Artificial intelligence-based air conditioning control methods, systems, and edge computing devices

Abstract

This invention discloses an artificial intelligence-based air conditioning control method, system, and edge computing device, comprising the following steps: collecting user behavior data and indoor environment data; constructing a state vector and adding time rhythm encoding to generate a fused state input vector; inputting the vector into a multimodal associative structure to generate activation input pairs, and combining a gating factor to determine whether to enhance the associative connection weights; extracting rhythm phase parameters to modulate weights, inputting them into a neural network associative path for propagation, and generating an air conditioning control parameter output set; performing a difference analysis between this set and user manual operation records to construct a cognitive conflict index; when the index exceeds a threshold, weights are weakened and associative connections are updated; based on the updated weight set, propagation is re-established to output control parameters and set the air conditioning operating state, while recording edge feedback. This invention achieves personalized, adaptive, and rhythm-driven response in air conditioning control, improving control accuracy and user experience.

Description

Technical Field

[0001] This invention relates to the field of edge computing technology, and in particular to an air conditioning control method, system and edge computing device based on artificial intelligence. Background Technology

[0002] Most existing air conditioning control systems rely on preset temperature thresholds and environmental sensor data for single-feedback adjustment, lacking a deep understanding and modeling of individual user behavior characteristics, usage habits and lifestyles. In some AI-based solutions, although models such as neural networks are introduced for predicting and optimizing air conditioning control parameters, most methods still use static input features and single-cycle feedback mechanisms, making it difficult to form a stable user preference mapping and dynamic control strategy.

[0003] In terms of data processing, traditional solutions typically model user behavior data and indoor environment data separately, failing to achieve semantic fusion and contextual association. This limits the model's adaptability to complex scene changes. Some methods attempt to introduce attention mechanisms or time series modeling structures to improve the model's perception capabilities, but they often neglect the establishment of a linkage structure between behavior and environmental data, lack modeling support for the relationship between behavior and rhythm, and cannot effectively respond to changes in an individual's control intentions at different time periods.

[0004] Furthermore, existing neural networks rely heavily on backpropagation mechanisms during weight updates, lacking the ability to immediately correct sudden cognitive biases or control conflicts. They cannot quickly adapt to user behavior when control deviations occur, reducing the accuracy and comfort of the interaction. The current model structure also does not fully integrate associative memory mechanisms and bio-inspired weight adjustment rules, making it difficult to support the continuous associative expression of complex intentions and the transfer of control strategies for periodic behaviors. This results in slow response and insufficient control accuracy of the air conditioning system, leading to low user satisfaction.

[0005] Therefore, how to provide air conditioning control methods, systems, and edge computing devices based on artificial intelligence is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an air conditioning control method, system, and edge computing device based on artificial intelligence. This invention utilizes a fusion of state input vectors, rhythm modulation mechanisms, and Heblin learning rules to construct an association-driven neural network linkage structure, thereby achieving dynamic perception and intelligent control of user behavior and environmental status. It has the advantages of accurate response, strong rhythm adaptability, and high control coherence.

[0007] According to embodiments of the present invention, an air conditioning control method, system, and edge computing device based on artificial intelligence include the following steps:

[0008] Collect user's current behavior data and indoor environment data, construct behavior state vector and environment state vector respectively, and add current time rhythm encoding to generate fused state input vector;

[0009] The fusion state input vector is input into the multimodal associative input construction structure to generate neuron activation input pairs, which are then linked and matched with the context gating factor to determine whether to perform weight enhancement operations on the associative connections in the neural network.

[0010] After completing the associative connection weight enhancement, the rhythm phase parameters at the current moment are extracted, and rhythm modulation processing is performed on the established associative weight set to generate a time rhythm modulation weight set.

[0011] The fusion state input vector is input into the neural network associative pathway structure based on the Heb learning rule, and associative propagation calculation is performed according to the time rhythm modulation weight set to generate the air conditioning control parameter output set;

[0012] A cognitive conflict index is constructed by analyzing the differences between the set of air conditioner control parameter outputs and the user's manual operation records during the current control cycle.

[0013] When the cognitive conflict index exceeds a set threshold, the corresponding neuron activation input pair is identified, and the weights of the associated associative connections are weakened and adjusted to obtain an updated set of weights.

[0014] Based on the updated set of associative connection weights, the associative propagation calculation is re-executed on the current fusion state input vector, and the updated set of air conditioning control parameters is output. This set is then input to the edge control device to perform the setting operation of the air conditioning operating parameters and record the state data fed back by the edge control device.

[0015] Optionally, the user's current behavior data includes the user's current spatial location and limb movement status, and the indoor environment data includes indoor temperature and infrared heat flux intensity.

[0016] Optionally, the step of generating a fusion state input vector by adding the current time rhythm code includes: obtaining the timestamp information corresponding to the current moment, constructing a rhythm time vector containing hourly code, daily cycle code and weekly cycle code based on the timestamp information, and concatenating the rhythm time vector with the behavior state vector and the environmental state vector composed of the user's current spatial position state, limb movement state, indoor temperature and infrared heat flux intensity to generate a fusion state input vector.

