A Dynamic Heat Dissipation Control Method and System for a Green Clean Energy Intelligent Energy Storage Cabinet

Abstract

The present invention discloses a dynamic heat dissipation control method and system for a green clean energy intelligent energy storage cabinet, which relates to the technical field of temperature control of energy storage cabinets. Time periods are divided and nodes are set at equal intervals for real-time collection of actual data. An information-driven joint prediction network is used to consider the randomness and intermittency of wind power generation and photovoltaic power generation, and node wind power prediction and node photovoltaic power prediction are carried out. A time-frequency conversion trend enhancement network is used to enhance the temporal trend of the load power for node load prediction. A multi-modal integrated learning network is used to consider the correlation between the substation power and the temperature of the energy storage cabinet, and accurate node temperature prediction is performed. Whether to use the PID algorithm to confirm the predicted cooling power for temperature reduction regulation is decided through node state judgment, and the constraint supply-demand balance equation corresponding to the state is selected to predict the substation power, improving the utilization rate of renewable energy. Using the predicted cooling power for temperature reduction regulation also effectively reduces energy consumption, ensuring the stability, safety and environmental protection of the energy storage cabinet.

Description

Technical Field

[0001] The present invention relates to the technical field of temperature control for energy storage cabinets, and particularly to a dynamic heat dissipation control method and system for a green clean energy intelligent energy storage cabinet. Background Art

[0002] In the context of the rapid development of green clean energy, energy storage cabinets, as key energy management devices, have become the core for the efficient utilization of renewable energy such as wind energy and solar energy. However, during the operation of these energy storage cabinets, a large amount of heat is generated due to battery charging and discharging and equipment loads. If the heat dissipation is not timely or the control is improper, it will not only reduce the equipment efficiency and lifespan, but may also pose safety risks.

[0003] The existing invention patent with the authorized announcement number CN118363413B proposes a dual-prevention energy storage cabinet temperature control method and system, including: real-time monitoring of the temperature inside the energy storage cabinet and the battery module through multi-point temperature sensors and a wireless sensor network, and processing the temperature data using a data fusion algorithm. The system combines an improved fuzzy control algorithm and a neural network prediction method to predict the temperature and determine whether it reaches a preset threshold, and then generates a control decision to drive the temperature regulation equipment to work, improving the safety of the energy storage cabinet and the accuracy of temperature monitoring.

[0004] However, the existing technology has the following drawbacks:

[0005] 1. Ignoring the intermittency and randomness of renewable energy, it fails to effectively control the supply-demand balance inside the energy storage cabinet while achieving temperature control, easily causing energy waste or energy shortage problems, and it is difficult to meet the sustainable operation requirements of the energy storage cabinet.

[0006] 2. When the existing technology uses advanced algorithms or neural networks for temperature prediction, it ignores that the temperature change of the energy storage cabinet actually depends on the charge-discharge power. Only considering the change trend of temperature data at known moments cannot effectively represent the change of temperature data at future moments, resulting in a decrease in the accuracy of temperature prediction for future moments.

[0007] Therefore, there is an urgent need for an efficient and intelligent dynamic heat dissipation control method to cope with the complex changes of renewable energy and the weak trend of temperature data in time series. Summary of the Invention

[0008] Aiming at the deficiencies of the existing technology, the present invention proposes a dynamic heat dissipation control method and system for a green clean energy intelligent energy storage cabinet to perform energy-saving, environmental-friendly, accurate and efficient temperature control of the energy storage cabinet, ensuring the stability and safety of the energy storage cabinet.

[0009] The technical solution for achieving the object of the present invention is as follows:

[0010] A dynamic heat dissipation control method for a green clean energy intelligent energy storage cabinet, comprising the following specific steps:

[0011] Set the temperature threshold and the optimal temperature , define time periods and set at node intervals within the time periods nodes, being the total number of time period nodes, and collect actual data in real time at each node;

[0012] Define known time periods and unknown time periods , when the system time advances to node , initiate the regulation of unknown time period nodes;

[0013] Use an information-driven joint prediction network to perform node wind power prediction and node photovoltaic power prediction, and use a time-frequency conversion trend enhancement network to perform node load prediction;

[0014] Use a multi-modal integrated learning network to comprehensively consider the substation power and the temperature of the energy storage cabinet, perform node temperature prediction and node status judgment, and make corresponding node preprocessing based on the judgment result;

[0015] Sequentially execute the regulation of unknown time period nodes until all nodes in the unknown time period have been regulated and then stop. When the system time advances to node , set new known time periods and new unknown time periods to continue initiating the regulation of new unknown time period nodes.

[0016] Furthermore, using an information-driven joint prediction network to perform node wind power prediction and node photovoltaic power prediction includes the following specific steps:

[0017] Respectively select the power and environmental vectors of the nodes before to construct a power sequence and an environmental sequence ;

[0018] Linearly combine the power sequence and the environmental sequence , and process through the SELU function to adjust the activation feature distribution to obtain an activation feature sequence ;

[0019] Use the GLU layer to further process the activation feature sequence to dynamically adjust the contribution degree of each activation feature vector, generating a contribution sequence ;

[0020] Superimposed contribution sequence And the power sequence And input it into the batch normalization layer to condense the logic, intermittency, and randomness of wind power generation or photovoltaic power generation in time series, and obtain a dynamic data sequence ;

[0021] Pass the environmental sequence Through the LSTM layer to capture the time series relationship in the environmental sequence And generate the predicted environmental vector of the node Through dimensionality reduction by the linear layer ;

[0022] Adopt the cross-attention mechanism to obtain the query matrix , predicted key matrix And predicted value matrix , and adjust the predicted value matrix Based on the correlation between the query matrix And the predicted key matrix To capture the time series evolution relationship, and further generate the predicted power of the node Through dimensionality reduction by the linear layer .

