A method and system for coordinated control of energy storage cooling capacity and peak-valley arbitrage.

By generating a thermoelectric prediction tensor and constructing a thermal demand Fibonacci stack, the system dynamically optimizes the cold energy allocation and peak-valley electricity price arbitrage of the energy storage system, solving the problem of the disconnect between battery temperature gradient and arbitrage strategy in traditional energy storage systems, and achieving stable control of battery temperature and improved economic efficiency.

CN122338283APending Publication Date: 2026-07-03HUNAN XILAIKE ENERGY STORAGE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610560571.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-27
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

The traditional energy storage system's thermal management strategy is disconnected from the peak-valley electricity price arbitrage strategy, resulting in the formation of temperature gradients inside the battery pack, which affects battery performance and lifespan, and fails to dynamically optimize the distribution of cooling capacity and arbitrage profits.

Method used

By collecting multi-dimensional operational data of energy storage systems and real-time transaction data of the electricity market, a thermoelectric prediction tensor is generated using a temporal convolutional network. The spatiotemporal distribution characteristics of heat load are analyzed, a heat demand Fibonacci stack is constructed, an arbitrage strategy skip table is generated, and independent heat exchangers and intelligent refrigerant boxes are driven to perform refined temperature control operations.

Benefits of technology

It achieves stable battery temperature maintenance within the optimal operating range, reduces cooling energy consumption costs, extends battery cycle life, and improves the operational economy and safety of the energy storage system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122338283A_ABST
    Figure CN122338283A_ABST
Patent Text Reader

Abstract

This invention relates to the field of energy storage system control technology, and discloses a collaborative control method and system for energy storage cooling capacity and peak-valley arbitrage. The method includes: collecting multi-dimensional operational data of the energy storage system and electricity market transaction data; generating a thermoelectric prediction tensor using a temporal convolutional network; analyzing the thermoelectric prediction tensor to calculate the cooling urgency weight of the battery cluster and constructing a heat demand Fibonacci stack; extracting the optimal cooling task and generating an arbitrage strategy skip list based on peak-valley electricity price constraints; traversing the skip list to obtain a control instruction set and driving independent heat exchangers and intelligent refrigerant boxes to perform refined temperature control operations. This invention solves the problems of lag in traditional heat dissipation control and difficulty in coordinating peak-valley electricity price arbitrage with temperature control through thermoelectric prediction, dynamic sorting of heat demand, and matching of cooling capacity supply and demand. It achieves stable battery temperature maintenance, maximized peak-valley arbitrage profits, and fully silent operation of the energy storage system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of energy storage system control technology, and relates to a collaborative control method and system for energy storage cooling capacity and peak-valley arbitrage. Background Technology

[0002] Energy storage systems typically use batteries as the core energy storage unit, equipped with corresponding temperature control devices and energy management systems to smooth grid load fluctuations and improve energy utilization efficiency. The temperature control device maintains the battery's operating temperature through air cooling or liquid cooling, while the energy management system controls charging and discharging according to preset strategies to achieve the storage and release of electrical energy.

[0003] During routine operation and maintenance of energy storage stations in industrial parks, engineers discovered that after continuous charge-discharge cycles, some individual battery cells exhibited significant capacity decay, and the temperature difference between adjacent batteries gradually increased. The root cause of this phenomenon lies in the traditional cooling system's use of a uniform airflow or liquid flow distribution method, which failed to dynamically adjust according to the actual heat generation of batteries in different locations, leading to a temperature gradient within the battery pack. Simultaneously, the energy management system did not fully consider the impact of temperature on battery performance when formulating charge-discharge strategies. The continuously high-temperature environment accelerated internal side reactions within the batteries, gradually deactivating active materials and resulting in irreversible capacity decay. This capacity decay not only reduces the usable capacity of the energy storage system but also makes it difficult to maintain a consistent depth of charge and discharge for the battery pack. In the event of a sudden power outage, it may be unable to provide sufficient backup power, affecting the continuous operation of critical equipment in the industrial park. Summary of the Invention

[0004] To overcome the aforementioned deficiencies in the prior art, this invention provides a method and system for coordinated control of energy storage cooling capacity and peak-valley arbitrage. In the prior art, the thermal management strategy and peak-valley electricity price arbitrage strategy of energy storage systems are isolated from each other, making it difficult to dynamically coordinate and optimize cooling capacity allocation and arbitrage benefits while ensuring battery safety. This invention aims to achieve dynamic optimization of battery thermal management and peak-valley electricity price arbitrage, thereby improving the economic efficiency and safety of energy storage system operation.

[0005] A method for coordinated control of energy storage cooling capacity and peak-valley arbitrage includes:

[0006] Collect multi-dimensional operational data of energy storage systems and real-time transaction data of the electricity market, and use temporal convolutional networks for feature extraction and fusion analysis to generate thermoelectric prediction tensors;

[0007] We analyze the spatiotemporal distribution characteristics of heat load in the thermoelectric prediction tensor, calculate the cooling urgency weight of each battery cluster, and construct a heat demand Fibonacci stack.

[0008] The minimum extraction operation is performed on the heat demand Fibonacci stack to extract the optimal cooling task. Combined with peak and valley electricity price constraints, the supply and demand of cooling capacity are matched to generate an arbitrage strategy skip list.

[0009] The arbitrage strategy skip list is traversed to obtain the control instruction set for the current time window, driving the independent heat exchanger and intelligent refrigerant box to perform refined temperature control operations.

[0010] Furthermore, the steps for generating the thermoelectric prediction tensor include:

[0011] Collect multi-dimensional operational data of energy storage systems and construct an energy storage operation dataset;

[0012] Collect real-time electricity market transaction data and construct an electricity price data set;

[0013] Based on the energy storage operation data set and the electricity price data set, a unified time axis multi-source sample matrix is ​​obtained through the time alignment method;

[0014] Based on a unified time-axis multi-source sample matrix, a thermoelectric feature derivation matrix is ​​obtained through thermoelectric feature derivation processing.

[0015] Based on the thermoelectric feature-derived matrix, feature extraction and fusion analysis are performed through a temporal convolutional network to obtain the thermoelectric prediction tensor.

[0016] Furthermore, the time alignment method includes the following steps:

[0017] A unified sampling time axis is determined based on the running timestamp collection sequence and the electricity price timestamp sequence;

[0018] Based on each acquisition sequence in the energy storage operation data set, the acquisition sequence is linearly interpolated according to a unified sampling time axis to obtain an operation data aligned sequence mapped to the unified sampling time axis.

[0019] Based on each sequence in the electricity price dataset, the sequences are subjected to a step hold operation along a unified sampling time axis to obtain an aligned sequence of electricity price data mapped onto the unified sampling time axis;

[0020] The alignment sequences of each operational data and the alignment sequences of electricity price data are horizontally spliced ​​together at the corresponding positions on a unified sampling time axis to generate a unified time axis multi-source sample matrix with time as the row and different physical quantities and electricity prices as the column.

[0021] Furthermore, the process of thermoelectric feature derivation includes:

[0022] Based on the current and voltage columns in the unified time axis multi-source sample matrix, the instantaneous power value is obtained by multiplying the current and voltage values ​​at each time point. The instantaneous power values ​​at all time points are arranged in chronological order to form an instantaneous power value sequence, which is then used as an instantaneous power derived column.

[0023] Based on the battery temperature column and ambient temperature column in the unified time axis multi-source sample matrix, the temperature difference value is obtained by calculating the difference between the battery temperature value and the ambient temperature value at each time point. The temperature difference values ​​at all time points are arranged in chronological order to form a temperature difference value sequence, and the temperature difference value sequence is used as a temperature difference derived column.

[0024] Based on the charge / discharge rate column and battery temperature column in the unified time axis multi-source sample matrix, and based on the sum of the product of the square of the charge / discharge rate and the battery internal resistance coefficient, and the product of the difference between the battery temperature and the ambient temperature and the heat dissipation coefficient, the heat generation rate estimate at each time point is calculated. The heat generation rate estimates at all time points are arranged in chronological order to form a heat generation rate estimate sequence, and the heat generation rate estimate sequence is used as a heat generation rate derived column.

[0025] Based on the real-time node electricity price column and the time-of-use electricity price prediction column in the unified time axis multi-source sample matrix, the difference between the real-time node electricity price and the time-of-use electricity price prediction value at each time point is calculated and divided by the time-of-use electricity price prediction value to obtain the electricity price deviation value. The electricity price deviation values ​​at all time points are arranged in chronological order to form an electricity price deviation value sequence, and the electricity price deviation value sequence is used as the electricity price deviation derivative column.

[0026] The instantaneous power derivative column, temperature difference derivative column, heat generation rate derivative column, and electricity price deviation derivative column are horizontally concatenated with the original columns of the unified time axis multi-source sample matrix to generate a thermoelectric feature derivative matrix.

[0027] Furthermore, the process of feature extraction and fusion analysis using a temporal convolutional network includes:

[0028] The thermoelectric feature derivation matrix is ​​sliced ​​according to a preset sliding window length to generate multiple thermoelectric feature input segments;

[0029] Multiple thermoelectric feature input segments are input into the causal convolutional layer of a temporal convolutional network for temporal feature extraction, resulting in a temporal local feature tensor.

[0030] The temporal local feature tensor is input into the residual connection module of the temporal convolutional network for feature fusion to obtain the temporal fused feature tensor;

[0031] The temporal fusion feature tensor is input into the output mapping layer of the temporal convolutional network to generate the thermoelectric prediction tensor.

[0032] Furthermore, the process of constructing a thermal demand Fibonacci stack includes:

[0033] Based on the thermoelectric prediction tensor, a spatiotemporal distribution matrix of heat load is constructed;

[0034] Based on the spatiotemporal distribution matrix of heat load and the battery thermal inertia model, the predicted sequence of temperature rise rate for each battery cluster is calculated.

[0035] Based on the temperature rise rate prediction sequence, the electricity price cost prediction vector, and the current state of the battery, the cooling urgency weight of each battery cluster is calculated.

[0036] Based on the cooling urgency weight of each battery cluster, a thermal demand Fibonacci stack is constructed.

[0037] Define a key-value reduction operation for thermal demand Fibonacci stacks to adjust priorities in response to sudden temperature changes in the battery cluster or when forecast data is updated.

[0038] Furthermore, the calculation process for the cooling urgency weight includes:

[0039] Obtain the current temperature deviation of each battery cluster;

[0040] Based on the predicted temperature rise rate sequence, the peak temperature rise rate of each battery cluster is calculated;

[0041] Based on the electricity price cost prediction vector, calculate the electricity price cost sensitivity coefficient of each battery cluster;

[0042] Based on the current temperature deviation, peak temperature rise rate, and electricity cost sensitivity coefficient, the cooling urgency weight of each battery cluster is calculated.

[0043] Furthermore, the process of generating the arbitrage strategy skip list includes:

[0044] Define the minimum extraction operation for the thermal demand Fibonacci heap, which is used to obtain and remove the thermal demand node with the highest cooling priority;

[0045] Obtain the current electricity price range identifier and cooling capacity storage status;

[0046] Extract the optimal cooling task, and generate cooling control instructions by combining the electricity price range identifier and the cooling capacity storage status;

[0047] Initialize the arbitrage strategy skip list and define the skip list operations;

[0048] The process involves repeatedly extracting the minimum operation and generating the instruction to create an arbitrage strategy skip list.

[0049] Furthermore, the process for performing refined temperature control includes:

[0050] Determine the current time window and perform an interval query to obtain the control instruction set;

[0051] Parse the current time window control instruction set and generate device-level control signals;

[0052] Drive the independent heat exchanger at the plug-in stage to perform directional cold energy transfer;

[0053] Drive the intelligent refrigerant box to perform cold energy storage and release control;

[0054] Perform battery operating temperature maintenance control;

[0055] Perform fully silent operation status monitoring.

[0056] An energy storage cooling capacity-peak arbitrage coordinated control system is provided for implementing the aforementioned energy storage cooling capacity-peak arbitrage coordinated control method. The system includes:

[0057] Thermoelectric prediction tensor generation module is used to collect multi-dimensional operation data of energy storage system and real-time transaction data of electricity market, and use temporal convolutional network to perform feature extraction and fusion analysis to generate thermoelectric prediction tensor.