[0017] Optionally, the process of determining whether to perform weight boosting on associative connections in the neural network includes:

[0018] The fusion state input vector is input into the activation mapping structure to generate the neuron activation vector at the current time step. Based on the set activation pairing rules, neuron activation input pairs are constructed in the neuron activation vector, denoted as (x...). i ,y j );

[0019] The fused state input vector is simultaneously input into the context-gated computation structure, and the current rhythm correlation score r is calculated based on the rhythm time encoding. t ∈[0,1], the behavior stability index s is calculated based on the continuous change sequence of the user behavior state vector and the environment state vector already recorded by the system. t ∈[0,1], calculate the historical linkage frequency value f based on the activation count of the input pair in the past time period. ij ∈N;

[0020] Weighted calculation generates matching confidence value m ij (t);

[0021] Where α+β+γ=1 is the system-set adjustment coefficient, f max This represents the historical maximum activation frequency limit.

[0022] Match confidence value m ij (t) triggers the set associative connection threshold θ c Compare ∈(0,1), if m ij (t)≥θ c If the neuron's activation input pair satisfies the associative connection weight enhancement condition, then it is determined that the neuron's activation input pair satisfies the associative connection weight enhancement condition.

[0023] Optionally, the process of generating the time rhythm modulation weight set includes:

[0024] Based on the time rhythm encoding in the fusion state input vector, the time rhythm phase parameter corresponding to the current time step is extracted and represented as a sine-cosine encoding vector within a unit period, which describes the relative position of the current time step in the rhythm period.

[0025] Traverse the established set of associative connection weights, for each pair of activated input pairs, extract the neuron activation values ​​of the connection pairs in the first K time steps in the historical state sequence, calculate the average activation level, activation variance and maximum activation difference of the connection on the time axis, and construct a connection historical activity index.

[0026] The behavior state vector and environment state vector at the current time step are mapped to the corresponding neuron connection channels respectively. The Euclidean distance between the feature evolution trajectories of the current input and the activated input pair in the historical state sequence is calculated as the context matching degree index. At the same time, the difference between the current indoor temperature and the equipment control state is extracted as the external disturbance response degree index.

[0027] By integrating rhythm phase parameters, context matching index, and external interference response index, a modulation factor triplet (θ) is constructed. t ,ρ ij ,σ ij In the input modulation rule function, weight adjustment and classification are performed based on different combination modes. The specific classification rules are as follows:

[0028]

[0029] Among them, w ij For Lenovo to connect the original weight, For the modulated weights, θ t ρ is the phase parameter of the current rhythm. ij σ is a context matching metric. ij λ1 and λ2 are the external interference response index, δ1, δ2 and ∈ are the modulation gain coefficients, and δ1, δ2 and ∈ are the threshold parameters;

[0030] All the connection results that have completed weight modulation constitute the time-rhythm modulation weight set.

[0031] Optionally, the process of inputting the fused state input vector into the neural network associative pathway structure based on the Hebbian learning rule and performing associative propagation calculation according to the time rhythm modulation weight set includes:

[0032] The fusion state input vector is input to the associative connection layer, which consists of several neuron nodes. The neuron nodes are fully connected by associative connection edges. Each associative connection edge is accompanied by an adjustable weight, and the initial weight is provided by the time rhythm modulation weight set.

[0033] The dimensions of the fused state input vector are mapped to the input neuron nodes according to the structural dimensions. Let the input activation value at the current time step be a. i Calculate the current activation state 'a' of the target neuron node at the other end of the connection edge. j These constitute neuronal activation pairs;

[0034] According to the Hebbian learning rule, a weight adjustment operation is performed on each associative connection edge, as follows:

[0035] If a i >θ and a j >θ, perform weight enhancement, and update weight Δw ij =η·a i ·a j ;

[0036] If a i >θ and a j If the value is less than or equal to θ, perform weight reduction and update the weight Δw.ij =-ε·a i ·(1-a j );

[0037] If a i If the weight of the current connected edge is less than or equal to θ, no weight update is performed.

[0038] Where η represents the weight boosting learning rate parameter, ε represents the weight weakening learning rate parameter, and θ represents the neuron activation threshold.

[0039] The updated connection weight is denoted as... The weight matrix consisting of all updated connection edge weights is input into the associative propagation module;

[0040] In the associative propagation module, a weighted summation calculation is performed on the output neuron nodes to generate the control response value of each output node;

[0041] The control response values ​​of all output nodes are combined into an air conditioning control parameter output set. The output set includes parameter values ​​in several dimensions, each corresponding to a different air conditioning operating parameter dimension, including the target temperature set value, the air supply mode selection value, and the operating fan speed level.

[0042] Optionally, the characteristic feature is that the construction process of the cognitive conflict index includes:

[0043] The air conditioning control parameter output set is divided into control command vectors of corresponding dimensions according to the control item components;

[0044] Perform time synchronization processing on the user's manual operation records within the current control cycle, and extract the operation vectors corresponding to each control item;

[0045] The control command vector and operation vector are compared dimension by dimension according to the control item, the numerical deviation is calculated, and the difference vector is formed.

[0046] Based on the adjustment frequency of each control item in the historical period, the deviation weighting coefficient of the control item is determined;

[0047] The difference vector and the deviation weight coefficient are weighted and summed to output the cognitive conflict index for the current control cycle.

