Industrial air conditioner and waste heat recovery collaborative intelligent optimization control method and related equipment

Abstract

The invention provides an industrial air conditioner and waste heat recovery collaborative intelligent optimization control method and related equipment, and is applied to the technical field of data processing. Industrial air conditioner and waste heat recovery system data are fused through the Internet of Things and edge calculation, a standardized parameter sequence is generated through wavelet noise reduction and is converted into an energy flow network atlas, high-energy-consumption working conditions are subdivided through fuzzy clustering, and a dynamic energy efficiency coupling model is constructed. And based on the model, constructing an optimization equation by using an improved particle swarm algorithm, and obtaining the energy consumption of the key node in combination with an energy gradient utilization matrix. And an energy-saving potential coefficient is generated through the multi-dimensional feature matrix, a production plan grouping operation mode is combined, a self-adaptive optimization control model is constructed by using deep reinforcement learning, and finally a real-time regulation and control instruction and an energy efficiency scheme are generated to realize system collaborative optimization.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an intelligent optimization control method and related equipment for the coordinated operation of industrial air conditioning and waste heat recovery. Background Technology

[0002] In industrial production, industrial air conditioning is one of the core pieces of equipment for ensuring a stable production environment. Its energy consumption typically accounts for 15%-30% of total industrial energy consumption. Furthermore, industrial production processes generate a significant amount of waste heat (such as heat dissipation from production equipment and exothermic reactions). If this waste heat is not effectively recovered, it not only wastes energy but may also exacerbate environmental thermal pollution. Currently, in industrial settings, industrial air conditioning and waste heat recovery systems often operate independently, lacking a coordinated control mechanism. This leads to significant contradictions and deficiencies in energy utilization between the two systems, as detailed below:

[0003] Existing technologies rely heavily on single sensors or traditional monitoring methods to collect data on industrial air conditioning operating parameters (such as return air temperature, compressor power, and cooling capacity) and waste heat recovery system data (such as waste heat medium flow rate, heat exchange efficiency, and medium temperature). They do not employ multi-source data fusion technology and do not effectively filter out interference noise such as equipment vibration, resulting in errors in the raw data (e.g., compressor power signal fluctuations of ±1.5kW due to vibration interference). This makes it impossible to provide accurate data support for subsequent energy efficiency analysis and control.

[0004] Industrial production conditions are complex and dynamic (such as fluctuations in production load and changes in ambient temperature). Existing technologies struggle to accurately segment high-energy-consuming conditions and often employ single-mechanism modeling or data-driven modeling methods without combining the advantages of both to construct a coupled model. This results in the model failing to accurately reflect the energy coupling relationship between the air conditioning and waste heat systems. For example, it is impossible to quantify the impact of waste heat quality (temperature gradient, flow stability) on air conditioning energy consumption under different operating conditions, making it difficult to achieve dynamic optimization of system energy efficiency.

[0005] On the one hand, the waste heat recovery system does not utilize waste heat in stages according to its quality, such as treating high-temperature waste heat (80-100℃) and low-temperature waste heat (30-50℃) equally, resulting in the underutilization of high-grade waste heat. On the other hand, the air conditioning system and the waste heat recovery system lack coordinated control. For example, when the air conditioning load increases sharply, the waste heat recovery system does not adjust the medium flow rate in time to increase the waste heat supply, resulting in the air conditioning consuming a large amount of extra electricity. Furthermore, the energy loss of key nodes (such as pipeline transmission and heat exchangers) is not accurately identified (such as the fluctuation error of the pipeline heat loss coefficient exceeding 0.005kW / m·h), further reducing the overall energy efficiency of the system.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0008] According to one aspect of this application, an intelligent optimization control method for the coordinated operation of industrial air conditioning and waste heat recovery is provided, comprising: acquiring industrial air conditioning operating parameters and waste heat recovery system data; using multi-source data fusion technology based on IoT sensors and edge computing gateways, employing wavelet denoising algorithm to eliminate equipment vibration interference noise, and generating a standardized operating parameter sequence; converting the standardized operating parameter sequence into a system energy flow network map, processing it with mechanism-data hybrid modeling software, selecting fuzzy clustering algorithm to subdivide high-energy-consumption operating conditions, and constructing a dynamic energy efficiency coupling model; based on the dynamic energy efficiency coupling model, setting the system energy consumption as an operating condition-time dependent function, using an improved particle swarm optimization algorithm to couple load prediction correction terms to construct an optimization objective equation, extracting conversion coefficients of different waste heat qualities to construct an energy cascade utilization matrix, and performing coordinated calculation of the energy consumption distribution of the air conditioning-waste heat system. Obtain energy loss values ​​at key nodes; combine these values ​​with production scenarios to extract peak energy consumption, waste heat utilization rate, and load matching degree, classify energy efficiency levels, and construct a multi-dimensional feature matrix containing operating parameters, load characteristics, and environmental conditions; based on this multi-dimensional feature matrix, compare real-time energy efficiency with historical data to identify abnormal signals such as sudden increases in energy consumption and insufficient waste heat utilization, and generate energy-saving potential coefficients using the slope of the energy efficiency-load curve; combine these energy-saving potential coefficients with production plans to group operating modes, use deep reinforcement learning algorithms to screen key control factors, and integrate energy efficiency parameters, equipment status, and production demand information to construct an adaptive optimization control model; based on the control strategy output of the adaptive optimization control model, and combined with the dynamic correlation information between air conditioning load and waste heat supply, generate real-time control instructions and control information for energy efficiency improvement schemes.

[0009] Another aspect of this application discloses an intelligent optimization control device for the coordinated operation of industrial air conditioning and waste heat recovery, comprising: a data acquisition module for acquiring industrial air conditioning operating parameters and waste heat recovery system data; based on multi-source data fusion technology using IoT sensors and edge computing gateways, employing wavelet denoising algorithms to eliminate equipment vibration interference noise and generating a standardized operating parameter sequence; a processing module for converting the standardized operating parameter sequence into a system energy flow network graph, processing it using mechanism-data hybrid modeling software, employing fuzzy clustering algorithms to subdivide high-energy-consumption operating conditions, and constructing a dynamic energy efficiency coupling model; based on the dynamic energy efficiency coupling model, setting system energy consumption as an operating condition-time dependent function, employing an improved particle swarm optimization algorithm to couple load prediction correction terms to construct an optimization objective equation, extracting conversion coefficients of different waste heat qualities to construct an energy cascade utilization matrix, and further processing the energy consumption distribution of the air conditioning-waste heat system. Collaborative computation is performed to obtain energy loss values ​​at key nodes. Based on these values, peak energy consumption, waste heat utilization rate, and load matching degree are extracted according to production operating scenarios to classify energy efficiency levels and construct a multi-dimensional feature matrix containing operating parameters, load characteristics, and environmental conditions. Using this multi-dimensional feature matrix, real-time energy efficiency is compared with historical data to identify abnormal signals such as sudden increases in energy consumption and insufficient waste heat utilization. An energy-saving potential coefficient is generated using the slope of the energy efficiency-load curve. This coefficient is then combined with the production plan to group operating modes. A deep reinforcement learning algorithm is used to screen key control factors, and energy efficiency parameters, equipment status, and production demand information are integrated to construct an adaptive optimization control model. Based on the control strategy output of the adaptive optimization control model, and combined with the dynamic correlation information between air conditioning load and waste heat supply, real-time control instructions and control information for energy efficiency improvement schemes are generated.

[0010] According to another aspect of this application, an electronic device includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-described intelligent optimization control method for coordinated industrial air conditioning and waste heat recovery by executing the executable instructions.

[0011] According to another aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a second processor, implements the above-described intelligent optimization control method for coordinated industrial air conditioning and waste heat recovery.

[0012] This application provides an intelligent optimization control method and related equipment for the coordinated operation of industrial air conditioning and waste heat recovery. Centering on the coordinated optimization of industrial air conditioning and waste heat recovery, it integrates multi-source data through the Internet of Things and edge computing, generates standardized parameter sequences through wavelet denoising, converts them into an energy flow network graph, and then uses fuzzy clustering to subdivide high-energy-consumption operating conditions, constructing a dynamic energy efficiency coupling model driven by a mechanism-data hybrid approach. Based on the model, an improved particle swarm optimization algorithm is used to construct optimization equations, and the energy consumption of key nodes is obtained by combining energy cascade utilization matrices. Then, energy-saving potential coefficients are generated through a multi-dimensional feature matrix, and combined with the production plan group operation mode, deep reinforcement learning (Actor-Critic architecture) is used to screen control factors, constructing an adaptive optimization control model. Finally, real-time control commands and energy efficiency schemes are generated, forming a closed loop of "data acquisition-modeling-optimization-control".

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0014] Figure 1 The flowchart illustrates an intelligent optimization control method for the coordinated operation of industrial air conditioning and waste heat recovery, provided in an embodiment of this application.