[0023] Furthermore, the node load prediction of the node Adopting the time-frequency conversion trend enhancement network includes the following specific steps:

[0024] Select the load power of the previous Nodes to construct a load power sequence ;

[0025] Pass the load power sequence Through the MLP layer and Fourier transform Expand the dimension and map it to the complex domain to obtain the frequency-domain load feature matrix ;

[0026] Multiply the frequency-domain load feature matrix With the complex matrix To emphasize the key low-frequency component information and suppress the high-frequency component information, and generate the frequency-domain core feature vector ;

[0027] Through the inverse Fourier transform Convert the frequency-domain core feature vector Back to the time domain to obtain an enhanced load power sequence beneficial to node load prediction ;

[0028] Adopt the LSTM layer to capture the enhanced load power sequence ​ Trending and periodic in the time domain, and generating the predicted load power of the th node .

[0029] Furthermore, the node temperature prediction using a multi-modal integrated learning network includes the following specific steps: Select the absolute values of the energy storage cabinet temperature and the substation power of the

[0030] nodes before this node to construct a temperature sequence and an absolute substation power sequence ;

[0031] Calculate the average temperature of the temperature sequence and the absolute substation average respectively with the absolute substation power sequence ;

[0032] Obtain the Pearson correlation coefficient characterizing the influence correlation between the predicted energy storage cabinet temperature of the node and the absolute value of the predicted substation power ;

[0033] Normalize the Pearson correlation coefficient by linear translation to obtain the embedding coefficient ;

[0034] Based on the embedding coefficient embed the absolute value of the predicted substation power into the predicted energy storage cabinet temperature to generate the combined temperature of the node ;

[0035] Synchronously input the combined temperature into three base learners to learn the time series trend, and use a meta-learner to inherit the outputs of the three base learners to generate the predicted energy storage cabinet temperature of the node .

[0036] Furthermore, based on the judgment result of the node state judgment of the node , perform the corresponding node preprocessing including:

[0037] If the predicted energy storage cabinet temperature is less than or equal to the temperature threshold , obtain the node based on the normal state constraint supply-demand balance equation Predicted power conversion and node Predicted stored electricity ;

[0038] If the predicted temperature of the energy storage cabinet is greater than the temperature threshold , use the PID algorithm to obtain the predicted refrigeration power of node , and obtain the predicted power conversion of node based on the supply-demand balance equation under the dangerous state constraint , and the predicted stored electricity of node and node , and update the predicted temperature of the energy storage cabinet to the optimal temperature . .

[0039] Furthermore, the predicted power conversion of node and the predicted stored electricity of node and node obtained based on the supply-demand balance equation under the normal state constraint include the following specific steps: including the following specific steps:

[0040] Take the predicted wind power of node and the predicted photovoltaic power as the predicted supply power, and take the predicted load power of node as the normal predicted consumption power; as the normal predicted consumption power;

[0041] Based on the theoretical normal supply-demand identity, obtain the theoretical power conversion of node , and the theoretical normal supply-demand identity is the supply-demand balance that the energy storage cabinet needs to follow theoretically at any node. The specific formula is as follows: , the theoretical normal supply-demand identity is the supply-demand balance that the energy storage cabinet needs to follow theoretically at any node. The specific formula is as follows:

[0042] ,

[0043] where the theoretical power conversion represents the theoretical charging power or theoretical discharging power of the energy storage cabinet at node ;

[0044] Obtain the predicted stored electricity of node , calculate the theoretical stored electricity of node based on the theoretical power conversion , and compare it with the storage upper limit and the storage lower limit ; and the storage lower limit ;

[0045] If the theoretical stored power is greater than or equal to the lower limit of storage and at the same time less than or equal to the upper limit of storage , the theoretical power conversion and the theoretical stored power are the predicted power conversion and the predicted stored power of node ;

[0046] If the theoretical stored power is less than the lower limit of storage , the predicted power conversion is set to , is the node interval, the predicted stored power , and the predicted power purchase is;

[0047] If the theoretical stored power is greater than the upper limit of storage , the predicted power conversion is set to , is the node interval, the predicted stored power , and the predicted power sale is;

[0048] The theoretical normal supply-demand identity is rewritten as a normal-constrained supply-demand balance equation, and the specific formula is as follows:

[0049] ,

[0050] ,

[0051] where is the predicted transaction power of node .

[0052] Furthermore, when the energy storage cabinet transfers to the dangerous state at node , the predicted cooling power is obtained by using the PID algorithm, including the following specific steps:

[0053] Obtain the optimal temperature and the predicted temperature of the energy storage cabinet at node , and calculate the adjustment deviation ;

[0054] Calculate the proportional term to represent the macroscopic adjustment of the energy storage cabinet temperature;

[0055] Calculate the integral term To represent the micro-adjustment of the energy storage cabinet temperature, so that the energy storage cabinet temperature further approaches the optimal temperature on the basis of macro-adjustment ;

[0056] Calculate the differential term , and control the micro-adjustment amplitude through the gradient to prevent over-adjustment;

[0057] Obtain the node 's refrigeration adjustment power , and superimpose it with the minimum operating power of the refrigeration equipment to obtain the predicted refrigeration power of the node . .

[0058] Furthermore, the difference between the dangerous state constrained supply-demand balance equation and the normal state constrained supply-demand balance equation is that the power consumption in the dangerous state is the sum of the predicted load power of the node and the predicted refrigeration power . The specific formula of the dangerous state constrained supply-demand balance equation is as follows:

[0059] ,

[0060] ,

[0061] wherein, is the predicted transaction power of the node .

[0062] A dynamic heat dissipation control system for a green clean energy intelligent energy storage cabinet, including a clock module, a prediction and regulation module, an energy management module, and a temperature control module;

[0063] The clock module sets known time periods and unknown time periods based on the system time, and sets nodes for the known time period and the unknown time period respectively, is the total number of time period nodes; ;

[0064] The prediction and regulation module stores actual data, executes unknown time period node regulation at the last node of the known time period, generates and sends a power conversion instruction to the energy management module, and at the same time decides whether to send a refrigeration start instruction to the temperature control module;

[0065] The energy management module includes a data docking unit and an instruction execution unit. The data docking unit collects actual data in real time at each node and reports it to the prediction and regulation module. The instruction execution unit receives the power conversion instruction and controls charging and discharging in real time according to the power conversion instruction;

[0066] The temperature control module receives the refrigeration start instruction, uses the PID algorithm to obtain the predicted refrigeration power, and feeds it back to the prediction and regulation module, and cools down at the corresponding node with the predicted refrigeration power.