[0048] Optionally, the step of identifying the corresponding neuron activation input pair when the cognitive conflict index exceeds a set threshold, performing a weakening adjustment operation on the associated associative connection weights, and obtaining an updated weight set includes:

[0049] Once the cognitive conflict index exceeds a set threshold, retrieve the neuron activation input pairs activated by the fusion state input vector within the current control cycle;

[0050] For each pair of neuron activation inputs, obtain the connection edge weights in the associative connection weight set and extract their corresponding temporal rhythm modulation coefficients.

[0051] Based on the activation input, the direction of the control parameter output deviation in the current cycle is compared with the direction of the manual correction operation. If the direction of the control parameter output deviation is consistent with the direction of the manual correction operation, the connection weight is determined to be the cognitive conflict contribution edge.

[0052] Perform a weakening operation on all cognitive conflict contributing edges, specifically including: using the rhythm modulation coefficient as the basis for attenuation weight, superimposing the normalized value of the cognitive conflict degree index as a weakening magnitude factor, and updating the current weight value of the connecting edge.

[0053] The updated set of associative connection weights is normalized and cached in the neural network structure, replacing the original set of associative connection weights.

[0054] An artificial intelligence-based air conditioning control system according to an embodiment of the present invention includes:

[0055] The behavior state construction module is used to collect user behavior data and construct behavior state vectors.

[0056] The environmental state construction module is used to collect indoor environmental data and construct an environmental state vector.

[0057] The rhythm coding generation module is used to generate the current time rhythm code, which is then concatenated with the state vector to generate a fused state input vector;

[0058] The multimodal associative input building module is used to generate neuron activation input pairs and perform matching judgments in conjunction with context gating factors;

[0059] The associative connection weight enhancement module is used to determine and enhance the associative connection edge weights of neuron activation input pairs that meet the matching conditions.

[0060] The rhythm modulation processing module is used to extract rhythm phase parameters, perform rhythm modulation on the associative weight set, and generate a modulation weight set.

[0061] The associative path propagation module is used to input the fused state input vector into the neural network structure and perform associative propagation calculations.

[0062] The air conditioning control parameter generation module is used to output the set of air conditioning control parameters for the current control cycle.

[0063] The cognitive conflict analysis module is used to compare user manual operation records and construct cognitive conflict indicators.

[0064] The Lenovo weight reduction adjustment module is used to identify cognitive conflict contributing edges and perform connection weight reduction operations;

[0065] The edge control execution module is used to receive air conditioning control parameters, set operating parameters, and record feedback status data.

[0066] An edge computing device stores a computer program that, when executed by a processor, enables the processor to perform an artificial intelligence-based air conditioning control method.

[0067] The beneficial effects of this invention are:

[0068] (1) Realize intelligent control response based on rhythm modulation: This invention extracts the rhythm phase parameters at the current moment and performs rhythm modulation processing on the associative connection weights to construct a set of control weights that conform to the user's work and rest rhythm, thereby improving the adaptability of the air conditioning control response to changes in time rhythm and enhancing the matching of the control output with the life rhythm.

[0069] (2) Enhance the dynamic adjustment capability of the associative neural network: Based on the Heb learning rule, this invention constructs an associative pathway structure and strengthens or weakens the associative connection edges on the basis of neuron activation input pairs, so as to realize the continuous learning and adaptive adjustment of the control model to user behavior differences and cognitive conflicts, and optimize the control path weight structure.

[0070] (3) A self-updating mechanism for realizing closed-loop user operation feedback: This invention judges the deviation between control output and user manual behavior by constructing a cognitive conflict degree index, and dynamically adjusts the associative connection weight accordingly, so as to realize the control model's feedback perception and weight reconstruction of environmental changes and user preferences, and ensure the long-term effectiveness and stability of the control strategy. Attached Figure Description

[0071] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0072] Figure 1 This is an overall flowchart of the air conditioning control method based on artificial intelligence proposed in this invention;

[0073] Figure 2 This is a schematic diagram of the module connections of the air conditioning control system based on artificial intelligence proposed in this invention;

[0074] Figure 3 This is a diagram of the neural network associative pathway structure based on Heb learning for the air conditioning control method based on artificial intelligence proposed in this invention. Detailed Implementation

[0075] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0076] refer to Figure 1-3 An artificial intelligence-based air conditioning control method, system, and edge computing device includes the following steps:

[0077] Collect user's current behavior data and indoor environment data, construct behavior state vector and environment state vector respectively, and add current time rhythm encoding to generate fused state input vector;

[0078] The fusion state input vector is input into the multimodal associative input construction structure to generate neuron activation input pairs, which are then linked and matched with the context gating factor to determine whether to perform weight enhancement operations on the associative connections in the neural network.

[0079] After completing the associative connection weight enhancement, the rhythm phase parameters at the current moment are extracted, and rhythm modulation processing is performed on the established associative weight set to generate a time rhythm modulation weight set.

[0080] The fusion state input vector is input into the neural network associative pathway structure based on the Heb learning rule, and associative propagation calculation is performed according to the time rhythm modulation weight set to generate the air conditioning control parameter output set;

[0081] A cognitive conflict index is constructed by analyzing the differences between the set of air conditioner control parameter outputs and the user's manual operation records during the current control cycle.