[0015] Figure 2 This illustration shows a schematic diagram of the structure of an intelligent optimization control device for industrial air conditioning and waste heat recovery in coordination, provided in an embodiment of this application. Detailed Implementation

[0016] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0017] The following is combined Figure 1 This application describes an intelligent optimization control method for the coordinated operation of industrial air conditioning and waste heat recovery, based on exemplary embodiments thereof. It should be noted that the application scenarios described below are merely illustrative for the purpose of understanding the spirit and principles of this application, and the embodiments of this application are not limited in any way. Rather, the embodiments of this application are applicable to any suitable scenario.

[0018] In one implementation, Figure 1 A schematic flowchart illustrating an intelligent optimization control method for coordinated industrial air conditioning and waste heat recovery according to an embodiment of this application is shown, including:

[0019] S101 acquires industrial air conditioning operating parameters and waste heat recovery system data. Based on the multi-source data fusion technology of IoT sensors and edge computing gateways, it uses wavelet denoising algorithm to eliminate equipment vibration interference noise and generate a standardized operating parameter sequence.

[0020] In one implementation, when acquiring the operating parameters of the industrial air conditioning system and the data of the waste heat recovery system, it is necessary to comprehensively cover the core operating indicators of both types of systems. The operating parameters of the industrial air conditioning system include, in addition to return air temperature (e.g., 25℃±0.5℃), refrigeration compressor power (e.g., 15kW), and fan speed (e.g., 1450r / min), supply air temperature (e.g., 18℃±0.3℃), condenser inlet and outlet water temperatures (e.g., 30℃ / 35℃), and cooling capacity (e.g., 50kW). These parameters directly reflect the cooling efficiency and operating status of the air conditioning system. The data of the waste heat recovery system includes, in addition to the flow rate of the waste heat medium (e.g., hot water) (e.g., 8m³ / h), medium inlet temperature (e.g., 60℃), and heat exchanger efficiency (e.g., 85%), medium outlet temperature (e.g., 45℃), waste heat recovery amount (e.g., 12kW), and pipeline pressure loss (e.g., 0.1MPa), which can accurately reflect the efficiency of waste heat recovery and system losses.

[0021] Based on multi-source data fusion technology using IoT sensors and edge computing gateways, a sensing and processing network covering the entire system is constructed. Specifically, temperature sensors (accuracy ±0.1℃) are installed in the return and supply air ducts of the air conditioning unit, a power sensor (sampling frequency 1kHz) is installed at the compressor output, and a speed sensor is installed at the fan motor. Flow sensors (range 0-20m³ / h) and temperature sensors are installed at the inlet and outlet pipes of the heat exchanger in the waste heat recovery system, respectively, and an efficiency monitoring module is installed on the heat exchanger shell. These IoT sensors transmit the collected, dispersed data to the edge computing gateway in real time. The gateway first performs format verification on the data (e.g., removing outliers from temperature sensors that exceed the -50℃ to 100℃ range), then converts different types of parameters (temperature, flow rate, power, etc.) into unified data frames in JSON format. Timestamp alignment (error ≤10ms) is used to achieve spatiotemporal synchronization between air conditioning operating parameters and waste heat recovery data, ensuring the correlation of multi-source data in both time and space dimensions.

[0022] When using wavelet denoising algorithms to eliminate equipment vibration interference noise, precise processing is applied to the noise characteristics of different parameters. Taking the compressor power signal as an example, the original signal contains high-frequency noise generated by the compressor's mechanical vibration (such as piston reciprocating motion and motor operation). The signal is decomposed into eight scales using a db4 wavelet basis, with scales 7 and 8 corresponding to high-frequency noise components above 1000Hz. This component exhibits irregular and drastic fluctuations (instantaneous amplitude can reach ±1.2kW). After removing this noise, the signal is reconstructed using inverse wavelet transform, narrowing the power data fluctuation range from the original ±1.5kW to within ±0.3kW. This effectively preserves the trend characteristics of the power signal as the load changes (e.g., when the production load increases, the power steadily increases from 15kW to 18kW), avoiding noise interference with subsequent energy efficiency analysis.

[0023] When generating a standardized operating parameter sequence, the data is standardized and structured. First, the processed data is divided into time series at 1-minute intervals, with each time node corresponding to a complete set of parameters (e.g., 10:00:00 includes return air temperature of 25.2℃, compressor power of 15.1kW, and waste heat medium flow rate of 7.9m³ / h). Second, the parameter units are standardized, such as temperature based on ℃, power based on kW, and flow rate based on m³ / h, to avoid calculation errors caused by unit confusion. Finally, the min-max normalization method is used to map each parameter value to the 0-1 range (e.g., compressor power range of 10-20kW, 15kW corresponds to 0.5), forming a three-dimensional sequence data containing "timestamp (2024-05-20 10:00:00), parameter identifier ('air conditioner_compressor power'), and standardized value (0.5)". This sequence can be directly imported into mechanism-data hybrid modeling software to provide a high-quality data foundation for the subsequent construction of the system energy flow network map.

[0024] S102, after converting the standardized operating parameter sequence into a system energy flow network map, it is processed by mechanism-data hybrid modeling software, and fuzzy clustering algorithm is used to subdivide high-energy-consumption operating conditions to construct a dynamic energy efficiency coupling model.

[0025] In one implementation, air conditioning energy consumption data, waste heat medium flow rate data, and heat exchange efficiency parameters from a standardized operating parameter sequence are extracted and transformed to generate energy flow node association groups, network transmission loss coefficients, and parameter spatiotemporal matching degrees. From the standardized sequence, the hourly power consumption of the air conditioning compressor (e.g., 120 kWh), the hourly circulation flow rate of the waste heat medium (hot water) (e.g., 480 m³), ​​and the real-time heat exchange efficiency of the heat exchanger (e.g., 82%) are extracted. Using a directed graph model, physical units such as air conditioning units, waste heat exchangers, and pipes are mapped to energy flow nodes, generating node association groups (e.g., energy transfer links such as "compressor → condenser → waste heat exchanger → user end"). The network transmission loss coefficient is calculated based on pipe material and length (e.g., the heat loss coefficient for steel pipes is set to 0.03 kW / m·h). Through time series alignment and spatial location matching, the parameter spatiotemporal matching degree is obtained (e.g., the time matching degree between the peak air conditioning energy consumption and the peak waste heat generation is 85%, and the spatial transmission distance matching degree is 90%).

[0026] The energy flow node association groups, network transmission loss coefficients, and parameter spatiotemporal matching degrees are processed to generate mechanistic modeling constraints, data fitting error thresholds, and clustering feature extraction dimensions. Based on the energy conservation law in the node association groups, mechanistic modeling constraints are set (e.g., compressor output energy = cooling capacity + waste heat emission + transmission loss, with an allowable error range ≤ 5%). Based on the fluctuation range of the network transmission loss coefficient, data fitting error thresholds are set (e.g., the fitting error of the heat loss coefficient ≤ 0.002 kW / m·h). Combining the key influencing factors of parameter spatiotemporal matching degrees, the clustering feature extraction dimensions are determined to be 3-dimensional (time synchronization dimension, spatial transmission dimension, and energy loss dimension).

[0027] Based on mechanistic modeling constraints, data fitting error thresholds, and clustering feature extraction dimensions, high-energy-consuming paths, waste heat utilization bottlenecks, and sudden change signals in the system energy flow network graph are classified and subdivided to generate a high-energy-consuming operating condition feature library. In the system energy flow network graph, high-energy-consuming paths are identified as "compressor high-load operation → excessive pipeline transmission loss → waste heat exchanger efficiency below 70%". These paths are further subdivided by the proportion of daily operating time (e.g., paths occurring for ≥6 hours per day are classified as Category A, and those occurring for 3-6 hours are classified as Category B). For waste heat utilization bottlenecks (e.g., insufficient heat exchange due to insufficient medium flow), the paths are subdivided by the size of the flow gap (e.g., gaps ≥20% are considered severe bottlenecks, and gaps of 10%-20% are considered general bottlenecks). For sudden change signals in operating conditions (e.g., a sudden increase in air conditioning energy consumption of more than 20% due to a sudden increase in production load), the paths are subdivided by the magnitude and duration of the change (e.g., magnitudes ≥30% and durations ≥10 minutes are considered emergency changes). The above subdivision results are integrated into a high-energy-consuming operating condition feature library containing 12 categories of features. Each category of features includes parameters such as energy flow intensity, duration, and associated node status.