[0058] The thermal demand Fibonacci stack construction module is used to analyze the spatiotemporal distribution characteristics of heat load in the thermoelectric prediction tensor, calculate the cooling urgency weight of each battery cluster, and construct the thermal demand Fibonacci stack.

[0059] The arbitrage strategy skip list generation module is used to perform minimum extraction operations on the heat demand Fibonacci stack, extract the optimal cooling task, combine peak and valley electricity price constraints to match cooling supply and demand, and generate arbitrage strategy skip lists.

[0060] The refined temperature control execution module is used to traverse the arbitrage strategy skip table to obtain the control instruction set of the current time window, and drive the independent heat exchanger and intelligent refrigerant box to perform refined temperature control operations.

[0061] This invention addresses key challenges in heat dissipation control of energy storage systems, including difficulties in coordinating heat demand forecasting with electricity price fluctuations, low efficiency in dynamic sorting of large-scale battery clusters, mismatch between cooling supply and demand, and noise control. Through multi-dimensional data fusion and intelligent scheduling strategies, it effectively improves the operational economy and reliability of energy storage systems. Specifically, this invention uses a temporal convolutional network to extract features from multi-dimensional operational data of the energy storage system and real-time electricity market transaction data, generating a thermoelectric prediction tensor. This effectively captures the lag characteristics of battery heating and the periodic patterns of electricity prices, transforming traditional passive feedback control into proactive prediction. This avoids increased energy consumption and battery life loss caused by delayed heat dissipation after temperature overheating. A heat demand Fibonacci stack, constructed based on the spatiotemporal distribution characteristics of heat load and cooling urgency weights, leverages the constant-level amortized time complexity of the heat demand Fibonacci stack to reduce key values, solving the computational bottleneck of dynamic sorting of heat dissipation demands for massive battery clusters and ensuring millisecond-level response time for cooling task priority adjustment. This invention utilizes an arbitrage strategy to efficiently store and query cooling control commands via a skip table. Combined with peak-valley electricity price constraints, it generates commands for cold storage and release, enabling cross-period arbitrage by storing cold during off-peak hours and releasing cold during peak hours, thus reducing cooling energy costs. The coordinated control of the intelligent refrigerant box and the independent heat exchanger at the interlocking level achieves completely silent heat dissipation by releasing stored cold energy during noise-sensitive periods, while simultaneously maintaining the battery temperature stably within the optimal operating range of 25-33 degrees Celsius, slowing down battery capacity decay and extending cycle life. This invention achieves deep coupling between thermal management and economic dispatch in energy storage systems, providing key technical support for the efficient, economical, and low-noise operation of large-scale energy storage systems. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 This is a flowchart of a collaborative control method for energy storage cooling capacity and peak-valley arbitrage in this invention;

[0064] Figure 2 This is a schematic diagram of the spatiotemporal distribution matrix of heat load in this invention;

[0065] Figure 3 This is a schematic diagram of the thermal demand Fibonacci stack in this invention;

[0066] Figure 4 This is a schematic diagram of the skip table for the arbitrage strategy in this invention;

[0067] Figure 5This is a schematic diagram illustrating the collaborative operation of the intelligent refrigerant box and the independent heat exchanger at the plug-in stage in this invention;

[0068] Figure 6 This is a functional block diagram of an energy storage cooling capacity-peak arbitrage collaborative control system according to the present invention. Detailed Implementation

[0069] 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.

[0070] Example 1

[0071] Please see Figure 1 As shown, this embodiment provides a collaborative control method for energy storage cooling capacity and peak-valley arbitrage, including:

[0072] S1: Collect multi-dimensional operational data of energy storage systems and real-time transaction data of the electricity market, and use temporal convolutional networks to perform feature extraction and fusion analysis to generate thermoelectric prediction tensors;

[0073] This step addresses the challenge of coordinating the prediction of battery heat dissipation demand with electricity price fluctuations during energy storage system operation. Based on multi-dimensional operational data of the energy storage system and real-time electricity market transaction data, a thermoelectric prediction tensor is constructed that comprehensively reflects the evolution trend of battery heat load and the changing patterns of electricity cost. This thermoelectric prediction tensor, as the final output of S1, will be used in S2 to analyze the spatiotemporal distribution characteristics of heat load and calculate the cooling urgency weights for each battery cluster.

[0074] Specifically, the process of generating the thermoelectric prediction tensor includes:

[0075] S11: Collect multi-dimensional operational data of the energy storage system and construct an energy storage operation data set.

[0076] In this step, based on the sensor network and battery management system deployed in the energy storage system, voltage acquisition sequences, current acquisition sequences, battery temperature acquisition sequences, ambient temperature acquisition sequences, charge / discharge rate acquisition sequences, and operating timestamp acquisition sequences are synchronously acquired, and an energy storage operation dataset is constructed based on these acquisition sequences. The voltage acquisition sequence refers to the discrete voltage reading sequence output by the terminal voltage sensors of each battery cell in the battery cluster during the operating cycle, which is periodically read by the voltage sampling module of the battery management system. The current acquisition sequence refers to the charge / discharge current reading sequence output by the main circuit current sensor of the battery cluster, which is acquired by a Hall current sensor or shunt. The battery temperature acquisition sequence refers to the battery temperature reading sequence fed back by temperature sensors arranged on the surface of each battery cell or inside the battery module, which is acquired by a multi-channel temperature acquisition module. The battery temperature acquisition sequence includes temperature measurement data at different locations within each battery cluster, used to characterize the temperature distribution differences within the battery cluster. The ambient temperature acquisition sequence refers to the ambient temperature reading sequence fed back by ambient temperature sensors deployed in the energy storage room or energy storage container, which is periodically read by an environmental monitoring module. The charge / discharge rate acquisition sequence refers to the time series of charge / discharge rates calculated based on the current acquisition sequence and the battery's rated capacity. Specifically, the absolute value of the current at each sampling moment is divided by the battery's rated capacity to obtain the corresponding charge / discharge rate value, i.e., charge / discharge rate value = |I(t)| / Q_nom, where I(t) is the current value at the sampling moment, and Q_nom is the battery's rated capacity, expressed in ampere-hours (AHs). This value characterizes the battery's operating intensity at different times. The operating timestamp acquisition sequence refers to the sequence of time points recorded by the energy storage system's main controller during each sampling, uniformly generated by the system clock module. Combining these acquisition sequences according to battery cluster identifier, signal type, and time sequence yields an energy storage operation data set with time as the primary key, providing a foundational data source for subsequent time alignment and feature extraction.

[0077] S12: Collect real-time electricity market transaction data and construct an electricity price data set.

[0078] In this step, communication with the power trading platform or grid dispatch system's data interface is used to obtain real-time node price sequences, time-of-use price prediction sequences, and price timestamp sequences, and a price data set is constructed based on these sequences. The real-time node price sequence refers to the real-time marginal price reading sequence of each trading node published by the power market, obtained through the power market data interface according to the trading cycle. The real-time node price is expressed in yuan per kilowatt-hour, reflecting the market clearing price corresponding to the current power supply and demand relationship. The time-of-use price prediction sequence refers to the sequence of time-of-use price prediction values ​​within a preset future time window generated by the power trading platform based on historical power market data and the power trading platform's built-in load forecasting model. The load forecasting model is an existing mature forecasting service model of the power trading platform. The length of the preset time window is determined according to the dispatching cycle of the energy storage system, preferably 24 hours, and is obtained through the power trading platform's forecasting service interface. The price timestamp sequence refers to the time point sequence corresponding to the release of power market data, used to mark the time attribute of each price data point. The above sequences are organized in chronological order to obtain a set of electricity price data with time as the primary key, which provides data support in the electricity price dimension for subsequent time alignment with energy storage operation data.

[0079] S13: Based on the energy storage operation data set and the electricity price data set, a unified time axis multi-source sample matrix is ​​obtained through the time alignment method.

[0080] Specifically, the time alignment method includes the following steps:

[0081] S131: Determine a unified sampling time axis based on the operating timestamp acquisition sequence and the electricity price timestamp sequence. The unified sampling time axis is a set of time points evenly divided within the operating cycle of the energy storage system according to a set minimum sampling interval. The set minimum sampling interval is determined based on the smaller value between the control response cycle of the energy storage system and the trading cycle of the electricity market, preferably 1 second, and is used as a unified time reference for all signal alignment.

[0082] S132: Based on each acquisition sequence in the energy storage operation data set, perform linear interpolation on the acquisition sequence according to a unified sampling time axis to obtain an aligned sequence of operation data mapped onto the unified sampling time axis. The linear interpolation operation refers to finding the two nearest sampling points before and after the target time point in the corresponding acquisition sequence of the energy storage operation data set for each time point on the unified sampling time axis, and calculating the interpolation result corresponding to the target time point on the unified sampling time axis according to the time difference ratio. Specifically, the calculation method is to divide the time difference between the target time point and the previous sampling point by the time interval between the two sampling points, multiply it by the difference between the acquisition values ​​of the two sampling points, and finally add the acquisition value of the previous sampling point to obtain the interpolation result.

[0083] Preferably, the calculation formula for the linear interpolation operation is as follows:

[0084]

[0085] in, This represents the target time point on the unified sampling time axis. and These represent the two most recent sampling time points before and after the target time point in the original acquisition sequence, and satisfy the following conditions: , and These represent the collected values ​​at the corresponding sampling time points. This represents the interpolation result at the target time point. By constraining the variation amplitude of adjacent samples, the pseudo-oscillations generated by interpolation are limited, thus balancing data smoothness and dynamic response fidelity.

[0086] S133: Based on each sequence in the electricity price dataset, perform a step-holding operation on the sequence according to the unified sampling time axis to obtain an aligned electricity price data sequence mapped onto the unified sampling time axis. The step-holding operation refers to finding the nearest electricity price release time point before the target time point on the unified sampling time axis in the corresponding electricity price sequence for each time point on the unified sampling time axis, and using the electricity value corresponding to the target time point on the unified sampling time axis as the electricity value at the current time point. This step-holding operation conforms to the actual characteristic that time-of-use electricity prices in the electricity market remain unchanged during the trading cycle.

[0087] S134: The aligned sequences of each operational data point and the aligned sequences of electricity price data are horizontally concatenated at corresponding positions on a unified sampling time axis to generate a unified time axis multi-source sample matrix with time as the row and different physical quantities and electricity price quantities as the columns. Each row of the unified time axis multi-source sample matrix corresponds to a time point on the unified sampling time axis, and each column corresponds to a specific data dimension. The data dimensions include the voltage, current, battery temperature, ambient temperature, charge / discharge rate of each battery cluster, as well as the real-time node electricity price and time-of-use electricity price prediction values. The unified time axis multi-source sample matrix serves as the input to S14 to construct the heat generation rate derived sequence and the electricity price cost derived sequence, enabling the correlation analysis of battery heating characteristics and electricity price fluctuation patterns under the same time benchmark.

[0088] S14: Based on the unified time axis multi-source sample matrix, the thermoelectric feature derivation matrix is ​​obtained through thermoelectric feature derivation processing.

[0089] Specifically, the process of thermoelectric feature derivation includes:

[0090] S141: Based on the current and voltage columns in the unified time-axis multi-source sample matrix, the instantaneous power value is obtained by multiplying the current and voltage values ​​at each time point. The instantaneous power values ​​at all time points are arranged in chronological order to form an instantaneous power value sequence, which is then used as a derived column of instantaneous power. The instantaneous power value is calculated as the product of the current and voltage values ​​and is used to characterize the energy conversion intensity of the battery at the corresponding time point.

[0091] Preferably, the formula for calculating the instantaneous power value is as follows:

[0092]

[0093] in, Indicates time The instantaneous power value, in watts. Indicates time The current value, in amperes. Indicates time The voltage value is expressed in volts.

[0094] S142: Based on the battery temperature column and ambient temperature column in the unified time-axis multi-source sample matrix, calculate the temperature difference value by subtracting the ambient temperature value from the battery temperature value at each time point. Arrange the temperature difference values ​​of all time points in chronological order to form a temperature difference value sequence, and use this temperature difference value sequence as a derived column of the temperature difference. The temperature difference value is calculated as the difference between the battery temperature value and the ambient temperature value, and is used to characterize the thermal potential difference of the battery relative to the environment.