[0082] When the cognitive conflict index exceeds a set threshold, the corresponding neuron activation input pair is identified, and the weights of the associated associative connections are weakened and adjusted to obtain an updated set of weights.

[0083] Based on the updated set of associative connection weights, the associative propagation calculation is re-executed on the current fusion state input vector, and the updated set of air conditioning control parameters is output. This set is then input to the edge control device to perform the setting operation of the air conditioning operating parameters and record the state data fed back by the edge control device.

[0084] In this embodiment, the user's current behavior data includes the user's current spatial location and limb movement status, and the indoor environment data includes indoor temperature and infrared heat flux intensity.

[0085] In this embodiment, the step of generating a fusion state input vector by adding the current time rhythm code includes: obtaining the timestamp information corresponding to the current moment, constructing a rhythm time vector containing hourly code, daily cycle code and weekly cycle code based on the timestamp information, and concatenating the rhythm time vector with the behavior state vector and the environment state vector composed of the user's current spatial position state, limb movement state, indoor temperature and infrared heat flux intensity to generate a fusion state input vector.

[0086] In this embodiment, the process of determining whether to perform weight enhancement on the associative connections in the neural network includes:

[0087] The fusion state input vector is input into the activation mapping structure to generate the neuron activation vector at the current time step. Based on the set activation pairing rules, neuron activation input pairs are constructed in the neuron activation vector, denoted as (x...). i ,y j The fusion state input vector contains multi-dimensional input information composed of behavioral state vector, environmental state vector and current time rhythm encoding. The activation mapping structure, as an intermediate expression component connecting the fusion state input and the neuronal association mechanism, mainly functions to activate neuronal nodes with response characteristics based on the characteristics of the fusion state input vector, and generate the neuron activation vector at the current moment. The activation mapping structure includes a set of projection channels with trainable weights. Each neuron in the channel corresponds to a sub-dimension or combination of dimensions in the fusion vector. The activation value of the neuron is calculated by mapping through the linear transformation between the fusion state input vector and the channel weights, combined with a non-linear activation function, and outputs a neuron activation vector with fixed dimensions. Each element in the activation vector represents the response intensity of a neuron in the current input state, which is used to describe the influence distribution of the current state on the potential association path in the neural network.

[0088] The defined activation pairing rules are used to select neuron pairs with a linkage relationship from the neuron activation vectors, and construct the set of neuron activation input pairs that participate in the subsequent weight enhancement judgment. The activation pairing rules include the following constraint logic:

[0089] Activation threshold screening: Set a lower limit for response intensity, retain only neurons with activation values ​​greater than the threshold as candidate nodes, and remove neurons that do not respond significantly to the current state;

[0090] Associative structure constraint: Based on the associative connection topology among candidate neurons, determine whether there are trainable associative connections, and retain only node pairs with structural connections;

[0091] Response intensity difference limit: Limits the activation difference between two neurons within a preset range to avoid pairing imbalance caused by severe inconsistency in activation levels;

[0092] State similarity matching: Calculate the content similarity of the input subspace corresponding to the activated neurons, and retain only neuron pairs with consistent semantic representation directions;

[0093] Pairing priority sorting: Sort all neuron pairs that meet the screening criteria from high to low according to their average activation values, and select the top N pairs as the neuron activation input pairs at the current time.

[0094] The fused state input vector is simultaneously input into the context-gated computation structure, and the current rhythm correlation score r is calculated based on the rhythm time encoding. t ∈[0,1], the behavior stability index s is calculated based on the continuous change sequence of the user behavior state vector and the environment state vector already recorded by the system. t ∈[0,1], calculate the historical linkage frequency value f based on the activation count of the input pair in the past time period. ij ∈N;

[0095] Weighted calculation generates matching confidence value m ij (t):

[0096]

[0097] Where α+β+γ=1 is the system-set adjustment coefficient, f max This represents the historical maximum activation frequency limit.

[0098] Match confidence value m ij (t) triggers the set associative connection threshold θ c Compare ∈(0,1), if m ij (t)≥θ c If the neuron's activation input pair satisfies the associative connection weight enhancement condition, then it is determined that the neuron's activation input pair satisfies the associative connection weight enhancement condition.

[0099] In this embodiment, the generation process of the time rhythm modulation weight set includes:

[0100] Based on the time rhythm encoding in the fusion state input vector, the time rhythm phase parameter corresponding to the current time step is extracted and represented as a sine-cosine encoding vector within a unit period, which describes the relative position of the current time step in the rhythm period.

[0101] Traverse the established set of associative connection weights, for each pair of activated input pairs, extract the neuron activation values ​​of the connection pairs in the first K time steps in the historical state sequence, calculate the average activation level, activation variance and maximum activation difference of the connection on the time axis, and construct a connection historical activity index.

[0102] The behavior state vector and environment state vector at the current time step are mapped to the corresponding neuron connection channels respectively. The Euclidean distance between the feature evolution trajectories of the current input and the activated input pair in the historical state sequence is calculated as the context matching degree index. At the same time, the difference between the current indoor temperature and the equipment control state is extracted as the external disturbance response degree index.