[0067] Furthermore, the prediction and regulation module includes a storage unit, a wind power prediction unit, a photovoltaic power prediction unit, a load prediction unit, a temperature prediction unit, and an instruction generation unit;

[0068] The storage unit receives and stores the actual data reported by the energy management module;

[0069] The wind power prediction unit uses an information-driven joint prediction network to generate the predicted wind power environment vector and the predicted wind power of the node; ;

[0070] The photovoltaic power prediction unit uses an information-driven joint prediction network to generate the predicted photovoltaic power environment vector and the predicted photovoltaic power of the node in the unknown time period; ;

[0071] The load prediction unit uses a time-frequency conversion trend enhancement network to generate the predicted load power of the node in the unknown time period; ;

[0072] The temperature prediction unit obtains the predicted substation power and the temperature update label , and based on the temperature update label decides whether to update the predicted temperature of the energy storage cabinet , and uses a multi-modal integrated learning network to generate the predicted temperature of the energy storage cabinet of the node based on the predicted temperature of the energy storage cabinet ; ;

[0073] The instruction generation unit performs a transmission operation based on a state-driven response strategy.

[0074] Furthermore, the instruction generation unit performing a transmission operation based on a state-driven response strategy includes the following specific steps:

[0075] Obtain the predicted wind power , the predicted photovoltaic power , the predicted load power and the predicted temperature of the energy storage cabinet of the node; ;

[0076] Execute node state judgment by comparing the predicted temperature of the energy storage cabinet and the temperature threshold ;

[0077] If the temperature of the energy storage cabinet is predicted is less than or equal to the temperature threshold , the predicted power conversion is obtained by using the normal state constrained supply-demand balance equation , and the predicted power conversion is sequentially placed into the power conversion command, and the temperature update label is set to 0, and the predicted power conversion and the temperature update label are fed back to the temperature prediction unit;

[0078] If the temperature of the energy storage cabinet is predicted is greater than the temperature threshold , the nodes and the predicted temperature of the energy storage cabinet are packaged to generate a refrigeration start command and sent to the temperature control module;

[0079] The predicted refrigeration power of the node is obtained , and the predicted power conversion is obtained by using the dangerous state constrained supply-demand balance equation , and the predicted power conversion is sequentially placed into the power conversion command, and the temperature update label is set to 1, and the predicted power conversion and the temperature update label are fed back to the temperature prediction unit;

[0080] Until the predicted power conversions of the nodes in the unknown time period have all been placed into the power conversion command and sent to the energy management module.

[0081] Compared with the prior art, the remarkable advantages of the present invention are as follows:

[0082] 1. Considering the intermittency and randomness of wind power generation and photovoltaic power generation, an information-driven joint prediction network is specifically designed. By capturing the time series correlation of wind power generation and photovoltaic power generation and combining the influence of power generation environment factors, the wind power generation power and photovoltaic power generation power are accurately predicted to facilitate the energy regulation of the energy storage cabinet;

[0083] 2. Considering the problem that the change trend of the load power is not obvious in the time series, a time-frequency conversion trend enhancement network is designed. The load power sequence is converted from the time domain to the frequency domain, and the product operation is used to filter the noise influence, and then switched back to the time domain to enhance the time series trend, assisting the LSTM to achieve accurate complex power prediction to facilitate the energy regulation of the energy storage cabinet;

[0084] 3. Consider the impact of the charge and discharge behavior of the energy storage cabinet on its temperature, and specifically design a multi-modal integrated learning network to accurately predict the temperature of the energy storage cabinet jointly from multiple perspectives. At the same time, use node state judgment to achieve temperature early warning, and decide whether to accurately cool the energy storage cabinet to the optimal temperature at the corresponding node with the predicted cooling power obtained by the PID algorithm, so as to realize effective temperature control adjustment of the energy storage cabinet with the minimum energy consumption and ensure the safety of the operation of the energy storage cabinet;

[0085] 4. Design the supply-demand balance equations with normal and dangerous state constraints, comprehensively consider the supply power and consumption power of the energy storage cabinet under different node states, the pre-planned charge and discharge behavior of the energy storage cabinet and the trading behavior with the main power grid, ensure the energy balance of the operation of the energy storage cabinet, avoid waste of renewable energy, and improve the environmental protection of the energy storage cabinet. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 It is a flowchart of a dynamic heat dissipation control method for a green clean energy intelligent energy storage cabinet;

[0087] Figure 2 It is a diagram of the information-driven joint prediction network model in the present invention;

[0088] Figure 3 It is a diagram of the time-frequency conversion trend enhancement network model in the present invention;

[0089] Figure 4 It is a diagram of the multi-modal integrated learning network model in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0090] The present invention will be further described in detail below with reference to the drawings and embodiments.

[0091] Embodiment 1

[0092] As Figure 1 shown, a specific embodiment of the present invention discloses a dynamic heat dissipation control method for a green clean energy intelligent energy storage cabinet, including the following specific steps:

[0093] Set the temperature threshold and the optimal temperature , define a time period to represent a single day, and set nodes at node intervals in the time period, being the total number of time period nodes. At each node, real-time data is collected through sensors and data interfaces. Among them, the temperature threshold is the highest temperature acceptable when the energy storage cabinet is in a normal state. When the temperature of the energy storage cabinet is greater than the temperature threshold , the energy storage cabinet enters a dangerous state, the operating efficiency decreases, and safety accidents may be caused by overheating. The optimal temperature is is the most suitable temperature for the energy storage cabinet to operate and is also the target temperature for the refrigeration equipment to cool down when the energy storage cabinet enters the dangerous state;

[0094] Define the day when the system time is located as the known period , and the next day of the day when the system time is located is the unknown period , where and respectively represent the th nodes of the known period and the unknown period, ;

[0095] When the system time advances to the last node of the known period , start the node regulation of the unknown period. The node regulation of the unknown period refers to performing node wind power prediction, node photovoltaic power prediction, node load prediction, node state judgment, and node preprocessing on a single node of the unknown period;

[0096] Use the information-driven joint prediction network to capture the non-linear dependence correlation of wind power and photovoltaic power in time series, consider the characteristics of power generation intermittency and randomness caused by the wind power environment vector and the photovoltaic power environment vector, and perform node wind power prediction and node photovoltaic power prediction;

[0097] Use the time-frequency conversion trend enhancement network to extract the trend change and periodic change of the load power in time series from the frequency domain perspective, and perform node load prediction;

[0098] Use the multi-modal integrated learning network to comprehensively consider the influence of the substation power on the temperature of the energy storage cabinet, perform node temperature prediction, compare with the temperature threshold to perform node state judgment, and make corresponding node preprocessing based on the judgment result;

[0099] Based on the node time sequence, sequentially execute the node regulation of the unknown period until all nodes of the unknown period are all regulated and then stop;

[0100] When the system time advances to the first node of the unknown period , convert the unknown period into a new known period, set the new unknown period as , wait until the last node of the new known period and then continue to start the node regulation of the new unknown period.