[0095] Preferably, the formula for calculating the temperature difference is as follows:

[0096]

[0097] in, Indicates time The temperature difference, in degrees Celsius. Indicates time The battery temperature value, Indicates time The ambient temperature value.

[0098] S143: Based on the charge / discharge rate column and battery temperature column in the unified time-axis multi-source sample matrix, and the sum of the product of the square of the charge / discharge rate and the battery internal resistance coefficient, and the product of the difference between the battery temperature and the ambient temperature and the heat dissipation coefficient, the estimated heat generation rate for each time point is calculated. All the estimated heat generation rate values ​​for all time points are arranged in chronological order to form a heat generation rate estimation sequence, which is then used as a derived column for heat generation rate. The heat generation rate estimation is calculated by multiplying the square of the charge / discharge rate by the battery internal resistance coefficient, plus the difference between the battery temperature and the ambient temperature multiplied by the heat dissipation coefficient. The battery internal resistance coefficient is determined by looking up a table based on the battery type and state of charge, while the heat dissipation coefficient is determined based on the thermal design parameters of the energy storage module.

[0099] Preferably, the formula for calculating the estimated heat generation rate is as follows:

[0100]

[0101] in, Indicates time The estimated heat generation rate, in watts. This represents the battery's internal resistance coefficient, measured in ohms. It is determined by looking up a table based on the battery type and state of charge, with a preferred value range of 0.5 milliohms to 5 milliohms. Indicates time The charge / discharge rate, expressed in 1 / hour. This indicates the battery's rated capacity, measured in ampere-hours (Ahs). The heat dissipation coefficient is expressed in watts per degree Celsius and is determined based on the thermal design parameters of the energy storage box. The preferred value range is 0.1 to 1.0 watts per degree Celsius.

[0102] The heat generation rate derivative column is used to characterize the heat generation rate of the battery under different operating conditions.

[0103] S144: Based on the real-time node electricity price column and the time-of-use electricity price prediction column in the unified time axis multi-source sample matrix, the electricity price deviation value is obtained by dividing the difference between the real-time node electricity price and the time-of-use electricity price prediction value at each time point by the time-of-use electricity price prediction value. The electricity price deviation values ​​at all time points are arranged in chronological order to form an electricity price deviation value sequence, which is used as a derived column of electricity price deviation. The electricity price deviation value is calculated by subtracting the time-of-use electricity price prediction value at the corresponding time point from the real-time node electricity price value in the time-of-use electricity price prediction value column of the unified time axis multi-source sample matrix, and then dividing by the time-of-use electricity price prediction value. This value is used to characterize the degree of fluctuation of the real-time electricity price relative to the expected electricity price.

[0104] Preferably, the formula for calculating the electricity price deviation value is as follows:

[0105]

[0106] in, Indicates time The electricity price deviation value is a dimensionless value. Indicates time The real-time node electricity value, expressed in yuan per kilowatt-hour. Indicates time The time-of-use electricity price forecast for the corresponding time period, in yuan per kilowatt-hour; when When the value is zero, the electricity price deviation value is set to the preset maximum deviation constant.

[0107] S145: The instantaneous power derived column, temperature difference derived column, heat generation rate derived column, and electricity price deviation derived column are horizontally concatenated with the original columns of the unified time-axis multi-source sample matrix to generate a thermoelectric feature derived matrix. The number of columns in the thermoelectric feature derived matrix is ​​the sum of the original number of columns and the number of derived columns in the unified time-axis multi-source sample matrix. The thermoelectric feature derived matrix serves as the input to S15 for feature extraction and fusion analysis of the temporal convolutional network.

[0108] S15: Based on the thermoelectric feature derivation matrix, feature extraction and fusion analysis are performed through a temporal convolutional network to obtain the thermoelectric prediction tensor.

[0109] Specifically, the feature extraction and fusion analysis process of the temporal convolutional network includes:

[0110] S151: The thermoelectric feature derivation matrix is ​​sliced ​​according to a preset sliding window length to generate multiple thermoelectric feature input segments. The preset sliding window length is determined based on the battery thermal inertia time constant and the electricity market trading cycle, preferably 1800 time steps. When the sampling interval is 1 second, it corresponds to a 30-minute historical data window. The battery thermal inertia time constant refers to the characteristic time required for the battery temperature to respond to changes in charging and discharging power. It is determined through battery thermal model parameter identification experiments, and the preferred value range is 600 seconds to 1800 seconds, i.e., 10 minutes to 30 minutes. The thermoelectric feature input segments are three-dimensional tensor structures, with the first dimension being the segment index, the second dimension being the number of time steps within the window, and the third dimension being the number of feature dimensions.

[0111] S152: The thermoelectric feature input segment is input into the causal convolutional layer of the temporal convolutional network for temporal feature extraction. The causal convolutional layer refers to a convolutional layer structure in which the convolutional kernel only performs convolution operations on the input at the current time and previous time steps, to ensure that the prediction result does not depend on future information. The temporal convolutional network adopts a dilated causal convolution structure. The dilated causal convolution refers to introducing a dilation coefficient by inserting a fixed number of zero-value intervals between adjacent weight parameters of the convolutional kernel on the basis of causal convolution. The value of the dilation coefficient determines the number of intervals between adjacent weight parameters, so that the receptive field of the convolutional kernel grows exponentially with the number of network layers, thereby capturing temporal dependencies over a longer time span without increasing the number of network parameters.

[0112] Preferably, the calculation formula for the dilated causal convolution is as follows:

[0113]

[0114] in, Indicates time The convolution output, Indicates time Input value, The convolution kernel is represented by the first... Each weight parameter is automatically learned and determined during the training of the temporal convolutional network using the backpropagation algorithm and gradient descent optimization method. Indicates the kernel size. This represents the coefficient of thermal expansion.

[0115] The expansion coefficient increases exponentially with the number of network layers, and the... The formula for calculating the expansion coefficient of a layer is shown below:

[0116]

[0117] in, Indicates the first The coefficient of thermal expansion of the layer, This represents the network layer index, starting from 1.

[0118] The formula for calculating the receptive field size of the dilated causal convolution is as follows:

[0119]

[0120] in, Indicates the size of the receptive field. Indicates the total number of network layers. This represents the kernel size, preferably 3. Through layer-by-layer processing of the causal convolutional layers, local temporal patterns are extracted from the thermoelectric feature input segment, resulting in a temporal local feature tensor. The receptive field size is calculated as follows: the increment of the receptive field for each causal convolutional layer is equal to the kernel size. Subtract 1 and then multiply by the expansion coefficient of the corresponding layer. The total increment is obtained by summing the receptive field increments of all layers. Due to the coefficient of thermal expansion The summation of a geometric series yields... Adding the initial time point yields the final receptive field size. .

[0121] S153: Input the temporal local feature tensor into the residual connection module of the temporal convolutional network for feature fusion. The residual connection module refers to a network structure that adds the output of the convolutional layer to the input element by element, which is used to alleviate the gradient vanishing problem during the training of deep networks and to preserve the original input information.

[0122] Preferably, the calculation formula for the residual connection module is as follows:

[0123]

[0124] in, This represents the output feature tensor of the residual connection module. This represents the output feature tensor of the convolutional layer. This represents the input feature tensor of the residual connection module. The modified linear unit activation function is defined as follows: When the dimensions of the input feature tensor and the convolutional output feature tensor are inconsistent, a 1×1 convolution transformation is performed on the input feature tensor to match the dimensions before the addition operation is performed. Through the processing of the residual connection module, the multi-scale temporal features extracted by different convolutional layers are fused to obtain a temporal fused feature tensor.

[0125] S154: Input the temporal fusion feature tensor into the output mapping layer of the temporal convolutional network to generate a thermoelectric prediction tensor. The output mapping layer is a fully connected layer structure that maps the temporal fusion feature tensor to an output space consisting of a preset prediction temporal length and a preset prediction dimension.

[0126] Preferably, the calculation formula for the output mapping layer is as follows:

[0127]

[0128] in, This represents the output of the thermoelectric prediction tensor. This represents a one-dimensional vector obtained after flattening the temporal fusion feature tensor. This represents the weight matrix of the output mapping layer. This represents the bias vector of the output mapping layer; the weight matrix and bias vector are automatically learned and determined during the training process of the temporal convolutional network based on the backpropagation algorithm and gradient descent optimization method.

[0129] The thermoelectric prediction tensor is restored to a three-dimensional structure after a reshaping operation. The ReLU activation function is used to ensure non-negativity for the output of the heat load prediction dimension, while no activation function is used for the output of the electricity cost prediction dimension to preserve positive and negative price signals.

[0130] The preset prediction time domain length is determined based on the energy storage system's scheduling cycle and peak / valley electricity price periods, preferably 24 hours. The preset prediction dimensions include a heat load prediction dimension and an electricity price cost prediction dimension. The heat load prediction dimension corresponds to the predicted heat load value of each battery cluster at each future time step, and the electricity price cost prediction dimension corresponds to the predicted electricity price cost value at each future time step. The thermoelectric prediction tensor is a three-dimensional tensor structure, with the first dimension being the battery cluster index, the second dimension being the future time step index, and the third dimension containing two components: the predicted heat load value and the predicted electricity price cost value.

[0131] Specifically, traditional energy storage systems often employ a feedback control strategy based on the current temperature, meaning that cooling equipment is only activated when the battery temperature exceeds a set threshold. This approach suffers from significant hysteresis because there is a non-linear hysteresis between battery heating and charging / discharging behavior. When the temperature sensor detects overheating, activating cooling at this point not only increases energy consumption but also affects battery cycle life. Furthermore, traditional methods do not incorporate electricity price factors into the heat dissipation control decision, potentially leading to frequent activation of cooling equipment during peak electricity price periods, increasing operating costs. S1, by constructing a thermoelectric prediction tensor, achieves joint prediction of future heat load and electricity price costs, transforming heat dissipation control from a passive response to proactive prediction. The dilated causal convolutional structure of the temporal convolutional network effectively captures the hysteresis characteristics of battery heating relative to the charge / discharge rate, as well as the periodic patterns of electricity price fluctuations, enabling the system to anticipate future periods of thermal shock and low electricity price windows. The construction of the thermoelectric feature derivative matrix transforms the original physical measurements into derived features with clear thermodynamic and economic significance, making the learning objective of the temporal convolutional network more explicit. The three-dimensional structural design of the thermoelectric prediction tensor can simultaneously express information from multiple cell clusters, multiple time steps, and multiple prediction dimensions, providing a complete data benchmark for subsequent cooling task prioritization and peak-valley arbitrage strategy formulation, and solving the problem of the disconnect between heat demand prediction and electricity price prediction.

[0132] S2: Analyze the spatiotemporal distribution characteristics of heat load in the thermoelectric prediction tensor, calculate the cooling urgency weight of each battery cluster, and construct a heat demand Fibonacci stack.

[0133] This step addresses the challenge of efficiently and dynamically prioritizing the heat dissipation demands of massive battery clusters in large-scale energy storage power stations. Based on the thermoelectric prediction tensor output from S1, it constructs a thermal demand Fibonacci stack capable of supporting millisecond-level responses to local hotspot changes by analyzing the spatiotemporal distribution characteristics of heat load and calculating cooling urgency weights. This thermal demand Fibonacci stack, as the final output of S2, will be invoked in S3 to extract the optimal cooling task and perform cooling supply and demand matching in conjunction with peak-valley electricity price constraints.

[0134] Specifically, the process of constructing the thermal demand Fibonacci stack includes:

[0135] S21: Construct the spatiotemporal distribution matrix of heat load based on the thermoelectric prediction tensor.