[0103] By integrating rhythm phase parameters, context matching index, and external interference response index, a modulation factor triplet (θ) is constructed. t ,ρ ij ,σ ij In the input modulation rule function, weight adjustment and classification are performed based on different combination modes. The specific classification rules are as follows:

[0104]

[0105] Among them, w ij For Lenovo to connect the original weight, For the modulated weights, θ t ρ is the phase parameter of the current rhythm. ij σ is a context matching metric. ij λ1 and λ2 are the external interference response index, δ1, δ2 and ∈ are the modulation gain coefficients, and δ1, δ2 and ∈ are the threshold parameters;

[0106] All the connection results that have completed weight modulation constitute the time-rhythm modulation weight set.

[0107] In this embodiment, the process of inputting the fused state input vector into the neural network associative pathway structure based on the Hebbian learning rule and performing associative propagation calculation according to the time rhythm modulation weight set includes:

[0108] The fusion state input vector is input to the associative connection layer, which consists of several neuron nodes. The neuron nodes are fully connected by associative connection edges. Each associative connection edge is accompanied by an adjustable weight, and the initial weight is provided by the time rhythm modulation weight set.

[0109] The dimensions of the fused state input vector are mapped to the input neuron nodes according to the structural dimensions. Let the input activation value at the current time step be a. i Calculate the current activation state 'a' of the target neuron node at the other end of the connection edge. j These constitute neuronal activation pairs;

[0110] According to the Hebbian learning rule, a weight adjustment operation is performed on each associative connection edge, as follows:

[0111] If a i >θ and a j >θ, perform weight enhancement, and update weight Δwij =η·a i ·a j ;

[0112] If a i >θ and a j If the value is less than or equal to θ, perform weight reduction and update the weight Δw. ij =-ε·a i ·(1-a j );

[0113] If a i If the weight of the current connected edge is less than or equal to θ, no weight update is performed.

[0114] Where η represents the weight boosting learning rate parameter, ε represents the weight weakening learning rate parameter, and θ represents the neuron activation threshold.

[0115] The updated connection weight is denoted as... The weight matrix consisting of all updated connection edge weights is input into the associative propagation module;

[0116] In the associative propagation module, a weighted summation calculation is performed on the output neuron nodes to generate the control response value of each output node;

[0117] The control response values ​​of all output nodes are combined into an air conditioning control parameter output set. The output set includes parameter values ​​in several dimensions, each corresponding to a different air conditioning operating parameter dimension, including the target temperature set value, the air supply mode selection value, and the operating fan speed level.

[0118] In this embodiment, the construction process of the cognitive conflict index includes:

[0119] The air conditioning control parameter output set is divided into control command vectors of corresponding dimensions according to the control item components;

[0120] Perform time synchronization processing on the user's manual operation records within the current control cycle, and extract the operation vectors corresponding to each control item;

[0121] The control command vector and operation vector are compared dimension by dimension according to the control item, the numerical deviation is calculated, and the difference vector is formed.

[0122] Based on the adjustment frequency of each control item in the historical period, the deviation weighting coefficient of the control item is determined;

[0123] The difference vector and the deviation weight coefficient are weighted and summed to output the cognitive conflict index for the current control cycle.

[0124] In this embodiment, the step of identifying the corresponding neuron activation input pair when the cognitive conflict index exceeds a set threshold, performing a weakening adjustment operation on the associated associative connection weights, and obtaining an updated weight set includes:

[0125] Once the cognitive conflict index exceeds a set threshold, retrieve the neuron activation input pairs activated by the fusion state input vector within the current control cycle;

[0126] For each pair of neuron activation inputs, obtain the connection edge weights in the associative connection weight set and extract their corresponding temporal rhythm modulation coefficients.

[0127] Based on the activation input, the direction of the control parameter output deviation in the current cycle is compared with the direction of the manual correction operation. If the direction of the control parameter output deviation is consistent with the direction of the manual correction operation, the connection weight is determined to be the cognitive conflict contribution edge.

[0128] Perform a weakening operation on all cognitive conflict contributing edges, specifically including: using the rhythm modulation coefficient as the basis for attenuation weight, superimposing the normalized value of the cognitive conflict degree index as a weakening magnitude factor, and updating the current weight value of the connecting edge.

[0129] The updated set of associative connection weights is normalized and cached in the neural network structure, replacing the original set of associative connection weights.

[0130] An artificial intelligence-based air conditioning control system according to an embodiment of the present invention includes:

[0131] The behavior state construction module is used to collect user behavior data and construct behavior state vectors.

[0132] The environmental state construction module is used to collect indoor environmental data and construct an environmental state vector.

[0133] The rhythm coding generation module is used to generate the current time rhythm code, which is then concatenated with the state vector to generate a fused state input vector;

[0134] The multimodal associative input building module is used to generate neuron activation input pairs and perform matching judgments in conjunction with context gating factors;

[0135] The associative connection weight enhancement module is used to determine and enhance the associative connection edge weights of neuron activation input pairs that meet the matching conditions.

[0136] The rhythm modulation processing module is used to extract rhythm phase parameters, perform rhythm modulation on the associative weight set, and generate a modulation weight set.

[0137] The associative path propagation module is used to input the fused state input vector into the neural network structure and perform associative propagation calculations.

[0138] The air conditioning control parameter generation module is used to output the set of air conditioning control parameters for the current control cycle.

[0139] The cognitive conflict analysis module is used to compare user manual operation records and construct cognitive conflict indicators.