[0101] Further, the data is divided into actual data and predicted data. The actual data is obtained through real-time acquisition, and the predicted data is obtained through network or equation prediction. The data includes the temperature of the energy storage cabinet, the power conversion, the stored electricity, the load power, the wind power, the photovoltaic power, the wind power environment vector, and the photovoltaic power environment vector. Among them, the load power is the sum of the internal load power required by the self-powered equipment in the energy storage cabinet at a single node and the external load power required by the enterprise's electrical equipment at a single node. The wind power environment vector includes the wind speed, air pressure, and temperature of the environment where the wind turbine operates at a single node, and the photovoltaic power environment vector includes the sunlight intensity and temperature of the environment where the photovoltaic module operates at a single node.

[0102] As Figure 2 shown, further, an information-driven joint prediction network is used for node The node wind power prediction and node photovoltaic power prediction of the node include the following specific steps:

[0103] Select the power of the nodes before a node to construct a power sequence . It should be noted that the nodes to are in the known time period, and the power is the actual power collected in real time. The nodes to are in the unknown time period, and the power is the predicted power generated by the information-driven joint prediction network. The power is the wind power or the photovoltaic power, depending on whether the information-driven joint prediction network is performing node wind power prediction or node photovoltaic power prediction;

[0104] Select the environment vectors of the nodes before a node to construct an environment sequence . It should be noted that the nodes to are in the known time period, and the environment vectors are the actual environment vectors collected in real time. The nodes to are in the unknown time period, and the environment vectors are the predicted environment vectors generated by the information-driven joint prediction network. The environment vector is the wind power environment vector or the photovoltaic power environment vector, depending on whether the information-driven joint prediction network is performing node wind power prediction or node photovoltaic power prediction;

[0105] Use the activation and stabilization layer to process the power sequence and the environment sequence . The activation and stabilization layer will process the power sequence and the environment sequence Perform a linear combination, adjust the activation feature distribution through the SELU function with a self-normalizing tree structure to automatically tend to a standard distribution with zero mean and unit variance, and obtain an activation feature sequence , the activation feature sequence of the node in the activation feature vector The specific calculation formula is as follows:

[0106] ,

[0107] where , and are the linear weight vectors corresponding to the power and the environmental vector respectively, both have the same dimension as the environmental vector , is the bias vector of the node , represents the Hadamard product;

[0108] Further process the activation feature sequence using the GLU layer to dynamically adjust the contribution degree of each activation feature vector in the activation feature sequence to avoid key information being overwhelmed by non-key information, and generate a contribution sequence , the contribution sequence of the node in the contribution The specific formula is as follows:

[0109] ,

[0110] where Sigmoid is the activation function, represents the transpose of the activation feature vector , and are learnable linear weight vectors, and are learnable bias values;

[0111] Stack the contribution sequence and the power sequence and input them into the batch normalization layer to obtain a dynamic data sequence , the dynamic data sequence of the node in the dynamic data , LayerNorm represents the batch normalization operation, and the dynamic data sequence condenses the logic, intermittency, and randomness of wind power generation or photovoltaic power generation in time series;

[0112] The environmental sequence Pass through the LSTM layer to capture the environmental sequence in the temporal relationship, and reduce the dimension through a linear layer to generate the node predicted environmental vector The predicted environmental vector is the predicted wind power environmental vector or the predicted photovoltaic environmental vector , depending on whether the information-driven joint prediction network is performing node wind power prediction or node photovoltaic prediction;

[0113] Adopt the cross-attention mechanism instead of the self-attention mechanism. The self-attention mechanism is to generate the query matrix , key matrix and value matrix respectively through three different linear modulations of the dynamic data sequence and key matrix to adjust the value matrix by calculating the correlation degree of the query matrix in order to retain the key information in the dynamic data sequence . The cross-attention mechanism is to generate the predicted key matrix and the predicted value matrix through two different linear modulations of the predicted environmental vector of the node . Calculate the correlation degree between the query matrix and the predicted key matrix to adjust the predicted value matrix by the correlation degree to capture the temporal evolution relationship between the dynamic data sequence and the predicted environmental vector . Further reduce the dimension through a linear layer to generate the predicted power of the node or the predicted photovoltaic power , depending on whether the information-driven joint prediction network is performing node wind power prediction or node photovoltaic prediction. The specific formula of the cross self-attention mechanism is as follows:

[0114] ,

[0115] where Softmax is the activation function, is the scaling factor of the cross-attention mechanism, equal to the number of rows of the predicted key matrix , and are the learnable linear weight matrix and bias value respectively.