[0136] In this step, data on the heat load prediction dimension is extracted from the thermoelectric prediction tensor to construct a heat load spatiotemporal distribution matrix. This matrix is ​​a two-dimensional matrix structure, where row indices correspond to battery cluster identifiers, column indices correspond to future time step identifiers, and matrix elements represent the predicted heat load values ​​for the corresponding battery clusters at the corresponding future time steps. This heat load spatiotemporal distribution matrix characterizes the heat load evolution of each battery cluster in the energy storage system within the future prediction time domain, providing thermodynamic data support for subsequent calculations of cooling urgency weights. Simultaneously, data on the electricity price cost prediction dimension is extracted from the thermoelectric prediction tensor to construct an electricity price cost prediction vector. This vector is a one-dimensional vector structure, where vector indices correspond to future time step identifiers, and vector elements represent the predicted electricity price cost values ​​for the corresponding future time steps. This electricity price cost prediction vector characterizes the electricity price fluctuation trend within the future prediction time domain, providing economic data support for subsequent calculations of cooling urgency weights. See also... Figure 2 This is a schematic diagram of the spatiotemporal distribution matrix of heat load provided in this embodiment. For example... Figure 2 As shown, this visually illustrates the spatiotemporal evolution of the energy storage system's future operating state. The main matrix at the top uses the "future time axis" (T1 to T8) as the horizontal axis and the "physical location of battery clusters" (B1 to B5) as the vertical axis. The color intensity of the squares in the spatiotemporal distribution matrix of the heat load represents the predicted heat load level. For example, during the period from T3 to T5, battery cluster B2 is dark blue, indicating that the area where battery cluster B2 is located will face a high heat load impact during this period. The bar chart below simultaneously displays the predicted electricity price cost at the corresponding time, transitioning from green (low electricity price) to red (high electricity price). In actual control logic, simple temperature prediction or electricity price prediction is insufficient to form optimal decisions. Figure 2The thermodynamic dimension (heat load) and the economic dimension (electricity price) are juxtaposed and correlated on the same time base. Through this visual mapping, the system can easily identify high-heat-high-price risk periods (such as cluster B4 at time T7) and low-heat-low-price cold storage windows (such as time T1). This provides intuitive data support for calculating the cooling urgency weight in step S2, enabling subsequent control strategies to avoid forced cooling during peak electricity price periods and achieve peak-shaving and valley-filling thermal management.

[0137] S22: Based on the spatiotemporal distribution matrix of heat load and the battery thermal inertia model, calculate the predicted sequence of temperature rise rate for each battery cluster.

[0138] Specifically, the calculation process for the temperature rise rate prediction sequence includes:

[0139] S221: Establish a battery thermal inertia model. The battery thermal inertia model is a dynamic characteristic model describing the battery's temperature response to changes in heat load, and it is modeled using a first-order inertial element. The construction process of the battery thermal inertia model includes: firstly, obtaining the mass and specific heat capacity parameters of each battery cell in the battery cluster, and using the product of the battery mass and specific heat capacity as the thermal capacity coefficient, which characterizes the temperature rise when the battery absorbs a unit of heat; then, based on the thermal design parameters of the energy storage box and the heat exchanger configuration, determining the thermal resistance coefficient through thermal resistance calibration experiments or thermal simulation calculations, which characterizes the battery's ability to dissipate heat to the surrounding environment. Specifically, the thermal resistance coefficient is obtained by measuring the difference between the battery surface temperature and the ambient temperature under steady-state heat dissipation conditions and dividing it by the battery's steady-state heat generation power; finally, based on the principle of energy conservation, substituting the thermal capacity coefficient and thermal resistance coefficient into the first-order linear differential equation to construct the battery thermal inertia model, so that the model input is the heat generation rate and the output is the change in battery temperature over time.

[0140] The battery thermal inertia model is described by a first-order linear differential equation, the mathematical expression of which is shown below:

[0141]

[0142] in, This represents the heat capacity coefficient, measured in joules per degree Celsius, and is calculated using the following formula: , This indicates the battery's mass, expressed in kilograms. This indicates the specific heat capacity of the battery, measured in joules per kilogram per degree Celsius, with a preferred value range of 800 to 1200 joules per kilogram per degree Celsius. This represents the thermal resistivity, expressed in degrees Celsius per watt. It is determined based on the thermal design parameters of the energy storage tank, with a preferred value range of 0.1 to 1.0 degrees Celsius per watt. Indicates time The battery temperature; Indicates time Ambient temperature; Indicates time The rate of heat generation. Left side of the equation. This represents the heat required for a change in battery temperature, i.e., the product of the thermal capacity coefficient and the rate of temperature change; the first term on the right-hand side of the equation. This indicates the rate of heat generation inside the battery; the second term on the right side of the equation. This represents the rate at which the battery dissipates heat to the environment, calculated by dividing the temperature difference between the battery and the environment by the thermal resistivity. When the rate of heat generation inside the battery exceeds the rate at which it dissipates heat to the environment, the battery temperature rises; conversely, when the rate at which it dissipates heat to the environment exceeds the rate at which it generates heat, the battery temperature falls.

[0143] The transfer function form of the battery thermal inertia model is shown below:

[0144]

[0145] in, Represents the Laplace operator. This represents the thermal time constant, with the unit being seconds.

[0146] S222: Based on the predicted heat load sequence corresponding to each battery cluster in the spatiotemporal distribution matrix of heat load, and combined with the thermal capacity coefficient and thermal resistance coefficient of the battery thermal inertia model, the predicted temperature rise rate of each battery cluster at each future time step is calculated. The predicted temperature rise rate is calculated by dividing the predicted heat load by the thermal capacity coefficient and then subtracting the current temperature deviation value divided by the product of the thermal resistance coefficient and the thermal capacity coefficient. The current temperature deviation value is the difference between the current battery temperature and the target operating temperature, and the target operating temperature is set as the median of the battery's optimal operating range.

[0147] Preferably, the formula for calculating the predicted temperature rise rate is as follows:

[0148]

[0149] in, Indicates the first A battery cluster in the future The predicted rate of temperature rise, in degrees Celsius per second. The first element in the spatiotemporal distribution matrix of heat load represents the... A battery cluster in the future The predicted heat load value, Indicates the first The thermal capacity coefficient of each battery cluster Indicates the first The thermal resistance coefficient of each battery cluster Indicates the first The current temperature of each battery cluster, This indicates the target operating temperature, with a preferred value of 29 degrees Celsius, which is the midpoint of the battery's optimal operating range of 25 to 33 degrees Celsius.

[0150] S223: Arrange the predicted temperature rise rate values ​​of each battery cluster at each future time step in chronological order to obtain a temperature rise rate prediction sequence for each battery cluster. The temperature rise rate prediction sequence is a one-dimensional vector structure, the vector length of which is equal to the dimension of the future time step index in the thermoelectric prediction tensor, that is, the number of time steps within the future prediction time range covered by the thermoelectric prediction tensor, and the vector elements are the predicted temperature rise rate values ​​for the corresponding time steps.

[0151] S23: Calculate the cooling urgency weight of each battery cluster based on the temperature rise rate prediction sequence, the electricity price cost prediction vector, and the current state of the battery.

[0152] Specifically, the calculation process for the cooling urgency weight includes:

[0153] S231: Obtain the current temperature deviation of each battery cluster. The current temperature deviation is calculated by subtracting the target operating temperature from the current battery temperature and then dividing by the allowable temperature fluctuation range. The allowable temperature fluctuation range is determined based on the upper and lower limits of the battery's optimal operating range, which is set to 25 degrees Celsius to 33 degrees Celsius. The allowable temperature fluctuation range is half the width of the battery's optimal operating range.

[0154] Preferably, the formula for calculating the current temperature deviation is as follows:

[0155]

[0156] in, Indicates the first The current temperature deviation of each battery cluster is a dimensionless value. Indicates the first The current temperature of each battery cluster, This indicates the target operating temperature, with a value of 29 degrees Celsius. This indicates the allowable temperature fluctuation range, calculated using the following formula: Degrees Celsius. The current temperature deviation is a dimensionless value; the larger the absolute value, the further the battery's current temperature is from the target operating temperature, and the more urgent the need for heat dissipation. This indicates the upper limit of the battery's optimal operating temperature range, with a value of 33 degrees Celsius. This indicates the lower limit temperature of the battery's optimal operating range, with a value of 25 degrees Celsius.

[0157] S232: Calculate the peak temperature rise rate for each battery cluster based on the temperature rise rate prediction sequence. The peak temperature rise rate is the maximum value in the temperature rise rate prediction sequence, used to characterize the most severe temperature rise that the battery cluster may experience in the future prediction time domain.

[0158] Preferably, the formula for calculating the peak temperature rise rate is as follows:

[0159]

[0160] in, Indicates the first Peak temperature rise rate of each battery cluster, This represents the number of time steps within the prediction time domain.

[0161] S233: Calculate the electricity cost sensitivity coefficient for each battery cluster based on the electricity cost prediction vector. The electricity cost sensitivity coefficient is calculated by dividing the predicted electricity cost value at the time step corresponding to the peak temperature rise rate in the temperature rise rate prediction sequence by the mean of the electricity cost prediction vector. The electricity cost sensitivity coefficient is a dimensionless value. A value greater than 1 indicates that the electricity price during the peak temperature rise period is higher than the average electricity price. If cooling is started at this time, it will generate economic costs higher than the average electricity price level. Therefore, it is necessary to store cooling capacity in advance during off-peak electricity periods when the electricity price is lower than the average electricity price.

[0162] Preferably, the formula for calculating the electricity price cost sensitivity coefficient is as follows:

[0163]

[0164] in, Indicates the first The electricity price cost sensitivity coefficient of an individual battery cluster Indicates the first The time step corresponding to the peak temperature rise rate of each battery cluster This represents the predicted electricity cost at that time step. The arithmetic mean of the electricity price cost prediction vector is calculated using the following formula: .in This represents the total number of time steps within the prediction time range. Indicates the first The predicted electricity cost for each time step. This means summing the electricity cost forecasts for all time steps within the forecast period and dividing by the total number of time steps. Obtain the arithmetic mean of the electricity price cost prediction vector. .

[0165] S234: Calculate the cooling urgency weight for each battery cluster based on the current temperature deviation, peak temperature rise rate, and electricity price sensitivity coefficient. The cooling urgency weight is calculated by multiplying the current temperature deviation by a temperature deviation weight factor, adding the peak temperature rise rate by a temperature rise rate weight factor, and then adding the electricity price sensitivity coefficient by an electricity price sensitivity weight factor. The temperature deviation weight factor, temperature rise rate weight factor, and electricity price sensitivity weight factor are preset trade-off coefficients, determined using the analytic hierarchy process (AHP) based on the energy storage system's optimization preferences for temperature stability and economic cost. Specifically, a hierarchical structure is constructed with cooling urgency as the target layer and temperature deviation, temperature rise rate, and electricity price sensitivity coefficient as criterion layers. A judgment matrix is ​​constructed by comparing the relative importance of each criterion pairwise. The eigenvector corresponding to the largest eigenvalue of the judgment matrix is ​​calculated and normalized to obtain the weight coefficient of each criterion. Simultaneously, a consistency ratio check ensures the logical consistency of the judgment matrix.

[0166] Preferably, the formula for calculating the cooling urgency weight is as follows:

[0167]

[0168] in, Indicates the first The cooling urgency weight of each battery cluster This indicates the temperature deviation weighting factor. This represents the weighting factor for the rate of temperature rise. This represents the electricity price sensitivity weighting factor. Determined using the analytic hierarchy process (AHP), the optimal value is [value to be filled in]. , , Satisfying the normalization condition The cooling urgency weight is negative to ensure that a smaller value indicates a higher cooling priority, consistent with the minimum stack characteristic of a Fibonacci stack based on heat demand. The cooling urgency weight is a composite index reflecting both the urgency of heat dissipation for the battery cluster and arbitrage opportunities; a smaller value indicates a higher cooling priority for the battery cluster.

[0169] S24: Construct a thermal demand Fibonacci stack based on the cooling urgency weight of each battery cluster.

[0170] Specifically, the construction process of the thermal demand Fibonacci stack includes:

[0171] S241: Define the data structure for the hot demand node. The data structure for the hot demand node includes the following fields: a battery cluster identifier field, used to uniquely identify the battery cluster corresponding to the current hot demand node; a key value field, used to store the cooling urgency weight of the battery cluster identified by the battery cluster identifier field, where a smaller value indicates a higher priority; a parent node pointer field, used to point to the parent node of the current hot demand node in the heap structure; a child node pointer field, used to point to any child node of the current hot demand node; left sibling pointer field and right sibling pointer field, used to link all child nodes of the same parent node into a doubly linked circular list; a degree number field, used to record the number of child nodes of the current hot demand node; and a flag field, used to indicate whether the current hot demand node has lost any child nodes after becoming a child node of a certain node.

[0172] S242: Initialize the hot-demand Fibonacci heap. The hot-demand Fibonacci heap is a mergeable heap data structure, consisting of two core components: a root linked list and a minimum node pointer. The root linked list is a doubly linked circular list storing all root nodes, and the minimum node pointer points to the node with the smallest key value in the root linked list. During initialization, the root linked list is empty, and the minimum node pointer is empty.