[0028] A feature library of high-energy-consumption operating conditions is fused and coupled to generate a dynamic energy-efficiency coupling model that includes energy efficiency correlation equations, dynamic operating condition conversion rules, and system collaborative constraints. The model is a mechanism-data hybrid-driven multivariate coupling model, with three layers: a physical layer (corresponding to actual equipment energy transfer), a feature layer (corresponding to operating condition characteristic parameters), and a decision layer (corresponding to energy efficiency optimization objectives). The structure adopts a neural network architecture of "input layer-hidden layer-output layer." The input layer has 8 nodes (including air conditioning energy consumption, waste heat flow, etc.), the hidden layer has 2 layers (16 neurons per layer), and the output layer has 3 nodes (energy efficiency value, conversion probability, and constraint satisfaction). The relevant parameters include the energy efficiency correlation equation, which is set as E=αP+βQ-γL. (Where E is the overall system energy efficiency, P is the air conditioning power, Q is the waste heat recovery amount, L is the transmission loss, and α, β, and γ are coefficients obtained through data fitting, which are 0.6, 0.3, and 0.1, respectively); The dynamic conversion rule of the operating conditions adopts the Markov chain model, and the probability of conversion from the high-energy-consumption condition A to the condition B is set to 0.3 (when the transmission loss is reduced by 10%); The system collaborative constraints are set as "air conditioning cooling capacity ≥ production cooling load demand", "waste heat recovery amount ≥ secondary energy demand", and "total energy consumption ≤ industry benchmark value of 150kWh / ton of product".

[0029] S103, based on the dynamic energy efficiency coupling model, sets the system energy consumption as a condition-time dependent function, uses an improved particle swarm optimization algorithm to couple the load prediction correction term to construct the optimization objective equation, extracts the conversion coefficients of different waste heat qualities to construct the energy cascade utilization matrix, and performs collaborative calculation on the energy consumption distribution of the air conditioning-waste heat system to obtain the energy loss value of key nodes.

[0030] In one implementation, the operating condition-related parameters, time-series energy consumption data, and system collaborative constraints in the dynamic energy efficiency coupling model are extracted and adapted to generate energy consumption function variable sets, time weight coefficients, and operating condition transition thresholds. Operating condition-related parameters are extracted from the physical and feature layers of the dynamic energy efficiency coupling model. These parameters include, in addition to the air conditioning cooling capacity of 40kW and waste heat generation of 25kW corresponding to 80% production load, the air conditioning cooling capacity of 25kW and waste heat generation of 15kW at 50% production load, and the air conditioning power compensation value of 5kW at an ambient temperature of 35℃. These parameters reflect the matching relationship of system energy under different operating conditions. The time-series energy consumption data covers the entire cycle, including hourly energy consumption from 8:00 to 18:00 on weekdays (peak of 150kWh at 12:00, secondary peak of 130kWh at 10:00), and a valley of 30kWh at 1:00 AM, as well as weekend energy consumption data (e.g., an average energy consumption of 60kWh on Saturdays) to reflect the energy consumption patterns at different times.

[0031] In addition to the waste heat recovery being no less than 30% of the total air conditioning energy consumption, the system coordination constraints also include hard constraints such as the deviation between the air conditioning cooling capacity and the production cooling load being ≤5% and the pipeline transmission loss rate being ≤8%. The parameters are mapped to the 0-1 range through min-max normalization. Variables strongly correlated with energy consumption are selected using Pearson correlation coefficients, generating an energy consumption function variable set (four core variables: production load, ambient temperature, air conditioning power, and waste heat medium flow rate). Time weighting coefficients are calculated based on the proportion of energy consumption to total daily energy consumption in each time period. During the daytime production period (8:00-18:00), the energy consumption accounts for 85%, so the weight is set to 0.8. During the nighttime non-production period (18:00-8:00 the next day), the energy consumption accounts for 15%, so the weight is set to 0.2. Furthermore, the weight is increased to 0.85 for the peak production period (10:00-14:00). Based on the analysis of historical operating condition transition thresholds, in addition to production load ≥70% being high load and ≤30% being low load, a medium load condition (30%-70%) is added. An auxiliary threshold is set in conjunction with ambient temperature (e.g., when the ambient temperature is ≥32℃ under high load conditions, an enhanced heat dissipation mode is triggered).

[0032] The energy consumption function variable set, time weight coefficient, and operating condition transition threshold are processed to generate the optimization objective equation structure, particle swarm optimization (PSO) algorithm parameter range, and load prediction correction rules. Based on the energy consumption function variable set (production load x, ambient temperature y, air conditioning power z, waste heat medium flow rate w), the optimization objective equation structure is constructed in conjunction with the system collaborative objective. In addition to minf(x,y,z,w)=αz-βq+γl (α=0.6, β=0.3, γ=0.1), a constraint term h(x,y,w) (such as a penalty coefficient when the waste heat medium flow rate is insufficient) is introduced to make the equation more closely match the actual optimization requirements. The PSO algorithm parameter range is dynamically adjusted according to the operating condition characteristics. The population size is set to 100 under high load conditions and 50 under low load conditions. The performance weight is set to 0.9 in the early stage of iteration (1-30 times) (global search) and 0.5 in the later stage (71-100 times) (local convergence). The learning factors c1 and c2 are adjusted to 2.2 in the high potential optimization range (energy saving space ≥15%) to enhance the learning ability. The load prediction correction rule is set in stages. When the actual load deviates from the predicted load by 5%-10%, the correction coefficient k=1.1; when the deviation is ≥10%, k=1.2. In addition, a secondary correction is triggered in combination with the operating condition transition threshold (such as when switching from low load to medium load, an additional compensation coefficient of 0.05 is introduced).

[0033] Based on the optimized objective equation structure, particle swarm optimization algorithm parameter range, and load prediction correction rules, the temperature gradient, flow stability, and energy conversion efficiency of different waste heat qualities are classified and quantified to generate an energy cascade utilization matrix. In addition to the three temperature gradient levels of high (80-100℃), medium (50-80℃), and low (30-50℃), the high level is further subdivided into 80-90℃ and 90-100℃, and the medium level into 50-65℃ and 65-80℃, to more accurately reflect the waste heat quality. Flow stability is evaluated by the fluctuation coefficient within 30 minutes: ≤5% is stable (e.g., waste heat medium flow rate 8m³ / h ± 0.4m³ / h), 5%-10% is relatively stable (8m³ / h ± 0.8m³ / h), and >10% is unstable (8m³ / h ± 1.0m³ / h), and the fluctuation frequency is recorded (if the unstable state occurs ≥3 times per hour, it is downgraded). Energy conversion efficiency is subdivided by heat exchanger type: plate heat exchangers ≥85% are high-efficiency, 75%-85% are medium-efficiency; tubular heat exchangers ≥80% are high-efficiency, 65%-80% are medium-efficiency, ensuring that the quantitative standards match the equipment characteristics. The comprehensive benefit value of waste heat at each stage is calculated based on the optimization objective equation. The weights are iteratively optimized using the particle swarm optimization algorithm to generate a 3×3×3 energy cascade utilization matrix. The matrix element value is 0.95 for waste heat at 90-100℃, stable flow rate, and efficient conversion by plate heat exchangers; 0.85 for waste heat at 80-90℃, relatively stable flow rate, and efficient conversion by tubular heat exchangers; and 0.3 for waste heat at 30-50℃, unstable flow rate, and inefficient conversion. The matrix element value directly reflects the priority of waste heat utilization.

[0034] The energy cascade utilization matrix is ​​subjected to collaborative calculation and iterative optimization to generate key node energy loss values ​​that include instantaneous energy consumption values, cumulative energy loss, and the proportion of system energy efficiency. An improved particle swarm optimization algorithm is used to perform 100 iterations of calculation on the matrix. The first 50 iterations are based on the search for the global optimal solution, and the latter 50 iterations dynamically adjust the search direction in conjunction with load prediction correction rules (e.g., in the 60th iteration, due to a 12% load prediction deviation, a correction coefficient k=1.2 is introduced to adjust the particle velocity). The collaborative calculation covers 12 key nodes (compressor, condenser, waste heat exchanger, air supply duct, etc.), and the instantaneous energy consumption value of each node is accurate to the minute (e.g., the instantaneous energy consumption of the compressor at 14:00:00 is 22kW, and at 14:01:00 it is 21.8kW). The cumulative energy loss is statistically analyzed on a daily, weekly, and monthly basis (e.g., the cumulative loss of the pipeline transmission node is 8.5kWh on the day and 58kWh on the week). The system energy efficiency ratio is calculated in conjunction with the energy input-output ratio of the nodes (e.g., the waste heat exchanger has an input energy of 50kWh and an output utilization of 42kWh, with an energy efficiency ratio of 28%). The final output of key node energy loss values ​​is accompanied by loss cause labels (such as a sudden increase of 2kW in compressor instantaneous energy consumption labeled as "load fluctuation", and excessive cumulative pipeline loss labeled as "insulation layer aging"), forming a complete evaluation system that includes real-time monitoring, historical accumulation, and loss attribution, providing precise targets for subsequent energy efficiency optimization.