[0116] Such asFigure 3 As shown, further, a time-frequency conversion trend enhancement network is adopted for node load prediction, which includes the following specific steps:

[0117] Select the load powers of the nodes before to construct a load power sequence . It should be noted that for nodes to in the known time period, the load powers are all the actual load powers collected in real time. For nodes to in the unknown time period, the load powers are all the predicted load powers generated by the time-frequency conversion trend enhancement network;

[0118] Input the load power sequence into the MLP layer for dimension expansion to generate a time-domain load feature matrix , and then use the Fourier transform to map the time-domain load feature matrix to the complex domain to obtain a frequency-domain load feature matrix , where and are the weight vector and bias matrix of the MLP layer respectively, and the dimension of the frequency-domain load feature matrix is ;

[0119] Multiply the frequency-domain load feature matrix with a complex matrix of dimension . The complex matrix can be regarded as a filter, which is used to emphasize the key low-frequency component information in the frequency-domain load feature matrix and suppress the high-frequency component information representing noise to generate a frequency-domain core feature vector . Specifically, , the specific calculation formula for the th frequency-domain core feature in the frequency-domain core feature vector

[0120] is as follows:

[0121] where is the frequency-domain load feature component in the th row and th column of the frequency-domain load feature matrix , and is the Column value of the layer;

[0122] Through the inverse Fourier transform Convert the frequency-domain core feature vector Back to the time domain to obtain the enhanced load power sequence , since the periodicity of the data is manifested as significant peaks at specific frequencies at the frequency domain end, and the trend features and noise are often located in the low-frequency components and high-frequency components respectively. Compared with the load power sequence , the enhanced load power sequence Not only takes into account the periodicity, but also filters out the noise influence through the product operation in the frequency domain, significantly enhancing the trend, which is more conducive to node load prediction;

[0123] Use the LSTM layer to capture the trend and periodicity of the enhanced load power sequence in the time domain, and generate the predicted load power of the th node through the linear layer for dimensionality reduction .

[0124] As Figure 4 shown, further, the node temperature prediction using the multi-modal integrated learning network includes the following specific steps: Select the energy storage cabinet temperature of the

[0125] previous nodes to construct the temperature sequence , it should be noted that the nodes from node to node are in the known period, and the energy storage cabinet temperature is the actual energy storage cabinet temperature collected in real time. The nodes from node to node

[0126] Select the substation power of the previous nodes and take the absolute value to construct the absolute substation power sequence , it should be noted that the nodes from node to node to node are in the unknown period, and the substation power is the predicted substation power calculated based on the predicted data;

[0127] Calculate the temperature sequence and the absolute substation power sequence Average temperature and absolute power change average value , where represents the averaging operation;

[0128] Obtain the node predicted temperature of the energy storage cabinet and the absolute value of the predicted power change Pearson correlation coefficient between them , to characterize the influence correlation of the absolute value of the predicted power change on the predicted temperature of the energy storage cabinet . The specific calculation formula of the Pearson correlation coefficient is as follows:

[0129] ,

[0130] where and respectively represent the temperature of the energy storage cabinet and the power change of the node . If the node is in the known time period, they are the actual temperature of the energy storage cabinet and the actual power change. If the node is in the unknown time period, they are the predicted temperature of the energy storage cabinet and the predicted power change. Since the value range of the Pearson correlation coefficient is in , the Pearson correlation coefficient is now processed by linear translation normalization to obtain the embedding coefficient ;

[0131] Based on the embedding coefficient , the absolute value of the predicted power change of the node is embedded into the predicted temperature of the energy storage cabinet to generate the combined temperature of the node ;

[0132] The combined temperature is synchronously input into three different base learners to learn the time series trend, and the first predicted temperature of the energy storage cabinet , the second predicted temperature of the energy storage cabinet and the third predicted temperature of the energy storage cabinet of the node are obtained. In this embodiment, the base learners are selected as LightGBM, BP neural network and XGBoost, which are widely used and have good prediction performance;

[0133] The first predicted temperature of the energy storage cabinet , the second predicted temperature of the energy storage cabinet and the third predicted temperature of the energy storage cabinet Further input the meta-learner, which inherits the time series trends learned by the three base learners and further conducts learning and prediction to generate nodes for predicting the temperature of the energy storage cabinet In this embodiment, since the base learners have learned most of the time series trends from different perspectives, the meta-learner selects the LR model with a fast convergence speed, high stability, and strong generalization ability.

[0134] Further, compare the predicted temperature of the energy storage cabinet at node with the temperature threshold and execute the node status judgment of node . Based on the judgment result, perform the corresponding node preprocessing, including:

[0135] If the predicted temperature of the energy storage cabinet is less than or equal to the temperature threshold , the energy storage cabinet is in a normal state at node . Execute the normal node preprocessing, perform node energy storage prediction based on the normal constraint supply-demand balance equation, and obtain the predicted substation power at node and the predicted stored electricity at node ;

[0136] If the predicted temperature of the energy storage cabinet is greater than the temperature threshold , the energy storage cabinet transfers to a dangerous state at node . Use the PID algorithm to obtain the predicted cooling power of the refrigeration equipment at node . When the system time advances to node , turn on the refrigeration equipment with the predicted cooling power to cool the energy storage cabinet to the optimal temperature . Perform node energy storage prediction based on the dangerous state constraint supply-demand balance equation, and obtain the predicted substation power at node and the predicted stored electricity at node , and update the predicted temperature of the energy storage cabinet to the optimal temperature . Furthermore, obtaining the predicted substation power

[0137] at node and the predicted stored electricity at node based on the normal constraint supply-demand balance equation includes the following specific steps:

[0138] Get Node Predicted wind power and predicted photovoltaic power , sum and calculate the energy storage cabinet at the node The predicted supply power of the node Forecasted load power and as a node The normal predicted power consumption;

[0139] Get nodes based on the theoretical normal supply and demand identity Theoretical transformer power , the theoretical normal supply and demand identity is the supply and demand balance that the energy storage cabinet should theoretically follow at any node, that is, at the node Theoretical transformer power It should be equal to the difference between the predicted supply power and the normal predicted consumption power. The specific formula of the theoretical normal supply and demand identity is as follows:

[0140] ,

[0141] Among them, the theoretical transformer power Indicates that the energy storage cabinet is at the node The theoretical charging power or theoretical discharging power of the transformer is It is the opposite of the theoretical discharge power, indicating that the energy storage cabinet needs to discharge at the theoretical discharge power to meet the power demand of the load equipment. When the predicted supply power is greater than or equal to the normal predicted consumption power, the theoretical transformer power Equal to the theoretical charging power, indicating that the predicted supply power still has a surplus after meeting the power demand of the load equipment, and the surplus is stored in the energy storage cabinet at the theoretical charging power;