[0173] S243: Traverse each battery cluster, create a hot demand node for each cluster, and perform an insertion operation. The insertion operation is as follows: create a new hot demand node, write the battery cluster identifier into the battery cluster identifier field, write the cooling urgency weight into the key value field, initialize the degree field to zero, and initialize the flag field to false; add the new hot demand node to the root linked list, specifically by inserting the new hot demand node into the root linked list where the smallest node pointer points to the left of the current node; compare the key value of the new hot demand node with the key value of the current smallest node, and if the key value of the new hot demand node is less than the key value of the current smallest node, update the smallest node pointer to point to the new hot demand node. The time complexity of this insertion operation is constant, which can efficiently handle the creation of demand nodes for massive battery clusters in large-scale energy storage power stations.

[0174] S244: After inserting the thermal demand nodes for all battery clusters, a thermal demand Fibonacci stack is obtained. This thermal demand Fibonacci stack contains thermal demand nodes equal to the number of battery clusters. Each thermal demand node stores the cooling urgency weight of its corresponding battery cluster through a key-value field. The smallest node pointer in the thermal demand Fibonacci stack points to the thermal demand node with the smallest cooling urgency weight, i.e., the highest cooling priority. See also... Figure 3 This is a schematic diagram of the thermal demand Fibonacci stack provided in this embodiment. Figure 3 As shown, this reveals the logical structure of the dynamic ranking of heat dissipation requirements of massive battery clusters in large-scale energy storage power stations. Figure 3The core is a doubly linked circular "root list" connecting several root nodes (such as battery cluster 3, battery cluster 1, and battery cluster 6). The numbers stored inside each root node (such as 1.8 and 3.2) represent the calculated "cooling urgency weight," with smaller values ​​indicating a more urgent cooling requirement. The "minimum node pointer" always points to the node with the smallest weight in the root list (here, battery cluster 3), ensuring that the operation of obtaining the optimal cooling task is completed in constant time. In large-scale energy storage scenarios, the number of battery clusters may be hundreds or thousands, and their temperature and operating status change rapidly. Traditional sorting structures (such as binary heaps) are inefficient when frequently updating priorities. This embodiment adopts a Fibonacci heap structure, whose unique "lazy merging" mechanism makes the node key value reduction operation have an amortized time complexity of O(1). This means that when a battery cluster experiences a sudden surge in temperature due to a large current charge / discharge (resulting in a sharp decrease in weight), the system can respond in milliseconds and adjust the position of the battery cluster with the surging temperature in the thermal demand Fibonacci stack. This ensures that step S3 can always extract the battery cluster with the lowest cooling urgency weight (i.e., the one whose temperature is closest to or has exceeded the safe operating range and has the most urgent heat dissipation demand) for processing as soon as possible, thus guaranteeing the thermal safety of the system.

[0175] S25: Defines the key-value reduction operation for thermal demand Fibonacci stacks, used for priority adjustment in response to sudden changes in battery cluster temperature or updates to forecast data.

[0176] Specifically, the process of the key-value reduction operation includes:

[0177] S251: Input the node to be adjusted and the new key value. The node to be adjusted is the heat demand node in the heat demand Fibonacci stack that needs to update the cooling urgency weight. The new key value is the recalculated cooling urgency weight, and the new key value must be less than or equal to the current key value of the node to be adjusted.

[0178] S252: Determine if the node to be adjusted has a parent node. If the node to be adjusted does not have a parent node, that is, the node to be adjusted is itself the root node, then directly update the key value field of the node to be adjusted to the new key value, and compare the new key value with the key value of the node pointed to by the minimum node pointer of the hot demand Fibonacci heap. The minimum node pointer is continuously maintained during the initialization of the hot demand Fibonacci heap and each insertion operation. If the new key value is less than the key value of the node pointed to by the minimum node pointer, then update the minimum node pointer to point to the node to be adjusted.

[0179] S253: If the node to be adjusted has a parent node, compare the new key value with the parent node's key value. If the new key value is greater than or equal to the parent node's key value, directly update the key value field of the node to be adjusted to the new key value. If the new key value is less than the parent node's key value, perform a cut-off operation and a cascading cut-off operation.

[0180] S254: The cut-off operation is as follows: remove the node to be adjusted from the child node list of the parent node, decrement the degree number field of the parent node by one, add the node to be adjusted to the root list, set the parent node pointer of the node to be adjusted to null, and set the flag field of the node to be adjusted to false.

[0181] S255: The process of the cascading cut-off operation is as follows: obtain the original parent node of the node to be adjusted; if the original parent node has its own parent node, check the flag field of the original parent node; if the flag field is false, set the flag field to true; if the flag field is true, recursively perform the cut-off operation and the cascading cut-off operation on the original parent node.

[0182] S256: After completing the cut-off operation and cascading cut-off operation, update the key-value field of the node to be adjusted to the new key-value field. Compare the new key-value field with the key-value field of the current smallest node. If the new key-value field is less than the key-value field of the current smallest node, update the pointer of the smallest node. The amortized time complexity of the key-value reduction operation is constant. It can efficiently adjust the thermal demand Fibonacci stack structure when the temperature of the battery cluster changes abruptly, causing a change in the cooling urgency weight, and always keep the thermal demand node corresponding to the battery cluster that needs the most cooling in a quickly accessible position.

[0183] Specifically, in large-scale energy storage power stations, the number of battery clusters may reach hundreds or even thousands, and the heat dissipation requirements of each cluster dynamically adjust with changes in charge / discharge status and environmental conditions. Traditional priority queue implementations, such as binary heaps, suffer from efficiency bottlenecks when handling frequent priority change operations because the key-value reduction operation of a binary heap requires an upward adjustment, the time complexity of which is proportional to the height of the thermal demand Fibonacci heap. When the energy storage system needs to respond in real time to changes in local hot spots and quickly reorder cooling tasks, the adjustment overhead of the binary heap may lead to control response delays. S2 uses a thermal demand Fibonacci heap as the data structure for thermal demand sorting. By leveraging the fact that the key-value reduction and merging operations of the Fibonacci heap have constant-level amortized time complexity, it solves the computational bottleneck problem of dynamically sorting the heat dissipation requirements of massive battery cells. The cooling urgency weighting is calculated by weighting and integrating temperature deviation, temperature rise rate, and electricity price sensitivity coefficient. This ensures that the ranking of the thermal demand Fibonacci stack not only reflects the current urgency of heat dissipation but also embodies future thermal shock risks and peak-valley arbitrage opportunities, laying the foundation for subsequent on-demand allocation of cooling capacity. The construction of the thermal demand Fibonacci stack organizes the previously dispersed cooling needs of individual battery clusters into a unified priority structure, enabling S3 to obtain the optimal cooling task through a simple top-of-stack extraction operation, avoiding the computational overhead of comparing all battery clusters.

[0184] S3: Perform minimum extraction operation on the heat demand Fibonacci stack to extract the optimal cooling task, combine peak and valley electricity price constraints to match cooling supply and demand, and generate arbitrage strategy skip list;

[0185] This step addresses the difficulty in optimizing energy consumption costs caused by the mismatch between cooling generation and usage times. Based on the thermal demand Fibonacci stack output from S2, it obtains the optimal cooling task by performing a minimum extraction operation and combines peak-valley electricity price constraints to match cooling supply and demand, constructing an arbitrage strategy skip list that supports fast interval lookups and insertions. This arbitrage strategy skip list, as the final output of S3, will be invoked in S4 to obtain the control instruction set for the current time window and drive the execution of refined temperature control operations.

[0186] Specifically, the process of generating the arbitrage strategy skip list includes:

[0187] S31: Defines the minimum extraction operation for the thermal demand Fibonacci heap, used to retrieve and remove the thermal demand node with the highest cooling priority.

[0188] Specifically, the process of extracting the minimum operation includes:

[0189] S311: Obtain the heat demand node pointed to by the smallest node pointer of the heat demand Fibonacci stack, denoted as the current smallest node. The current smallest node corresponds to the battery cluster with the smallest cooling urgency weight, i.e., the highest cooling priority.

[0190] S312: Add all child nodes of the current smallest node to the root linked list. Specifically, traverse the child node linked list of the current smallest node, set the parent node pointer of each child node to null, and insert the child node into the root linked list.

[0191] S313: Remove the current smallest node from the root list.

[0192] S314: Perform a heap structure consolidation operation. The purpose of the heap structure consolidation operation is to merge root nodes with the same degree in the root linked list, so that the degree of each root node in the root linked list is different. The process of the heap structure consolidation operation is as follows: Create a degree index array, the length of which is the maximum possible degree in the root linked list plus one, and initialize the array elements to empty; traverse each root node in the root linked list. For the current root node, obtain the degree of the current root node as an index to query the degree index array. If the corresponding position in the array is empty, store the current root node in the corresponding position in the array. If the corresponding position in the array is not empty, perform a link operation between the current root node and the root nodes stored in the array. The link operation compares the key value of the current root node and the root nodes stored in the array, and takes the root node with the larger key value as the child node of the root node with the smaller key value. After linking, set the corresponding position in the degree index array to empty, and continue to check whether higher degree positions of the linked nodes need to be linked; after traversal, reconstruct the root linked list according to the degree index array.

[0193] S315: Traverse the merged root list, find the root node with the smallest key value, and update the pointer of the smallest node to point to the root node with the smallest key value.

[0194] S316: Returns the battery cluster identifier and cooling urgency weight stored in the current minimum node as the extracted optimal cooling task.

[0195] S32: Obtain the current electricity price range identifier and cold storage status.

[0196] Specifically, the method for obtaining the electricity price range identifier is as follows: The current time is compared with the time-of-use electricity price period division table to determine the electricity price range to which the current time belongs. The electricity price range includes three types: off-peak range, flat range, and peak range. The time-of-use electricity price period division table is determined according to the time-of-use electricity price policy of the electricity market. The method for obtaining the cooling capacity storage status is as follows: The current cooling capacity storage amount and maximum cooling capacity storage capacity fed back by the cooling capacity storage monitoring module of the intelligent refrigerant box are read, and the cooling capacity storage ratio is calculated. The cooling capacity storage ratio is the quotient of the current cooling capacity storage amount divided by the maximum cooling capacity storage capacity.

[0197] Preferably, the formula for calculating the cold storage ratio is as follows:

[0198]

[0199] in, This represents the proportion of cold storage capacity; it is a dimensionless value ranging from 0 to 1. This indicates the current cold storage capacity, in kilojoules. This indicates the maximum cold storage capacity, expressed in kilojoules.

[0200] S33: Generate cooling control instructions based on the optimal cooling task, electricity price range identifier, and cooling capacity storage status.

[0201] Specifically, the process for generating the cooling capacity control command includes:

[0202] S331: Determine whether the electricity price range identifier is an off-peak electricity range. If the electricity price range identifier is an off-peak electricity range, further determine whether the cold storage ratio is less than a preset cold storage upper limit threshold. The cold storage upper limit threshold is determined based on the safe operation requirements of the intelligent refrigerant box and the cold storage efficiency curve, with a preferred value of 0.9. If the cold storage ratio is less than the cold storage upper limit threshold, generate a cold storage command. The cold storage command includes three action elements: starting the refrigeration compressor, setting the target refrigerant temperature, and opening the cold storage valve. If the cold storage ratio is greater than or equal to the cold storage upper limit threshold, determine whether the heat demand Fibonacci stack is non-empty and whether the optimal cooling task can be obtained by extracting the minimum operation. If the heat demand Fibonacci stack is non-empty, obtain the optimal cooling task and generate a direct cooling command. The direct cooling command includes three action elements: starting the refrigeration compressor, setting the target refrigerant temperature, and opening the cooling supply valve.