[0140] The Lenovo weight reduction adjustment module is used to identify cognitive conflict contributing edges and perform connection weight reduction operations;

[0141] The edge control execution module is used to receive air conditioning control parameters, set operating parameters, and record feedback status data.

[0142] An edge computing device stores a computer program that, when executed by a processor, enables the processor to perform an artificial intelligence-based air conditioning control method.

[0143] Example 1:

[0144] To verify the feasibility of this invention in practice, it was applied to a smart home system integration scenario. The air conditioning equipment needs to automatically adjust its operating parameters based on user behavior and environmental conditions to improve comfort and reduce energy consumption. Traditional air conditioning control systems often rely on single environmental parameters such as temperature or humidity settings and execute start-stop operations through static strategies. They cannot respond to changes in user behavior or time rhythm factors, resulting in delayed control response, energy redundancy, and decreased user satisfaction. In this scenario, user behavior has obvious time periodic characteristics. For example, changes in state such as morning activities, afternoon naps, and nighttime sleep have clear adjustment needs for the air conditioning operating mode. However, the existing control logic cannot identify the inherent relationship between behavior and time rhythm, resulting in untimely switching of air conditioning operating modes, leading to phenomena such as overcooling or insufficient comfort.

[0145] To address the aforementioned issues, the neural network air conditioning control method based on Hebbian learning rules proposed in this invention is deployed in this smart home system. By collecting user activity data (such as movement trajectories within the room and frequency of voice interaction commands) and environmental state data (room temperature, humidity, CO2 concentration, and light intensity) at different time periods, a fusion state input vector is constructed using temporal rhythm encoding. The multimodal associative input construction structure designed in this invention, combined with a context gating mechanism, dynamically generates currently activated neuron pairs and calculates the enhancement degree of associative connection weights according to Hebbian learning rules, modulating the signal propagation intensity along the neural network associative path. Under the action of the temporal rhythm modulation mechanism, control preferences at different time periods are dynamically adjusted through weight sets to achieve joint control of behavioral response and rhythmic drive. Based on this, a cognitive conflict index is constructed. When the user's manual operation frequently deviates from the current output control result, the associated neuron input pairs are automatically identified and the corresponding weights are adjusted to achieve adaptive memory correction. Finally, the control output is input to the edge control device for command-driven compression mechanisms, air supply modes, fan speed settings, and other control operations, completing closed-loop control.

[0146] The test compared an air conditioning system using an intelligent control system with one that did not use the control strategy of this invention. Each system operated for 72 hours in three typical environments: hot sunny days, moderately warm cloudy days, and high humidity at night. Temperature control response time, user satisfaction scores, and average energy consumption were recorded. User satisfaction scores were collected via handheld devices and normalized to a range of 0-1. The results showed that the air conditioning system using the control strategy of this invention responded faster during periods of significant environmental fluctuation, effectively identifying behavioral states before and after sleep and before and after leaving home, and switching to the corresponding mode. Simultaneously, energy consumption was significantly reduced compared to traditional strategies, and the frequency of user intervention was significantly decreased.

[0147] To verify the control effectiveness and energy-saving effect of this invention in a real-world scenario, the operational performance indicators collected during the testing process are shown in the table below:

[0148] Table 1: Comparative Operational Data Statistics of Intelligent Air Conditioning Control Systems

[0149]

[0150] Based on the data results in the "Statistical Table of Comparative Operation Data of Intelligent Air Conditioning Control System", a comprehensive analysis is conducted on the performance of the method of the present invention and the traditional control strategy under different environmental conditions, demonstrating the technical advantages of the present invention in terms of response efficiency, user experience and energy-saving performance.

[0151] Under "high temperature and sunny weather" conditions, the average response time of the method of this invention is 5.2 seconds, which is much lower than the 13.6 seconds of the traditional strategy. This indicates that the system can quickly identify environmental changes and output corresponding control parameters. Under the same conditions, the frequency of manual intervention by users is 1.1 times / day, while that of the traditional strategy is 4.8 times / day. This shows that the invention can match user expectations well through the associative path generation and rhythm modulation mechanism, reducing the need for manual adjustment. In terms of satisfaction score, the method of this invention reached 0.92, which is significantly higher than the 0.75 of the traditional strategy. At the same time, the average daily energy consumption is reduced by 1.28 kWh, which verifies the energy-saving benefits of the system under high load conditions.

[0152] In a "medium-temperature cloudy" environment, the average response time of the method of this invention is 4.8 seconds, while that of the traditional strategy is 11.9 seconds; the frequency of user intervention is 0.9 times / day and 3.7 times / day, respectively, and the satisfaction scores are 0.94 and 0.76. The invention still demonstrates rapid response and high adaptability under such medium load conditions, and maintains precise linkage control of behavior and rhythm even when the air conditioning load demand is relatively low, further reducing unnecessary energy consumption. The average daily energy consumption is reduced to 4.52 kWh, which is 1.08 kWh less than that of the traditional strategy.