[0142] Get Node Predicted power storage , based on the theoretical transformer power Compute Node Theoretical power storage , is the node interval, since the energy storage cabinet has a storage limit and savings floor , when the stored power is equal to the storage limit When the energy storage cabinet is fully charged, it cannot be charged any further. When the energy storage cabinet is depleted, it is considered that the energy storage cabinet is exhausted and cannot continue to discharge. With savings cap and savings floor To enforce energy storage constraint judgments;

[0143] If the theoretical stored electricity is greater than or equal to the storage lower limit and at the same time is less than or equal to the storage upper limit , it indicates that the theoretical power conversion is feasible. The theoretical power conversion and the theoretical stored electricity are the predicted power conversion and the predicted stored electricity of node ; ;

[0144] If the theoretical stored electricity is less than the storage lower limit , it indicates that the available electricity of the energy storage cabinet cannot support the energy storage cabinet to discharge at the absolute value of the theoretical power conversion . Then the predicted power conversion is set to , and the predicted stored electricity . The insufficient electricity needs to be purchased from the main power grid, and the predicted power purchase is ;

[0145] If the theoretical stored electricity is greater than the storage upper limit , it indicates that the rechargeable electricity of the storage cabinet cannot support the energy storage cabinet to charge at the theoretical power conversion . Then the predicted power conversion is set to , and the predicted stored electricity . The excess electricity can be sold to the main power grid, and the predicted power selling is ;

[0146] At this time, the theoretical normal supply-demand identity is rewritten as a normal constraint supply-demand balance equation, and the specific formula is as follows:

[0147] ,

[0148] ,

[0149] where is the predicted trading power of node . If the predicted trading power is greater than or equal to 0, the predicted trading power is equal to the predicted power purchase. If the predicted trading power is less than 0, the predicted trading power is equal to the opposite of the predicted power selling.

[0150] Furthermore, the energy storage cabinet is at node When transitioning to the dangerous state, the PID algorithm is used to obtain the predicted refrigeration power , including the following specific steps:

[0151] Obtain the optimal temperature and the predicted temperature of the energy storage cabinet at node , and calculate the adjustment deviation ;

[0152] Calculate the proportional term , and the proportional term represents the macroscopic adjustment of the energy storage cabinet temperature, which is used to approach the optimal temperature at one time through a single macroscopic adjustment , where is the preset proportional coefficient, and obtain the 0th adjustment deviation ;

[0153] Calculate the integral term , where is the total number of microscopic adjustments, is the th adjustment deviation after the th microscopic adjustment, is the preset integral proportional coefficient, and the integral term represents the microscopic adjustment of the energy storage cabinet temperature, and through times of microscopic adjustments, the energy storage cabinet temperature further approaches the optimal temperature on the basis of the macroscopic adjustment ;

[0154] Calculate the derivative term , where is the total number of microscopic adjustments, is the difference between the th adjustment deviation and the th adjustment deviation, which can be approximately replaced by the gradient of the adjustment deviation, is the preset derivative proportional coefficient, and the derivative term controls the amplitude of the microscopic adjustment of the PID algorithm through the gradient to prevent over-adjustment;

[0155] Obtain the refrigeration adjustment power at node , then the predicted refrigeration power at node , where is the minimum operating power when the refrigeration equipment starts.

[0156] Furthermore, the predicted power transformation at node and the predicted stored electricity at node are obtained by the dangerous state constraint supply-demand balance equation and the normal state constraint supply-demand balance equation ​​ Logically consistent. When the energy storage cabinet of a node enters the dangerous state, the predicted supply power of the node is no different from the normal state, which is the sum of the predicted wind power and the predicted photovoltaic power of the node . The difference is that the power consumption in the dangerous state needs to add the predicted refrigeration power of the refrigeration equipment on the basis of the predicted load power of the node , while the power consumption in the normal state is only the predicted load power . Therefore, the specific formula for the dangerous state to constrain the supply-demand balance equation is as follows:

[0157] ,

[0158] ,

[0159] wherein, is the predicted transaction power of the node .

[0160] Embodiment 2

[0161] The present invention also discloses a dynamic heat dissipation control system for a green clean energy intelligent energy storage cabinet, which is used to execute a dynamic heat dissipation control method for a green clean energy intelligent energy storage cabinet proposed in Embodiment 1, including a clock module, a prediction and regulation module, an energy management module, and a temperature control module;

[0162] The clock module records the system time, regards the day where the system time is located as the known period, regards the day after the day where the system time is located as the unknown period, and sets nodes for the known period and the unknown period respectively at node intervals , being the total number of period nodes;

[0163] The prediction and regulation module stores actual data, executes node regulation for the unknown period at the last node of the known period, generates and sends a substation power instruction for the unknown period to the energy management module, and at the same time decides whether to send a refrigeration start instruction to the temperature control module;

[0164] The energy management module includes a data docking unit and an instruction execution unit. The data docking unit collects actual data in real time through sensors or intelligent devices at each node of the known period and reports it to the prediction and regulation module. The instruction execution unit receives the substation power instruction and controls the charging and discharging of the energy storage cabinet in real time according to the substation power instruction when the system time reaches the corresponding node of the unknown period;

[0165] The temperature control module receives the refrigeration start instruction, uses the PID algorithm to obtain the predicted refrigeration power and feedbacks it to the prediction and regulation module. When the system time reaches the corresponding node in the refrigeration start instruction, the energy storage cabinet is cooled with the predicted refrigeration power.