[0203] S332: Determine if the electricity price range is a peak power range. If the electricity price range is a peak power range, further determine if there is an optimal cooling task to be processed. If there is an optimal cooling task and the cold storage ratio is greater than the preset lower limit threshold for cold storage, a cooling release command is generated. The cooling release command includes three action elements: shutting down the refrigeration compressor, opening the cold storage release valve, and starting the refrigerant circulation pump. The lower limit threshold for cold storage is determined based on the minimum effective cold release capacity of the intelligent refrigerant box, preferably with a value of 0.1. If the cold storage ratio is less than or equal to the lower limit threshold for cold storage and the cooling urgency weight of the optimal cooling task is less than the preset emergency heat dissipation threshold, an emergency cooling command is generated. The emergency cooling command includes three action elements: starting the refrigeration compressor, setting the target refrigerant temperature, and opening the cooling supply valve. The emergency heat dissipation threshold is determined based on the risk level of the battery temperature exceeding the safe operating range, preferably with a value of -2.0.

[0204] S333: Determine if the electricity price range is within a flat range. If the electricity price range is within a flat range, a decision is made based on the cooling urgency weight and cold storage ratio of the optimal cooling task. If the cooling urgency weight is greater than the preset delayable heat dissipation threshold and the cold storage ratio is greater than the lower limit threshold of cold storage, a cooling command is generated; if the cooling urgency weight is greater than the delayable heat dissipation threshold and the cold storage ratio is less than or equal to the lower limit threshold of cold storage, a direct cooling command is generated; if the cooling urgency weight is less than or equal to the delayable heat dissipation threshold, a standby command is generated. The standby command includes two action elements: maintaining the current valve state and monitoring temperature changes. The delayable heat dissipation threshold is determined based on the safety margin of the battery temperature deviating from the target operating temperature and the controllability of the temperature rise rate. When the cooling urgency weight is greater than the delayable heat dissipation threshold, it indicates that the heat dissipation demand is still within a controllable range and can be delayed until a more favorable electricity price period. The preferred value is -0.5.

[0205] S334: Add a timestamp and target battery cluster identifier to the generated cooling control command to construct a cooling control command record. The data structure of the cooling control command record includes the following fields: execution timestamp field, recording the time when the current cooling control command should be executed; target battery cluster identifier field, recording the battery cluster targeted by the current cooling control command; command type field, recording which type of command it is: cold storage command, cold release command, direct cooling command, emergency cooling command, or standby command; action parameter field, recording the specific action elements and parameter values ​​included in the current cooling control command.

[0206] S34: Initialize the arbitrage strategy skip list and define the skip list operations.

[0207] Specifically, the data structure definition of the arbitrage strategy skip list includes:

[0208] S341: Define the skip list node data structure. The skip list node data structure includes the following fields: a key field, which stores the execution timestamp corresponding to the current skip list node, used to determine the node's sequential position in the arbitrage strategy skip list; a data field, which stores the cooling control instruction record corresponding to the current skip list node; and a hierarchy pointer array field, which stores the pointers to the successor nodes of the current skip list node at each hierarchy, with the array length equal to the hierarchy height of the current skip list node.

[0209] S342: Define the skip list structure. The arbitrage strategy skip list consists of three core components: a head node, the current maximum level, and a node counter. The head node is the entry node of the arbitrage strategy skip list. The key-value field of the head node is set to the minimum possible timestamp, and the length of the level pointer array is set to the preset maximum number of levels. The current maximum level records the maximum level height of existing nodes in the arbitrage strategy skip list. The node counter records the total number of cooling control instruction records stored in the skip list.

[0210] S343: Define the skip list insertion operation. The skip list insertion operation is as follows: Input the cooling control instruction record to be inserted, and extract the execution timestamp field as the insertion key; Starting from the highest level of the head node, search to the right layer by layer until a position is found where the key value is less than the insertion key value and the successor node key value is greater than or equal to the insertion key value, and record the predecessor node of each layer; Use a random level generation function to determine the level height of the new node. The generation rule of the random level generation function is to repeatedly try to increase the level with a preset probability until the attempt fails. The preset probability is determined according to the space efficiency and query efficiency balance requirements of the skip list in the arbitrage strategy; Create a new node, write the insertion key value into the key value field, and write the cooling control instruction record into the data field; According to the level height of the new node, insert the new node between the predecessor node and the successor node at the corresponding position in each layer.

[0211] Preferably, the random level generation function uses a geometric distribution to determine the node level, and the probability distribution formula is as follows:

[0212]

[0213] in, Represents a node-level random variable. Indicates a specific level value. This represents the probability of promotion, with a preferred value of 0.5. This represents the maximum number of levels, calculated using the following formula: , This indicates the number of nodes expected to be stored in the arbitrage strategy skip list; Floor function (round down). When actually generating the hierarchy, the initial hierarchy is set to 1, and the following operations are performed iteratively: generating a hierarchy that follows a uniform distribution. random numbers ,like And the current level value is less than If the value is positive, the level is incremented by 1; otherwise, the loop terminates and the current level value is returned.

[0214] S344: Define the skip list range query operation. The process of the skip list range query operation is as follows: Input the start and end timestamps of the query range; starting from the highest level of the head node, search layer by layer to the right until a position is found where the key value is less than the start timestamp and the successor node's key value is greater than or equal to the start timestamp; starting from the bottom level of the position where the key value is less than the start timestamp and the successor node's key value is greater than or equal to the start timestamp, traverse along the successor pointer until a node with a key value greater than the end timestamp is encountered; collect the data fields of all nodes whose key values ​​are within the query range during the traversal into the query result list, and return the query result list.

[0215] The average time complexity of the skip list insertion operation and the range query operation is... ,in This represents the number of nodes expected to be stored in the skip list for the arbitrage strategy.

[0216] S35: Execute the extraction of minimum operations and instruction generation in a loop to generate an arbitrage strategy skip list.

[0217] Specifically, the process of iteratively executing the extraction of minimum operations and instruction generation to generate the arbitrage strategy skip list includes:

[0218] S351: Determine if the hot demand Fibonacci heap is empty. If the hot demand Fibonacci heap is empty, the loop ends, and the arbitrage strategy skip list is completed.

[0219] S352: If the heat demand Fibonacci stack is not empty, then perform the minimum extraction operation to obtain the optimal cooling task.

[0220] S353: Based on the optimal cooling task, the current electricity price range identifier, and the cooling capacity storage status, generate cooling capacity control instruction records according to the process described in S33.

[0221] S354: Perform a skip list insertion operation to insert the cooling control instruction record into the arbitrage strategy skip list.

[0222] S355: Determine whether the key values ​​of other nodes in the thermal demand Fibonacci stack need to be updated. The update determination is based on the following criteria: if the currently generated cooling control command is a cooling storage command or a cooling release command, the cooling storage state will change, and the cooling urgency weight of the affected battery cluster needs to be recalculated and the key value reduction operation needs to be performed; if there is new temperature monitoring data or prediction data update, the cooling urgency weight of the corresponding battery cluster needs to be recalculated and the key value reduction operation needs to be performed.

[0223] S356: Return to S351 and continue the loop until the heat demand Fibonacci heap is empty or the preset scheduling time domain boundary is reached. See also Figure 4 This is the intended representation of the arbitrage strategy jump provided in this embodiment. For example... Figure 4 As shown, this demonstrates how discrete control commands are organized into an efficient time-series index. The timeline at the bottom marks key electricity price switching points (e.g., the end of off-peak electricity at 08:00 and the start of peak electricity at 12:00). The main data structure employs a multi-layered indexing mechanism. Layer L0 contains all specific command data (e.g., "cold storage preparation," "cooling to B1"), while upper-level indexes such as L1 and L2 provide channels for quick navigation. Figure 4The highlighted yellow area illustrates a "range query" operation, where the system quickly identified all control command sets within the current time window (11:55-12:05). Thermal management of energy storage systems is a continuous and real-time process. Control commands often appear in clusters during specific time periods (such as the instant of charge / discharge switching), with time intervals between commands potentially shortening to milliseconds (e.g., during charge / discharge switching). Ordinary linked lists require linear traversal when searching for commands within a specific time period, resulting in low efficiency. The skip list structure constructed in this embodiment supports logarithmic-level search efficiency, enabling step S4 to instantly retrieve the sequence of actions to be executed within the current tens of milliseconds through a simple range query. This structure not only ensures the timing accuracy of command execution but also significantly reduces the computational load on the controller, achieving a smooth transition from macro-level strategy planning to micro-level device actions.

[0224] Specifically, the cooling capacity generation of energy storage systems mainly relies on the operation of refrigeration compressors, whose energy consumption typically accounts for over 60% of the auxiliary energy consumption of energy storage systems. If the refrigeration compressors are activated for cooling during peak electricity price periods, it not only increases direct cooling energy costs but also occupies electricity resources that could be used for peak-valley arbitrage. S3 constructs an arbitrage strategy skip list to match discrete cooling demands with peak-valley electricity price constraints, enabling a cross-time-period arbitrage strategy of cooling and storing cold during off-peak hours and releasing cold capacity during peak hours. Extracting the minimum operation efficiently retrieves the optimal cooling task to be processed from the thermal demand Fibonacci stack, with the amortized time complexity proportional to the logarithm of the number of nodes in the thermal demand Fibonacci stack, ensuring processing efficiency in large-scale energy storage power plants. The arbitrage strategy skip list adopts a probabilistically balanced multi-level linked list structure, supporting logarithmic insertion and query time complexity, while also supporting efficient interval query operations, enabling S4 to quickly obtain all control instructions to be executed within the current time window. The generation logic of the cooling control command comprehensively considers three dimensions: electricity price range, cooling storage status, and cooling urgency weight. It maximizes peak-valley arbitrage benefits while ensuring battery temperature safety, and solves the problem of the disconnect between heat dissipation control and energy consumption cost optimization in traditional energy storage systems.

[0225] S4: Traverse the arbitrage strategy skip list to obtain the control instruction set of the current time window, and drive the independent heat exchanger and intelligent refrigerant box to perform fine temperature control operations.

[0226] This step focuses on noise control and energy consumption optimization during the heat dissipation process of the energy storage system. Based on the arbitrage strategy jump table output by S3, it obtains the control command set for the current time window through interval query, driving the independent heat exchanger at the plug-in level and the intelligent refrigerant box to collaboratively perform refined temperature control operations, achieving fully silent heat dissipation of the energy storage battery and maintaining the optimal operating temperature. See also Figure 5 This is a schematic diagram illustrating the collaborative operation of the intelligent refrigerant box and the independent heat exchanger at the plug-in stage provided in this embodiment. Figure 5 The diagram illustrates the physical architecture by which this invention achieves thermoelectric decoupling and precise temperature control. The "intelligent refrigerant box" on the left integrates a "cold storage medium container" (containing phase change material) and a "refrigeration compressor," serving as the central cold source. On the right are "energy storage boxes" distributed within each battery module, each equipped with an "independent heat exchanger" and an "electronic expansion valve." Through a circulation system consisting of "cooling supply pipes" and "return water pipes," the refrigerant carries the heat generated by the battery back to the refrigerant box. Figure 5 The magnified details show that the refrigerant flowing through the independent heat exchanger successfully reduced the battery temperature from 35°C to 28°C. The core advantage of the intelligent refrigerant box and the independent heat exchanger architecture working together lies in achieving the spatial and temporal separation of cooling capacity "production" and "consumption". During off-peak hours, the compressor operates to store cooling capacity in the phase change material (production); during peak hours or noise-sensitive periods, the compressor stops, and the system uses a circulating pump to precisely deliver the stored cooling capacity to the cells with cooling needs (consumption). Combined with the opening adjustment of the electronic expansion valve, the system not only achieves completely silent operation, avoiding the high noise of traditional air cooling, but also provides precise "point-to-point" cooling for the thermal inconsistencies of different battery clusters, effectively extending the overall cycle life of the battery pack.

[0227] Specifically, the process for performing refined temperature control includes:

[0228] S41: Determine the current time window and perform an interval query to obtain the control instruction set.

[0229] Specifically, the process of acquiring the control instruction set includes:

[0230] S411: Get the current timestamp of the system and record it as the current timestamp.

[0231] S412: Calculate the end timestamp of the current time window according to the preset control cycle length. The end timestamp is the sum of the current timestamp and the control cycle length. The control cycle length is determined according to the control response time of the energy storage system and the rate of change of battery temperature, and is preferably between 100 milliseconds and 1 second.

[0232] S413: Using the current timestamp as the start timestamp and the end timestamp as the query interval boundary, perform an interval query operation of the arbitrage strategy skip list to obtain all cooling control instruction records within the current time window, forming the current time window control instruction set. The current time window control instruction set is a list structure, where the list elements are cooling control instruction records, and the list is sorted in ascending order according to the execution timestamp field value in the cooling control instruction records.