[0035] S104 combines the energy loss values ​​of key nodes, extracts peak energy consumption, waste heat utilization rate, and load matching degree according to production operating scenarios, classifies energy efficiency levels, and constructs a multi-dimensional feature matrix containing operating parameters, load characteristics, and environmental condition information.

[0036] In one implementation, the instantaneous peak loss, cumulative loss ratio, and system energy efficiency deviation are extracted and quantified from the energy loss values ​​of key nodes to generate energy consumption characteristic parameters, waste heat utilization efficiency values, and load matching deviation coefficients. Instantaneous peak loss is extracted at the minute level from 12 key nodes (compressor, condenser, waste heat exchanger, air duct, etc.). Besides the 22kW compressor at 14:00 and the 3.5kW duct transmission node at 10:00, it also includes the 5.2kW condenser at 16:00 and the 2.8kW waste heat exchanger at 12:00, ensuring the capture of peak losses at each node. The cumulative loss ratio is calculated based on the period (day, week, month). For example, if the compressor's cumulative loss is 220kWh on a given day, the total system loss is 4kWh. 89kWh, accounting for 45%, and the cumulative loss of waste heat exchanger is 88kWh, accounting for 18%. The weekly cumulative percentage is also recorded (compressor 42%, waste heat exchanger 19%) to reflect the trend. The system energy efficiency deviation is calculated by the ratio of the actual energy efficiency to the benchmark value (such as the industry level 2 energy efficiency standard of 3.2kW・h / kW・h). When the actual energy efficiency is 2.94kW・h / kW・h, the deviation is (2.94-3.2) / 3.2=-0.08. If the actual energy efficiency is 3.36, the deviation is +0.05. The standardization process uses the min-max method to map the instantaneous loss peak (range 0-25kW) to the 0-1 interval (0.88 for 22kW and 0.14 for 3.5kW). The waste heat utilization efficiency value is calculated as "waste heat recovery amount / waste heat generation amount" (e.g., generation amount 120kWh, recovery amount 90kWh, efficiency value 0.75). The load matching deviation coefficient is calculated as "(air conditioning load - waste heat supply) / air conditioning load" (load 200kW, supply 176kW, deviation 0.12).

[0037] Energy consumption characteristic parameters, waste heat utilization efficiency values, and load matching deviation coefficients are processed to generate production operating condition scenario classification labels, energy efficiency level classification thresholds, and characteristic parameter normalization rules. In addition to S1 (continuous high load, 8:00-18:00, load 80%-90%), S2 (low load intermittent, 0:00-6:00, load 20%-30%), and S3 (fluctuating load, 6:00-8:00 and 18:00-20:00, load fluctuation ±15%), a new S4 (extreme load, such as load exceeding 110% during high-temperature periods in summer) is added, with corresponding labels. Energy efficiency level classification thresholds are further subdivided according to waste heat utilization efficiency values: Level A (≥0.8, sufficient recovery), Level B (0.6-0.8, basically...). The energy consumption is categorized into three levels: Grade A (compliant), Grade B (0.4-0.6, underutilized), and Grade C (<0.4, serious waste). A reward and penalty system is also established (e.g., Grade A receives a 20-point energy efficiency bonus, Grade D loses 30 points). The normalization rules for characteristic parameters are clearly defined: the load matching deviation coefficient is mapped to the range of -1 to 1 (positive deviation indicates excess waste heat, e.g., 0.12 represents 12% excess; negative deviation indicates underutilization, e.g., -0.08 represents 8% underutilization). Energy consumption characteristic parameters are retained to three decimal places to ensure accuracy, and all rules are adapted to different seasons (in summer, when air conditioning load is high, the deviation coefficient threshold is relaxed to ±0.15).

[0038] Based on production scenario classification labels, energy efficiency level thresholds, and feature parameter normalization rules, the fluctuation range of operating parameters, dynamic load change trends, and environmental condition influence weights are classified and associated to generate a multi-dimensional feature association dataset. For scenario S1, the fluctuation range of air conditioning return air temperature (24-26℃, standard deviation 0.3℃), dynamic trend of production load (80%-90%, hourly fluctuation ≤2%), and influence weight of ambient temperature (0.3, for every 1℃ increase in temperature, air conditioning energy consumption increases by 2%) are extracted, and associated with energy efficiency level B (efficiency value 0.72) and load matching deviation coefficient 0.05 (excess 5%). For scenario S2, the return air temperature (22-23℃, standard deviation 0.2℃), load trend (20%-30%, intermittent shutdown 1-2 times / h), and influence weight of ambient humidity (0.2, when humidity > 60%, heat exchange efficiency decreases by 3%) are extracted, and associated with level A (efficiency value 0.85) and deviation coefficient -0.03 (less than 3%). Scenario S3 extracts return air temperature (23-25℃, standard deviation 0.5℃), load trend (fluctuation range ±15%), and environmental wind speed influence weight (0.15, when wind speed > 3m / s, heat dissipation efficiency increases by 5%), with association level B (0.78) and deviation coefficient 0.08; Scenario S4 extracts return air temperature (26-28℃), load trend (110%-120%), and comprehensive environmental influence weight (0.4), with association level C (0.55) and deviation coefficient -0.12. The dataset contains 100 samples (S1: 40 groups, S2: 20 groups, S3: 30 groups, S4: 10 groups), each group has 8 features (2 energy consumption feature parameters, 1 efficiency value, 1 deviation coefficient, 2 operating parameter ranges, 1 trend, and 1 weight).

[0039] The multidimensional feature association dataset was integrated and dimensionality-reduced to generate a multidimensional feature matrix containing dynamic ranges of operating parameters, load feature classification vectors, and environmental condition influence coefficients. The integrated dynamic ranges of operating parameters, excluding compressor power (15-20kW), include fan speed (1200-1450r / min), waste heat medium flow rate (6-10m³ / h), and heat exchanger inlet / outlet temperature difference (15-25℃). The load feature classification vectors use unique thermal encoding: high load [1,0,0,0], medium load [0,1,0,0], low load [0,0,1,0], and extreme load [0,0,0,1]. The environmental condition influence coefficients are refined into temperature (0.35), humidity (0.25), wind speed (0.1), atmospheric pressure (0.05), and light intensity (0.25, with a significant impact in summer). Dimensionality reduction was achieved using principal component analysis (PCA). A linear transformation was performed on the 8-dimensional features, and five principal components with eigenvalues ​​> 1 were selected (cumulative contribution rate 86.7%). The first principal component (contribution rate 32.1%) primarily reflects load intensity, while the second principal component (21.3%) reflects waste heat utilization efficiency. The final multidimensional feature matrix is ​​100×5, with each row corresponding to a scenario. For example, a sample in scenario S1 might have values ​​[0.82, 0.65, 0.12, 0.35, 1]. This matrix can be directly input into subsequent models for energy-saving potential analysis.

[0040] S105, based on a multi-dimensional feature matrix, compares real-time energy efficiency with historical data to identify abnormal signals such as sudden increases in energy consumption and insufficient utilization of waste heat, and generates an energy-saving potential coefficient using the slope of the energy efficiency-load curve.

[0041] In one implementation, the dynamic range of operating parameters, load characteristic classification vector, and environmental condition influence coefficients in the multidimensional feature matrix are extracted and quantified to generate real-time energy efficiency assessment values, historical energy efficiency benchmark values, and characteristic parameter deviations. The historical energy efficiency benchmark values ​​are weighted and calibrated based on the matching degree between the production conditions and environmental conditions during the same period. The dynamic range of operating parameters extracted from the multidimensional feature matrix includes, in addition to the air conditioning compressor power of 15-20kW and the waste heat medium flow rate of 6-10m³ / h, fan speed of 1200-1450r / min and heat exchanger inlet and outlet temperature difference of 15-25℃. The real-time energy efficiency assessment value is obtained through comprehensive calculation of multiple parameters. For example, combining a cooling capacity of 50kW and an input power of 18kW, the calculation is 2.78kW·h / kW·h, rounded to two decimal places. Historical data extraction covers the complete cycle of the same period in the past 3 months (e.g., Monday to Friday 8:00-18:00). The matching degree of production conditions is refined to the number of equipment in operation (e.g., currently and historically both have 3 compressors running, matching degree 0.9) and production shifts (e.g., both are day shifts, matching degree 1.0), with an average of 0.95. The matching degree of environmental conditions covers temperature (currently 28℃ and historically 27℃, matching degree 0.95) and humidity (currently 60% and historically 58%, matching degree 0.98), with an average of 0.96. The historical baseline value is then calibrated by weighting production conditions (weight 0.6) and environmental conditions (weight 0.4) (3.0 before calibration, 3.0 × (0.95 × 0.6 + 0.96 × 0.4) = 2.86 after calibration). The deviation of characteristic parameters is calculated for each key parameter. For example, the deviation between the real-time waste heat medium flow rate of 7.5 m³ / h and the interval average of 8 m³ / h is (7.5-8) / 8=-0.06, and the deviation between the real-time ambient temperature coefficient of 0.35 and the matrix average of 0.33 is 0.06.