[0153] In the "high humidity at night" scenario, the response time of the method of this invention is 6.1 seconds, while that of the traditional strategy is 14.5 seconds. The frequency of user intervention is 0.7 times / day and 4.1 times / day, respectively, and the satisfaction score reaches 0.89, which is a significant improvement compared to the 0.69 of the traditional strategy. The rhythm factors in this scenario are more complex. This invention effectively adjusts the wind speed and operating mode in the nighttime rest scenario by integrating time phase encoding and cognitive conflict feedback mechanism, ensuring user comfort while reducing energy consumption from 6.67kWh to 5.14kWh.

[0154] Based on the comparative data under the three typical environmental conditions mentioned above, it can be seen that the method of the present invention is superior to traditional control methods in terms of control response timeliness, behavioral adaptability, user interaction experience, and operational efficiency. In particular, with the support of cognitive conflict feedback and dynamic weight adjustment mechanism, the system exhibits higher intelligent decision-making ability and rhythm perception ability, and has engineering application value.

[0155] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An air conditioning control method based on artificial intelligence, characterized in that, Includes the following steps: Collect user's current behavior data and indoor environment data, construct behavior state vector and environment state vector respectively, and add current time rhythm encoding to generate fused state input vector; The fusion state input vector is input into the multimodal associative input construction structure to generate neuron activation input pairs, which are then linked and matched with the context gating factor to determine whether to perform weight enhancement operations on the associative connections in the neural network. After completing the associative connection weight enhancement, the rhythm phase parameters at the current moment are extracted, and rhythm modulation processing is performed on the established associative weight set to generate a time rhythm modulation weight set. The fusion state input vector is input into the neural network associative pathway structure based on the Heb learning rule, and associative propagation calculation is performed according to the time rhythm modulation weight set to generate the air conditioning control parameter output set; A cognitive conflict index is constructed by analyzing the differences between the set of air conditioner control parameter outputs and the user's manual operation records during the current control cycle. When the cognitive conflict index exceeds a set threshold, the corresponding neuron activation input pair is identified, and the weights of the associated associative connections are weakened and adjusted to obtain an updated set of weights. Based on the updated set of associative connection weights, the associative propagation calculation is re-executed on the current fusion state input vector, and the updated set of air conditioning control parameters is output. This set is then input to the edge control device to perform the setting operation of the air conditioning operating parameters and record the state data fed back by the edge control device.

2. The air conditioning control method based on artificial intelligence according to claim 1, characterized in that, The user's current behavior data includes the user's current spatial location and limb movement status, and the indoor environment data includes indoor temperature and infrared heat flux intensity.

3. The air conditioning control method based on artificial intelligence according to claim 2, characterized in that, The step of generating a fusion state input vector by adding the current time rhythm code includes: obtaining the timestamp information corresponding to the current moment, constructing a rhythm time vector containing hourly code, daily cycle code and weekly cycle code based on the timestamp information, and concatenating the rhythm time vector with the behavior state vector and the environment state vector composed of the user's current spatial position state, limb movement state, indoor temperature and infrared heat flux intensity to generate a fusion state input vector.

4. The air conditioning control method based on artificial intelligence according to claim 3, characterized in that, The process of determining whether to perform weight boosting on associative connections in a neural network includes: The fusion state input vector is input into the activation mapping structure to generate the neuron activation vector at the current time step. Based on the set activation pairing rules, neuron activation input pairs are constructed in the neuron activation vector, denoted as (x...). i ,y j ); The fused state input vector is simultaneously input into the context-gated computation structure, and the current rhythm correlation score r is calculated based on the rhythm time encoding. t ∈[0,1], the behavior stability index s is calculated based on the continuous change sequence of the user behavior state vector and the environment state vector already recorded by the system. t ∈[0,1], calculate the historical linkage frequency value f based on the activation count of the input pair in the past time period. ij ; Weighted calculation generates matching confidence value m ij (t); Match confidence value m ij (t) triggers the set associative connection threshold θ c Compare ∈(0,1), if m ij (t)≥θ c If the neuron's activation input pair satisfies the associative connection weight enhancement condition, then it is determined that the neuron's activation input pair satisfies the associative connection weight enhancement condition.

5. The air conditioning control method based on artificial intelligence according to claim 4, characterized in that, The process of generating the time rhythm modulation weight set includes: Based on the time rhythm encoding in the fusion state input vector, the time rhythm phase parameter corresponding to the current time step is extracted and represented as a sine-cosine encoding vector within a unit period, which describes the relative position of the current time step in the rhythm period. Traverse the established set of associative connection weights, for each pair of activated input pairs, extract the neuron activation values ​​of the connection pairs in the first K time steps in the historical state sequence, calculate the average activation level, activation variance and maximum activation difference of the connection on the time axis, and construct a connection historical activity index. The behavior state vector and environment state vector at the current time step are mapped to the corresponding neuron connection channels respectively. The Euclidean distance between the feature evolution trajectories of the current input and the activated input pair in the historical state sequence is calculated as the context matching degree index. At the same time, the difference between the current indoor temperature and the equipment control state is extracted as the external disturbance response degree index. By integrating rhythm phase parameters, context matching index, and external interference response index, a modulation factor triplet (θ) is constructed. t ,ρ ij ,σ ij In the input modulation rule function, weight adjustment and classification are performed based on different combination modes. The specific classification rules are as follows: Among them, w ij For Lenovo to connect to the original weight, For the modulated weights, θ t ρ is the phase parameter of the current rhythm. ij σ is a context matching metric. ij λ1 and λ2 are the external interference response index, δ1, δ2, and ∈ are the modulation gain coefficients, and δ1, δ2, and ∈ are the threshold parameters. All the connection results that have completed weight modulation constitute the time-rhythm modulation weight set.