[0233] S42: Parse the current time window control instruction set and generate device-level control signals.

[0234] Specifically, the device-level control signal generation process includes:

[0235] S421: Traverse each cooling capacity control instruction record in the current time window control instruction set.

[0236] S422: For a cooling capacity control instruction record of type cold storage instruction, the following device-level control signals are generated: refrigeration compressor start signal, including a target refrigerant temperature setpoint, which is determined based on the cold storage efficiency and refrigerant physical properties, preferably between 5 and 10 degrees Celsius; cold storage valve open signal, including a valve opening percentage, which is determined based on the matching relationship between refrigeration power and cold capacity storage rate; cooling supply valve close signal; refrigerant circulation pump start signal, including a pump speed setpoint, which is determined based on the refrigerant flow rate requirement.

[0237] S423: For cooling control command records with command type "cooling release command", generate the following device-level control signals: refrigeration compressor stop signal; cold storage release valve open signal, including valve opening percentage, the valve opening percentage being determined according to the heat dissipation load requirements of the target battery cluster; cooling valve open signal, the target address being set to the plug-in level independent heat exchanger interface corresponding to the target battery cluster identifier; refrigerant circulation pump start signal, including pump speed setting value.

[0238] S424: For cooling capacity control command records with command type of direct cooling command or emergency cooling command, generate the following equipment-level control signals: refrigeration compressor start signal, including target refrigerant temperature setpoint; cold storage valve close signal; cooling supply valve open signal, with the target address set to the plug-in level independent heat exchanger interface corresponding to the target battery cluster identifier; refrigerant circulation pump start signal, including pump speed setpoint.

[0239] S425: For cooling control command records with a command type of standby command, generate the following device-level control signals: maintain current device status signal; temperature monitoring continue signal; set monitoring interval and alarm threshold.

[0240] The unified method for determining the valve opening percentage and pump speed setpoint is explained below:

[0241] The values ​​of the two types of control parameters mentioned above are based on the existing conventional valve flow-opening calibration curves and pump flow-speed calibration curves in the field of energy storage temperature control. The corresponding closed-loop fine-tuning method of the parameters adopts the PID control technology commonly used in this field, and the relevant calibration and adjustment methods are all existing technologies in this field.

[0242] The valve opening percentage is determined as follows: For the cold storage valve corresponding to the cold storage command, the value is determined by matching the remaining duration of the current off-peak electricity period, the remaining storable cold capacity of the smart refrigerant box, and the real-time output power of the refrigeration compressor, prioritizing the completion of the target cold storage capacity before the end of the off-peak electricity period; For the cold storage release valve and the battery cluster plug-in cooling valve corresponding to the cold release command, direct cooling command, and emergency cooling command, the value is determined by matching the cooling urgency weight of the target battery cluster and the real-time heat load demand, with a larger basic opening for battery clusters with higher cooling priority, and the flow rate is allocated according to the cooling priority when multiple battery clusters are cooling at the same time, and the valve is fully opened first in emergency cooling scenarios to ensure the maximum cold capacity supply.

[0243] The pump speed setpoint is determined by matching the total target refrigerant flow rate corresponding to all currently open valves, while setting constraint boundaries for different operating scenarios: a minimum speed threshold is set in cold storage mode to avoid pump cavitation; a maximum speed threshold is set during noise-sensitive periods and peak power release mode to meet the requirements of fully silent operation; and the rated speed is prioritized in emergency cooling scenarios to ensure maximum refrigerant circulation flow rate. The minimum speed threshold is set based on the constraint relationship between the refrigerant circulation pump speed, the effective net positive suction head at the pump inlet, the refrigerant vaporization pressure margin, and the pressure drop along the pipeline. This involves solving for the minimum allowable speed that simultaneously satisfies the requirement that the effective net positive suction head at the pump inlet is not lower than the cavitation resistance safety limit and that the refrigerant flow state can maintain stable and continuous delivery. This minimum allowable speed is then used as the minimum speed threshold. The maximum speed threshold is set based on the constraint relationship between the refrigerant circulation pump speed and the system radiated noise sound pressure level, pipeline vibration amplitude and cold energy delivery gain per unit time. The maximum allowable speed is calculated when the system radiated noise sound pressure level is not higher than the upper limit of the silent operation safety and the refrigerant flow state can maintain stable and continuous delivery. The maximum allowable speed is then used as the maximum speed threshold.

[0244] S426: The generated device-level control signals are grouped according to execution timestamps and device identifiers to form a device-level control signal scheduling table. The data structure of the device-level control signal scheduling table is a nested list, where the outer list index corresponds to the execution time sequence, and the inner list elements are the control signals that need to be sent to each device at the corresponding execution time.

[0245] S43: Drives the independent heat exchanger at the plug-in stage to perform directional cold energy transfer.

[0246] Specifically, the execution process of the directional cold energy transfer includes:

[0247] S431: Based on the target address of the cooling valve opening signal in the equipment-level control signal scheduling table, determine the target independent heat exchanger that needs to receive cooling capacity.

[0248] S432: Send a flow regulation signal to the target plug-in stage independent heat exchanger, the flow regulation signal including the target refrigerant flow rate value and the regulation response time.

[0249] Preferably, the target refrigerant flow rate value The calculation formula is as follows:

[0250]

[0251] in, The heat dissipation power requirement for the target battery cluster (equal to the heat load value of the target battery cluster) ), For the specific heat capacity of the refrigerant, The temperature difference between the refrigerant inlet and outlet (set value) ).

[0252] S433: After receiving the flow regulation signal, the independent heat exchanger in the plug-in stage adjusts the opening of the matching electronic expansion valve to achieve the target refrigerant flow rate.

[0253] S434: Monitor the actual refrigerant flow rate and battery temperature changes of the plug-in stage independent heat exchanger. If the deviation between the actual refrigerant flow rate and the target refrigerant flow rate is greater than a preset flow deviation threshold, the deviation is calculated by dividing the absolute value of the difference between the actual refrigerant flow rate and the target refrigerant flow rate by the target refrigerant flow rate. Then, a flow regulation deviation signal is output to trigger the electronic expansion valve opening compensation adjustment. If the battery temperature does not decrease as expected, a heat dissipation effect deviation signal is output for subsequent control strategy correction. The flow deviation threshold is determined according to the heat dissipation accuracy requirements of the plug-in stage independent heat exchanger and the adjustment accuracy of the electronic expansion valve, and is preferably 10%.

[0254] S44: Drives the intelligent refrigerant box to perform cold energy storage and release control.

[0255] Specifically, the execution process of the cold energy storage and release control includes:

[0256] S441: Determine the current operating mode of the intelligent refrigerant box based on the cold storage valve opening signal or the cold storage release valve opening signal in the equipment-level control signal scheduling table.

[0257] S442: If the current working mode is cold storage mode, the intelligent refrigerant box performs the following operations: start the refrigeration compressor unit to cool the refrigerant to the target refrigerant temperature; open the cold storage valve to allow the cooled refrigerant to flow through the cold storage medium container and transfer the cold energy to the phase change cold storage material; monitor the temperature and phase state of the phase change cold storage material, and when the phase change cold storage material is completely solidified or reaches the target cold storage temperature, adjust the output power of the refrigeration compressor to maintain the cold energy storage state.

[0258] S443: If the current operating mode is the cooling release mode, the intelligent refrigerant tank performs the following operations: stops the refrigeration compressor unit or keeps it in low-power standby mode; opens the cold storage release valve, allowing the return refrigerant from the interlocking independent heat exchanger to flow through the cold storage medium container and absorb the cold energy released by the phase change cold storage material; starts the refrigerant circulation pump, delivering the refrigerant after absorbing the cold energy to the target interlocking independent heat exchanger. In the cooling release mode, the refrigeration compressor unit is in a stopped or low-power state, and the heat dissipation function of the energy storage system relies entirely on the stored cold energy, without generating compressor operating noise.

[0259] S444: Monitors changes in the cooling capacity of the intelligent refrigerant box, updates the cooling capacity storage status, and feeds it back to the control system for subsequent cooling supply and demand matching decisions.

[0260] S45: Performs battery operating temperature maintenance control.

[0261] Specifically, the execution process of the battery operating temperature maintenance control includes:

[0262] S451: Continuously collect real-time temperature data for each battery cluster and compare the relationship between the real-time temperature and the target operating temperature range. The target operating temperature range is 25 degrees Celsius to 33 degrees Celsius. The target operating temperature range is the optimal operating temperature range for the battery, within which the battery's electrochemical reaction efficiency is highest and its capacity decay is slowest.

[0263] S452: If the real-time battery temperature is 25 degrees Celsius lower than the lower limit of the target operating temperature range, determine whether the current state is in cooling mode. If in cooling mode, reduce the target refrigerant flow rate of the corresponding plug-in independent heat exchanger or suspend the supply of refrigerant to the plug-in independent heat exchanger; if not in cooling mode, maintain the current state.

[0264] S453: If the real-time battery temperature is 33 degrees Celsius higher than the upper limit of the target operating temperature range, determine the current cold storage status and electricity price range identifier. If the cold storage ratio is greater than the lower limit threshold, increase the target refrigerant flow rate of the corresponding independent heat exchanger. If the cold storage ratio is less than or equal to the lower limit threshold and the electricity price range identifier is the peak electricity range, trigger the generation of an emergency cooling command and update the arbitrage strategy skip table.

[0265] S454: If the real-time battery temperature is within the target operating temperature range, the refrigerant flow rate will be finely adjusted according to the specific position of the real-time battery temperature within the target operating temperature range, so that the real-time battery temperature tends to stabilize towards the median value of the target operating temperature range.

[0266] S46: Perform fully silent operation status monitoring.

[0267] Specifically, the execution process of the fully silent operation status monitoring includes:

[0268] S461: Determine whether the current time period is a noise-sensitive period. The noise-sensitive period is determined based on the noise restriction requirements of the scenario where the energy storage system is located. The noise-sensitive period includes nighttime hours or quiet periods in application scenarios such as parks, buildings, and data centers.

[0269] S462: If the current period is a noise-sensitive period, check the operating status of the refrigeration compressor unit. If the refrigeration compressor unit is running and it is not in an emergency cooling situation, send a compressor stop signal to the intelligent refrigerant box and switch to pure cooling mode to dissipate heat by relying on stored cold energy.

[0270] S463: Monitor the noise generation status of each device, including the refrigeration compressor unit, refrigerant circulation pump, and electronic expansion valve. The noise generated by the refrigerant circulation pump and electronic expansion valve under normal operating conditions is lower than the ambient background noise, which refers to the background noise level in the energy storage system's environment when no equipment is running, typically 30 to 40 decibels, and does not affect the completely silent operation. If the refrigeration compressor unit must be started due to emergency cooling needs, the start-up time and runtime are recorded for subsequent optimization of the cold storage strategy.

[0271] S464: Output a fully silent operation status indicator. This fully silent operation status indicator is a Boolean value. It is true when the refrigeration compressor unit is stopped and the energy storage system relies on stored cold energy for heat dissipation; otherwise, it is false. This fully silent operation status indicator is used to provide feedback on the current noise control status to the energy storage system operation and maintenance personnel and the park management system.

[0272] Specifically, traditional energy storage systems typically rely on air cooling or direct cooling for heat dissipation. Air cooling requires high-power fans, which often generate noise levels exceeding 65 decibels and account for over 40% of the system's auxiliary energy consumption. Direct cooling requires continuous compressor operation, also producing noise and energy consumption. In noise-sensitive applications such as parks, buildings, and data centers, traditional cooling methods struggle to simultaneously meet temperature control and quiet operation requirements. S4 utilizes an arbitrage strategy to obtain a pre-planned set of control commands through interval lookups in a jump table. During noise-sensitive periods, it prioritizes utilizing the stored cooling capacity in the intelligent refrigerant tank, achieving silent heat dissipation through refrigerant circulation and avoiding the activation of the refrigeration compressor or fans during these times. The independent heat exchanger design at the module level allows each energy storage module to independently receive and regulate the cooling supply. Refrigerant flow regulation is achieved through electronic expansion valves rather than mechanical regulating valves, ensuring quiet operation during the regulation process. The collaborative operation of the intelligent refrigerant box and the independent heat exchanger at the plug-in level achieves thermoelectric decoupling. The energy required for heat dissipation is stored as cooling during off-peak hours and released for use during peak hours. This reduces operating energy costs and ensures that the battery operates within its optimal operating temperature range of 25°C to 33°C. Maintaining a stable battery operating temperature effectively extends battery cycle life, slows down the rate of battery capacity decay, and reduces the total lifecycle cost of the energy storage system. The time-axis organization of the arbitrage strategy skip table ensures the continuity and predictability of control command execution, avoiding the response lag and energy consumption fluctuations that may result from discrete control strategies.