[0042] The system processes real-time energy efficiency assessment values, historical energy efficiency benchmark values, and characteristic parameter deviations to generate energy efficiency comparison difference values, abnormal signal identification thresholds, and fluctuation trend judgment rules. The abnormal signal identification thresholds are dynamically set according to production load intensity levels. In addition to the direct difference between real-time and historical values ​​(2.8-2.86=-0.06), the energy efficiency comparison difference also calculates a relative difference (-0.06 / 2.86≈-0.021) to reflect the deviation ratio. The abnormal signal identification thresholds are further subdivided into more refined load ranges: ultra-high load (100%-120%) ±0.35, high load (80%-100%) ±0.3, medium-high load (65%-80%) ±0.25, medium load (50%-65%) ±0.2, low load (30%-50%) ±0.18, and ultra-low load (<30%) ±0.15. Each range is also adjusted for ambient temperature (e.g., the threshold increases by 0.05 during high summer temperatures). The fluctuation trend judgment rules add a combination of duration and amplitude conditions. For example, if energy efficiency decreases for five consecutive time points (one point every 10 minutes) with a cumulative decrease exceeding 0.6, or if the decrease at a single time point exceeds 0.3 and is accompanied by a characteristic parameter deviation exceeding ±0.1, it is considered a significant downward trend.

[0043] Based on energy efficiency comparison differences, abnormal signal identification thresholds, and fluctuation trend judgment rules, the magnitude of energy consumption surges, waste heat utilization gaps, and energy efficiency-load curve shapes are classified and sloped to generate an energy-saving potential assessment dataset. The slope calculation incorporates a decay coefficient based on the equipment's operating years for correction. Energy consumption surges are categorized by duration: instantaneous surges (1-5 minutes, e.g., from 150kWh to 200kWh), short-term surges (5-30 minutes), and sustained surges (>30 minutes), with the surge triggering factors (e.g., sudden increase in production load, equipment failure). Waste heat utilization gaps are subdivided by gap percentage: mild gaps (<10%, e.g., demand 140kW, actual 128kW), moderate gaps (10%-20%, e.g., actual 120kW), and severe gaps (>20%, e.g., actual 100kW), with the gap duration recorded simultaneously (e.g., severe gaps lasting 40 minutes). The energy efficiency-load curve shape is classified according to the slope range: strong negative correlation (slope < -0.03), weak negative correlation (-0.03 to -0.01), no significant correlation (-0.01 to 0.01), and positive correlation (> 0.01). When calculating the slope, the attenuation coefficient is 1.0 for equipment operating for less than 1 year, 0.95 for 1-3 years, 0.9 for 3-5 years, and 0.85 for more than 5 years (e.g., for equipment operating for 4 years, the original slope is -0.02, corrected to -0.018). The dataset contains the above classification results and the original calculated values, totaling 12 feature dimensions.

[0044] The energy-saving potential assessment dataset is integrated and quantified to generate an energy-saving potential coefficient that includes instantaneous energy-saving space, cumulative energy-saving potential, and dynamic adjustment coefficient. The dynamic adjustment coefficient is dynamically corrected based on the real-time load rate of the waste heat recovery equipment. Instantaneous energy-saving space is calculated according to the type of surge: for instantaneous surges, 50% of the difference between the peak value and the baseline value (200-150=50kWh) is taken as the space (25kWh); for short-term surges, 60% of the difference between the average and the mean is taken; and for sustained surges, 80% is taken. Cumulative energy-saving potential is statistically calculated on a daily, weekly, and monthly basis. For example, if the cumulative energy deficit is 80kWh for mild, 120kWh for moderate, and 200kWh for severe on a given day, the weighted average (weights 0.3, 0.5, and 0.8) is calculated as (80×0.3+120×0.5+200×0.8)=234kWh, which is converted into a potential value of 234 / total daily energy consumption of 1000=0.234. The dynamic adjustment coefficient is first set to a base value based on the load rate (0.8 when the load rate is <50%, 1.0 when the load rate is 50%-80%, and 1.2 when the load rate is >80%), and then correlated with the equipment health (e.g., a health score of 90 corresponds to a correction factor of 0.95). The final coefficient = base value × health factor (e.g., 1.0 × 0.95 = 0.95 when the load rate is 80%). The energy-saving potential coefficient is presented in the form of a three-dimensional array (e.g., [25kWh, 0.234, 0.95]), with a label indicating the calculation basis (e.g., "based on a 33% instantaneous surge and a severe shortage of 40kW").

[0045] S106 combines the energy-saving potential coefficient with the production plan, groups the operating modes, uses a deep reinforcement learning algorithm to screen key control factors, and integrates energy efficiency parameters, equipment status and production demand information to build an adaptive optimization control model.

[0046] In one implementation, key production parameters affecting the system's operating mode are extracted based on the production plan. These operating modes are then categorized into high-potential, medium-potential, and low-potential groups according to their energy efficiency optimization levels, with each group corresponding to a unique optimization priority label. When extracting key production parameters from the production plan, the core influencing factors across the entire production process are covered, ensuring a strong correlation between the parameters and the system's operating mode. Key production parameters extracted from the production plan include, in addition to daily production shifts (e.g., three shifts: morning shift 8:00-16:00, afternoon shift 16:00-24:00, night shift 0:00-8:00), product capacity (e.g., 800 tons / day, including capacity allocation for each time period: morning shift 300 tons, afternoon shift 350 tons, night shift 150 tons), and number of operating equipment (e.g., 3 air conditioning units, including 2 main units and 1 standby unit), production process type (e.g., mechanical processing requires stable cooling load, chemical processing requires fluctuating cooling load), and raw material input (e.g., 500 tons of raw materials per day correspond to a 15% increase in cooling load demand).

[0047] When classifying energy efficiency optimization levels, the group determination criteria are refined by combining seasonal characteristics with the dynamic correlation of production load. The high-potential group (energy-saving potential coefficient ≥ 0.3) includes not only the summer high-temperature period (ambient temperature ≥ 30℃) operation mode, but also the peak production period operation mode (e.g., morning shift 10:00-14:00, afternoon shift 18:00-22:00). In this scenario, the air conditioning cooling load demand exceeds 180kW, and the waste heat generation surges simultaneously, resulting in significant energy-saving potential. It corresponds to the priority label "P1," and is given priority in allocating computing resources and control permissions during optimization. The medium-potential group (0.15-0.3) includes not only the spring and autumn regular mode (ambient temperature 15-25℃), but also the off-peak production period operation mode (e.g., morning shift 8:00-10:00, afternoon shift 16:00-18:00). The cooling load demand is 120-180kW, and the waste heat utilization rate is maintained at 70%-80%. It corresponds to the priority label "P2," and the optimization strategy focuses on balancing stability and energy saving. The low potential group (<0.15) includes not only the low-load mode in winter (ambient temperature <15℃), but also the operation mode during off-peak production periods (such as night shift 0:00-6:00) and equipment maintenance periods. The cooling load demand is <120kW, the waste heat utilization rate is over 80%, and the energy-saving space is limited. It corresponds to the priority label "P3". Optimization is mainly to ensure basic operation needs.

[0048] A comparative analysis was conducted on the energy-saving potential coefficients, energy efficiency parameters, and production demand information of different potential groups to generate characteristic difference data between groups. When comparing and analyzing the core parameters of different potential groups, the analysis was carried out from three dimensions: data distribution, influencing factors, and optimization direction, to ensure the accurate quantification of characteristic differences. The core parameters for the high-potential group (P1) include an average energy-saving potential coefficient of 0.35, an average energy loss value of 120kWh at key nodes (compressor node loss accounts for 45% and pipeline transmission node loss accounts for 25%), a waste heat utilization rate of 65%, a production cooling load of 200kW, as well as equipment operating time (average 16 hours per day) and an average ambient temperature of 32℃. The corresponding parameters for the medium-potential group (P2) are an average energy-saving potential coefficient of 0.22, an average energy loss value of 80kWh at key nodes (compressor account for 40% and pipeline account for 20%), a waste heat utilization rate of 75%, a production cooling load of 150kW, an average daily equipment operating time of 12 hours, and an average ambient temperature of 22℃. The corresponding parameters for the low-potential group (P3) are an average energy-saving potential coefficient of 0.1, an average energy loss value of 50kWh at key nodes (compressor account for 35% and pipeline account for 15%), a waste heat utilization rate of 85%, a production cooling load of 80kW, an average daily equipment operating time of 8 hours, and an average ambient temperature of 10℃.