6. The air conditioning control method based on artificial intelligence according to claim 5, characterized in that, The process of inputting the fused state input vector into the neural network associative pathway structure based on the Hebbian learning rule, and performing associative propagation calculation according to the time rhythm modulation weight set includes: The fusion state input vector is input to the associative connection layer, which consists of several neuron nodes. The neuron nodes are fully connected by associative connection edges. Each associative connection edge is accompanied by an adjustable weight, and the initial weight is provided by the time rhythm modulation weight set. The dimensions of the fused state input vector are mapped to the input neuron nodes according to the structural dimensions. Let the input activation value at the current time step be a. i Calculate the current activation state 'a' of the target neuron node at the other end of the connection edge. j These constitute neuronal activation pairs; According to the Hebbian learning rule, a weight adjustment operation is performed on each associative connection edge, as follows: If a i >θ and a j >θ, perform weight enhancement, and update weight Δw ij =η·a i ·a j ; If a i >θ and a j If the value is less than or equal to θ, perform weight reduction and update the weight Δw. ij =-ε·a i ·(1-a j ); If a i If the weight of the current connected edge is less than or equal to θ, no weight update is performed. Where η represents the weight boosting learning rate parameter, ε represents the weight weakening learning rate parameter, and θ represents the neuron activation threshold. The updated connection weight is denoted as... The weight matrix consisting of all updated connection edge weights is input into the associative propagation module; In the associative propagation module, a weighted summation calculation is performed on the output neuron nodes to generate the control response value of each output node; The control response values ​​of all output nodes are combined into an air conditioning control parameter output set. The output set includes parameter values ​​in several dimensions, each corresponding to a different air conditioning operating parameter dimension, including the target temperature set value, the air supply mode selection value, and the operating fan speed level.

7. The air conditioning control method based on artificial intelligence according to claim 6, characterized in that, The process of constructing the cognitive conflict index includes: The air conditioning control parameter output set is divided into control command vectors of corresponding dimensions according to the control item components; Perform time synchronization processing on the user's manual operation records within the current control cycle, and extract the operation vectors corresponding to each control item; The control command vector and operation vector are compared dimension by dimension according to the control item, the numerical deviation is calculated, and the difference vector is formed. Based on the adjustment frequency of each control item in the historical period, the deviation weighting coefficient of the control item is determined; The difference vector and the deviation weight coefficient are weighted and summed to output the cognitive conflict index for the current control cycle.

8. The air conditioning control method based on artificial intelligence according to claim 7, characterized in that, When the cognitive conflict index exceeds a set threshold, the process of identifying the corresponding neuron activation input pairs, weakening and adjusting the associated associative connection weights to obtain an updated weight set includes: Once the cognitive conflict index exceeds a set threshold, retrieve the neuron activation input pairs activated by the fusion state input vector within the current control cycle; For each pair of neuron activation inputs, obtain the connection edge weights in the associative connection weight set and extract their corresponding temporal rhythm modulation coefficients. Based on the activation input, the direction of the control parameter output deviation in the current cycle is compared with the direction of the manual correction operation. If the direction of the control parameter output deviation is consistent with the direction of the manual correction operation, the connection weight is determined to be the cognitive conflict contribution edge. Perform a weakening operation on all cognitive conflict contributing edges, specifically including: using the rhythm modulation coefficient as the basis for attenuation weight, superimposing the normalized value of the cognitive conflict degree index as a weakening magnitude factor, and updating the current weight value of the connecting edge. The updated set of associative connection weights is normalized and cached in the neural network structure, replacing the original set of associative connection weights.

9. An artificial intelligence-based air conditioning control system, applied to any one of the artificial intelligence-based air conditioning control methods according to claims 1 to 8, characterized in that, include: The behavior state construction module is used to collect user behavior data and construct behavior state vectors. The environmental state construction module is used to collect indoor environmental data and construct an environmental state vector. The rhythm coding generation module is used to generate the current time rhythm code, which is then concatenated with the state vector to generate a fused state input vector; The multimodal associative input building module is used to generate neuron activation input pairs and perform matching judgments in conjunction with context gating factors; The associative connection weight enhancement module is used to determine and enhance the associative connection edge weights of neuron activation input pairs that meet the matching conditions. The rhythm modulation processing module is used to extract rhythm phase parameters, perform rhythm modulation on the associative weight set, and generate a modulation weight set. The associative path propagation module is used to input the fused state input vector into the neural network structure and perform associative propagation calculations. The air conditioning control parameter generation module is used to output the set of air conditioning control parameters for the current control cycle. The cognitive conflict analysis module is used to compare user manual operation records and construct cognitive conflict indicators. The Lenovo weight reduction adjustment module is used to identify cognitive conflict contributing edges and perform connection weight reduction operations; The edge control execution module is used to receive air conditioning control parameters, set operating parameters, and record feedback status data.

10. An edge computing device, characterized in that, The edge computing device stores a computer program that, when executed by a processor, enables the processor to perform the artificial intelligence-based air conditioning control method according to any one of claims 1 to 8.

Images