[0273] For example, the workflow of the above method is illustrated using a park energy storage power station as an application scenario. This park energy storage power station comprises several battery clusters deployed near park buildings, with strict requirements on nighttime operating noise. The system first collects voltage, current, and temperature data for each battery cluster, as well as ambient temperature data, through a sensor network. Simultaneously, it obtains real-time node electricity prices and future time-of-use electricity price predictions from an electricity trading platform. After time alignment and thermoelectric feature derivation processing, the collected data is input into a temporal convolutional network to generate a thermoelectric prediction tensor containing predicted heat load and electricity cost values ​​for each future time step. The system extracts the spatiotemporal distribution matrix of heat load and the electricity cost prediction vector from the thermoelectric prediction tensor, calculates the predicted temperature rise rate sequence for each battery cluster using a battery thermal inertia model, and then calculates a cooling urgency weight that comprehensively reflects temperature deviation, temperature rise rate, and electricity price sensitivity coefficient, constructing a heat demand Fibonacci stack. The system continuously extracts the highest priority cooling task from the heat demand Fibonacci stack, generates cooling control instructions based on the current electricity price range and cooling capacity storage status, and inserts arbitrage strategy skip lists. When the control cycle arrives, the system retrieves the control instruction set for the current time window from the arbitrage strategy jump table through interval queries, parses and generates a device-level control signal scheduling table, and drives the intelligent refrigerant box and each independent heat exchanger at the sub-box level to perform refined temperature control operations. During off-peak electricity hours, the system starts the refrigeration compressor to store cold energy in the phase change cold storage material of the intelligent refrigerant box. During peak electricity hours or nighttime noise-sensitive periods, the system shuts down the refrigeration compressor, releases the stored cold energy, and adjusts the refrigerant circulation to achieve silent battery cooling, maintaining the battery temperature stably within the optimal operating temperature range. This maximizes peak-valley electricity price arbitrage profits while meeting the park's quiet operation requirements.

[0274] Example 2

[0275] This embodiment, based on Embodiment 1, provides a collaborative control system for energy storage cooling capacity and peak-valley arbitrage, such as... Figure 6 As shown, it includes:

[0276] Thermoelectric prediction tensor generation module is used to collect multi-dimensional operation data of energy storage system and real-time transaction data of electricity market, and use temporal convolutional network to perform feature extraction and fusion analysis to generate thermoelectric prediction tensor.

[0277] The thermal demand Fibonacci stack construction module is used to analyze the spatiotemporal distribution characteristics of heat load in the thermoelectric prediction tensor, calculate the cooling urgency weight of each battery cluster, and construct the thermal demand Fibonacci stack.

[0278] The arbitrage strategy skip list generation module is used to perform minimum extraction operations on the heat demand Fibonacci stack, extract the optimal cooling task, combine peak and valley electricity price constraints to match cooling supply and demand, and generate arbitrage strategy skip lists.

[0279] The refined temperature control execution module is used to traverse the arbitrage strategy skip table to obtain the control instruction set of the current time window, and drive the independent heat exchanger and intelligent refrigerant box to perform refined temperature control operations.

[0280] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the claims.

Claims

1. A method for coordinated control of energy storage cooling capacity and peak-valley arbitrage, characterized in that, The method includes: Collect multi-dimensional operational data of energy storage systems and real-time transaction data of the electricity market, and use temporal convolutional networks for feature extraction and fusion analysis to generate thermoelectric prediction tensors; We analyze the spatiotemporal distribution characteristics of heat load in the thermoelectric prediction tensor, calculate the cooling urgency weight of each battery cluster, and construct a heat demand Fibonacci stack. The minimum extraction operation is performed on the heat demand Fibonacci stack to extract the optimal cooling task. Combined with peak and valley electricity price constraints, the supply and demand of cooling capacity are matched to generate an arbitrage strategy skip list. The arbitrage strategy skip list is traversed to obtain the control instruction set for the current time window, driving the independent heat exchanger and intelligent refrigerant box to perform refined temperature control operations.

2. The energy storage cooling capacity-peak arbitrage coordinated control method according to claim 1, characterized in that, The steps for generating the thermoelectric prediction tensor include: Collect multi-dimensional operational data of energy storage systems and construct an energy storage operation dataset; Collect real-time electricity market transaction data and construct an electricity price data set; Based on the energy storage operation data set and the electricity price data set, a unified time axis multi-source sample matrix is ​​obtained through the time alignment method; Based on a unified time-axis multi-source sample matrix, a thermoelectric feature derivation matrix is ​​obtained through thermoelectric feature derivation processing. Based on the thermoelectric feature-derived matrix, feature extraction and fusion analysis are performed through a temporal convolutional network to obtain the thermoelectric prediction tensor.

3. The energy storage cooling capacity-peak arbitrage coordinated control method according to claim 2, characterized in that, The time alignment method includes the following steps: A unified sampling time axis is determined based on the running timestamp collection sequence and the electricity price timestamp sequence; Based on each acquisition sequence in the energy storage operation data set, the acquisition sequence is linearly interpolated according to a unified sampling time axis to obtain an operation data aligned sequence mapped to the unified sampling time axis. Based on each sequence in the electricity price dataset, the sequences are subjected to a step hold operation along a unified sampling time axis to obtain an aligned sequence of electricity price data mapped onto the unified sampling time axis; The alignment sequences of each operational data and the alignment sequences of electricity price data are horizontally spliced ​​together at the corresponding positions on a unified sampling time axis to generate a unified time axis multi-source sample matrix with time as the row and different physical quantities and electricity prices as the column.

4. The energy storage cooling capacity-peak arbitrage coordinated control method according to claim 2, characterized in that, The process of thermoelectric feature derivation includes: Based on the current and voltage columns in the unified time axis multi-source sample matrix, the instantaneous power value is obtained by multiplying the current and voltage values ​​at each time point. The instantaneous power values ​​at all time points are arranged in chronological order to form an instantaneous power value sequence, which is then used as an instantaneous power derived column. Based on the battery temperature column and ambient temperature column in the unified time axis multi-source sample matrix, the temperature difference value is obtained by calculating the difference between the battery temperature value and the ambient temperature value at each time point. The temperature difference values ​​at all time points are arranged in chronological order to form a temperature difference value sequence, and the temperature difference value sequence is used as a temperature difference derived column. Based on the charge / discharge rate column and battery temperature column in the unified time axis multi-source sample matrix, the sum of the product of the square of the charge / discharge rate and the battery internal resistance coefficient, and the product of the difference between the battery temperature and the ambient temperature and the heat dissipation coefficient, the estimated heat generation rate at each time point is calculated. The estimated heat generation rate at all time points is arranged in chronological order to form a heat generation rate estimation value sequence, and the heat generation rate estimation value sequence is used as the heat generation rate derived column.

5. The energy storage cooling capacity-peak arbitrage coordinated control method according to claim 4, characterized in that, Also includes: Based on the charge / discharge rate column and battery temperature column in the unified time axis multi-source sample matrix, and based on the sum of the product of the square of the charge / discharge rate and the battery internal resistance coefficient, and the product of the difference between the battery temperature and the ambient temperature and the heat dissipation coefficient, the heat generation rate estimate at each time point is calculated. The heat generation rate estimates at all time points are arranged in chronological order to form a heat generation rate estimate sequence, and the heat generation rate estimate sequence is used as a heat generation rate derived column. The instantaneous power derivative column, temperature difference derivative column, heat generation rate derivative column, and electricity price deviation derivative column are horizontally concatenated with the original columns of the unified time axis multi-source sample matrix to generate a thermoelectric feature derivative matrix.

6. The energy storage cooling capacity-peak arbitrage coordinated control method according to claim 2, characterized in that, The process of feature extraction and fusion analysis using temporal convolutional networks includes: The thermoelectric feature derivation matrix is ​​sliced ​​according to a preset sliding window length to generate multiple thermoelectric feature input segments; Multiple thermoelectric feature input segments are input into the causal convolutional layer of a temporal convolutional network for temporal feature extraction, resulting in a temporal local feature tensor. The temporal local feature tensor is input into the residual connection module of the temporal convolutional network for feature fusion to obtain the temporal fused feature tensor; The temporal fusion feature tensor is input into the output mapping layer of the temporal convolutional network to generate the thermoelectric prediction tensor.

7. The energy storage cooling capacity-peak arbitrage coordinated control method according to claim 1, characterized in that, The process of constructing a thermal demand Fibonacci stack includes: Based on the thermoelectric prediction tensor, a spatiotemporal distribution matrix of heat load is constructed; Based on the spatiotemporal distribution matrix of heat load and the battery thermal inertia model, the predicted sequence of temperature rise rate for each battery cluster is calculated. Based on the temperature rise rate prediction sequence, the electricity price cost prediction vector, and the current state of the battery, the cooling urgency weight of each battery cluster is calculated. Based on the cooling urgency weight of each battery cluster, a thermal demand Fibonacci stack is constructed. Define a key-value reduction operation for thermal demand Fibonacci stacks to adjust priorities in response to sudden temperature changes in the battery cluster or when forecast data is updated.

8. The energy storage cooling capacity-peak arbitrage coordinated control method according to claim 7, characterized in that, The calculation process for the cooling urgency weight includes: Obtain the current temperature deviation of each battery cluster; Based on the predicted temperature rise rate sequence, the peak temperature rise rate of each battery cluster is calculated; Based on the electricity price cost prediction vector, calculate the electricity price cost sensitivity coefficient of each battery cluster; Based on the current temperature deviation, peak temperature rise rate, and electricity cost sensitivity coefficient, the cooling urgency weight of each battery cluster is calculated.

9. The energy storage cooling capacity-peak arbitrage coordinated control method according to claim 1, characterized in that, The process of generating a skip list for arbitrage strategies includes: Define the minimum extraction operation for the thermal demand Fibonacci heap, which is used to obtain and remove the thermal demand node with the highest cooling priority; Obtain the current electricity price range identifier and cooling capacity storage status; Extract the optimal cooling task, and generate cooling control instructions by combining the electricity price range identifier and the cooling capacity storage status; Initialize the arbitrage strategy skip list and define the skip list operations; The process involves repeatedly extracting the minimum operation and generating the instruction to create an arbitrage strategy skip list.

10. The energy storage cooling capacity-peak arbitrage coordinated control method according to claim 1, characterized in that, The process for performing refined temperature control includes: Determine the current time window and perform an interval query to obtain the control instruction set; Parse the current time window control instruction set and generate device-level control signals; Drive the independent heat exchanger at the plug-in stage to perform directional cold energy transfer; Drive the intelligent refrigerant box to perform cold energy storage and release control; Perform battery operating temperature maintenance control; Perform fully silent operation status monitoring.

11. An energy storage cooling capacity-peak arbitrage coordinated control system, used to implement the energy storage cooling capacity-peak arbitrage coordinated control method according to any one of claims 1-10, characterized in that, The system includes: Thermoelectric prediction tensor generation module is used to collect multi-dimensional operation data of energy storage system and real-time transaction data of electricity market, and use temporal convolutional network to perform feature extraction and fusion analysis to generate thermoelectric prediction tensor. The thermal demand Fibonacci stack construction module is used to analyze the spatiotemporal distribution characteristics of heat load in the thermoelectric prediction tensor, calculate the cooling urgency weight of each battery cluster, and construct the thermal demand Fibonacci stack. The arbitrage strategy skip list generation module is used to perform minimum extraction operations on the heat demand Fibonacci stack, extract the optimal cooling task, combine peak and valley electricity price constraints to match cooling supply and demand, and generate arbitrage strategy skip lists. The refined temperature control execution module is used to traverse the arbitrage strategy skip table to obtain the control instruction set of the current time window, and drive the independent heat exchanger and intelligent refrigerant box to perform refined temperature control operations.