[0166] Furthermore, the prediction and regulation module includes a storage unit, a wind power prediction unit, a photovoltaic power prediction unit, a load prediction unit, a temperature prediction unit, and an instruction generation unit;

[0167] The storage unit receives and stores the actual data reported by the energy management module;

[0168] The wind power prediction unit uses an information-driven joint prediction network to generate the predicted wind power environment vector and the predicted wind power at nodes in the unknown time period, and sends the predicted wind power to the instruction generation unit;

[0169] The photovoltaic power prediction unit uses an information-driven joint prediction network to generate the predicted photovoltaic power environment vector and the predicted photovoltaic power at nodes in the unknown time period, and sends the predicted wind power to the instruction generation unit;

[0170] The load prediction unit uses a time-frequency conversion trend enhancement network to generate the predicted load power at nodes in the unknown time period and sends it to the instruction generation unit;

[0171] The temperature prediction unit obtains the predicted substation power and the temperature update label . If the temperature update label is 1, then the predicted temperature of the energy storage cabinet at the node is updated to the optimal temperature . If the temperature update label is 0, then the predicted temperature of the energy storage cabinet is directly called, and the multi-modal integrated learning network is used to generate the predicted temperature of the energy storage cabinet at nodes in the unknown time period and sends it to the instruction generation unit;

[0172] The instruction generation unit performs transmission operations based on the state-driven response strategy. The transmission operations include sending the predicted substation power

[0172] at the node, sending the substation power instruction, and sending the refrigeration start instruction. ​

[0173] Furthermore, the instruction generation unit executes the transmission operation based on the state-driven response strategy, including the following specific steps:

[0174] Obtain the predicted wind power of the node during the unknown period , the predicted photovoltaic power , the predicted load power and the predicted temperature of the energy storage cabinet ;

[0175] Execute the node state judgment by comparing the predicted temperature of the energy storage cabinet with the temperature threshold ;

[0176] If the predicted temperature of the energy storage cabinet is less than or equal to the temperature threshold , then the energy storage cabinet is in a normal state at the node . Obtain the predicted substation power using the normal state constraint supply-demand balance equation , place the predicted substation power sequentially into the substation power instruction, set the temperature update label to 0, and feedback the predicted substation power and the temperature update label to the temperature prediction unit;

[0177] If the predicted temperature of the energy storage cabinet is greater than the temperature threshold , then the energy storage cabinet transfers to a dangerous state at the node . Package the node and the predicted temperature of the energy storage cabinet to generate a refrigeration start instruction and send it to the temperature control module;

[0178] Until obtaining the predicted refrigeration power of the node fed back by the temperature control module, obtain the predicted substation power using the dangerous state constraint supply-demand balance equation , place the predicted substation power sequentially into the substation power instruction, set the temperature update label to 1, and feedback the predicted substation power and the temperature update label to the temperature prediction unit;

[0179] Until the predicted substation powers of all nodes in the unknown period have been placed into the substation power instruction and sent to the energy management module.

[0180] The present invention discloses a dynamic heat dissipation control method and system for a green clean energy intelligent energy storage cabinet. By dividing time periods and equally spacing nodes to collect real-time actual data, an information-driven joint prediction network is adopted to consider the randomness and intermittency of wind power generation and photovoltaic power generation, and accurate node wind power prediction and node photovoltaic power prediction are carried out. A time-frequency conversion trend enhancement network is used to enhance the time series trend of the load power for accurate node load prediction. A multi-modal integrated learning network is used to consider the correlation between the substation power of the previous node and the temperature of the energy storage cabinet for accurate node temperature prediction. Whether to use the PID algorithm to confirm the predicted cooling power is determined through node state judgment to achieve cooling with a determined power for the determined node, and the constraint supply-demand balance equation corresponding to the state is selected to predict the substation power. The utilization rate of renewable energy is greatly improved, energy waste is avoided, and the energy consumption is effectively reduced by cooling regulation with a determined predicted cooling power at the node. On the premise of ensuring the stability and safety of the energy storage cabinet, the environmental friendliness is improved.

[0181] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A green and clean energy intelligent energy storage cabinet dynamic heat dissipation control method, characterized in that: The specific steps include: Set temperature thresholds and optimal temperatures, define time periods and set nodes, and collect actual data at the nodes; Define known time periods and unknown time periods. When the system time advances to the last node in the known time period, start the unknown time period node control. The information-driven joint prediction network is used to perform node wind power prediction and node photovoltaic prediction in unknown time periods, and the time-frequency conversion trend enhancement network is used to perform node load prediction in unknown time periods. A multimodal integrated learning network is used to perform node temperature prediction and node status judgment in unknown time periods, and corresponding node preprocessing is performed based on the judgment results. The process stops after all nodes in the unknown period have executed the unknown period node control. When the system time advances to the last node in the unknown period, the known period and the unknown period are updated to continue to start the unknown period node control. Among them, the information-driven joint prediction network performs node wind power prediction and node photovoltaic prediction, including the following specific steps: respectively select the power and environment vectors of several nodes before a single node to construct a power sequence and an environment sequence; adjust the distribution of the linear combination of the power sequence and the environment sequence through the SELU function to obtain an activation feature sequence; use the GLU layer to dynamically adjust the contribution degree of each activation feature vector in the activation feature sequence to generate a contribution sequence; superimpose the contribution sequence and the power sequence and input the batch normalization layer to obtain a dynamic data sequence; the environment sequence passes through the LSTM layer to capture the time series relationship, and generates a predicted environment vector of a single node through linear layer dimensionality reduction; use the cross attention mechanism to adjust the prediction value matrix based on the correlation between the query matrix and the prediction key matrix, and generate the predicted power of a single node through linear layer dimensionality reduction; The time-frequency conversion trend enhancement network for node load prediction includes the following specific steps: selecting the load power of several nodes before a single node to construct a load power sequence; expanding the load power sequence through an MLP layer and Fourier transform and mapping it to generate a frequency domain load feature matrix; performing a multiplication operation on the frequency domain load feature matrix and a complex matrix to suppress high-frequency component information and generate a frequency domain core feature vector; converting the frequency domain core feature vector to generate an enhanced load power sequence through an inverse Fourier transform, and using a combination of an LSTM layer and a linear layer to generate a predicted load power of a single node; The multimodal integrated learning network performs node temperature prediction, including the following specific steps: selecting the energy storage cabinet temperature and the absolute value of the transformer power of several nodes before a single node to respectively construct a temperature sequence and an absolute transformer power sequence; respectively calculating the temperature mean and the absolute transformer mean of the temperature sequence and the absolute transformer power sequence; obtaining the Pearson correlation coefficient of the previous node, and generating the embedding coefficient of the previous node through linear translation normalization processing, where the previous node refers to the node before the single node and adjacent to the single node; embedding the absolute value of the transformer power of the previous node into the energy storage cabinet temperature of the previous node based on the embedding coefficient of the previous node to generate the joint temperature of the previous node; synchronously inputting the joint temperature of the previous node into multiple base learners, and using a meta-learner to inherit the outputs of multiple base learners to generate the predicted energy storage cabinet temperature of a single node; The corresponding node preprocessing is performed based on the judgment result of the node state judgment, including: if the predicted energy storage cabinet temperature of a single node is less than or equal to a temperature threshold, the predicted transformer power of the single node and the predicted storage power of the next node are obtained based on the normal state constraint supply and demand balance equation, and the next node refers to a node after the single node and adjacent to the single node; if the predicted energy storage cabinet temperature of a single node is greater than the temperature threshold, the predicted cooling power of the single node is obtained by using a PID algorithm, the predicted transformer power of the single node and the predicted storage power of the next node are obtained based on the dangerous state constraint supply and demand balance equation, and the predicted energy storage cabinet temperature of the single node is updated to the optimal temperature.

