65; when a data point's score exceeds this threshold, it is marked as a potential abnormal pattern. The detection process focuses on abrupt changes in frequency domain energy feature parameters, such as a sudden increase of more than 50% in energy values ​​of a specific frequency band without external interference, lasting for more than five minutes. The system automatically records the timestamps, elevation region codes, and associated feature parameter lists of abnormal events. An abnormal pattern report is generated weekly, including pattern type statistics, duration distribution, and feature correlation analysis. All detected unexpected fluctuation patterns are stored in a separate database and indexed in chronological order for subsequent analysis.

[0066] The update mechanism of the trend change feature matrix is ​​linked to the pattern recognition results. When the recognition unit detects a new unexpected fluctuation pattern, the system extracts the feature parameter data of the corresponding time window of the event. Based on the precise timestamp of the event, the corresponding column vector in the trend change feature matrix is ​​located. A dedicated flag bit is set in the storage structure of this column vector, with an initial value of zero. It is updated to a specific encoded value when an abnormal event is detected. The encoding rule uses an eight-bit binary number to represent the anomaly type: the first three bits identify the altitude region, such as 001 representing 500-1000 meters; the middle three bits identify the feature category, such as 010 representing pressure characteristics; and the last two bits identify the anomaly intensity, such as 01 representing a moderate anomaly. The matrix update operation is performed in memory, and is synchronized to disk storage after every 100 updates. The system maintains an anomaly flag bit index table, supporting fast retrieval by time range or anomaly type. When constructing the pressure control space, the pressure regulation module incorporates the anomaly flag bit as an additional dimension into the spatial structure. This dimension of data is not used in regular distance calculations. It only triggers a special processing procedure when anomaly markers are detected in the current real-time data: automatically increasing the sensitivity coefficient of the real-time pressure adjustment factor by 20% and activating the historical retrieval function for similar event handling plans. The pattern recognition unit performs a full data review quarterly, re-evaluating marked anomalies: if the same pattern repeatedly appears in subsequent operations and is confirmed as an inherent characteristic of the system, the pattern is removed from the anomaly database and the feature parameter baseline range is updated. Simultaneously, the isolated forest model is retrained to ensure that the anomaly detection criteria remain consistent with the actual evolution of the system.

[0067] The dynamic correction unit implements stability control for model weight adjustments. After each weight correction, the system runs the new model in the shadow control system for 24 hours. During this period, the control parameters generated by the new model are compared in real time with the parameters of the old model in actual operation. When the difference in key parameters is found to exceed 10% for an extended period, the deployment of the new model is automatically paused and a manual review request is generated. After the review is approved, a small-scale pilot run of 72 hours is required. Only after confirming that there are no abnormalities can the model be applied nationwide. The model version rollback mechanism retains the three most recent stable versions. If any version experiences control instability, the system can switch to a historical version within one minute.

[0068] The pattern recognition unit is equipped with a feature importance evaluation module, which counts the frequency of each feature parameter in abnormal events and calculates the frequency weight value. High-weight features obtain a higher sampling probability in the Isolation Forest algorithm, improving the detection sensitivity of related anomalies. The weight coefficients are recalculated quarterly based on newly accumulated anomaly data, dynamically adjusting the detection strategy.

[0069] The storage structure of the trend change feature matrix is ​​optimized using block compression technology. The matrix is ​​divided into weekly data blocks along the time dimension, with each block containing 1080 column vectors. Lossy compression is performed on each data block, retaining three decimal places of precision, with the compression ratio controlled at 50%. During decompression, the data stream is restored through linear interpolation to ensure that key trend information is not lost. Anomaly markers are stored independently as a bitmap index, supporting fast bitwise operation retrieval. This design significantly reduces storage requirements while ensuring data availability.

[0070] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0071] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A centralized energy storage power station heat dissipation system, characterized in that, include: The data acquisition module acquires historical operating data of the energy storage power station under different altitude conditions and generates a set of historical cooling parameters based on the historical operating data. The feature analysis module performs feature extraction processing on the historical cooling parameter set to obtain a historical feature parameter set, and establishes a trend change feature matrix based on the trend of the historical feature parameter set over time. The pressure regulation module constructs a pressure control space based on the trend change feature matrix, calculates the parameter clustering degree of historical abnormal events of the energy storage power station in the pressure control space, and determines a set of feature early warning indicators based on the parameter clustering degree. The heat dissipation execution module extracts real-time feature parameters based on the real-time operation data of the energy storage power station, establishes a real-time feature state vector, and calculates the positional deviation between the real-time feature state vector and the feature warning indicator set within the pressure control space to generate a real-time pressure adjustment factor. Specifically, this includes: acquiring coolant pressure data in real time through a pressure sensor; performing wavelet packet transform processing on the coolant pressure data to form a real-time feature state vector; calculating the spatial Manhattan distance between the real-time feature state vector and the feature warning indicator set; and generating a real-time pressure adjustment factor based on the spatial Manhattan distance. The pressure regulation module generates a pressure control model based on the real-time pressure regulation factor, the historical feature parameter set, and the trend change feature matrix, and outputs the coordinated control parameters of the variable frequency screw pump and the gear pump; the heat dissipation execution module adjusts the working pressure of the coolant based on the coordinated control parameters.