[0049] Further detailed correlation analysis was conducted on the data showing differences between groups to clarify the causes of these differences and identify areas for optimization. Compared to the medium-potential group, the high-potential group had a 59% higher energy-saving potential (0.35-0.22) / 0.22≈59%), mainly due to the high ambient temperature causing the air conditioning compressor to operate at high load, resulting in a surge in energy consumption; the waste heat utilization rate was 13% lower (75%-65%), stemming from the fact that the amount of waste heat generated during high-temperature periods exceeded the processing capacity of the recovery equipment, leading to some waste heat; and the production cooling load was 33% higher (200-150) / 150≈33%, influenced by the combined effects of peak product production capacity and process heat dissipation requirements. The medium-potential group and the low-potential group have a higher energy-saving potential of 120% (0.22-0.1) / 0.1=120%, because the production load and ambient temperature are moderate, and there is still some room for energy consumption optimization; the waste heat utilization rate is low of 11.8% (85%-75%) / 75%≈11.8%, which is affected by frequent equipment start-ups and shutdowns; the production cooling load is high of 87.5% (150-80) / 80=87.5%, which is related to capacity increase and continuous process operation.

[0050] The system integrates and processes characteristic difference data, energy-saving potential coefficients, energy efficiency parameters, equipment status, and production demand information to generate a multi-dimensional model training dataset. Energy efficiency parameters include energy loss values ​​at key nodes and waste heat utilization rates. Data integration must ensure sample coverage and dimensional integrity to provide comprehensive support for model training. The number of samples is allocated according to the optimization needs and data availability of the potential groups. The high-potential group, due to its complex scenarios and high optimization requirements, collects 400 samples (including data from various temperature ranges in summer and peak production periods); the medium-potential group collects 300 samples (including data from different weather conditions in spring and autumn and off-peak production periods); and the low-potential group collects 300 samples (including data from different low-temperature ranges in winter, production troughs, and maintenance periods).

[0051] The sample dimensions must cover all factors affecting system control, including energy-saving potential coefficient, energy loss values ​​of 12 key nodes (compressor, condenser, waste heat exchanger, etc.), waste heat utilization rate, equipment health (scored on a 10-point scale, such as 9 points for good and 6 points for needing maintenance), production cooling load, environmental parameters (temperature, humidity, wind speed), equipment operating parameters (compressor frequency, fan speed, waste heat medium flow rate), and production process parameters (raw material input rate, product output rate), for a total of 20 dimensions. In the data preprocessing stage, min-max normalization was used to map continuous parameters such as energy loss (0-200kWh) and production cooling load (50-250kW) to the 0-1 range; discrete parameters such as equipment health (1-10 points) and production shifts (1-3 representing morning, noon and evening shifts) were encoded using one-heat encoding; parameters that are greatly affected by external factors, such as ambient temperature (-5-40℃), were processed by Z-score normalization. Finally, a multi-dimensional model training dataset with 1000 valid samples was formed, which was divided into a training set (70%), a validation set (20%), and a test set (10%) for subsequent model construction.

[0052] Based on a multi-dimensional model training dataset, a deep reinforcement learning algorithm is used to screen key regulatory factors that significantly affect control effectiveness, and a preliminary adaptive optimization control model including a policy network and a value network is constructed. The model selection adopts the Actor-Critic architecture from deep reinforcement learning, adapting to the system's dynamic optimization requirements, and its hierarchical design aligns with data processing and decision-making logic. The state input layer receives data from 20 feature dimensions of the multi-dimensional dataset, transforming the standardized parameters into a tensor format recognizable by the model. The feature extraction layer uses a fully connected neural network to mine deep correlations between features, capturing the implicit mapping relationship between energy-saving potential coefficients and regulatory factors. The decision output layer is divided into policy output (Actor) and value evaluation (Critic), respectively realizing the generation of control commands and the evaluation of control effectiveness.

[0053] In terms of model structure, both the policy network (Actor) and the value network (Critic) employ a 3-layer fully connected neural network. The input layer has 20 neurons, corresponding to 20 feature dimensions; the first hidden layer has 40 neurons, using the ReLU activation function to enhance nonlinear fitting ability, and the second hidden layer has 20 neurons, also using the ReLU activation function; the output layer, the policy network, has 8 neurons, corresponding to 8 key control factors (compressor frequency, fan speed, waste heat medium flow rate, heat exchanger start / stop status, air conditioning supply air temperature, condenser water temperature, pipe valve opening, and standby equipment start / stop), and uses the Softmax activation function to output the action probability of each control factor; the value network has 1 neuron, using a linear activation function to output the value assessment value of the current state.

[0054] Model parameters were set based on system characteristics and training requirements. The initial learning rate was set to 0.001 to ensure rapid parameter updates in the early stages of training; the discount factor was set to 0.95 to emphasize long-term optimization benefits; the experience replay pool capacity was set to 10,000 to store historical training samples and break data correlations; the batch sampling size was set to 32 to balance training efficiency and stability; and the target network update frequency was set to once every 100 steps to avoid training oscillations. Key control factors were screened through feature importance analysis and gradient descent. The compressor frequency (contributing 35%, directly affecting air conditioning energy consumption), waste heat medium flow rate (contributing 25%, determining waste heat recovery efficiency), heat exchanger start / stop status (contributing 20%, controlling the waste heat utilization path), and air conditioning supply air temperature (contributing 10%, affecting cooling load matching) were ultimately identified as core control factors, with the remaining factors serving as auxiliary control items.

[0055] Dynamic scenario simulations and parameter tuning were performed on the preliminary adaptive optimization control model to generate the final adaptive optimization control model. The dynamic scenario simulations needed to cover various operating condition changes the system might face, comprehensively testing the adaptability and stability of the preliminary model. Thirty dynamic scenarios were built using MATLAB / Simulink, including scenarios such as a 20% increase in production load (e.g., raw material input suddenly increasing from 500 tons / day to 600 tons / day), a 15% decrease in waste heat medium flow rate (e.g., pipeline blockage causing flow rate to drop from 10 m³ / h to 8.5 m³ / h), a 5°C increase in ambient temperature (e.g., extreme high temperatures in the summer afternoon), sudden equipment failure (e.g., one air conditioning compressor shutting down), production process switching (e.g., switching from machining to chemical production), and grid voltage fluctuations (±10%). Each scenario simulation lasted 2 hours, and model output data was collected at 120 time points for comparison with the actual optimal data.

[0056] During the parameter tuning phase, precise adjustments were made based on scenario testing results to improve model performance and robustness. To address the initial model's issues of control lag and overfitting under high-load surges, the learning rate was adjusted from 0.001 to 0.0005 to slow down parameter updates and avoid oscillations; the number of hidden layer neurons was increased from 40 / 20 to 50 / 25 to enhance the model's ability to fit complex scenarios; the L2 regularization coefficient was increased by 0.001 to suppress overfitting by penalizing excessively large weights; the experience replay pool capacity was expanded to 15,000 to incorporate more abnormal scenario samples and improve model generalization; and the target network update frequency was adjusted to once every 80 steps to accelerate the response speed to dynamic scenarios.

[0057] The final model performance needs to be verified from three dimensions: control effect, convergence speed, and stability. In the high-potential group scenario, the control effect is improved by 18% compared to the initial model. For example, during the high-temperature period in summer, the air conditioning energy consumption is reduced from 150kWh to 123kWh, and the waste heat utilization rate is increased from 65% to 76%. In the medium-potential group, the control effect is improved by 12%. In the spring and autumn normal mode, the energy consumption is reduced from 100kWh to 88kWh, and the waste heat utilization rate is increased from 75% to 84%. In the low-potential group, the control effect is improved by 8%. In the winter low-load mode, the energy consumption is reduced from 60kWh to 55.2kWh, and the waste heat utilization rate is maintained above 85%. The convergence speed is 25% faster than the initial model, and the number of training iterations is reduced from 8000 to 6000 to reach a stable state. In 20 continuous dynamic scenario tests, the model output deviation is ≤5%, which meets the requirements for stable operation of industrial systems. Finally, an adaptive optimization control model that can be directly applied to the coordinated control of industrial air conditioning and waste heat recovery is formed.

[0058] S107, based on the control strategy output of the adaptive optimization control model, combined with the dynamic correlation information of air conditioning load and waste heat supply, generates real-time control instructions and control information for energy efficiency improvement schemes.

[0059] In one implementation, control strategies for different operating modes are extracted from the final adaptive optimization control model. The strategies need to cover the target thresholds and dynamic adjustment logic of the core control factors. Examples are as follows: For the high-potential group (P1, peak summer production), the model output control strategy includes a compressor frequency target of 50Hz (±2Hz dynamically adjusted), a waste heat medium flow rate of 10m³ / h (±0.5m³ / h), heat exchangers running all day, and air conditioning supply temperature of 18℃ (±0.3℃); For the medium-potential group (P2, spring and autumn off-peak), the strategy is a compressor frequency of 40Hz (±3Hz), a waste heat medium flow rate of 8m³ / h (±0.8m³ / h), heat exchangers running intermittently according to production shifts (early shift 8:00-16:00), and supply air temperature of 19℃ (±0.5℃); For the low-potential group (P3, winter off-peak), the strategy is a compressor frequency of 30Hz (±2Hz), a waste heat medium flow rate of 6m³ / h (±0.3m³ / h), heat exchangers running only during the day shift (10:00-14:00), and supply air temperature of 20℃ (±0.5℃). Meanwhile, the strategy also includes linkage rules for auxiliary control factors, such as the condenser water temperature needing to be reduced by 0.5°C for every 1°C increase in ambient temperature, and the valve opening of the pipeline being adjusted synchronously with the fluctuation of the waste heat medium flow rate (if the flow rate decreases by 10%, the opening rate increases by 5%).