2. A green clean energy intelligent energy storage cabinet dynamic heat dissipation control method as claimed in claim 1, characterized in that: The method of obtaining the predicted substation power of a single node and the predicted storage power of the next node based on the normal state constraint supply and demand balance equation includes the following specific steps: The sum of the predicted wind power and the predicted photovoltaic power of a single node is taken as the predicted supply power, and the predicted load power of a single node is taken as the normal predicted consumption power; The theoretical substation power of a single node is obtained based on the theoretical normal supply and demand identity. The theoretical normal supply and demand identity is the supply and demand balance that the energy storage cabinet should follow in theory at any node. Get the predicted power storage of a single node, calculate the theoretical power storage of the next node based on the theoretical transformer power, and compare it with the upper and lower limits of power storage; If the theoretical storage power of the next node is greater than or equal to the storage lower limit and less than or equal to the storage upper limit, the theoretical transformer power of the single node and the theoretical storage power of the next node are used as the predicted transformer power of the single node and the predicted storage power of the next node respectively; If the theoretical storage power of the next node is less than the storage lower limit or greater than the storage upper limit, reset the predicted transformer power of the single node and the predicted storage power of the next node to determine the predicted transaction power of the single node; The theoretical normal supply and demand identity is rewritten as the normal constrained supply and demand balance equation. The normal constrained supply and demand balance equation adds the predicted transaction power of a single node to the theoretical normal supply and demand identity to maintain the supply and demand balance.

3. A green clean energy intelligent energy storage cabinet dynamic heat dissipation control method as claimed in claim 1, characterized in that: The method of using the PID algorithm to obtain the predicted cooling power of a single node includes the following specific steps: Obtain the optimal temperature and the predicted energy storage cabinet temperature of a single node, and calculate the adjustment deviation of a single node; A proportional term is calculated for a single node to represent the macro-adjusted temperature; The integral term of a single node is calculated to approach the optimal temperature through multiple micro adjustments; Calculate the differential term of a single node to control the range of a single micro-adjustment and prevent over-adjustment; The cooling adjustment power of a single node is obtained based on the proportional term, integral term and differential term of a single node, and the predicted cooling power of a single node is obtained by superimposing it with the minimum operating power.

4. A green clean energy intelligent energy storage cabinet dynamic heat dissipation control system, used to execute a green clean energy intelligent energy storage cabinet dynamic heat dissipation control method as described in any one of claims 1-3, characterized in that: Including clock module, prediction and control module, energy management module and temperature control module; The clock module sets a known period and an unknown period based on the system time, and sets nodes for the known period and the unknown period respectively; The prediction and control module stores actual data, performs node control of unknown time period at the last node of known time period, generates and sends power conversion power instruction to energy management module, and decides whether to send cooling start instruction to temperature control module; The energy management module collects actual data at each node and reports it to the prediction and control module, and controls charging and discharging according to the received substation power instructions; The temperature control module receives a refrigeration start instruction, uses a PID algorithm to obtain a predicted refrigeration power and feeds it back to the prediction and control module, and performs temperature reduction at the corresponding node with the predicted refrigeration power.

5. A green clean energy intelligent energy storage cabinet dynamic heat dissipation control system as claimed in claim 4, characterized in that: The prediction and control module includes a wind power prediction unit, a photovoltaic prediction unit, a load prediction unit, a temperature prediction unit and an instruction generation unit; The wind power prediction unit uses an information-driven joint prediction network to generate a predicted wind power environment vector and predicted wind power of a single node; The photoelectric prediction unit uses an information-driven joint prediction network to generate a predicted photoelectric environment vector and predicted photoelectric power of a single node; The load prediction unit generates a predicted load power of a single node using a time-frequency conversion trend enhancement network; The temperature prediction unit obtains the predicted transformer power and temperature update tag of the previous node, decides whether to update the predicted energy storage cabinet temperature of the previous node based on the temperature update tag, and uses a multi-modal integrated learning network to generate a predicted energy storage cabinet temperature of a single node; The instruction generation unit performs the transfer operation based on the state-driven response strategy.

6. A green clean energy intelligent energy storage cabinet dynamic heat dissipation control system as claimed in claim 5, characterized in that: The instruction generation unit performs the transmission operation based on the state-driven response strategy, including the following specific steps: Obtain the predicted wind power, predicted photovoltaic power, predicted load power and predicted energy storage cabinet temperature of a single node; Node status judgment is performed by comparing the predicted energy storage cabinet temperature of a single node with the temperature threshold; If the predicted energy storage cabinet temperature of a single node is less than or equal to the temperature threshold, the predicted substation power of the single node is obtained by using the normal constraint supply and demand balance equation and inserted into the substation power instruction, the temperature update tag of the single node is set to 0, and the predicted substation power and temperature update tag of the single node are fed back to the temperature prediction unit; If the predicted energy storage cabinet temperature of a single node is greater than the temperature threshold, the single node and the predicted energy storage cabinet temperature of the single node are packaged to generate a cooling start instruction and send it to the temperature control module; Obtain the predicted cooling power of a single node, use the dangerous state constraint supply and demand balance equation to obtain the predicted transformer power of a single node and insert it into the transformer power instruction, set the temperature update tag of the single node to 1, and feed back the predicted transformer power and temperature update tag of the single node to the temperature prediction unit; Until the predicted substation power of all nodes in the unknown period has been placed in the substation power instruction and sent to the energy management module.

Images