2. The centralized energy storage power station heat dissipation system according to claim 1, characterized in that, The data acquisition module obtains historical operating data of the energy storage power station under different altitude conditions and generates a historical cooling parameter set based on the historical operating data. Specifically, this includes: collecting historical operating data of the energy storage power station in a preset altitude range through a barometer and an altitude sensor; performing data validity screening based on the variation range and fluctuation characteristics of coolant temperature, pressure, and flow parameters in the historical operating data; and classifying and integrating the screened historical operating data according to parameter type to generate a historical cooling parameter set that includes ambient air pressure parameters, coolant boiling point parameters, and pipeline pressure drop parameters.

3. The centralized energy storage power station heat dissipation system according to claim 2, characterized in that, The feature analysis module performs feature extraction processing on the historical cooling parameter set to obtain a historical feature parameter set. Based on the time-varying trend of the historical feature parameter set, a trend change feature matrix is ​​established. Specifically, this includes: performing wavelet packet decomposition on the historical cooling parameter set to extract frequency domain energy feature parameters; establishing the historical feature parameter set according to the altitude gradient; calculating the change trend values ​​of the historical feature parameter set within a specified time window; and arranging the change trend values ​​in a time series to form a trend change feature matrix, where rows represent different historical feature parameters and columns represent corresponding time nodes.

4. The centralized energy storage power station heat dissipation system according to claim 3, characterized in that, The pressure regulation module constructs a pressure control space based on the trend change feature matrix, calculates the parameter clustering degree of historical abnormal events of the energy storage power station in the pressure control space, and determines a feature early warning index set based on the parameter clustering degree. Specifically, this includes: determining the dimension of the pressure control space based on the number of historical feature parameters in the trend change feature matrix; calculating the clustering density value of historical coolant boiling events in the pressure control space based on a density clustering algorithm; setting a density threshold based on the clustering density value; and selecting historical feature parameters corresponding to feature positions that exceed the density threshold as the feature early warning index set.

5. The centralized energy storage power station heat dissipation system according to claim 1, characterized in that, The pressure regulation module generates a pressure control model based on the real-time pressure regulation factor, the historical feature parameter set, and the trend change feature matrix, and outputs coordinated control parameters for the variable frequency screw pump and the gear pump. Specifically, this includes: inputting the real-time pressure regulation factor and the historical feature parameter set into a gradient boosting decision tree model; outputting the speed regulation coefficient of the variable frequency screw pump and the flow compensation coefficient of the gear pump based on the historical trend values ​​of the trend change feature matrix; and generating coordinated control parameters based on the speed regulation coefficient and the flow compensation coefficient.

6. The centralized energy storage power station heat dissipation system according to claim 5, characterized in that, The heat dissipation execution module adjusts the working pressure of the coolant according to the coordinated control parameters, specifically including: controlling the output pressure of the variable frequency screw pump according to the speed adjustment coefficient; adjusting the auxiliary flow of the gear pump based on the flow compensation coefficient; and delivering pressurized coolant to the liquid cooling plate through a double-layer vacuum insulated pipeline.

7. The centralized energy storage power station heat dissipation system according to claim 6, characterized in that, The heat dissipation execution module also includes an abnormal response unit, specifically including: activating the PTC heater to preheat the coolant when the real-time pressure adjustment factor exceeds a preset threshold; and dynamically adjusting the power level of the PTC heater according to the temperature difference parameters of the plate heat exchanger.

8. The centralized energy storage power station heat dissipation system according to claim 7, characterized in that, The pressure regulation module further includes a dynamic correction unit, specifically comprising: receiving the actual coolant pressure data from the heat dissipation execution module in real time; calculating the root mean square error between the actual pressure data and the target pressure value; and correcting the weight parameters of the gradient boosting decision tree model based on the root mean square error.

9. The centralized energy storage power station heat dissipation system according to claim 8, characterized in that, The feature analysis module also includes a pattern recognition unit, specifically comprising: using the isolated forest algorithm to detect unexpected fluctuation patterns in the historical cooling parameter set; and updating the anomaly marker bits of the trend change feature matrix according to the unexpected fluctuation patterns.

Images