[0060] Real-time data collection of air conditioning load and waste heat supply is used to clarify their matching relationship and dynamic trends through correlation analysis, providing a basis for adjusting control commands. For example: Using IoT sensors, air conditioning cooling load (e.g., currently 190kW, up 10kW from 10 minutes ago) and waste heat generation (e.g., currently 140kW, up 8kW from 10 minutes ago) are collected every 5 minutes. The matching degree is calculated (140 / 190≈73.7%, lower than the model strategy's target matching degree of 80%). Analyzing the dynamic correlation trend reveals that the air conditioning load has increased at a rate of 5kW / 10 minutes in the past 30 minutes, while waste heat supply has increased at a rate of 4kW / 10 minutes. The load growth is faster than the waste heat supply, and the matching degree is expected to drop below 70% after 15 minutes. Simultaneously, ambient temperature (currently 32℃, up 1℃ from 1 minute ago) and production load (currently 85%, within the high-potential load range) are collected. It is determined that the load growth is due to increased heat dissipation from the production process, while the lagging growth in waste heat supply is due to a delayed response from the waste heat recovery equipment.

[0061] By combining dynamic correlation information with the initial model strategy, the core control factors are dynamically corrected to generate directly executable real-time control commands, ensuring coordinated matching between air conditioning load and waste heat supply. For example, in the current high-potential scenario with an air conditioning load of 190kW and a waste heat supply of 140kW (matching degree 73.7%), the initial strategy is corrected as follows: the compressor frequency is increased from 50Hz to 52Hz (increasing cooling capacity to cope with load growth), and the waste heat medium flow rate is increased from 10m³ / h to 11m³ / h (accelerating waste heat recovery and improving supply capacity); the heat exchanger remains on all day, and its heat exchange efficiency target value is increased from 85% to 88% (achieved by increasing the heat exchange area ratio); the air conditioning supply air temperature remains unchanged at 18℃, but the linked condenser water temperature is decreased from 30℃ to 29.5℃ (offsetting the impact of rising ambient temperature on cooling efficiency); the pipe valve opening is increased from 80% to 85% (to reduce transmission losses in conjunction with the increased waste heat medium flow rate). The final generated real-time control command is output in a standardized format, including the command identifier (e.g., “P1-202405201430”), control factor name, target value, execution time (e.g., “execute before 14:35 on 2024-05-20”), and actuator (e.g., “compressor 1#, waste heat pump 2#, heat exchanger 1#”).

[0062] Based on the expected effects of real-time control commands, and combined with historical system operating data and energy efficiency optimization targets, an energy efficiency improvement plan control information is formulated, including short-term adjustments, medium-term optimizations, and long-term improvements. Examples are as follows: Short-term (1-24 hours) control information: Execute according to real-time control commands, monitoring air conditioning energy consumption (target from 150kWh to below 140kWh) and waste heat utilization rate (target from 65% to above 70%) hourly. If energy consumption exceeds the target by 5% or utilization rate is 3% below the target, a secondary correction is triggered (e.g., further increasing the waste heat medium flow rate by 0.5m³ / h); Medium-term (1-7 days) control information: For high-potential scenarios, adjust the timing of production shifts and waste heat recovery equipment operation (e.g., adjusting the peak production time of the middle shift). The optimization plan, implemented one hour earlier from 18:00 to 22:00 to coincide with peak waste heat generation, aims to increase daily waste heat utilization by 5% and reduce air conditioning energy consumption by 8%. Long-term (1-3 months) control information: Based on model operation data, it is recommended to maintain inefficient waste heat exchangers (currently 82% efficiency, lower than the industry average of 85%) (e.g., cleaning scale buildup on heat exchange tubes) and replace aging air conditioning fan motors (currently consuming 10% more energy than new motors). The plan anticipates a 12% improvement in overall system energy efficiency and annual energy savings of approximately 12,000 kWh after maintenance. The plan also includes threshold settings for energy efficiency monitoring indicators (e.g., triggering an early warning when air conditioning energy consumption exceeds 150 kWh / hour, and triggering an alarm when waste heat utilization is below 60%) and responsible departments (e.g., short-term adjustments are handled by the workshop maintenance team, while long-term improvements are the responsibility of the equipment management department).

[0063] In one implementation, such as Figure 2 As shown, this application also provides an intelligent optimization control device for the coordinated operation of industrial air conditioning and waste heat recovery, comprising:

[0064] The acquisition module 201 is used to acquire industrial air conditioning operating parameters and waste heat recovery system data. Based on the multi-source data fusion technology of IoT sensors and edge computing gateways, it uses wavelet denoising algorithm to eliminate equipment vibration interference noise and generate a standardized operating parameter sequence.

[0065] Processing module 202 is used to convert the standardized operating parameter sequence into a system energy flow network map, which is then processed by mechanism-data hybrid modeling software. A fuzzy clustering algorithm is used to subdivide high-energy-consumption operating conditions, constructing a dynamic energy efficiency coupling model. Based on this model, system energy consumption is defined as a condition-time dependent function. An improved particle swarm optimization algorithm is used to couple load prediction correction terms to construct an optimization objective equation. Conversion coefficients for different waste heat qualities are extracted to construct an energy cascade utilization matrix. The energy consumption distribution of the air conditioning-waste heat system is calculated collaboratively to obtain energy loss values ​​at key nodes. Combining these key node energy loss values, peak energy consumption, waste heat utilization rate, and load matching degree are extracted according to production operating scenarios, and the system is divided into... Energy efficiency rating is determined by constructing a multi-dimensional feature matrix that includes operating parameters, load characteristics, and environmental conditions. Based on this matrix, real-time energy efficiency is compared with historical data to identify abnormal signals such as sudden increases in energy consumption and insufficient waste heat utilization. An energy-saving potential coefficient is generated using the slope of the energy efficiency-load curve. This coefficient is then integrated with the production plan to group operating modes. A deep reinforcement learning algorithm is used to screen key control factors, and an adaptive optimization control model is constructed by fusing energy efficiency parameters, equipment status, and production demand information. Based on the control strategy output of the adaptive optimization control model, and combined with the dynamic correlation information between air conditioning load and waste heat supply, real-time control instructions and control information for energy efficiency improvement schemes are generated.

[0066] The computer-readable storage medium provided in the above embodiments of this application and the intelligent optimization control method for coordinated industrial air conditioning and waste heat recovery provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

Claims

1. A smart optimization control method for the coordinated operation of industrial air conditioning and waste heat recovery, characterized in that, include: The system acquires industrial air conditioning operating parameters and waste heat recovery system data. Based on the multi-source data fusion technology of IoT sensors and edge computing gateways, it uses wavelet denoising algorithm to eliminate equipment vibration interference noise and generate a standardized operating parameter sequence. After converting the standardized operating parameter sequence into a system energy flow network map, the model is processed by mechanism-data hybrid modeling software, and a fuzzy clustering algorithm is used to subdivide the high-energy-consumption operating conditions to construct a dynamic energy efficiency coupling model. Based on the dynamic energy efficiency coupling model, the system energy consumption is set as a condition-time dependent function. An improved particle swarm algorithm is used to couple the load prediction correction term to construct the optimization objective equation. The conversion coefficients of different waste heat qualities are extracted to construct the energy cascade utilization matrix. The energy consumption distribution of the air conditioning-waste heat system is calculated collaboratively to obtain the energy loss value of key nodes. By combining the energy loss values ​​of key nodes, the peak energy consumption, waste heat utilization rate and load matching degree are extracted according to the production working conditions, energy efficiency levels are classified, and a multi-dimensional feature matrix containing operating parameters, load characteristics and environmental conditions is constructed. Based on a multidimensional feature matrix, real-time energy efficiency is compared with historical data to identify abnormal signals such as sudden increases in energy consumption and insufficient utilization of waste heat. Energy-saving potential coefficients are generated using the slope of the energy efficiency-load curve. By combining the energy-saving potential coefficient with the production plan, grouping the operating modes, using deep reinforcement learning algorithms to screen key control factors, and integrating energy efficiency parameters, equipment status and production demand information, an adaptive optimization control model is constructed. Based on the output of the control strategy of the adaptive optimization control model, and combined with the dynamic correlation information of air conditioning load and waste heat supply, control information for real-time control instructions and energy efficiency improvement schemes is generated.

2. The method as described in claim 1, characterized in that, After converting the standardized operating parameter sequence into a system energy flow network map, it is processed by mechanism-data hybrid modeling software. Fuzzy clustering algorithm is then used to further subdivide high-energy-consumption operating conditions, constructing a dynamic energy efficiency coupling model, including: The air conditioning energy consumption data, waste heat medium flow data and heat exchange efficiency parameters in the standardized operating parameter sequence are extracted and transformed to generate energy flow node association groups, network transmission loss coefficients and parameter spatiotemporal matching degrees. The energy flow node association group, network transmission loss coefficient, and parameter spatiotemporal matching degree are processed to generate mechanism modeling constraints, data fitting error thresholds, and cluster feature extraction dimensions. Based on the constraints of mechanism modeling, the data fitting error threshold, and the clustering feature extraction dimension, the high-energy-consuming paths, waste heat utilization bottlenecks, and sudden operating condition signals in the system energy flow network graph are classified and subdivided to generate a high-energy-consuming operating condition feature library. The feature library of high-energy-consumption operating conditions is fused and coupled to generate a dynamic energy efficiency coupling model that includes energy efficiency correlation equations, dynamic operating condition transformation rules, and system collaborative constraints.

3. The method as described in claim 1, characterized in that, Based on a dynamic energy efficiency coupling model, the system energy consumption is defined as a condition-time dependent function. An improved particle swarm optimization algorithm is used to couple a load prediction correction term to construct an optimization objective equation. Conversion coefficients for different waste heat qualities are extracted to construct an energy cascade utilization matrix. The energy consumption distribution of the air conditioning-waste heat system is calculated collaboratively to obtain the energy loss values ​​at key nodes, including: The operating condition-related parameters, time series energy consumption data, and system collaborative constraints in the dynamic energy efficiency coupling model are extracted and adapted to generate energy consumption function variable groups, time weight coefficients, and operating condition transition thresholds. The energy consumption function variable set, time weight coefficient, and operating condition transition threshold are processed to generate the optimization objective equation structure, particle swarm algorithm parameter range, and load prediction correction rules. Based on the optimization objective equation structure, particle swarm algorithm parameter range, and load prediction correction rules, the temperature gradient, flow stability, and energy conversion efficiency of different waste heat qualities are classified and quantified to generate an energy cascade utilization matrix. The energy cascade utilization matrix is ​​subjected to collaborative calculation and iterative optimization to generate key node energy loss values ​​that include instantaneous energy consumption values, cumulative energy loss, and the proportion of system energy efficiency.

4. The method as described in claim 3, characterized in that, By combining energy loss values ​​at key nodes, peak energy consumption, waste heat utilization rate, and load matching degree are extracted according to production operating scenarios to classify energy efficiency levels. A multi-dimensional feature matrix containing operating parameters, load characteristics, and environmental condition information is constructed, including: The instantaneous peak loss, cumulative loss ratio, and system energy efficiency deviation in the energy loss values ​​of key nodes are extracted and quantified to generate energy consumption characteristic parameters, waste heat utilization efficiency values, and load matching deviation coefficients. The energy consumption characteristic parameters, waste heat utilization efficiency value, and load matching deviation coefficient are processed to generate production working condition scenario classification labels, energy efficiency level classification thresholds, and characteristic parameter normalization rules. Based on production working condition scenario classification labels, energy efficiency level classification thresholds, and feature parameter normalization rules, the fluctuation range of operating parameters, dynamic load change trends, and environmental condition influence weights are classified and associated to generate a multi-dimensional feature association dataset. The multidimensional feature association dataset is integrated and dimensionality reduction optimized to generate a multidimensional feature matrix containing the dynamic range of operating parameters, load feature classification vector, and environmental condition influence coefficient.

5. The method as described in claim 1, characterized in that, Based on a multidimensional feature matrix, real-time energy efficiency is compared with historical data to identify abnormal signals such as sudden increases in energy consumption and insufficient waste heat utilization. Energy-saving potential coefficients are generated using the slope of the energy efficiency-load curve, including: The dynamic range of operating parameters, load feature classification vector, and environmental condition influence coefficient in the multidimensional feature matrix are extracted and quantified to generate real-time energy efficiency assessment value, historical energy efficiency benchmark value, and feature parameter deviation. The historical energy efficiency benchmark value is weighted and calibrated by combining the matching degree between the production conditions and environmental conditions during the same period. The real-time energy efficiency assessment value, historical energy efficiency benchmark value, and characteristic parameter deviation are processed to generate energy efficiency comparison difference, abnormal signal identification threshold, and fluctuation trend judgment rule. The abnormal signal identification threshold is set with a dynamic threshold range according to the production load intensity level. Based on the energy efficiency comparison difference, abnormal signal identification threshold, and fluctuation trend judgment rules, the sudden increase in energy consumption, waste heat utilization gap, and energy efficiency-load curve shape are classified and slope calculated to generate an energy-saving potential assessment dataset. The slope calculation incorporates the attenuation coefficient of the equipment's operating years for correction. The energy-saving potential assessment dataset is integrated and quantified to generate an energy-saving potential coefficient that includes instantaneous energy-saving space, cumulative energy-saving potential, and dynamic adjustment coefficient. The dynamic adjustment coefficient is dynamically corrected by associating it with the real-time load rate of the waste heat recovery equipment.

6. The method as described in claim 1, characterized in that, By combining energy-saving potential coefficients with production plans, grouping operating modes, employing deep reinforcement learning algorithms to screen key control factors, and integrating energy efficiency parameters, equipment status, and production demand information, an adaptive optimization control model is constructed, including: Based on the production plan, key production parameters affecting the system's operating mode are extracted, and the operating modes are divided into high-potential, medium-potential, and low-potential groups according to their energy efficiency optimization levels. Each group corresponds to a unique optimization priority label. Comparative analysis of the energy-saving potential coefficients, energy efficiency parameters, and production demand information of different potential groups was conducted to generate characteristic difference data between the groups. The feature difference data, energy-saving potential coefficient, energy efficiency parameters, equipment status and production demand information are integrated and processed to generate a multi-dimensional model training dataset. Among them, the energy efficiency parameters include the energy loss value of key nodes and the waste heat utilization rate. Based on the multi-dimensional model training dataset, a deep reinforcement learning algorithm is used to screen out the key regulatory factors that have a significant impact on the control effect, and a preliminary adaptive optimization control model including a policy network and a value network is constructed. Dynamic scenario simulation and parameter tuning are performed on the preliminary adaptive optimization control model to generate the final adaptive optimization control model.

7. An intelligent optimization control device for the coordinated operation of industrial air conditioning and waste heat recovery, characterized in that, The device includes: The data acquisition module is used to acquire industrial air conditioning operating parameters and waste heat recovery system data. Based on the multi-source data fusion technology of IoT sensors and edge computing gateways, it uses wavelet denoising algorithm to eliminate equipment vibration interference noise and generate standardized operating parameter sequences. The processing module converts standardized operating parameter sequences into a system energy flow network map. This map is then processed by mechanism-data hybrid modeling software. A fuzzy clustering algorithm is used to further subdivide high-energy-consumption operating conditions, constructing a dynamic energy efficiency coupling model. Based on this model, system energy consumption is defined as a condition-time dependent function. An improved particle swarm optimization algorithm is used to couple load prediction correction terms to construct an optimization objective equation. Conversion coefficients for different waste heat qualities are extracted to construct an energy cascade utilization matrix. The energy consumption distribution of the air conditioning-waste heat system is calculated collaboratively to obtain energy loss values ​​at key nodes. Combining these key node energy loss values, peak energy consumption, waste heat utilization rate, and load matching degree are extracted according to production operating scenarios to classify energy efficiency. The system is structured by: constructing a multi-dimensional feature matrix containing information on operating parameters, load characteristics, and environmental conditions; comparing real-time energy efficiency with historical data to identify abnormal signals such as sudden increases in energy consumption and insufficient waste heat utilization; and generating an energy-saving potential coefficient using the slope of the energy efficiency-load curve. This energy-saving potential coefficient is then integrated with the production plan to group operating modes, and a deep reinforcement learning algorithm is used to screen key control factors. This process integrates energy efficiency parameters, equipment status, and production demand information to construct an adaptive optimization control model. Based on the control strategy output of the adaptive optimization control model, and combined with the dynamic correlation information between air conditioning load and waste heat supply, real-time control instructions and control information for energy efficiency improvement schemes are generated.

8. An electronic device, characterized in that, include: First processor; and memory for storing executable instructions of the first processor; The first processor is configured to execute the intelligent optimization control method for industrial air conditioning and waste heat recovery coordination as described in any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the second processor, it implements the intelligent optimization control method for coordinated industrial air conditioning and waste heat recovery as described in any one of claims 1 to 6.

Images